Targeting intensive glycaemic control versus targeting conventional glycaemic control for type 2 diabetes mellitus

  • Review
  • Intervention

Authors

  • Bianca Hemmingsen,

    Corresponding author
    1. Department 7812, Rigshospitalet, Copenhagen University Hospital, Copenhagen Trial Unit, Centre for Clinical Intervention Research, Copenhagen, Denmark
    • Bianca Hemmingsen, Copenhagen Trial Unit, Centre for Clinical Intervention Research, Department 7812, Rigshospitalet, Copenhagen University Hospital, Blegdamsvej 9, Copenhagen, DK-2100, Denmark. biancahemmingsen@hotmail.com. bh@ctu.rh.dk.

    Search for more papers by this author
  • Søren S Lund,

    1. Boehringer Ingelheim Pharma GmbH & Co. KG, Ingelheim, Germany
    Search for more papers by this author
  • Christian Gluud,

    1. Copenhagen Trial Unit, Centre for Clinical Intervention Research, Department 7812, Rigshospitalet, Copenhagen University Hospital, The Cochrane Hepato-Biliary Group, Copenhagen, Denmark
    Search for more papers by this author
  • Allan Vaag,

    1. Rigshospitalet and Copenhagen University, Department of Endocrinology, Diabetes and Metabolism, København N, Denmark
    Search for more papers by this author
  • Thomas P Almdal,

    1. Copenhagen University Hospital Gentofte, Department of Medicine F, Hellerup, Denmark
    Search for more papers by this author
  • Jørn Wetterslev

    1. Rigshospitalet, Copenhagen University Hospital, Copenhagen Trial Unit, Centre for Clinical Intervention Research, Department 7812, Copenhagen, Denmark
    Search for more papers by this author

Abstract

Background

Patients with type 2 diabetes mellitus (T2D) have an increased risk of cardiovascular disease and mortality compared to the background population. Observational studies report an association between reduced blood glucose and reduced risk of both micro- and macrovascular complications in patients with T2D. Our previous systematic review of intensive glycaemic control versus conventional glycaemic control was based on 20 randomised clinical trials that randomised 29 ,986 participants with T2D. We now report our updated review.

Objectives

To assess the effects of targeted intensive glycaemic control compared with conventional glycaemic control in patients with T2D.

Search methods

Trials were obtained from searches of The Cochrane Library, MEDLINE, EMBASE, Science Citation Index Expanded, LILACS, and CINAHL (all until December 2012).

Selection criteria

We included randomised clinical trials that prespecified targets of intensive glycaemic control versus conventional glycaemic control targets in adults with T2D.

Data collection and analysis

Two authors independently assessed the risk of bias and extracted data. Dichotomous outcomes were assessed by risk ratios (RR) and 95% confidence intervals (CI). Health-related quality of life and costs of intervention were assessed with standardized mean differences (SMD) and 95% Cl.

Main results

Twenty-eight trials with 34,912 T2D participants randomised 18,717 participants to intensive glycaemic control versus 16,195 participants to conventional glycaemic control. Only two trials had low risk of bias on all risk of bias domains assessed. The duration of the intervention ranged from three days to 12.5 years. The number of participants in the included trials ranged from 20 to 11,140. There were no statistically significant differences between targeting intensive versus conventional glycaemic control for all-cause mortality (RR 1.00, 95% CI 0.92 to 1.08; 34,325 participants, 24 trials) or cardiovascular mortality (RR 1.06, 95% CI 0.94 to 1.21; 34,177 participants, 22 trials). Trial sequential analysis showed that a 10% relative risk reduction could be refuted for all-cause mortality. Targeting intensive glycaemic control did not show a statistically significant effect on the risks of macrovascular complications as a composite outcome in the random-effects model, but decreased the risks in the fixed-effect model (random RR 0.91, 95% CI 0.82 to 1.02; and fixed RR 0.93, 95% CI 0.87 to 0.99; P = 0.02; 32,846 participants, 14 trials). Targeting intensive versus conventional glycaemic control seemed to reduce the risks of non-fatal myocardial infarction (RR 0.87, 95% CI 0.77 to 0.98; P = 0.02; 30,417 participants, 14 trials), amputation of a lower extremity (RR 0.65, 95% CI 0.45 to 0.94; P = 0.02; 11,200 participants, 11 trials), as well as the risk of developing a composite outcome of microvascular diseases (RR 0.88, 95% CI 0.82 to 0.95; P = 0.0008; 25,927 participants, 6 trials), nephropathy (RR 0.75, 95% CI 0.59 to 0.95; P = 0.02; 28,096 participants, 11 trials), retinopathy (RR 0.79, 95% CI 0.68 to 0.92; P = 0.002; 10,300 participants, 9 trials), and the risk of retinal photocoagulation (RR 0.77, 95% CI 0.61 to 0.97; P = 0.03; 11,212 participants, 8 trials). No statistically significant effect of targeting intensive glucose control could be shown on non-fatal stroke, cardiac revascularization, or peripheral revascularization. Trial sequential analyses did not confirm a reduction of the risk of non-fatal myocardial infarction but confirmed a 10% relative risk reduction in favour of intensive glycaemic control on the composite outcome of microvascular diseases. For the remaining microvascular outcomes, trial sequential analyses could not establish firm evidence for a 10% relative risk reduction. Targeting intensive glycaemic control significantly increased the risk of mild hypoglycaemia, but substantial heterogeneity was present; severe hypoglycaemia (RR 2.18, 95% CI 1.53 to 3.11; 28,794 participants, 12 trials); and serious adverse events (RR 1.06, 95% CI 1.02 to 1.10; P = 0.007; 24,280 participants, 11 trials). Trial sequential analysis for a 10% relative risk increase showed firm evidence for mild hypoglycaemia and serious adverse events and a 30% relative risk increase for severe hypoglycaemia when targeting intensive versus conventional glycaemic control. Overall health-related quality of life, as well as the mental and the physical components of health-related quality of life did not show any statistical significant differences.

Authors' conclusions

Although we have been able to expand the number of participants by 16% in this update, we still find paucity of data on outcomes and the bias risk of the trials was mostly considered high. Targeting intensive glycaemic control compared with conventional glycaemic control did not show significant differences for all-cause mortality and cardiovascular mortality. Targeting intensive glycaemic control seemed to reduce the risk of microvascular complications, if we disregard the risks of bias, but increases the risk of hypoglycaemia and serious adverse events.

Résumé scientifique

Contrôle glycémique intensif versus contrôle glycémique classique dans le diabète de type 2

Contexte

Les patients atteints de diabète de type 2 (DT2) présentent un risque accru de maladie cardio-vasculaire et de mortalité par rapport à la population générale. Les études observationnelles signalent une association entre la réduction de la glycémie et la réduction du risque de complications micro et macrovasculaires chez les patients atteints de DT2. Notre revue systématique précédente de contrôle glycémique intensif versus contrôle glycémique classique était basée sur 20 essais cliniques randomisés qui randomisaient 29 986 participants atteints de DT2. Nous rendons maintenant ompte de notre mise à jour de la revue.

Objectifs

Évaluer les effets de cibler un contrôle glycémique intensif par rapport à un contrôle glycémique classique chez les patients atteints de DT2.

Stratégie de recherche documentaire

Les essais ont été trouvés à partir de recherches de La Bibliothèque Cochrane, MEDLINE, EMBASE, Science Citation Index Expanded, LILACS et CINAHL (tous jusqu' à décembre 2012).

Critères de sélection

Nous avons inclus les essais cliniques randomisés qui spécifiaient objectifs de contrôle glycémique intensif versus contrôle glycémique conventionnel chez des adultes atteints de DT2.

Recueil et analyse des données

Deux auteurs ont indépendamment évalué les risques de biais et extrait des données. Les résultats dichotomiques ont été analysés à l'aide des risques relatifs (RR) et des intervalles de confiance (IC) à 95%. La qualité de vie liée à la santé et les coûts de l'intervention a été évaluée, avec des différences moyennes standardisées (DMS) et IC à 95%.

Résultats principaux

Vingt-huit essais totalisant 34 912 participants atteints de DT2 randomisés avec 18 717 participants pour un contrôle glycémique intensif versus 16,195 participants à un contrôle glycémique classique. Seuls deux essais avaient un faible risque de biais pour tous les domaines de risque de biais évalués. La durée de la d'intervention allait de trois jours à 12,5 ans. Le nombre de participants dans les essais inclus allait de 20 à 11 140. Il n'y avait aucune différence statistiquement significative entre contrôle glycémique intensif versus classique pour la mortalité, toutes causes confondues (RR 1.00, IC à 95%, de 0,92 à 1,08; 34 325 participants, 24 essais) ou de mortalité cardio-vasculaire (RR de 1,06, IC à 95%, de 0,94 à 1,21; 34 177 participants, 22 essais). L'analyse séquentielle des essais a montré qu' une réduction du risque relatif de 10% pouvait être réfutée pour la mortalité toutes causes confondues. Le contrôle glycémique intensif n'a pas montré un effet statistiquement significatif sur les risques de complications macrovasculaires en tant que critère composite de jugement dans le modèle à effets aléatoires, mais a diminué les risques dans le modèle à effets fixes (RR aléatoire de 0,91, IC à 95%, de 0,82 à 1,02; et fixé RR 0,93, IC à 95%, de 0,87 à 0,99; P =0,02; 32 846 participants, 14 essais). Le contrôle glycémique intensif versus classique semblait réduire les risques d'infarctus du myocarde non mortel (RR 0,87, IC à 95%, de 0,77 à 0,98; P =0,02; 30 417 participants, 14 essais), l'amputation d'une extrémité inférieure (RR 0,65, IC à 95%, de 0,45 à 0,94; P = 0,02; 11 200 participants, 11 essais), ainsi que le risque de développer un critère composite de jugement de maladies vasculaires microchirurgicales (RR de 0,88, IC à 95%, de 0,82 à 0,95; P = 0,0008; 25 927 participants, 6 essais), néphropathie (RR 0,75, IC à 95%, de 0,59 à 0,95; P = 0,02; 28,096 participants, 11 essais), la rétinopathie (RR 0,79, IC à 95%, de 0,68 à 0,92; P = 0,002; 10,300 participants, 9 essais), et le risque de la photocoagulation de la rétine (RR 0,77, IC à 95%, de 0,61 à 0,97; P =0,03; 11 212 participants, 8 essais). Aucun effet statistiquement significatif de contrôle glycémique intensif n'a pu être démontré sur les accident vasculaire cérébral (AVC) non mortels, la revascularisation cardiaque ou périphérique. Des analyses séquentielles d'essais n'ont pas permis de confirmer une réduction du risque d'infarctus du myocarde non mortel, mais a confirmé une réduction du risque relatif de 10% en faveur du contrôle glycémique intensif sur le résultat composite de maladies microvasculaires. Pour les autres critères de jugement microvasculaires, des analyses séquentielles d'essais n'ont pas pu établir de preuves solides pour une réduction du risque relatif de 10%. Le contrôle glycémique intensif augmente significativement le risque d'hypoglycémie légère, mais une hétérogénéité substantielle était observée; l'hypoglycémie sévère (RR 2,18, IC à 95%, de 1,53 à 3,11; 28 794 participants, 12 essais); et les événements indésirables graves (RR 1,06, IC à 95%, de 1,02 à 1,10; P = 0,007; 24 280 participants, 11 essais). L'analyse séquentielle des essais pour une augmentation du risque relatif de 10% montrait une preuve solide d'hypoglycémie légère et des événements indésirables graves et une augmentation du risque relatif de 30% pour l'hypoglycémie grave en focalisant le contrôle glycémique intensif versus le classique. L'ensemble de qualité de vie liée à la santé, ainsi que la santé mentale et physique liée à la qualité de vie en relation à la santé n'ont pas montré de différence statistiquement significative.

Conclusions des auteurs

Bien que nous avons été en mesure d'élargir le nombre de participants par 16% dans cette mise à jour, nous avons encore trouvé peu des données sur les résultats et le risque de biais des essais était considéré comme élevé. En ciblant le contrôle intensif en comparaison avec le contrôle classique il n'y avait pas de différences significatives en termes de mortalité toutes causes confondues et de mortalité cardio-vasculaire. En ciblant le contrôle glycémique intensif, il semblait avoir une réduction du risque de complications microvasculaires, si nous rejetons les risques de biais, mais une augmentation du risque d'hypoglycémie et d'événements indésirables graves.

Plain language summary

Targeting intensive glycaemic control versus targeting conventional glycaemic control for people with type 2 diabetes mellitus

People with type 2 diabetes mellitus (T2D) have an increased mortality and morbidity compared to the general population. T2D is characterised by several metabolic defects including impaired insulin secretion and action, causing chronic hyperglycaemia (high glucose levels in the blood). Chronic hyperglycaemia is strongly associated with increased risk of kidney, eye, and nerve complications (microvascular complications) as well as increased risk of stroke, heart disease, and amputations (macrovascular complications). Epidemiological studies suggest that reducing blood glucose in people with T2D may reduce the risk of death and morbidity. However, such studies do not represent a reliable methodology to assess the effects of interventions because of the inherent risk of imbalances (which may be hidden and therefore uncorrectable) between groups, other than those resulting from the interventions. It is still not clear whether targeting more intensive glycaemic control is better than conventional glycaemic control in terms of clinical outcomes based on evidence from randomised clinical trials (RCTs).

In this updated Cochrane systematic review, we identified 28 RCTs comparing intensive glycaemic control versus conventional glycaemic control in participants with T2D. A total of 18,717 participants randomised to intensive glycaemic control and 16,195 to conventional glycaemic control were included in the analyses. The trials were primarily conducted in Europe and Northern America. The mean duration of the intervention period varied from three days to 12.5 years. Only two trials were considered to have low risk of bias; we may, therefore, have evaluated RCTs with high risk of overestimating beneficial effects and underestimating harmful effects.

Our analyses did not show any statistically significant reduction in either death from any cause or death from heart disease when targeting intensive glycaemic control compared with conventional control. Intensive glycaemic control seemed to reduce the risk of non-fatal myocardial infarction, amputation of a lower extremity, and microvascular complications while increasing the risk of severe adverse events and hypoglycaemia. Targeting intensive glycaemic control did not appear to change the risk of non-fatal stroke, cardiac revascularization (a procedure to reconstruct damaged heart blood vessels), and peripheral revascularization. Health-related quality of life did not differ significantly when comparing targeting intensive with conventional glycaemic control.

There is a need for more powerful RCTs with low risk of bias to guide the choice of targeting intensive versus conventional glycaemic control in patients with T2D.

Résumé simplifié

Contrôle glycémique intensif versus contrôle glycémique classique dans les personnes avec le diabète de type 2

Les patients atteints de diabète de type 2 (DT2) présentent une augmentation de la morbidité et de la mortalité par rapport à la population générale. Le DT2 se caractérise par plusieurs déficiences métaboliques y compris une déficience de la sécrétion et de l'action de l'insuline entraînant une hyperglycémie chronique (une concentration élevée de glucose dans le sang). L'hyperglycémie chronique est fortement associée à un risque accru de complications rénales, oculaires et nerveuses (complications microvasculaires) ainsi qu'à une augmentation du risque d'accident vasculaire cérébral, de cardiopathie et d'amputation (complications macrovasculaires). Les études épidémiologiques suggèrent que la réduction de la glycémie chez les patients atteints de DT2 peut réduire le risque de décès et la morbidité. Cependant, ces études ne peuvent pas avoir une méthodologie fiable pour évaluer les effets des interventions en raison du risque inhérent de déséquilibres (qui peuvent être cachés et donc pas corrigibles) entre les groupes, autres que ceux résultant d'interventions. Sur la base de preuves issues d'essais cliniques randomisés (ECR), il n'est toujours pas clair si un contrôle glycémique plus intensif est plus efficace qu' un contrôle glycémique classique en termes de résultats cliniques.

Dans cette revue systématique Cochrane mise à jour, nous avons identifié 28 ECR comparant le contrôle glycémique intensif versus le contrôle glycémique conventionnel chez des participants atteints de DT2. Un total de 18 717 participants randomisés avec un contrôle glycémique intensif et 16 195 avec un contrôle glycémique classique ont été inclus dans les analyses. Les essais avaient été principalement réalisés en Europe et en Amérique du Nord. La durée moyenne de la période d'intervention allait de trois jours à 12,5 ans. Seuls deux essais ont été considérées à faible risque de biais; nous avons par conséquent évalué des ECR présentant un risque élevé de surestimer des effets bénéfiques et de sous-estimer des effets nocifs.

Nos analyses n'ont pas mis en évidence aucune réduction statistique significative de décès toutes causes confondues ou des décès à cause de cardiopathie associée au contrôle glycémique intensif par rapport au contrôle classique. Le contrôle glycémique intensif a semblé réduire le risque d'infarctus du myocarde non mortel, l'amputation d'une extrémité inférieure, et de complications microvasculaires mais augmentait le risque d'événements indésirables graves et d'hypoglycémie. Le contrôle glycémique intensif ne semblait pas affecter le risque d'accident vasculaire cérébral (AVC) non mortels, de revascularisation cardiaque (une procédure de reconstruction de vaisseaux cardiaques endomagée) et de revascularisation périphérique. La qualité de vie liée à la santé ne différait pas significativement en comparant le contrôle glycémique conventionnel et intensif.

Ils sont nécessaires davantage d'ECR de qualité présentant un faible risque de biais pour orienter le choix du contrôle glycémique intensif versus classique chez les patients atteints de DT2.

Notes de traduction

Traduit par: French Cochrane Centre
Traduction financée par: Pour la France : Minist�re de la Sant�. Pour le Canada : Instituts de recherche en sant� du Canada, minist�re de la Sant� du Qu�bec, Fonds de recherche de Qu�bec-Sant� et Institut national d'excellence en sant� et en services sociaux.

Laički sažetak

Intenzivna ili standardna kontrola glukoze u oboljelih od dijabetesa tipa 2

Oboljeli od dijabetesa tipa 2 imaju veću smrtnost i pobol u odnosu na opću populaciju. Dijabetes tipa 2 obilježen je nekolicinom metabolički poremećaja, koji uključuju poremećeno izlučivanje inzulina i kroničnu hiperglikemiju (povišena razina glukoze u krvi). Kronična hiperglikemija je povezana s povećanim rizikom od komplikacija koje zahvaćaju bubrege, oči i živce (komplikacije koje zahvaćaju male krvne žile, odnosno mikrovaskularne komplikacije), kao i povećanim rizikom od moždanog udara, srčanih bolesti i amputacija (komplikacije koje zahvaćaju velike krvne žile, odnosno makrovaskularne komplikacije). Epidemiološke studije pokazuju da smanjivanje razine glukoze u krvi u oboljelih od dijabetesa tipa 2 smanjuje rizik od smrti i pobola. Međutim, takve studije ne predstavljaju pouzdanu metodu za procjenu učinka neke intervencije zbog mogućeg rizika od pristranosti i razlika među skupinama koje se ne odnose na istraživanu intervenciju. Još uvijek nije jasno je li intenzivna kontrola glikemije (razine glukoze u krvi) bolja od konvencionalne kontrole glikemije za smanjenje smrtnosti ili srčanih bolesti, na temelju dokaza iz randomiziranih kontroliranih pokusa.

U ovom obnovljenom Cochrane sustavnom pregledu pronađeno je 28 studija koje su usporedile intenzivnu i standardnu kontrolu glikemije u osoba s dijabetesom tipa 2. Utim studijama ukupno 18.717 ispitanika nasumično je raspoređeno u skupinu s intenzivnom kontrolom glikemije i 16.195 u skupinu sa standardnom kontrolom glikemije. Istraživanja su provedena prvenstveno u Europi i Sjevernoj Americi. Srednje trajanje intervencije trajalo je od 3 dana do 12,5 godina. Samo su dva istraživanja procijenjena niskim rizikom od pristranosti, stoga su sve studije analizirane imajući u vidu da njihovi rezultati možda pretjerano prikazuju korisne učinke intervencije, a podcjenjuju moguće štetne učinke.

Analize u ovom sustavnom pregledu nisu pokazale statistički značajno smanjenje u broju srmti od svih uzroka niti smrti od srčanih bolesti u skupini s intenzivnom kontrolom glukoze u usporedbi s ispitanicima sa standardnom kontrolom glukoze. Intenzivna kontrola glikemije mogla bit smanjiti rizik od ne-fatalnog srčanog udara (infarkta miokarda), amputacije donjih udova i mikrovaskularnih komplikacija (u malim krvnim žilama), dok u isto vrijeme može povećati rizik od ozbiljnih nuspojava i hipoglikemije. Ne čini se da korištenje intenzivne kontrole glikemije mijenjaju rizik od ne-fatalnog moždanog udara, srčane revaskularizacije (postupka u kojem se uspostavlja tijek krvi u oštećenim krvnim žilama) i periferne revaskularizacije (takav postupak na perifernim krvnim žilama). Kvaliteta života povezana sa zdravljem nije se značajno razlikovala kad se usporede skupine s intenzivnom ili standardnom kontrolom glikemije.

Potrebno je više većih randomiziranih kontroliranih pokusa, visoke kvalitete, kako bi se utvrdilo što je korisnije za oboljele od dijabetesa tipa 2 - intenzivna ili standardna kontrola glukoze u krvnoj plazmi.

Bilješke prijevoda

Cochrane Hrvatska
Prevela: Livia Puljak
Ovaj sažetak preveden je u okviru volonterskog projekta prevođenja Cochrane sažetaka. Uključite se u projekt i pomozite nam u prevođenju brojnih preostalih Cochrane sažetaka koji su još uvijek dostupni samo na engleskom jeziku. Kontakt: cochrane_croatia@mefst.hr

Резюме на простом языке

Ориентация на интенсивный гликемический контроль против ориентации на обычный гликемический контроль для людей с сахарным диабетом 2 типа

Люди с сахарным диабетом 2 типа (СД2) имеют повышенную смертность и заболеваемость по сравнению с общей популяцией. СД2 характеризуется некоторыми метаболическими дефектами, такими, как нарушение секреции инсулина и его действия, что вызывает хроническую гипергликемию (высокий уровень глюкозы в крови). Хроническая гипергликемия жёстко связана с повышенным риском осложнений со стороны почек, глаз и нервов (микрососудистые осложнения), а также с повышенным риском развития инсульта, ишемической болезни сердца, и ампутаций (макрососудистые осложнения). Эпидемиологические исследования позволяют предполагать, что снижение уровня глюкозы крови у людей с СД2 может снизить риск смерти и заболеваемость. Тем не менее, такие исследования не представляют собой надежную методику оценки эффектов вмешательств из-за присущего им риска несбалансированности (которые могут быть скрыты и, следовательно, неисправимы) групп между собой, кроме тех, которые являются результатом вмешательств. До сих пор не ясно, будет ли более интенсивный контроль гликемии лучше, чем обычный гликемический контроль, с точки зрения клинических исходов, основанных на доказательствах из рандомизированных клинических испытаний (РКИ).

В этом обновленном Кокрейновском систематическом обзоре мы выявили 28 РКИ, сравнивающих интенсивный гликемический контроль с обычным гликемическим контролем у участников с СД2. Всего были включены в анализ 18717 участников, рандомизированных в группу интенсивного контроля гликемии и 16195 участников, рандомизированных в группу обычного контроля гликемии. В основном, исследования были проведены в Европе и в Северной Америке. Средняя продолжительность периода вмешательства варьировала от трех дней до 12,5 лет. Только два испытания были оценены как имеющие низкий риск смещения; следовательно, мы, возможно, оценили РКИ с высоким риском переоценки благотворного влияния и недооценки вредных последствий.

Наши анализы не показали какого-либо статистически значимого снижения ни смерти от любых причин, ни смерти от сердечно-сосудистых заболеваний при ориентации на интенсивный гликемический контроль по сравнению с обычным контролем. Интенсивный контроль гликемии, казалось, уменьшал риск нефатального инфаркта миокарда, ампутации нижней конечности и микрососудистых осложнений, увеличивая риск серьезных побочных эффектов и гипогликемии. Ориентация на интенсивный гликемический контроль, оказалось, не изменяет риска нефатального инсульта, реваскуляризации сердца (процедуры для восстановления поврежденных кровеносных сосудов сердца) и периферической реваскуляризации. Связанное со здоровьем качество жизни значимо не отличалось при сравнении ориентирования на интенсивный гликемический контроль с обычным.

Существует потребность в более мощных РКИ с низким риском смещения (систематической ошибки), чтобы рекомендовать выбор интенсивного или обычного контроля гликемии у пациентов с СД2.

Заметки по переводу

Перевод: Александрова Эльвира Григорьевна. Редактирование: Зиганшина Лилия Евгеньевна. Координация проекта по переводу на русский язык: Казанский федеральный университет. По вопросам, связанным с этим переводом, пожалуйста, свяжитесь с нами по адресу: lezign@gmail.com

Laienverständliche Zusammenfassung

Strenge Blutzuckerkontrolle im Vergleich zu herkömmlicher Blutzuckerkontrolle für Menschen mit Diabetes mellitus Typ 2

Menschen mit Diabetes mellitus Typ 2 (T2D) haben eine erhöhte Sterblichkeit und Morbidität im Vergleich zur allgemeinen Bevölkerung. T2D ist durch mehrere Stoffwechseldefekte gekennzeichnet, einschließlich einer beeinträchtigten Insulin-Sekretion und -Wirkung, die chronische Hyperglykämie (hoher Blutzuckerspiegel) verursachen können. Chronische Hyperglykämie wird stark mit einem erhöhten Risiko von Nieren-, Augen- und Nervenkomplikationen (mikrovaskuläre Komplikationen) sowie einem erhöhten Risiko für Schlaganfall, Herzerkrankungen und Amputationen (makrovaskuläre Komplikationen) verbunden. Epidemiologische Studien deuten darauf hin, dass die Verringerung des Blutzuckers bei Patienten mit T2D das Sterblichkeits- und Krankheitsrisiko reduzieren kann. Jedoch weisen solche Studien keine verlässliche Methodik auf, um die Wirkungen von Interventionen zu bewerten. Es besteht nämlich ein inhärentes Risiko für Ungleichgewicht (das versteckt und daher nicht korrigierbar ist) zwischen den Gruppen, mit Ausnahme derer, die aus den Interventionen resultieren. Es ist noch nicht klar, ob strenge Blutzuckerkontrolle besser ist als herkömmliche Blutzuckerkontrolle in Bezug auf klinischen Endpunkte, basierend auf Evidenz aus randomisierten klinischen Studien (RCTs).

In diesem aktualisierten systematischen Cochrane Review haben wir 28 RCTs identifiziert, die bei Teilnehmern mit T2D die strenge Blutzuckerkontrolle mit der herkömmlichen Blutzuckereinstellung vergleichen. Insgesamt wurden 18.717 Teilnehmer, die randomisiert der Gruppe mit der strengen Blutzuckerkontrolle zugeteilt wurden und 16.195 Teilnehmer, die der herkömmlichen Blutzuckerkontrolle zugeteilt wurden, in die Analysen einbezogen. Die Studien wurden vor allem in Europa und Nordamerika durchgeführt. Die mittlere Dauer des Interventionszeitraums variiert von drei Tagen bis 12,5 Jahre. Nur zwei Studien hatten ein geringes Risiko für Bias; wir haben daher RCTs mit hohem Risiko für die Überschätzung positiver Auswirkungen und für die Unterschätzung schädlicher Auswirkungen ausgewertet.

Unsere Analysen zeigten beim Vergleich von strenger Blutzuckerkontrolle mit herkömmlicher Kontrolle keine statistisch signifikante Verminderung der Todesfälle jeglicher Ursache oder der Todesfälle durch Herzerkrankungen. Die strenge Blutzuckerkontrolle schien das Risiko von nicht-tödlichen Herzinfarkten, Amputation der unteren Extremität und mikrovaskuläre Komplikationen zu reduzieren und gleichzeitig die Gefahr von schweren unerwünschten Wirkungen und Hypoglykämie (Unterzuckerung) zu erhöhen. Die strenge Blutzuckerkontrolle scheint das Risiko für nicht-tödliche Schlaganfälle, Revaskularisation des Herzens (ein Verfahren, um beschädigte Herzblutgefäße zu rekonstruieren) und periphere Revaskularisation nicht zu verändern. Gesundheitsbezogene Lebensqualität unterschied sich nicht signifikant bei einem Vergleich der strengen Blutzuckerkontrolle mit der herkömmlichen Blutzuckerkontrolle.

Es besteht ein Bedarf an aussagekräftigeren RCTs mit geringem Risiko für Bias, um bei der Entscheidung zwischen strenger Blutzuckerkontrolle oder herkömmlicher Kontrolle bei Patienten mit T2D zu beraten.

Anmerkungen zur Übersetzung

K. Kunzweiler, Koordination durch Cochrane Schweiz.

Summary of findings(Explanation)

Summary of findings for the main comparison. Intensive glycaemic control versus conventional glycaemic control for type 2 diabetes mellitus
  1. aDowngraded by one due to trial sequential analysis showing that more data are needed; funnel plot indicates small trial bias

    bDowngraded by one due to a relatively few number of participants reporting non-fatal stroke or end-stage renal disease

    cDowngraded by two due to trial sequential analysis showing that more data are needed, a relatively few number of participants reporting amputation of lower extremity and most of the events stemming from one trial only

Intensive glycaemic control versus conventional glycaemic control for type 2 diabetes mellitus
Patient or population: participants with type 2 diabetes mellitus
Settings: mostly outpatients
Intervention: intensive glycaemic control versus conventional glycaemic control
OutcomesIllustrative comparative risks* (95% CI)Relative effect
(95% CI)
No of participants
(studies)
Quality of the evidence
(GRADE)
Comments
Assumed riskCorresponding risk
Control Intensive glycaemic control versus conventional glycaemic control
All-cause mortality
Follow-up: median 24 months
95 per 1000 95 per 1000
(87 to 103)
RR 1
(0.92 to 1.08)
34325
(24)
⊕⊕⊕⊝
moderatea
-
Cardiovascular mortality
Follow-up: median 27 months
45 per 1000 48 per 1000
(42 to 55)
RR 1.06
(0.94 to 1.21)
34177
(22)
⊕⊕⊕⊝
moderatea
-
Non-fatal myocardial infarction
Follow-up: median 60 months
48 per 1000 41 per 1000
(37 to 47)
RR 0.87
(0.77 to 0.98)
30417
(14)
⊕⊕⊕⊝
moderatea
-
Non-fatal stroke
Follow-up: median 54.6 months
29 per 1000 29 per 1000
(25 to 35)
RR 1
(0.84 to 1.19)
30003
(13)
⊕⊕⊕⊝
moderateb
-
Amputation of lower extremity
Follow-up: median 65.1 months
13 per 1000 9 per 1000
(6 to 12)
RR 0.65
(0.45 to 0.94)
11200
(11)
⊕⊕⊝⊝
lowc
-
End-stage renal disease
Follow-up: median 93.6 months
16 per 1000 14 per 1000
(11 to 17)
RR 0.87
(0.71 to 1.06)
28145
(8)
⊕⊕⊕⊝
moderateb
-
Hypoglycaemia - Severe hypoglycaemia
Follow-up: median 12 months
29 per 1000 64 per 1000
(45 to 91)
RR 2.18
(1.53 to 3.11)
28794
(17)
⊕⊕⊕⊕
high
Trial sequential analysis showed firm evidence for a 30% relative risk increase with intensive glycaemic control.
*The basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).
CI: Confidence interval; RR: Risk ratio
GRADE Working Group grades of evidence
High quality: Further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: We are very uncertain about the estimate.

Background

Description of the condition

The prevalence of type 2 diabetes mellitus (T2D) is increasing worldwide (King 1998). Insulin resistance in peripheral tissues and inadequate compensatory insulin secretion are essential elements in the pathogenesis of T2D (LeRoith 2002). Reduced insulin secretion is caused by a decrease in the β-cell mass, dysfunction of existing β-cells, or both (LeRoith 2002). A consequence of this is chronic hyperglycaemia (elevated levels of plasma glucose) with disturbances of carbohydrate, fat, and protein metabolism (LeRoith 2002).

Chronic hyperglycaemia is strongly associated with microvascular (for example, nephropathy, retinopathy, and neuropathy) as well as macrovascular complications (for example, ischaemic heart disease, stroke, and ischaemia of the lower extremities). The mortality rate is increased among individuals with T2D compared to the non-diabetic population. The main cause of the increased mortality is macrovascular disease (Almdal 2004; de Marco 1999; Stamler 1993).

Description of the intervention

Since the discovery of insulin for the treatment of diabetes mellitus, the primary immediate goal in the treatment of diabetes mellitus has been to normalise or near normalise blood glucose (Bliss 2005). T2D is a progressive disease with β-cell function deteriorating over time (UKPDS 1998). Therefore, the glucose-lowering treatment must be intensified over that time in order to achieve near normal glycaemia. All T2D patients are initially advised to follow 'lifestyle' interventions including weight loss and increased physical activity (AACE/ACE Consensus Statement 2009; IDF 2012; Inzucchi 2012). However, in order to maintain optimal glycaemic control over time, the large majority of patients with T2D will require additional glucose-lowering pharmacological intervention. The most commonly used first-line glucose-lowering medications are oral glucose-lowering drugs, primarily metformin (which is thought to increase insulin sensitivity). Insulin secretagogues (sulphonylureas, glinides, or incretin-based therapies) that stimulate insulin secretion are also recommended and used among the first-line therapy options (AACE/ACE Consensus Statement 2009; IDF 2012; Inzucchi 2012). In addition to lowering blood glucose, sulphonylureas or glinides often increase the risk of hypoglycaemia and promote weight gain, whereas metformin or incretin-based therapies appear to have either neutral or beneficial effects (for example weight loss) when given as monotherapy (AACE/ACE Consensus Statement 2009; IDF 2012; Inzucchi 2012).

If lifestyle changes and the maximum tolerated dose of one oral glucose-lowering drug fail to achieve the glycaemic goal, other glucose-lowering drugs may be added (AACE/ACE Consensus Statement 2009; Hemmingsen 2013; IDF 2012; Inzucchi 2012). In the case of suboptimal glycaemic control, insulin treatment can be initiated. In contrast to other glucose-lowering medications, there is theoretically no upper limit to the dose of insulin above which further glucose-lowering will be absent. Hence, insulin can be used at all stages of the disease.

At present, the evidence forming the basis for the recommendations set out in the current guidelines for treating T2D mostly consists of the documented ability of the various interventions to reduce blood glucose, as well as data on adverse effects such as weight gain and hypoglycaemia. Only relatively few clinical trials of glucose-lowering interventions have reported mortality or disabling vascular complications in the eyes, kidney, heart etc. The effects on such outcomes are therefore not well understood and, to some extent, are even contradictory. For example, there has been great concern about the cardiovascular risk profile of rosiglitazone. The concerns resulted in a recent withdrawal of all rosiglitazone-containing anti-diabetic medicines by the European Medicines Agency (EMA 2010). The US Food and Drug Administration (FDA) also re-evaluated the use of rosiglitazone but decided to keep rosiglitazone on the market and place more restrictions on its manufacturer (Cohen 2010). Despite decades of research there is still no compelling evidence demonstrating clear beneficial or harmful effects on mortality or diabetic complications of other currently available glucose-lowering drugs (Inzucchi 2012).

The question of whether lowering or intending to lower blood glucose per se is beneficial in patients with T2D with respect to several patient-relevant outcomes, for example mortality and cardiovascular disease, remains unanswered. In patients with type 1 diabetes mellitus, a beneficial effect of intensive glycaemic control on cardiovascular disease and mortality has been suggested (DCCT/EDIC 2005). In persons with T2D, observational studies suggest that hyperglycaemia is associated with an increased risk of cardiovascular disease and mortality (UKPDS-35 2000), and a 10-year observational follow-up from the 'UK Prospective Diabetes Study' (UKPDS) suggested long-term beneficial effects of intensive glucose control on cardiovascular disease and mortality (UKPDS-80 2008). However, in patients with T2D, three recent randomised clinical trials have not been able to detect (or reject) reduced cardiovascular morbidity or mortality as a result of intensive glycaemic control when compared with conventional glycaemic control (ACCORD 2008; ADVANCE 2008; VADT 2009). In fact, the 'Action to Control Cardiovascular Risk in Diabetes' (ACCORD) trial showed increased mortality in the group allocated to intensive glycaemic control compared with the conventional glycaemic control group. Such an adverse effect was not observed in the 'Action in Diabetes and Vascular Disease: Preterax and Diamicron Modified Release Controlled Evaluation' (ADVANCE) trial or the 'Veterans Affairs Diabetes Trial' (VADT) despite very similar achieved levels of glycaemic control, about 6.5% to 7.0%, in all three trials. The cause of the increased mortality in the ACCORD trial has not been clarified but factors such as baseline glycaemic level, neuropathy, and aspirin use were shown to significantly influence the effect on mortality when targeting intensive glycaemic control. In contrast, suggested factors such as diabetes duration, age, hypoglycaemia, pre-existing cardiovascular disease, and drug interactions have not been shown to be of importance (Calles-Escandon 2010).

The three recently published trials used different glycaemic targets and glucose-lowering strategies to achieve these targets (ACCORD 2008; ADVANCE 2008; VADT 2009). Hence, the definition of intensive and conventional glycaemic control varied among the trials. The ACCORD trial and VADT used a target glycosylated haemoglobin A1c (HbA1c) for intensive glycaemic control of below 6.0%, compared to a target of below 6.5% in the ADVANCE trial. The definition of conventional glycaemic control was expressed as a target HbA1c of 7% to 8% in all except the ADVANCE trial, which referred to local guidelines (Table 1). The results from these trials have created a debate about the optimal choice of glycaemic target. At present (April 2013), the American Diabetes Association (ADA)/ European Association for the Study of Diabetes (EASD) recommend an HbA1c level of less than 7.0% as the standard glycaemic treatment goal, whereas the American Association of Clinical Endocrinologists (ACEE/ACE) recommends an HbA1c level of less than 6.5% (AACE/ACE Consensus Statement 2009; Inzucchi 2012).  

Table 1. Glycaemic control in trials
  1. aIn the ACCORD trial, SMBG targets were defined as “action required” thresholds (see protocol at http://www.accordtrial.org/public/protocol_2005-05-11.pdf)

    ACCORD: Action to Control Cardiovascular Risk in Diabetes, ADVANCE: Action in Diabetes and Vascular Disease: Preterax and Diamicron Modified Release Controlled Evaluation, DIGAMI: Diabetes and Insulin-Glucose Infusion in Acute Myocardial Infarction, HbA1c: glycosylated haemoglobin A1c, IDA: Insulin Diabetes Angioplasty; IQR: Interquartile range, REMBO: Rational Effective Multicomponent Therapy in the Struggle Against DiaBetes Mellitus in Patients With COngestve Heart Failure, SE: Standard error, SMBG: self monitoring of blood glucose; UGDP: University Group Diabetes Program, UKPDS: United Kingdom Prospective Diabetes Study, VA CSDM: Veterans Affairs Cooperative Study in Type 2 Diabetes Mellitus, VADT: Veterans Affairs Diabetes Trial

    C: control; I: intervention

Characteristic

Study ID

Glycaemic target, intensive treatmentGlycaemic target, conventional treatmentGlycaemic control at the end of follow-up, HbA1c (%) or other used glycaemic measurement (mmol/L) (mean(SD))Number of participants achieving treatment goalComment
ACCORD 2008

HbA1c < 6%

Fasting SMBGa < 5.6 mmol/L (100 mg/dL) or

2 hour blood glucose < 7.8 mmol/L (140 mg/dL)

HbA1c 7%-7.9%

Fasting SMBGa > 5.0 mmol/L (90 mg/dL)

I: 6.4 (0.7)

C: 7.5 (0.8)

I: NR

C: NR

Data are from the end of the intervention period.

SD is calculated from IQR

ADDITION-EuropeHbA1c < 7%, but change in antidiabetic medicine with HbA1c > 6.5%Not specified

I: 6.6 (1.0)

C: 6.7 (1.0)

I: 1143 out of 1513

C: 878 out of 1226

Number of participants achieving glycaemic target are the number of participants with HbA1c less than 7% at the end of follow-up. Number read from figure
ADVANCE 2008HbA1c ≤ 6.5%Glycaemic target of HbA1c defined from
local guidelines

I: HbA1c: 6.5 (1.0)

C: HbA1c: 7.2 (1.4)

I: 3133 (at the end of follow-up)

C: No specified target

 
Araki 2012

 

HbA1c < 6.9%

Prevent diabetic come and avoid remarkable hyperglycaemia (fasting blood glucose < 200 mg/dl or HbA1c < 9%)

I: 7.7 (1.10)

C: 7.8 (1.12)

I: 343

C: 304

The number of participants measured for the achieved glycaemic control was 325 in the intensive group and 338 in the conventional group
Bagg 2001

HbA1c < 7%

Before meal capillary glucose: 4-7 mmol/L, 2 hour blood glucose < 10 mmol/L

Avoid symptomatic hyperglycaemia and fortnightly fasting capillary glucose test > 17 mmol/L

I: HbA1c: 7.0 (0.4)

C: HbA1c: 10.2 (0.2)

I: 3

C: No specific target

 
Becker 2003Fasting capillary blood glucose < 6.5 mmol/LFasting capillary blood glucose < 8.5 mmol/L

I: 7.2 (1.2)

C: 7.4 (1.4)

I: NR

C: NR

 
Blonde 2009Fasting plasma glucose 3.9–5.0 mmol/LFasting plasma glucose 4.4–6.1 mmol/L

I: HbA1c: 6.9

C: HbA1c: 6.9

I: NR

C: NR

The glycaemic treatment goal in the trial was measured as fasting blood glucose levels. The proportion of participants achieving the target
Cao 2011Blood glucose target between 4.4 and 6.1 mmol/Lblood glucose target between 10.0 and 11.0 mmol/L

I: (Postoperative) 5.5 (0.8) mmol/L

C: (Postoperative) 9.9 (1.0) mmol/L

I: NR

C: NR

 
Cooray 2011Reduce the HbA1c with 1% unitContinue unchanged conventional glycaemic control

I: NR

C: NR

I: NR

C: NR

 
DIGAMI 2 2005Insulin infusion until stable normoglycaemia and at least for 24 hours. Subcutaneous insulin was initiated at the cessation of the infusion. The treatment goal for patients in group 1 was a fasting blood glucose level of 5-7 mmol/L and a non-fasting level of < 10 mmol/LStandard glucose control

I: HbA1c: 7.0 (1.0)
FBG (target): 8.0 (2.0)

C: HbA1c: 7.0 (1.3)
FBG: 8.6 (3.0)

I: NR

C: NR

HbA1c and fasting blood glucose is read from figure
Fantin 2011Targeting glucose levels between 80 and 100 mg/dL in pre meal periods and lower than 140 mg/dL in random glucose measurements250 mg/dl or less

I: FBG: 8.9 (3.5)

C: FBG: 11.1 (5.4)

I: NR

C: NR

Value for fasting blood glucose is calculated from text. mg/dl is calculated to mmol/L by dividing with 18
Guo 2008

HbA1c < 7.0%

Fasting plasma glucose 4-7 mmol/L

No treatment goal

I: HbA1c: 6.3 (0.9)
FPG (target): 7.1 (1.7)

C: HbA1c: 7.1 (0.9)
FPG: 8.3 (2.6)

I: 142

C: No specific target

 
IDA 2009

HbA1c < 6.5%

Fasting blood glucose 5-7 mmol/L, before meal < 10 mmol/L

Standard treatment

I: HbA1c (median): 6.3 (1.5)

C: HbA1c (median): 6.6 (0.9)

I: 37 (HbA1c)

C: No specific target

SD is calculated from IQR
Jaber 1996Fasting blood glucose ≤ 6.6 mmol/L, 2 hour postprandial glucose < 10 mmol/L, or to reach maximum daily doses of sulphonylureaNot defined

I: 9.2 (2.1)
FBG (target): 8.5 (2.3)

C: 12.1 (3.7)
FBG: 11.0 (3.9)

I: NR

C: NR

Measurement of glycated haemoglobin is not further specified
Kumamoto 2000

HbA1c < 7.0%

Fasting blood glucose (< 140 mg/dL), 2 hour postprandial glucose < 200 mg/dL, mean amplitude of glycaemic excursions < 100 mg/dL

Fasting blood glucose close to < 140 mg/dL without symptoms of hyperglycaemia or hypoglycaemia.

I: 7.2 (1.0)
FBG (target): 6.3 (1.6)

C: 9.4 (1.3)
FBG (target): 7.4 (1.6)

I: 14

C: 3

 
Lu 2010Fasting blood glucose < 6.1 mmol/L, postprandial 2 hour glucose < 7.8 mmol/LFasting blood glucose < 7.0 mmol/L, postprandial 2 hour glucose < 10.0 mmol/L.

I: 6.1 (0.5)

C: 7.8 (0.7)

I: NR

C: NR

 
Melidonis 2000Blood glucose 8.3-11.0 mmol/L in the first 48 hours after an acute coronary event, thereafter normoglycaemiaNo specific target

I: HbA1c not measured.
Plasma glucose (target): 6.6 mmol/L (0.5)

C: HbA1c not measured.
Plasma glucose: 10.5 mmol/L (2.1)

I: NR

C: NR

The reported plasma glucose value is for the last day of hospitalisation
Natarajan 2012Fasting capillary glucose levels of 4.0-5.4 mmol/LHbA1c < 8.4%

I: 6.9 (0.8)

C: 7.6 (1.5)

I: NR

C: NR

 
REMBO 2008

HbA1c < 7% in participants receiving sulphonylurea;

HbA1c < 6.5% in participants receiving insulin

Not specified, standard care

I (median): 6.7 (1.2)

C (median): 6.7 (1.2)

I: NR

C: NR

SD is calculated from IQR
Service 1983HbA1c to normal range, and to maintain 80 minute postprandial plasma glucose below 150 mg/dL (8.3 mmol/L)Eliminate symptoms, but not to a degree to reduce 80 minute postprandial plasma glucose below 150 mg/dL

I (median): 8.7

C (median): 9.4

I: 3

C: 4

 
Stefanidis 2003Near normal glycaemia (defined as 6.6-8.2 mmol/L)No specific target

I: 8 (1.1)
Plasma glucose (target): 6.9 mmol/L (1.8)

C: 8 (1.0)
Plasma glucose: 9.9 mmol/L (1.7)

I: 31

C: NR

We assume HbA1c is unchanged at the end of follow-up due to the short intervention period.

The reported plasma glucose value is for the last day of hospitalisation

Steno-2 2008HbA1c < 6.5%HbA1c < 7.5% (1993-1999), HbA1c < 6.5% (2000-2001)

I: 7.9 (1.2)

C: 9.0 (1.8)

I: 13

C: 3

Data are from the end of the intervention period (7.8 years of follow-up).

Number of patients achieved glycaemic target is read from figure.

UGDP 1975Maintain blood glucose in normal range (defined as fasting blood glucose < 110mg/100 mL, blood glucose < 210 mg/100 mL one hour after ingestion of 50 gm glucose and one and one-half hours after the morning insulin injection)Minimize the likelihood of hypoglycaemic reactions without reducing the insulin dose to pharmacologically inactive amounts

I: FBG (target): 6.7

C: FBG: 9.7

I: NR

C: NR

Value for fasting blood glucose is calculated from text. mg/dl is calculated to mmol/l by dividing with 18.

SD calculated from SE.

UKPDS 1998

Fasting blood glucose < 6 mmol/L

In insulin treated patients; pre-meal glucose 4-7 mmol/L

Fasting blood glucose < 15 mmol/L without symptoms of hyperglycaemia

I (median): 8.1 (1.8)
FPG (target, median): 8.6 mmol/l

C (median): 8.7 (1.6)
FPG (target, median): 9.8 mmol/L

I: NR

C: NR

HbA1c used is the median for the last 5 years follow-up period. SD calculated from IQR.

Fasting plasma glucose read from figure, 10 years after randomisation.

Data are from the UKPDS 33

VA CSDM 1995Maintain mean HbA1c < 7.5%.
Treament is adjusted with home blood glucose monitoring aiming, at fasting blood glucose of 4.48 to 6.44 mmol/L and other pre prandial levels ≤ 7.28 mmol/L
Avoiding excessive hyperglycaemia, or symptoms of excessive glucosuria, ketonuria, or hypoglycaemia (alert HbA1c < 12.9%)

I: HbA1c: 7.1 (0.7)

C: HbA1c: 9.2 (0.8)

I: 7 (maintained target)

C: No specified target

The value of HbA1c is estimated from figure after 24 months of follow-up.

The SD for HbA1c is calculated from SE

VADT 2009HbA1c ≤ 6%. The goal for HbA1c level was an absolute reduction of 1.5 percentage points in the intensive-therapy group, as compared with conventional intervention groupWell-being, avoidance of deterioration of HbA1c, keeping levels at 8–9% and preventing symptoms of glycosuria, hypoglycaemia, and ketonuria

I (median): 6.9 (0.9)

C (median): 8.4 (1.2)

I: NR

C: NR

HbA1c estimated from figure.

Data from glycaemic control are medians. SD calculated from IQR

Yang 2007Fasting blood glucose < 7.0 mmol/L, 2 hour postprandial glucose < 10 mmol/L, HbA1c < 7.0%Not specified

I: 6.5 (1.1)

C: 6.8 (1.4)

I: NR

C: NR

 
Zhang 2011 HbA1c of 6.5% or less, with the purpose of making sure the difference value of HbA1c between the two groups ≥1%Standard control with HbA1c target based on local guidelines

I: 6.3 (0.7)

C: 7.0 (0.7)

I: NR

C: NR

 

In relation to prevention of microvascular complications in T2D patients, maintenance of tight blood glucose control was identified to have a beneficial effect on diabetes-related microvascular complications both in randomised clinical trials and in observational studies (ADVANCE 2008; Hemmingsen 2011; Ohkubo 1995; UKPDS 1998; UKPDS-35 2000). However, among the trials there are inconsistencies with respect to which types of microvascular complications are prevented by intensive glycaemic control. For example, in the UKPDS trial the reduction in microvascular events was primarily due to the observed reduction in retinopathy, whereas in the ADVANCE trial it was due to a reduction in nephropathy, and in the Kumamoto trial it was related to both types of events (Kumamoto 2000).

Some trials investigate the effects of intensive glycaemic control combined with intensive control of other risk factors by using a so-called multimodal approach. These trials have, for example, investigated concomitant allocation to intensive treatments for blood pressure, lipids, and blood glucose in the same treatment arm. Hence, in such trials, it is not possible to estimate the effects of the individual treatment components (see, for example, Steno-2 2008). Other trials applied a so-called factorial design by investigating the effect of targeting several cardiovascular risk factors within each treatment arm in the same trial. With the applied stratification in those trials the influence of each intervention could be estimated (for example, ACCORD 2008). The blood pressure trial randomly assigned participants from the ACCORD trial to targeted intensive blood pressure control (systolic pressure less than 120 mm Hg) versus conventional blood pressure control (systolic pressure less than 140 mm Hg). Intensive blood pressure control did not reduce the risk of the composite macrovascular outcome (non-fatal myocardial infarction, non-fatal stroke, or cardiovascular death) compared with conventional blood pressure control. The lipid arm of the ACCORD trial investigated the effect of simvastatin in combination with fenofibrate treatment compared with simvastatin monotherapy. The conclusion of this combined fenofibrate with simvastatin treatment arm was that the lipid treatment did not influence the risk of cardiovascular events compared with simvastatin alone (ACCORD 2008).

Hence, as primarily suggested by the ACCORD trial, uncertainty remains about the putative beneficial effects of reducing blood glucose compared with the potential risks in T2D patients (ACCORD 2008). In particular, there have been major concerns regarding the extent to which intensive glycaemic control may increase the risk of cardiovascular disease and mortality. Although there seems to be general agreement that intensive glycaemic control reduces the risk of microvascular disease, there are inconsistencies among trials with respect to which types of microvascular disease are reduced. Also, guidelines differ with respect to the recommended optimal glycaemic level for patients with T2D.

Adverse effects of the intervention

The incidence of adverse effects appears to increase when more aggressive metabolic targets for glycosylated haemoglobin A1c (HbA1c) are applied (especially with the addition of insulin) (ACCORD 2008; ADVANCE 2008; UKPDS 1998). A larger number of glucose-lowering drugs, or larger doses of these drugs, are usually required to achieve more intensive glucose targets. Experimental and observational studies have suggested that hyperinsulinaemia, for example as caused by supraphysiologic doses of exogenous insulin, may lead to increased atherosclerosis (Muis 2005). This makes the distinction between the beneficial and harmful effects of the anti-diabetic drugs and of lowering glucose difficult.

The most common adverse reaction to glucose-lowering treatment is hypoglycaemia. The symptoms of mild episodes of hypoglycaemia are often well tolerated by patients, such as hunger, palpitations, tremor, and sweating. Mild hypoglycaemia often precedes severe hypoglycaemia, which can result in more serious symptoms such as confusion, coma, or even death (ADA Workgroup on Hypoglycemia 2005). A publication of the ACCORD trial found a link between symptomatic severe hypoglycaemia and increased risk of death (Bonds 2010); the ADVANCE trial did not find any relationship between repeated episodes of severe hypoglycaemia and death (Zoungas 2010). In addition, a cohort study has suggested an association between a history of severe hypoglycaemia and the risk of dementia among older patients with T2D (Whitmer 2009). Moreover, the different classes of anti-diabetic interventions have specific adverse reactions, for example, gastro-intestinal disturbances with metformin (Saenz 2005); weight gain and hypoglycaemia with sulphonylureas (Hemmingsen 2013); weight gain, oedema, bone fractures, and heart failure with glitazones (Richter 2006; Richter 2007). Weight gain and injection site reactions are among the common adverse effects of insulin (Horvath 2007). There has also been some concern about a potential increased risk of cancer in patients treated with insulin glargine compared with treatment with other types of insulin. Two cohort studies showed an increased risk of cancer-related death and all-cause mortality, whereas two other cohort studies could not find such a relationship (Currie 2009; Hemkens 2009; Jonasson 2009; SDRN 2009). The hypothesis that intensive glycaemic control could increase the risk of cancer compared with conventional glycaemic control in patients with T2D was, however, not supported in a recent meta-analysis (Johnson 2011). The focus of the present review was to primarily investigate the effect of targeting intensive versus conventional glucose control on mortality and vascular outcomes as well as adverse effects known to be causally related to the level of blood glucose control, for example, hypoglycaemia, in patients with T2D. We have not included a specific assessment of cancers. Because of the known complexities of cancer aetiology, type of cancer, etc., we believe the possible relationship between cancer and intensive glucose-lowering control should more adequately be dealt with in a separate review.

Likewise, a recent Cochrane review has assessed the effect of intensive glucose control on neuropathy in patients with diabetes. Therefore, neuropathy is not included as a separate outcome in the present systematic review (Callaghan 2012).

Why it is important to do this review

The dramatic increase in the number of T2D patients places serious demands on healthcare services. Cardiovascular disease is the main cause of the higher mortality in T2D patients. It is therefore relevant to clarify whether intervention regimens that target reduced blood glucose actually improve important patient outcomes such as mortality and cardiovascular disease. A previous meta-analysis, in 2006, suggested that improvement of glycaemic control may reduce macrovascular disease in T2D patients primarily due to a reduction in stroke and peripheral vascular events (Stettler 2006). Since the latter review, large-scale trials have been conducted comparing intensive versus conventional glycaemic control, that is ACCORD, ADVANCE, and VADT (ACCORD 2008; ADVANCE 2008; VADT 2009). Furthermore, several meta-analyses have been performed (for example, Mannucci 2009; Ray 2009; Turnbull 2009). Importantly, most of these meta-analyses included trials based on the achieved (that is, follow-up) rather than the target (that is, randomly allocated) differences in glycaemic control. Thus, trials without predefined differences in the targets of glycaemic control were included. For example, head-to-head anti-diabetic drug comparisons with a similar target for HbA1c levels of below 6.5% in both intervention groups were included, such as the 'PROspective pioglitAzone Clinical Trial In macroVascular Events' (PROactive) of add-on pioglitazone versus placebo (PROactive 2005). This chosen strategy of selection is potentially problematic since, in a clinical trial, the target and the achieved glycaemic levels represent different variables. The achieved glycaemic levels and the clinical outcomes are net results (that is, outcomes) of effects operating at baseline and during follow-up but they do not necessarily influence each other. In contrast, the different glycaemic targets, as part of the randomised treatment regimen, impact on the outcomes, whether as clinical outcomes or as achieved glycaemic levels, by potentially causing different changes to be made during the trial in the glucose-lowering treatments in each group. Thus, trial participants will always have an achieved glycaemic level but they will only have a target level if this has been predefined. This target level may either be similar or different between the groups. In other words, in a clinical trial it is probably not possible to randomise participants to an achieved glycaemic level; for example, in daily life it is unlikely that all participants can be kept to a given blood glucose concentration. For this reason, unlike the allocated therapies being completely separated by design, a complete separation between groups is usually not possible to obtain for the achieved levels of, for example, glucose. That is, unlike the allocated therapies, with respect to achieved levels of glucose some patients in either group will have the same level of intervention resulting in complexities about the relation to other outcomes. Hence, to some extent, achieved glycaemic levels represent observational data and preclude inferences about causality with respect to their influence on other outcomes. In contrast, target levels, as part of the randomised treatment strategy, can support inferences about causality. Therefore, to optimally address the clinical effects of aiming for intensive glycaemic control, which probably is the relevant question to address for the treatment guidelines as well as for the clinicians, it is necessary to meta-analyse trials based upon predefined differences in glycaemic targets.

Further, none of the meta-analyses until now have explored the required information size (the cumulative meta-analysis sample size) to detect or reject specific, clinically relevant intervention effects (Higgins 2010; Turner 2013; Wetterslev 2008; Wetterslev 2009). In 2011 we published our systematic review taking into consideration both the aimed level of glycaemic control as well as the required information size of the meta-analyses. We found no significant differences for all-cause mortality and cardiovascular mortality when targeting intensive glycaemic control compared with conventional glycaemic control. Targeting intensive glycaemic control reduced the risk of microvascular complications while increasing the risk of hypoglycaemia. Furthermore, intensive glycaemic control might reduce the risk of non-fatal myocardial infarction in trials exclusively dealing with glycaemic control in usual care settings.

Therefore, the balance of benefits and harms of tight glycaemic control are still unknown and need to be explored. The present systematic review focused on one of the most important and yet unsolved issues among the clinical questions, namely the clinical effect of targeting intensive glycaemic control per se in T2D patients. In contrast, the effects of achieved glycaemic levels or of specific glycaemic targets were not addressed in this review.

Objectives

To assess the effects of targeted intensive glycaemic control compared with conventional glycaemic control in patients with T2D.

Methods

Criteria for considering studies for this review

Types of studies

All randomised clinical trials of any design comparing targeted intensive glycaemic control with conventional glycaemic control in patients with T2D. Published and unpublished trials in all languages were included.

Types of participants

Adults aged 18 years and above with T2D were included. The diagnosis of T2D should have been established at randomisation into the trial using standard criteria (for example, ADA 1997; ADA 1999; ADA 2003; ADA 2008; ADAa 2010; NDDG 1979; WHO 1980; WHO 1985; WHO 1998; WHO 2011). Ideally, the diagnostic criteria should have been described. If necessary, the authors' definition of T2D was used.

Types of interventions

All included trials should have, prior to patient allocation, predefined in the protocol the different glycaemic targets for intensive and conventional glycaemic control. Trials using HbA1c equivalents (for example, total glycosylated haemoglobin) to compare predefined intensive versus conventional glycaemic treatment were included as well. Furthermore, if no HbA1c (or equivalent) target levels were predefined, trials targeting metabolic control as measured by fasting blood or plasma glucose or postprandial blood or plasma glucose also fulfilled the criteria for inclusion. Trials with a prespecified glycaemic target in the intensive group only were also included. However, as outlined, studies with different target levels for fasting or postprandial blood or plasma glucose but with similar HbA1c (or equivalent) target levels between interventions, or no specified target levels, did not fulfil the criterion for inclusion.

Types of outcome measures

Primary outcomes
  • All-cause mortality

  • Cardiovascular mortality (death from myocardial infarction, stroke, and peripheral vascular disease)

Secondary outcomes
  • Macrovascular complications (non-fatal myocardial infarction, non-fatal ischaemic stroke, non-fatal haemorrhagic stroke, amputation of lower extremity, and cardiac or peripheral revascularization)

  • Microvascular complications (manifestation and progression of nephropathy, end-stage renal disease, manifestation and progression of retinopathy, and retinal photocoagulation)

  • Adverse events (number of patients with any untoward medical occurrence not necessarily having a causal relationship with the treatment). We reported adverse events that led to treatment discontinuation separately. We defined serious adverse events according to the International Conference on Harmonisation (ICH) Guidelines, as any event that led to death, was life-threatening, required in-patient hospitalisation or prolongation of existing hospitalisation, resulted in persistent or significant disability, and any important medical event which may have jeopardised the patient or required intervention to prevent it (ICH 1997). All other adverse events were considered to be non-serious

  • Congestive heart failure

  • Hypoglycaemia, definitions may be heterogeneous between trials. Hypoglycaemia was defined as mild (controlled by patient), moderate (daily activities interrupted but self-managed), or severe (requiring assistance)

  • Health-related quality of life measured with validated instruments

  • Cost(s) of intervention

Macrovascular and microvascular outcomes were both assessed as a composite outcome and as each outcome separately.

Timing of outcome measurement

All outcome measures were assessed independently of the timing of the outcome measurements. The trials were divided according to their intervention periods into short (less than two years) and long (equal or greater than two years) duration.

Covariates, effect modifiers and confounders

Trials assessing multimodal treatment (more than one type of intervention, for example, blood glucose and blood pressure lowering as part of the intervention) together with intensive glycaemic control were included in the analyses. We planned that if the results of the interaction analyses between the interventions, with respect to the clinical outcomes, were not available in the publications from these trials the authors of the trials would be contacted to provide this information. These data would have been taken into account in the interpretation of the results of the meta-analyses. Furthermore, we planned that the presence of any such significant interactions would be subjected to sensitivity analysis (see 'Sensitivity analysis').

Search methods for identification of studies

Electronic searches

The following sources were searched to identify relevant trials:

  • The Cochrane Library (Issue 12, 2012);

  • MEDLINE (December 2012);

  • EMBASE (December 2012);

  • Science Citation Index Expanded (December 2010);

  • Latin American Caribbean Health Sciences Literature (LILACS) (December 2012);

  • Cumulative Index to Nursing & Allied Health Literature (CINAHL) (December 2012).

We intended to search 'The Chinese Biomedical Literature Database', but we did not get any response to our request from the Chinese Cochrane Centre.

For detailed search strategies please see Appendix 1. We used PubMed's 'My NCBI' (National Center for Biotechnology Information) e-mail alert service for identification of newly published studies using a basic search strategy (see Appendix 1).

We searched for ongoing trials and trial protocols for retrieved trials using the following databases:

  • Current Controlled Trials (www controlled-trials.com) (assessed March 2013);

  • ClinicalTrials.gov (www.clinicaltrials.gov) (assessed March 2013);

  • Centre Watch Clinical Trials Listing Service (www.centerwatch.com) (assessed March 2013);

  • World Health Organization (WHO) International Clinical Trials Registry Platform Search Portal (www.who.int/trialsearch) (assessed March 2013).

Searching other resources

In addition, we handsearched abstracts from major diabetes conferences (American Diabetes Association (ADA), European Association for the Study of Diabetes (EASD)). We contacted relevant pharmaceutical companies for unpublished clinical trial data relevant to the review for the first version of this review. However, the pharmaceutical companies either did not respond or responded that they published all trials. Therefore, the companies were not contacted for this update. The US Food and Drug Administration's (FDA) homepage was searched. We tried to identify additional trials by searching the reference lists of included trials and (systematic) reviews, meta-analyses, and health technology assessment reports.

Data extractions of all relevant non-English articles were obtained.

Additional key words that were of relevance were not identified during any of the electronic or other searches. It was not necessary to add additional key words.

Data collection and analysis

Selection of studies

Publications were excluded and full-text articles not retrieved if two of the authors (BH and AV or CG or CH or SL or TA) could determine with certainty from the titles and abstracts identified in the initial search that the trial was: performed in patients with type 1 diabetes mellitus, was not a randomised clinical trial, or did not compare targeted intensive glycaemic control versus targeted conventional glycaemic control. If a publication could not be excluded with certainty on the basis of the title, abstract, or both, the full text of the article was obtained. In cases of differences in opinion, JW or SL were consulted.

Full-text articles were retrieved if the study clearly fulfilled the inclusion criteria: (i) compared targeted intensive glycaemic control with targeted conventional glycaemic control; (ii) included patients with T2D; and (iii) was a randomised clinical trial. Inter-rater agreement for study selection was measured using the kappa statistic (Cohen 1960).

In some cases it was not possible to resolve disagreements without additional information and the authors of the articles were contacted.

A flow diagram of the number of studies identified and excluded at each stage was prepared in accordance with the PRISMA (Preferred reporting items for systematic reviews and meta-analyses) flow chart of study selection (Figure 1) (Liberati 2009).

Figure 1.

Study flow diagram.

Data extraction and management

Two review authors (BH and CH or TA) independently extracted information on each trial using standard data extraction forms. The forms included data concerning trial design, participants, interventions, and outcomes as detailed in the selection criteria described above. For details see: 'Characteristics of included studies'; Table 1; Table 2; Appendix 2; Appendix 3; Appendix 4; Appendix 5; Appendix 6; Appendix 7; Appendix 8; Appendix 9; Appendix 10; Appendix 11; Appendix 12; Appendix 13. Any relevant, missing information was sought from the original author(s) of the article (Appendix 14). Differences between review authors were resolved by discussion and involvement of a third author.

Table 2. Overview of study populations
  1. "-" denotes not reported

    aDuration of intervention and/or follow-up under randomised conditions until end of study
    b"During the first year, ten patients found participation too much of a burden, six moved and one died (7%). One outlier (a woman with a BMI of 59) was excluded from the analyses. Thus, 106 patients in group 6 and 108 patients in group 8 were included in the analyses." It means 231 patients were randomised. It is not possible from the articles to find out which group they were randomised to. Of all randomised patients 43 did not make the visit after 2 years
    cThe number is taken after 13.3 years of follow-up
    d"At the end of the trial, the vital status of 76 (2.0%) patients who had emigrated was not known; 57 and 19 in intensive and conventional groups, respectively, which reflects the 70/30 randomisation. A further 91 (2.4%) patients (65 in the intensive group) could not be contacted in the last year of the study for assessment of clinical endpoints." The n [finishing study] is calculated from the lost to follow-up (mortality) from the UKPDS 33. It is not clear from the UKPDS 34 1998 to clarify how the participants lost to follow-up are distributed. It is reported that 13 participants had unknown vital status, and that the number lost to follow-up for other outcomes is 56. Therefore only data for the UKPDS 33 1998 are used
    eMortality data were assessed on the participants lost to follow-up
    fDeaths occurring after withdrawal from the study were included in the analysis
    gTotals are not the sum of I and C for all columns, because not all trials provided data on the two intervention groups, but only the total

    N/A: not applicable

Characteristic Study IDIntervention(s) and controls[N] Screened/eligible[N] Randomised[N] finishing study (mortality)[N] finishing study (other outcomes)[N] Lost to follow-up mortality[N] Lost to follow-up other outcomes[%] Randomised finishing study (mortality)Follow-upa
1. ACCORD 2008I: targeting intensive glycaemic control

19,716

 

512849194707

110

Long-term follow-up: 209

403

Long-term follow-up: 421

95.9

 

Intervention: 3.5 years

Follow-up: 5 years

C: targeting conventional glycaemic control512349194727

102

Long-term follow-up: 204

372

Long-term follow-up: 396

96.0
total:10,25110,039947621277597.9 
2. ADDITION-Europe 2011I: targeting intensive glycaemic control1312 practices invited167 practices (1678 patients)161 practices (1678 participants)161 practices (1678 participants)0 practices (0 participants)0 practices (0 participants)

96.4% of practices

100% of participants

 
C: targeting conventional glycaemic control176 practices (1379 patients)157 practices (1377 participants)157 practices (1377 participants)0 practices (2 participants)0 practices (2 participants)

89.2 of practices

99.9% of participants

total:3057305530552299.9 

3. ADVANCE 2008

 

I: targeting intensive glycaemic control

12,877

 

557155645326724599.9

 

 

C: targeting conventional glycaemic control5569555952741029599.8
total:11,14011,12310,6001754099.8 
4. Araki 2012I: targeting intensive glycaemic controlNR5855283895719690.3 
C: targeting conventional glycaemic control5885413884720092.0
total:1173   106991.1 
5. Bagg 2001I: targeting intensive glycaemic control

More than 1000 patients

 

2117174481.0

 

 

C: targeting conventional glycaemic control22222200100.0
total:4339394490.7 

6. Becker 2003b

 

I: targeting intensive glycaemic control

296

 

106N/AN/AN/A N/AN/A

 

 

C: targeting conventional glycaemic control108N/AN/AN/A N/AN/A
total:231191188404382.7 
7. Blonde 2009I: targeting intensive glycaemic control472 patients were required for screening according to sample size estimation in publication, but number actually screened is not provided.122 121 107 1 15 99.1 
C: targeting conventional glycaemic control122 122 104 0 18 100.0 
total:24424321113399.6 
8. Cao 2011I: targeting intensive glycaemic controlDicrepancy in the number reported in publication: 210 or 2209292920010028 days
C: targeting conventional glycaemic control87878700100
total:17917917900100 
9. Cooray 2011I: targeting intensive glycaemic controlNR15N/AN/AN/AN/AN/A 
C: targeting conventional glycaemic control13N/AN/AN/AN/AN/A
total:28      

10. DIGAMI 2 2005

 

I: targeting intensive glycaemic control

 

 

47447447400100.0Lost to follow-up in extension period of the study (total follow-up: median: 4.1 years). Intensive group: 43, conventional group: 33
C: targeting conventional glycaemic control30630630600100.0
total: 78078078000100.0 
11. Fantin 2011I: targeting intensive glycaemic control10073535320391.4 
C: targeting conventional glycaemic control3535320391.4 
total:7070640691.4 

12. Guo 2008

 

I: targeting intensive glycaemic control

 

 

16616616600100.0

 

 

C: targeting conventional glycaemic control54545400100.0
total:22022022000100.0 
13. IDA 2009I: targeting intensive glycaemic control

 

 

515151012100.0

 

 

C: targeting conventional glycaemic control51515108100.0
total:102102102020100.0 

14. Jaber 1996

 

 

I: targeting intensive glycaemic control

892

 

2317176673.9

 

 

C: targeting conventional glycaemic control22222200100.0
total:4539396686.7 

15. Kumamoto 2000

 

I: targeting intensive glycaemic control

 

 

5553532296.4

 

 

C: targeting conventional glycaemic control5551514492.7
total:1101041046694.5 

16. Lu 2010

 

 

I: targeting intensive glycaemic control

 

 

21 N/A N/A N/A N/AN/A

 

 

C: targeting conventional glycaemic control20 N/A N/A N/A N/AN/A
total:41 N/A N/A N/A N/AN/A 
17. Melidonis 2000I: targeting intensive glycaemic control

179

 

24242400100.0

 

 

C: targeting conventional glycaemic control24242400100.0
total:48484800100.0 
18. Natarajan 2012I: targeting intensive glycaemic control19153636340294.4 
C: targeting conventional glycaemic control4242340881
total:78786801087.2 

19. REMBO 2008

 

 

I: targeting intensive glycaemic control

 

 

41414100100.0

 

 

C: targeting conventional glycaemic control40404000100.0
total:81818100100.0 
20. Service 1983I: targeting intensive glycaemic control

 

 

10882280.0

 

 

C: targeting conventional glycaemic control10101000100.0
total:2018182290.0 

21. Stefanidis 2003

 

 

I: targeting intensive glycaemic control

239

 

3631315586.1

 

 

C: targeting conventional glycaemic control3935354489.7
total:7566669988.0 
22. Steno-2 2008cI: targeting intensive glycaemic control

315

 

80807901100.0

 

 

C: targeting conventional glycaemic control80807802100.0
total:16016015703100.0 

23. UGDP 1975

 

I: targeting intensive glycaemic control

 

 

204191167133793.6

 

 

C: targeting conventional glycaemic control21020618242898.1
total:414397349176595.9 
24. UKPDS 1998dI: targeting intensive glycaemic control

5102

 

3071301429495712298.1

 

 

C: targeting conventional glycaemic control113811191093194598.3
total:4209413340427616798.2 

25. VA CSDM 1995e

 

 

I: Targeting intensive glycaemic control

289

 

75757104100.0

 

 

C: Targeting conventional glycaemic control78787800100.0
total:15315314904100.0 
26. VADT 2009fI: targeting intensive glycaemic control

2239

 

8928927720120100.0

 

 

C: targeting conventional glycaemic control8998997600139100.0
total:1791179115320259100.0 

27. Yang 2007

 

I: targeting intensive glycaemic control

116

 

57575700100.0

 

 

C: targeting conventional glycaemic control32323200100.0
total:89898900100.0 
28. Zhang 2011I: targeting intensive glycaemic control

 106

 

 48 N/A 48 N/A 0100 
C: targeting conventional glycaemic control 49 N/A 39 N/A 1079.6 
total:97N/A87N/A1089.7 

Grand totalg

 

 

All interventions

 

 

 

18,717 17,451

 

 

 

 

 

 

963 93.2

 

 

 

All controls 16,195 14,986 899 92.5
All interventions and controls 34,912 32,437 1860 92.9

We sent a request via e-mail to study authors of included trials to ask whether they were willing to answer questions regarding their trials. We present the results of this survey in Appendix 14. Furthermore, we sought relevant missing information on the trial from the primary or corresponding author(s) of the article, if required.

Dealing with duplicate publications and companion papers

In the case of duplicate publications and companion papers of a primary study, we tried to maximise the yield of information by simultaneous evaluation of all available data.

Assessment of risk of bias in included studies

Methodological quality was defined as the confidence that the design and the report of the randomised clinical trial restricted bias in the comparison of the interventions (Moher 1998). According to empirical evidence, the methodological quality of the trials was based on sequence generation; allocation concealment; blinding (participants, personnel, and outcome assessors); incomplete outcome data; selective outcome reporting; and other sources of bias (Gluud 2006; Higgins 2011; Kjaergard 2001; Lundh 2012; Moher 1998; Savovic 2012; Schulz 1995; Wood 2008).

Two authors (BH and CH or TA) independently assessed the risk of bias in each trial by means of The Cochrane Collaboration's tool (Higgins 2011). Any differences in opinion were resolved through discussion with JW or CG. The identified trials published in Russian and Chinese were judged for risk of bias by the data extractor, who also evaluated the trials.

The risk of bias components were classified as follows.

Sequence generation
  • Low risk of bias, if the allocation sequence was generated by a computer, a random number table, or similar.

  • Uncertain risk of bias, if the trial was described as randomised but the method used for the allocation sequence generation was not described.

  • High risk of bias, if a system involving dates, names, or admittance number was used for the allocation of patients (quasi-randomised). Such trials were not found but would have been excluded.

Allocation concealment
  • Low risk of bias, if the allocation of participants involved a central independent unit; on-site locked computer; or consecutively numbered, sealed envelopes.

  • Uncertain risk of bias, if the trial was described as randomised but the method used to conceal the allocation was not described.

  • High risk of bias, if the allocation sequence was known to the investigators who assigned participants or if the study was quasi-randomised. Such trials were not found but would have been excluded.

Blinding

It was not possible to blind the healthcare provider and the patients in the treatment groups. Assessment of blinding was therefore based on blinding of outcome assessors and was done separately for the objective and subjective outcomes of our review (see below). For the subgroup analyses of trials according to blinding, these were done based on the risk level of the trial judged by the following.

  • Low risk of bias, if the outcome assessors were blinded and the method of blinding was described.

  • Uncertain risk of bias, if the outcome assessors were blinded and the method of blinding was not described.

  • High risk of bias, if the outcome assessors were not blinded.

Incomplete data outcomes

Assessment of incomplete outcome data was done separately for the objective and subjective outcomes of our review (see below).

  • Low risk of bias, if any post-randomisation dropouts or withdrawals, if they occurred, were clearly described and the reasons for these dropouts were described.

  • Uncertain risk of bias, if it was not clear whether there were any dropouts or withdrawals or if the reasons for these dropouts were not clear.

  • High risk of bias, if the reasons for missing data were likely to be related to the outcomes: (1) 'as-treated' analysis was performed; (2) potentially inappropriate application of simple imputation; (3) potential for patients with missing outcomes to induce clinically relevant bias in effect estimate or effect size.

Selective outcome reporting

We assessed outcome reporting bias at trial level (Higgins 2011; Kirkham 2010) (Appendix 9; Appendix 10; Appendix 11).

  • Low risk of bias, if all the predefined (primary and secondary) outcomes mentioned in the trial's protocol or in the design article were reported and the reporting had been done in the prespecified way.

  • Uncertain risk of bias, if there was insufficient information to assess whether a risk of selective outcome reporting was present.

  • High risk of bias, if not all the prespecified outcomes were reported, if the primary outcomes were changed, or if some of the important outcomes were incompletely reported.

Sponsor bias
  • Low risk of bias, if the trial was unfunded or was not funded by an instrument, equipment, or drug manufacturer.

  • Uncertain risk of bias, if the source of funding was not clear.

  • High risk of bias, if the trial was funded by an instrument, equipment, or drug manufacturer.

Academic bias
  • Low risk of bias, if the author of the trial had not conducted previous trials addressing the same interventions.

  • Uncertain risk of bias, if it was not clear if the author had conducted previous trials addressing the same interventions.

  • High risk of bias, if the author of the trial had conducted previous trials addressing the same interventions.

Besides investigating each bias domain, we also evaluated the overall risk of bias. Too few trials had low risk of bias according to all domains. Accordingly, when sequence generation and allocation concealment were judged to be low risk of bias, the trial was classified as a trial with lower risk of bias.

We planned to explore the influence of the individual risk of bias criteria in subgroup analyses.

For blinding of outcome assessors (detection bias) and attrition bias (incomplete outcome data) we intended to evaluate risk of bias separately for subjective and objective outcomes (Savovic 2012).

We defined the following outcomes as objective.

  • All-cause mortality.

  • Cardiovascular mortality (death from myocardial infarction, stroke, and peripheral vascular disease).

  • Macrovascular complications (non-fatal myocardial infarction, non-fatal ischaemic stroke, non-fatal haemorraghic stroke, amputation of lower extremity, and cardiac or peripheral revascularization).

  • Microvascular complications (manifestation and progression of nephropathy, end-stage renal disease, manifestation and progression of retinopathy, and retinal photocoagulation).

  • Congestive heart failure.

  • Severe hypoglycaemia.

  • Cost(s) of intervention.

In many circumstances the assessment of our objective outcomes might be considered to be unbiased even if outcome assessors were aware of intervention assignments.

The remaining outcomes were classified as subjective.

  • Health-related quality of life.

  • Moderate and mild hypoglycaemia.

  • Adverse events.

Measures of treatment effect

Dichotomous data

Data on dichotomous outcomes were statistically summarised as relative risks (RR), equal to the ratio of absolute risks, the risk difference (RD), and equal to the difference in absolute risks. We produced 95% confidence intervals (CI) for each of these. The number needed to treat (NNT) was also calculated.

Continuous data

Continuous outcomes would have been summarised as difference in means (MD) with 95% CI if outcome data were provided on the same scale. For trials addressing the same outcome but using different outcome measures (for example, different scales measuring quality of life) standardised mean differences (SMD) were used.

Time-to-event data

Most trials recruited their participants over a defined recruitment period and they were followed up until a fixed date, beyond the end of recruitment. Therefore, the last recruited participants will be observed for a shorter period than those recruited first and will therefore be less likely to experience an event. Time-to-event outcomes (for example, time until death) were planned to be expressed as hazard ratios (HR) with 95% CI. The natural logarithm (ln) of the HR and its standard error (SE) were calculated. We preferred the unadjusted HR, when available, because adjusted HRs may represent very different ways of adjustment.

Dealing with missing data

Missing data were sought by contacting the trial authors. The impact of any missing data was discussed.

Intention-to-treat analysis is recommended in order to minimise bias in the analysis of the efficacy of randomised clinical trials. It pragmatically estimates the benefit of a change in treatment policy rather than the potential benefit for patients who receive the treatment exactly as planned (Hollis 1999). Full application of intention to treat is possible when complete outcome data are available for all randomised participants. Despite the fact that about half of all published reports of randomised clinical trials state that intention-to-treat analysis is used, handling of deviations from randomised allocation varies widely and many trials have missing data for the primary outcome variable (Hollis 1999). The methods used to deal with deviations from randomised allocation are generally inadequate, potentially leading to bias (Hollis 1999).

Performing an intention-to-treat analysis in a systematic review is not straightforward in practice since review authors must decide how to handle missing outcome data in the contributing trials (Gamble 2005). No consensus exists about how missing data should be handled in intention-to-treat analysis, and different approaches may be appropriate in different situations (Higgins 2011; Hollis 1999).

In the case of missing data, we applied 'complete-case analysis' for primary outcomes, which simply excludes all participants with the missing outcome from the analysis, as well as 'worst-best' and 'best-worst' case scenario analyses. The worst-best case scenario analyses assume that participants with unknown vital or event status receiving intensive glycaemic control were dead or had an event and all participants with unknown vital or event status receiving conventional intervention were alive or did not have an event. The best-worst case scenario analyses assume that participants with unknown vital or event status receiving intensive glycaemic control were alive or did not have an event and that all participants receiving the conventional intervention with unknown vital status were dead or had an event. We applied 'complete-case analysis' and 'worst-best' and 'best-worst' scenario analyses for the primary outcomes and for non-fatal myocardial infarction.

Dealing with duplicate publications

When more than one publication of an original trial was identified, we assessed those articles together to maximise data collection. In the case of substantial disagreements between older and newer articles the authors were contacted.

Assessment of heterogeneity

A priori, the authors evaluated the clinical diversity of the included trials. Heterogeneity was identified by visual inspection of the forest plots and by using a standard Chi2 test, with a significance level of α = 0.1. Heterogeneity was specifically examined with the I2 statistic. Values of I2 between 0% to 40% were graded as: heterogeneity might not be important. An I2 statistic between 30% to 60% was graded as representing moderate heterogeneity, I2 between 50% to 90% was graded as substantial heterogeneity, and I2 between 75% to 100% was graded as considerable heterogeneity (Higgins 2011). When heterogeneity was found, we attempted to determine potential reasons for it by examining individual trial characteristics and the subgroup characteristics for the main body of evidence.

Clinical heterogeneity was assessed by comparing the trials with regard to different clinical variables: patient characteristics, duration of disease, glycaemic target, other targeted metabolic variables, and outcome. When significant clinical, methodological, or statistical heterogeneity was found, we surveyed the individual trials to determine potential reasons for it.

We used both a random-effects model (DerSimonian 1986) and a fixed-effect model (DeMets 1987). In the case of discrepancy between the two models, we reported both results. We originally planned to report only the fixed-effect model, however, due to substantial heterogeneity among the included trials we decided to report primarily the random-effects model.

Between-trial heterogeneity was explored by meta-regression, depending on the available data. Therefore meta-regression was performed to explore a possible association between the intervention effects estimated in the trials and the following covariates that were selected in the protocol: average fasting blood glucose level at baseline, average HbA1c at baseline, average duration of diabetes at baseline, and duration of intervention. Meta-regression was performed using the software Comprehensive Meta Analysis (CMA 2005). The statistical method for the meta-regressions was a random-effects model meta-regression analysis based on unrestricted maximum likelihood. All log risk ratios were based on Mantel-Haenszel analyses.

Assessment of reporting biases

Funnel plots were drawn to provide visual assessment as to whether treatment effects were associated with trial size. There are a number of reasons for the asymmetry of a funnel plot, for example, methodological design of trials and publication bias (Higgins 2011).

Data synthesis

The median reported in the included trials was assumed to approximate to the arithmetic mean. Data were summarised statistically if they were: available, of sufficient quality, and were sufficiently similar (clinical homogeneity). Statistical analyses were performed according to the statistical guidelines referenced in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). Cluster-randomised clinical trials were handled in accordance with the Cochrane Handbook for Systematic Reviews of Interventions. When no cluster correlation coefficient or design effect was reported, the authors of the cluster-randomised trials were asked for the information. If no cluster correlation coefficient or design effect could be provided, estimates from similar trials were searched for.

Trial sequential analysis

Trial sequential analysis combines a calculation of the required information size (cumulated meta-analysis sample size to detect or reject a specific relative intervention effect) for meta-analysis with the threshold of statistical significance. It is a tool for quantifying the statistical reliability of data in a cumulative meta-analysis, adjusting significant values and confidence intervals for sparse data and repetitive testing on accumulating data. Trial sequential analysis was conducted on the primary and the secondary outcomes according to prespecified criteria in our protocol (see below) (Brok 2008; Brok 2009; Pogue 1997; Pogue 1998; Thorlund 2009; Wetterslev 2008; Wetterslev 2012). Meta-analysis may result in type one errors due to systematic errors (bias) or random errors due to repeated or early significance testing when updating a meta-analysis with new trials (Brok 2008; Brok 2009; Higgins 2010; Wetterslev 2008).

In a single trial, interim analysis increases the risk of type one errors. To avoid type one errors, group sequential monitoring boundaries are applied to decide whether a trial could be terminated early because of a sufficiently small P value, that is the cumulative Z-curve crosses the monitoring boundaries (Lan 1983). Sequential monitoring boundaries can be applied to meta-analysis as well, called trial sequential monitoring boundaries (Higgins 2010; Wetterslev 2008). In trial sequential analysis, the addition of each trial in a cumulative meta-analysis is regarded as an interim meta-analysis and helps to clarify whether additional trials are needed (Wetterslev 2008).

The idea in trial sequential analysis is that if the cumulative Z-curve crosses the boundary, a sufficient level of evidence for the anticipated intervention effect has been reached and no further trials may be needed. If the Z-curve does not cross the boundary, then there is insufficient evidence to reach a conclusion. To construct the trial sequential monitoring boundaries, the required information size is needed and is calculated as the least number of participants needed in a well-powered single trial and subsequently adjusted for heterogeneity among the included trials in the meta-analysis (Brok 2008; Brok 2009; Pogue 1997; Pogue 1998; Wetterslev 2008; Wetterslev 2009). We applied trial sequential analysis since it decreases the risk of type one error due to sparse data and potential multiple updating in a cumulative meta-analysis, and it provides us with important information in order to estimate the level of evidence of the experimental intervention. Additionally, trial sequential analysis provides us with important information regarding the need for additional trials and the required information size.

We applied trial sequential monitoring boundaries according to an information size based on an a priori effect corresponding to a number needed to treat (NNT) or harm (NNH) of 50 to 100 patients. This included a 10% relative risk reduction (RRR) for benefit and a 30% relative risk increase for harm using an overall type one error level of 5% (α = 0.05) and a type two error level of 20% (ß = 0.20 or power = 80%). We used the control event proportion estimated from the control groups of the included trials. For the heterogeneity adjustment of the required information size we used the diversity (D2) adjustment based on a D2 calculated from the included trials in the meta-analysis as I2 underestimates the heterogeneity among trials in a meta-analysis with regard to the required information size (Wetterslev 2009). Diversity is an estimate of the heterogeneity of the trials included in the meta-analysis, and represents the relative variance reduction when the meta-analysis model is changed from a random-effects model to a fixed-effect model (Wetterslev 2009).

We conducted trial sequential analysis on the primary outcomes. Moreover, it was applied to all secondary outcomes that showed significant effect estimates in both the random-effects and fixed-effect models.

Subgroup analysis and investigation of heterogeneity

Subgroup analyses were planned if one of the primary outcome measures demonstrated statistically significant differences between the intervention groups. In any other case, subgroup analyses were planned as a hypothesis generating exercise. The following subgroup analyses were planned:

  • anti-diabetic intervention used to achieve glycaemic target (drug classes compared to each other, monotherapy compared to combination therapy);

  • glycosylated haemoglobin A1c (HbA1c) target level less than 7.0% compared to HbA1c equal to or greater than 7.0%;

  • defined target in terms of HbA1c compared to non-HbA1c target;

  • cardiovascular disease at baseline compared to no cardiovascular disease at baseline;

  • peripheral revascularization and retinal photocoagulation (because the interventions depend on the local clinical practice);

  • age less than 65 years compared to age equal to or greater than 65 years.

All outcomes were analysed in the subgroups according to the type of intervention applied: trials exclusively dealing with glycaemic control in the usual care setting, glycaemic control as part of an acute intervention, glycaemic control initiated with surgical intervention, and multimodal intervention in a usual care setting. Trials exclusively dealing with glycaemic control in usual care were defined as those trials with random allocation to targeting intensive versus conventional glycaemic control without parallel (non-factorial) allocation to concomitant control of other risk factors than blood glucose, such as blood pressure or lipids. Factorial allocation to other regimens than glucose-lowering treatment, such as blood pressure or lipid-lowering treatment, was allowed in this group. Acute intervention should not be part of the treatment protocol. Multimodal intervention in usual care settings was defined as those trials with parallel (non-factorial) random allocation to concomitant control of other risk factors than blood glucose, such as blood pressure or lipids, where acute intervention should not be part of the protocol. Glycaemic control initiated with surgical intervention was defined as those trials in which all participants had a surgical procedure as one of the inclusion criteria. Acute intervention was defined as those trials where intensive versus conventional glycaemic control was initiated as part of an acute intervention during hospital admission for other reasons than control of diabetes, for example in participants with acute myocardial infarction. There was no requirement for the duration of the intervention in the acute intervention group, meaning that longer-term trials with follow-up over several years could be included. These four subgroups, according to the type of intervention, were mutually exclusive.

The following subgroup analyses were performed for the primary outcomes and non-fatal myocardial infarction.

  • Comparing trials with lower risk of bias regarding sequence generation and allocation concealment to trials with high or unclear risk of bias regarding sequence generation and allocation concealment.

  • Comparing trials with long study duration (> two years) to the trials with short study duration (≤ two years).

  • Comparing the trials using the filters: diagnostic criteria, language of publication, source of funding (industry versus other).

Tests of interaction were used to determine the effect of a subgroup on the intervention effect.

Heterogeneity examined by meta-regression

Meta-regression was conducted for the following covariates:

  • average duration of diabetes at baseline;

  • average fasting blood glucose at baseline;

  • average HbA1c at baseline;

  • average duration of the intervention.

Sensitivity analysis

We performed sensitivity analyses in order to explore the influence of the following factors on effect size:

  • repeating the analysis excluding the trials with longest duration or the largest trial to establish how much they influenced the results;

  • repeating the analysis including trials with zero events in the treatment groups with the trial sequential analysis program, applying an empirical continuity correction of 0.01 for zero events (Sweeting 2004);

  • repeating the analysis excluding only those trials assessing multimodal treatment with documented statistical interactions between the interventions on the clinical outcomes;

  • repeating the analysis excluding trials assessing acute effects of glycaemic control (less than 48 hours);

  • repeating the analysis excluding unpublished trials.

The sensitivity analysis including trials with zero events in the treatment groups with the trial sequential analysis program applying an empirical continuity correction was performed (Sweeting 2004).

The robustness of the results was tested by repeating the analysis using different measures of effects size (relative risk, odds ratio, etc.) and different statistical models (fixed-effect model and random-effects model) (Keus 2009).

Results

Description of studies

Results of the search

The initial search identified 12,201 records, of which 164 full papers were examined further. The other studies were excluded on the basis of their titles and abstracts because they clearly did not meet the inclusion criteria (Figure 1). After screening the full texts of the selected papers, 28 randomised trials described in 89 publications met our inclusion criteria. Twenty-six trials were completely or partly published in English, one in Russian (REMBO 2008), and two in Chinese (Yang 2007).

Abstracts from the American Diabetes Association (ADA) and European Association for the Study of Diabetes (EASD) conferences did not provide information on additional trials. A search of the US Food and Drug Administration (FDA) homepage did not reveal any additional trials. One relevant health technology assessment report was found (AHRQ 2007). Eighteen meta-analyses comparing intensive glycaemic control versus conventional glycaemic control in T2D patients were also retrieved (Boussageon 2011; Callaghan 2012; Castagno 2011; Coca 2012; Johnson 2011; Kelly 2009; Ma 2009; Mannucci 2009; Marso 2010; Ray 2009; Selvin 2004; Slinin 2012; Stettler 2006; Tkac 2009; Turnbull 2009; Wang 2009; Wu 2010; Zhang 2010). Neither the health technology assessment report nor the meta-analyses provided references to any additional trials. All authors of the included trials were sent a reference list and a request for information on additional trials, if possible. One publication was provided by an author (Genell Knatterud). Screening references of the University Group Diabetes Program (UGDP) provided the design article for this trial, which was not retrieved from the search (UGDP 1975). One trial was identified through screening for references for another systematic review (Blonde 2009). Through Internet searches for additional information on the included trials, the ADVANCE trial provided information on a web page from which information about the substudies was obtained. Trial protocols were retrieved for seven trials through the publication or the search of trial registers (ACCORD 2008; ADDITION-Europe 2011; ADVANCE 2008; Araki 2012; Blonde 2009; Fantin 2011; Natarajan 2012). One trial protocol was provided by the investigators (DIGAMI 2 2005).

Inter-rater agreement between the two trial selectors was 82% using the kappa statistic.

Searching trial registers for ongoing trials showed 10 trials with potential relevance (ADVANCE-ON; Chen 2009; DARE; GLUCOSURG1; HFDM; LIMBISCH; REMIT Pilot Trial; RESET-IT; VADT-FS 2008; Xiang 2010). The trials will be included when updating the review.

Missing data

We contacted all corresponding authors of the included trials for further details. Extraction schemes were sent to all the authors so that they could provide additional data or comment on the retrieved data (Appendix 14). Our request might not have reached all authors because of changes of contact information since the publication of the trials. Internet searches through PubMed or Google, or both, were made on these authors in order to find updated contact information. If no Internet address could be obtained, the authors were contacted through letters. Additional information about the UK Prospective Diabetes Study (UKPDS) (UKPDS 1998) was obtained from other meta-analyses (Kelly 2009; Ray 2009; Turnbull 2009).

Dealing with duplicate publications

Several of the included trials consisted of more than one publication. In one of the included trials a discrepancy was observed between two publications describing the same participants (Becker 2003). Unfortunately, we were unable to obtain contact information on the first authors of the duplicate articles. The article in the Netherlands Journal of Medicine 2003 described the details of the trial population reported on in Diabetes Care 1998. We corresponded with two of the other authors who unfortunately were not able to clarify the discrepancy between the publications (see 'Characteristics of included studies').

The UKPDS trial consists of several publications and we were uncertain about the overlap between the conventional treatment groups in UKPDS 33 and UKPDS 34 (UKPDS 1998). An author of both articles (Rury Holman) confirmed a complete overlap between the participants in the conventional treatment groups in UKPDS 33 and UKPDS 34. The intensively treated group receiving metformin in the UKPDS 34 was not a part of the intensively treated group in the UKPDS 33. Thus, where possible, all intensively treated patients from UKPDS were included whether allocated to insulin, sulphonylurea, or metformin; as were the conventional group from UKPDS 33. Data on the composite macrovascular outcome and hazard ratios for all-cause mortality and cardiovascular mortality in the UKPDS were obtained from the meta-analysis by Turnbull et al in which follow-up was truncated to five years and only data from UKPDS 33 were reported (Turnbull 2009). It was only possible to retrieve the reported number of retinopathies and nephropathies from UKPDS 33 and not from UKPDS 34. All other outcomes from UKPDS 33 and UKPDS 34 were reported after 10 years of follow-up.

Included studies

We included data from 28 trials. All were randomised clinical trials assessing the effects of intensive glycaemic control versus conventional glycaemic control in patients with T2D. A total of 34,912 participants were included, of which 18,717 were randomised to intensive glycaemic control and 16,195 were randomised to conventional glycaemic control (Table 2). For full details please see the table 'Characteristics of included studies'.

Trial designs

All 28 included trials were randomised clinical trials, of which four had a factorial design (ACCORD 2008; ADVANCE 2008; UKPDS 1998; Zhang 2011). None of the included trials had a cross-over design. The ACCORD and UKPDS used a partly factorial design since only a proportion of the patients were also randomised to other treatment arms besides the glucose control arm: blood pressure control, ACCORD (46% of participants) and UKPDS (27% of participants); lipid-lowering, ACCORD (54% of participants) and acarbose versus placebo in the UKPDS (59% of the participants) (ACCORD 2008; UKPDS 1998). In contrast, the ADVANCE trial and Zhang et al randomised all patients to the two treatment arms. The ADVANCE trial randomised patients to a glucose control arm as well as a blood pressure control arm (ADVANCE 2008). Zhang et al randomised patients to a glucose control arm as well as to perindopril or placebo (Zhang 2011). The tests for interaction between the allocation in the glucose trial and in the blood pressure or lipid trials on the primary outcome (composite of non-fatal stroke, non-fatal myocardial infarction, or cardiovascular death) in the ACCORD trial did not reach statistical significance (P = 0.08 for blood pressure, and P = 0.36 for lipids) (ACCORD 2008). However, there was evidence of an interaction between the intensive glucose-lowering group and the intensive blood pressure–lowering group with respect to all-cause mortality both before the transition (that is, after 3.7 years, which was the time when patients in the intensive glucose arm were switched to conventional glucose therapy because of increased mortality with the intensive glucose therapy) (P = 0.03 for interaction) and after 5 years at the end of the trial (P = 0.05 for interaction). This interaction was characterised by a marginally higher mortality in the intensive glucose-lowering group than in the standard glucose-lowering group among participants also assigned to the intensive blood pressure–lowering group (ACCORD 2008). That is, the risk of death associated with the intensive or conventional glucose-lowering intervention appeared to depend on the allocated intensity of blood pressure-lowering therapy, and vice versa (ACCORD 2008). In the ADVANCE trial, no significant interaction was observed on the primary outcomes of major microvascular and macrovascular events (individually and together) or mortality with allocation to the glucose and the blood pressure trials (P > 0.10) (ADVANCE 2008). In the UKPDS trial, there was no significant interaction for the change in HbA1c between allocation in the glucose and acarbose part of the trial (P = 0.43). Otherwise, interaction tests for the allocated interventions, in particular to intensive versus conventional glucose and blood pressure control, did not appear to have been reported from UKPDS (UKPDS 1998). Instead, the UKPDS trial reported trend tests for the relationship between the allocated glucose and blood pressure groups. These trend tests indicated additive effects between the allocated glucose and blood pressure groups for any diabetes-related outcome (P = 0.024), a tendency for diabetes-related death (P = 0.058), but not for all-cause mortality (P = 0.26). The nominal lowest risk for diabetes-related mortality or all-cause mortality was observed in the group allocated to intensive therapy of both glucose and blood pressure. Trend tests, however, do not allow for full assessment of interaction. In addition, the UKPDS investigated the achieved glucose and blood pressure levels across allocated interventions (that is, not comparing specific intervention groups) and showed a significant interaction for microvascular disease arising from less than additive effects of the two (P < 0.0001). That is, this indicated relatively less harmful effects of blood glucose and blood pressure at high than at low levels of the other factor (despite an about 16-fold higher risk with high than low levels of both factors). In contrast, there was no statistically significant interaction for diabetes-related death (P = 0.11) or all-cause mortality (P = 0.74) (despite an about 7- and 4-fold higher risk, respectively, with high versus low levels of both factors) (UKPDS 1998). Thus, in the ACCORD and ADVANCE trials the effects on the primary outcomes of either of the randomised interventions should have been independent of each other, whereas an interaction for mortality between glucose and blood pressure lowering was suggested by ACCORD (ACCORD 2008). There were insufficient statistical data to guide a conclusion of interactions between the glucose and blood pressure arms from the UKPDS. There were, however, no strong signals for harm in terms of mortality with combined intensive glucose and blood pressure control from that study (UKPDS 1998).

Besides the target of HbA1c below 6.5% in the ADVANCE trial and 7.0% in the REMBO (Rational Effective Multicomponent Therapy in the Struggle Against DiaBetes Mellitus in Patients With COngestve Heart Failure) trial, all participants in the intensively treated group received modified release gliclazide (ADVANCE 2008; REMBO 2008). In the UKPDS, besides the target of fasting plasma glucose below 6.0 mmol/L all intensively treated patients received dietary advice and they were randomly allocated to receive either insulin, sulphonylurea, or metformin, whereas all conventionally treated patients besides the target of fasting plasma glucose below 15 mmol/L and who were without symptoms of hyperglycaemia, received dietary advice only (UKPDS 1998). In five trials, the participants were randomised into intensive multimodal treatment of various risk factors, including differences in glycaemic treatment target values (ADDITION-Europe 2011; Araki 2012; Guo 2008; Steno-2 2008; Yang 2007). Two further trials had more than two intervention groups. We only extracted data from two intervention groups in these trials (DIGAMI 2 2005; UGDP 1975). In both trials we extracted the data from the most intensive treatment group and from the conventional treatment group. The 'Diabetes and Insulin-Glucose Infusion in Acute Myocardial Infarction' (DIGAMI) 2 trial had three intervention groups. We have used the data for group 1 (intensive insulin infusion at admission followed by a subcutaneous insulin-based long-term glucose control) as targeting intensive glycaemic control and group 3 (glycaemic control according to local guidelines) as targeting conventional glycaemic control. All other concomitant interventions were identical for the groups (DIGAMI 2 2005). The 'University Group Diabetes Program' (UGDP) randomised the participants to five different therapeutic regimens: insulin variable, insulin standard, tolbutamide, phenformin, or placebo. We chose to report the insulin variable (IVAR) group as the intensive group and the insulin standard (ISTD) group as the conventional group, because IVAR was the only group with a predefined glucose target and ISTD, as the only other group, was also allocated to insulin treatment (UGDP 1975).

The included trials were mainly conducted in Northern America and Europe. The number of study centres ranged from 1 to 215. Four trials were multinational (ACCORD 2008; ADDITION-Europe 2011; ADVANCE 2008; DIGAMI 2 2005).

The mean duration of the intervention period varied from three days (Stefanidis 2003) to 12.5 years (UGDP 1975). Four of the included trials reported a longer follow-up period than the intervention period (ACCORD 2008; DIGAMI 2 2005; Steno-2 2008; UKPDS 1998). For the Steno-2 trial, we have reported the outcomes for the longest follow-up time because the follow-up was complete for all participants (Steno-2 2008). We did not include the results of the longest follow-up for the UKPDS because a relatively small proportion of the randomised participants were included in the follow-up analyses (UKPDS-80 2008). The mortality and the microvascular complications reported from the ACCORD trial were from the longest follow-up, that is 1.5 years after termination of the glucose arm of the trial (ACCORD 2008). Even though the intensive intervention stopped after 3.7 years, the participants continued to be followed at least every four months until the originally planned end of trial, as the participants were followed in the lipid and blood pressure arms. Thus data on clinical outcomes continued to be collected for an additional 17 months (ACCORD 2008). For the DIGAMI 2 trial mortality data were from the extension period (median follow-up 4.1 years) (DIGAMI 2 2005). The Kumamoto trial had planned an intervention duration of six years (Kumamoto 2000). Only two of the included 110 participants changed their glycaemic intervention regimen after the predefined intervention period and the trial continued through the initiative of the participants. Because only two participants changed therapy, we reported all data except for mild hypoglycaemia (data not available) after 10 years of follow-up (Kumamoto 2000).

The 'Anglo-Danish-Dutch study of Intensive Treatment In PeOple with screeN detected diabetes in primary care' (ADDITION) trial was a cluster-randomised trial where the included practices were randomised to intensive versus conventional glycaemic control. The participants, but not the practices, were unaware of the assignment of the practice and, therefore, the ADDITION is the only included trial designed to blind participants (but not the investigators) (ADDITION-Europe 2011). The ADDITION-Europe reported the intracluster correlation coefficient for the primary outcome (estimated to be 0.002). The design effect was calculated to be 1.02 to correct for the clustering of the primary outcome, which was a composite outcome of macrovascular complications (Appendix 10). We could not find any information on the design effect on each component of the composite microvascular outcome, so we applied the design effect observed on the composite macrovascular outcome on each of the components. As not all data from the ADDITION-Europe are published yet, we included data in the analyses from two of the substudies embedded in ADDITION-Europe (ADDITION-Leicester; ADDITION-Netherlands). The two substudies were listed as included studies in order to be able to include them in the analyses, but when we stated that we included 28 trials and 34,912 participants we have only included the ADDITION-Europe (n = 3057 participants).

Trial participants

Eight trials did not describe how the T2D diagnosis was established (ADVANCE 2008; Araki 2012; Blonde 2009; Cooray 2011; Jaber 1996; Lu 2010; Natarajan 2012; REMBO 2008). In the UGDP trial, T2D diagnosis was based on the sum of four glucose values from a glucose tolerance test. As a result of this definition, participants with impaired glucose tolerance (according to later definitions) were included in the trial (UGDP 1975). According to the diagnostic criteria of T2D established in 1989 by the ADA and the World Health Organization (WHO), three participants included in our analysis in each of the intervention groups would have been diagnosed as having normal glucose levels. Moreover, 31 participants in the conventional treatment and 28 participants in the intensive treatment group fulfilled the criteria for impaired glucose tolerance. The main criterion for the T2D diagnosis in the UKPDS was based on the mean of two fasting plasma glucose values > 6 mmol/L (UKPDS 1998). This definition of T2D was different from the WHO criterion of fasting blood glucose of > 7.8 mmol/L (WHO 1985). Thus, all participants in the UKPDS trial did not necessarily have diabetes according to the WHO criteria (WHO 1985). All participants in the UGDP and UKPDS had a dietary run-in period of four weeks and three months, respectively. In the UGDP, participants who developed symptomatic hyperglycaemia during run-in were excluded. In the UKPDS, the participants with fasting blood glucose of 6.1 to 15.0 mmol/L after three months on a diet were randomised to UKPDS 33 and UKPDS 34. In the VADT trial, 127 participants failed to reach the diagnostic C-peptide level (VADT 2009).

The mean age of the participants of the included trials varied from 49.1 years to 71.8 years (Araki 2012; Guo 2008).

The duration of T2D at entry into the trials ranged between newly diagnosed to a mean disease duration of 17.2 years (Araki 2012). Established T2D diagnosis within one year before entry into the trial was an inclusion criterion in five trials (ADDITION-Europe 2011; Guo 2008; UGDP 1975; UKPDS 1998; Yang 2007). The risk profile among the trial participants with respect to cardiovascular disease was very different at entry in the included trials. Six trials had as an inclusion criterion cardiovascular disease (DIGAMI 2 2005; Fantin 2011; IDA 2009; Melidonis 2000; Natarajan 2012; REMBO 2008; Stefanidis 2003). In the REMBO trial all participants had congestive heart failure (REMBO 2008). Four trials had as a part of the inclusion criteria high risk of cardiovascular disease (besides T2D) (ACCORD 2008; ADVANCE 2008; Steno-2 2008; Zhang 2011). The Steno-2 and Lu et al had microalbuminuria as an inclusion criterion (Lu 2010; Steno-2 2008). The Kumamoto trial stratified the participants into two groups: a primary prevention population and a secondary intervention population. All participants in the primary prevention population had no microvascular disease at baseline whereas all in the secondary intervention population had either microalbuminuria or retinopathy (Kumamoto 2000).

Most exclusion criteria consisted of liver disease, kidney disease, or other severe concurrent illness.

Characteristics of interventions

The anti-diabetic interventions used in the trials often included add-on regimens consisting of several oral anti-diabetic interventions. If these regimens could not reach the glycaemic target, then insulin was initiated. The usual add-up regimen was identical in the intensive and conventional intervention groups of the trials, except in three trials where participants targeting intensive glucose control were given gliclazide (ADVANCE 2008; REMBO 2008), other sulphonylurea (glibenclamide, chlorpropamide, or glipizide), metformin, or insulin (UKPDS 1998). Gliclazide was discontinued in the participants randomised to the conventional glycaemic target in the ADVANCE trial (ADVANCE 2008). In the trial from Blonde et al, the participants in both the intensive and the conventional intervention group received insulin, both in combination with the per oral anti-diabetic the patients took at baseline. The combination of oral anti-diabetic interventions and insulin was allowed in most trials (see 'Interventions in trials', Appendix 2). Two trials only allowed insulin monotherapy in both the intensive intervention group and the conventional intervention group (Kumamoto 2000; UGDP 1975). One trial allowed combination therapy in the conventional group but only insulin in the intensive treatment group (Melidonis 2000). One trial allowed combination therapy in the intensive intervention group but not in the conventional intervention group (VA CSDM 1995). One trial did not specify in detail what the next treatment step would be in the intensive treatment group if the maximum dose of sulphonylurea could not attain the glycaemic target (Jaber 1996). Trials which had acute cardiovascular disease as an inclusion criterion had a treatment algorithm for insulin infusion for the intensive intervention group, starting at hospital admission (DIGAMI 2 2005; Melidonis 2000; Stefanidis 2003). All the participants with hospital admission at entry and targeting intensive glycaemic control had their blood glucose initially lowered with insulin (DIGAMI 2 2005; Melidonis 2000; Stefanidis 2003). Four trials had as an inclusion criterion that the participants were referred to operation (Cao 2011; Fantin 2011; IDA 2009; Natarajan 2012). One trial had an intervention period of 24 hours in which only monotherapy with insulin was applied in both intervention groups, but thereafter combination therapy was allowed (Fantin 2011). As shown, most trials did not report on the dose of insulin used in the two intervention groups (Appendix 2).

The median dose of insulin used in the intensive intervention group was 0.7 (range 0.5 to 1.0) units of insulin/day/kg body weight (ADVANCE 2008; Bagg 2001; Fantin 2011; Kumamoto 2000; Steno-2 2008; VA CSDM 1995; VADT 2009). The median dose of insulin in the conventional intervention group was 0.5 (range 0.4 to 0.8) units of insulin/day/kg body weight (see 'Interventions in trials', Appendix 2).

The treatment targets for glycaemic control varied between trials in both the intensive treatment groups and the conventional treatment groups. The ACCORD and VADT had the lowest HbA1c target level in the intensive intervention groups (both less than 6%) (ACCORD 2008; VADT 2009). Some of the trials did not predefine the glycaemic target in values of HbA1c but employed fasting glucose concentration as the treatment target (Becker 2003; Blonde 2009; Cao 2011; DIGAMI 2 2005; Fantin 2011; Guo 2008; Jaber 1996; Lu 2010; Natarajan 2012; UGDP 1975; UKPDS 1998). Two trials only defined targets for blood glucose, without further specification of when the blood glucose was taken (Melidonis 2000; Stefanidis 2003). One trial reported glycaemic control by HbA1c (Jaber 1996).

Many trials did not specify the target value for conventional glycaemic control. The Steno-2 trial intensified the glycaemic target in the conventional intervention group for the last two years of the intervention period. This change made the glycaemic target the same for the intensive and the conventional treatment group (Steno-2 2008).

The trial from Coa et al included participants with T2D and referred to operation for gastric cancer from ages 16 years and up (Cao 2011). However, based on the mean age of the participants and the standard deviations, it seems reasonable that a very few if any participants were aged less than 18 years (Appendix 4).

Outcome measures of included trials

For details see Summary of findings for the main comparison, Table 1, Appendix 2, Appendix 5, Appendix 6, Appendix 7, Appendix 9, Appendix 10, Appendix 12, and Appendix 13.

All-cause mortality or cardiovascular mortality was a predefined outcome or a part of the predefined composite outcome in 10 trials (ACCORD 2008; ADDITION-Europe 2011; ADVANCE 2008; Araki 2012; DIGAMI 2 2005; Steno-2 2008; UGDP 1975; UKPDS 1998; VA CSDM 1995; VADT 2009). The Kumamoto trial did not predefine mortality as an outcome in the planned intervention period of six years but assessed mortality as an outcome after 10 years (Kumamoto 2000) (Appendix 10).

Complications related to T2D, either microvascular or macrovascular, were a predefined outcome in 14 trials (ACCORD 2008; ADDITION-Europe 2011; ADVANCE 2008; Araki 2012; DIGAMI 2 2005; Fantin 2011; Kumamoto 2000; Steno-2 2008; UGDP 1975; UKPDS 1998; VA CSDM 1995; VADT 2009; Zhang 2011) (Appendix 10).

Moderate hypoglycaemia was mostly reported together with mild hypoglycaemia, and only three trials reported moderate hypoglycaemia separately (Fantin 2011; Melidonis 2000; Stefanidis 2003) (Appendix 7).

Patient satisfaction, general well-being or quality of life were assessed in only nine trials (ACCORD 2008; ADDITION-Europe 2011; Becker 2003; Jaber 1996; REMBO 2008; Steno-2 2008; UKPDS 1998; VA CSDM 1995; Zhang 2011) (Appendix 13). For the ADDITION-Europe population, only data from the subgroup of participants in the Netherlands and Leicester were available (ADDITION-Leicester; ADDITION-Netherlands). Data on quality of life for the total population in ADDITION-Europe will be published later (ADDITION-Europe 2011). Other trials had quality of life defined as an outcome but the results were not yet available (ADVANCE 2008; VADT 2009).

Some of the included trials (n = 15) did not predefine any of the outcomes we predefined as primary or secondary outcomes but assessed non-validated surrogate outcomes. It was possible to include data on some of our predefined outcomes in most of these trials (Bagg 2001; Becker 2003; Blonde 2009; Cao 2011; Guo 2008; IDA 2009; Jaber 1996; Lu 2010; Melidonis 2000; Natarajan 2012; REMBO 2008; Service 1983; Stefanidis 2003; Yang 2007). It was not possible to include outcome data from one of the included trials (Cooray 2011).

Excluded studies

Reasons for exclusion of studies are given in 'Characteristics of excluded studies'. Sixty studies described in 63 publications were excluded after further evaluation. In six cases, we contacted the authors of the articles to identify whether there were predefined differences in glycaemic target (Chan 2009; Crasto 2011; Du 2009; Olivarius 2001; Schaan 2011; Wexler 2012) or whether the trial was randomised or not (Kanaoka 2011). For the publications excluded due to the impossibility of separating data on participants with T2D from other included participants, the authors were asked about separate data on the participants with T2D. Main reasons for exclusion were: the study was not randomised (n = 17), participants were not patients with T2D or we could not separate data on those patients with T2D (n = 13), or no predefined differences in the glycaemic treatment target existed (n = 33).

Risk of bias in included studies

The risk of bias assessment of the included trials was performed using previously described criteria (please see section, 'Assessment of risk of bias in included studies'). For details of the judgements made for the individual trials, please see 'Characteristics of included studies', 'Risk of bias in included studies', Figure 2, and Figure 3. When a risk of bias domain could not be judged as low risk of bias, the authors were asked for additional information. For the update, we only contacted authors of the new trials and authors from updated trials, if necessary. The risk of bias assessment for blinding and incomplete outcome data are now specific at the outcome level (objective and subjective outcomes, please see 'Types of outcome measures'). Only two trials were judged as low risk of bias on all bias domains (Fantin 2011; UGDP 1975).

Figure 2.

Methodological quality graph: review authors' judgements about each methodological quality item presented as percentages across all included studies.

Figure 3.

Methodological quality summary: review authors' judgements about each methodological quality item for each included study.

Sequence generation

The generation of the allocation sequence was adequately described in only 17 trials (ACCORD 2008; ADDITION-Europe 2011; ADVANCE 2008; Araki 2012; Cao 2011; DIGAMI 2 2005; Fantin 2011; Guo 2008; IDA 2009; Melidonis 2000; Natarajan 2012; Service 1983; Stefanidis 2003; Steno-2 2008; UGDP 1975; UKPDS 1998; VADT 2009). One trial did not describe in the publication whether the participants were randomised or not (Cooray 2011). The authors replied that participants were randomised. The remaining trials were described as randomised but the method for sequence generation was not described (Bagg 2001; Becker 2003; Blonde 2009; Jaber 1996; Kumamoto 2000; Lu 2010; REMBO 2008; VA CSDM 1995; Yang 2007; Zhang 2011).

Allocation

The method used to conceal allocation was adequately described in only 14 trials (ACCORD 2008; ADDITION-Europe 2011; ADVANCE 2008; Araki 2012; Cao 2011; DIGAMI 2 2005; Fantin 2011; Guo 2008; IDA 2009; Natarajan 2012; Steno-2 2008; UGDP 1975; UKPDS 1998; VADT 2009). The method for allocation concealment was judged as unclear in 14 trials (Bagg 2001; Becker 2003; Blonde 2009; Cooray 2011; Jaber 1996; Kumamoto 2000; Lu 2010; Melidonis 2000; REMBO 2008; Service 1983; Stefanidis 2003; VA CSDM 1995; Yang 2007; Zhang 2011).

Blinding

Assessment of blinding was performed separately for the objective and subjective outcomes of our review. Therefore, trials which had adequately described blinding of the outcomes they predefined in the protocol to assess (for example, carotis intima thickness) might not necessarily be judged as low risk of bias according to blinding for the outcomes of our review. The method of blinding for our objective outcomes was adequately described in only 12 trials (ACCORD 2008; ADDITION-Europe 2011; ADVANCE 2008; Araki 2012; Cao 2011; DIGAMI 2 2005; Fantin 2011; Steno-2 2008; UGDP 1975; UKPDS 1998; VA CSDM 1995; VADT 2009). The assessment of our subjective outcomes was done by participants, who were not blinded except for the ADDITION trial (ADDITION-Europe 2011).

Incomplete outcome data

Assessment of incomplete outcome data was performed separately for the objective and subjective outcomes of our review. Incomplete data were addressed adequately for our objective outcomes in 19 of the included trials (ADDITION-Europe 2011; Bagg 2001; Blonde 2009; Cao 2011; DIGAMI 2 2005; Fantin 2011; Guo 2008; IDA 2009; Jaber 1996; Kumamoto 2000; Melidonis 2000; REMBO 2008; Service 1983; Stefanidis 2003; Steno-2 2008; UGDP 1975; UKPDS 1998; VA CSDM 1995; VADT 2009). For the remaining nine trials incomplete outcome data for the objective outcomes were judged as unclear (ACCORD 2008; ADVANCE 2008; Araki 2012; Becker 2003; Cooray 2011; Lu 2010; Natarajan 2012; Yang 2007; Zhang 2011).

Incomplete data were addressed adequately for our subjective outcomes in only 12 of the included trials (Blonde 2009; DIGAMI 2 2005; Fantin 2011; IDA 2009; Jaber 1996; Kumamoto 2000; Melidonis 2000; REMBO 2008; Stefanidis 2003; Steno-2 2008; UGDP 1975; UKPDS 1998; VA CSDM 1995; VADT 2009). For the remaining 16 trials incomplete outcome data for the objective outcomes were judged as unclear (ACCORD 2008; ADDITION-Europe 2011; ADVANCE 2008; Araki 2012; Bagg 2001; Becker 2003; Cao 2011; Cooray 2011; Guo 2008; Lu 2010; Natarajan 2012; Service 1983; Yang 2007; Zhang 2011).

Selective reporting

For outcomes reported and predefined to assess in the individual trials please see Characteristics of included studies, Appendix 9, Appendix 10. Selective outcome reporting bias was judged as adequate in only 15 trials (ACCORD 2008; ADDITION-Europe 2011; ADVANCE 2008; Araki 2012; Bagg 2001; Blonde 2009; DIGAMI 2 2005; Fantin 2011; Stefanidis 2003; Steno-2 2008; UGDP 1975; UKPDS 1998; VA CSDM 1995; VADT 2009). Twelve trials were judged as at unclear risk of selective reporting bias (Becker 2003; Cao 2011; Cooray 2011; Guo 2008; IDA 2009; Kumamoto 2000; Lu 2010; Melidonis 2000; REMBO 2008; Service 1983; Yang 2007). Thirteen of the included trials did not provide any protocol or trial document with definitions of the outcomes except for the publication with the results (Becker 2003; Cao 2011; Cooray 2011; Guo 2008; IDA 2009; Jaber 1996; Kumamoto 2000; Lu 2010; Melidonis 2000; REMBO 2008; Service 1983; Yang 2007; Zhang 2011). Five trials had incomplete reporting of outcomes which were predefined to be assessed in the method section of the trial publication or trial protocol (Araki 2012; Becker 2003; Cooray 2011; Jaber 1996; Zhang 2011) (Appendix 11). Four of these trials were therefore classified as at high risk of selective reporting (Becker 2003; Cooray 2011; Jaber 1996; Zhang 2011). The authors of Araki et al provided us with complete data on request, however some data were not yet analysed (Araki 2012). Six of the trials have still not reported all their secondary outcomes, but these might come with time as they were all recently finished and some of the authors have replied that they are still working on the data in order to get them published (ACCORD 2008; ADDITION-Europe 2011; ADVANCE 2008; Araki 2012; Natarajan 2012; VADT 2009). The prolongation of the intervention period (8 and 10 years of follow-up) was not originally planned and the outcomes were therefore not predefined to be assessed after 8 and 10 years (Kumamoto 2000).

Other trials had measured but just not reported outcomes of relevance for our review, as they were not a predefined outcome of the trial. In several cases it was possible to receive these additional data from the corresponding authors of the trials (Appendix 10; Appendix 14).

Other potential sources of bias

Most trials received funding from a private health insurance company or the medical industry to conduct the trial. Most authors had not previously performed other trials investigating the same intervention.

We divided the trials into those with a lower risk of bias and a high risk of bias based on the assessment of sequence generation and allocation concealment. Sequence generation and allocation concealment were assessed as at lower risk of bias in 13 trials (ACCORD 2008; ADDITION-Europe 2011; ADVANCE 2008; Araki 2012; Cao 2011; DIGAMI 2 2005; Fantin 2011; Guo 2008; IDA 2009; Natarajan 2012; Steno-2 2008; UGDP 1975; UKPDS 1998; VADT 2009).

Effects of interventions

See: Summary of findings for the main comparison Intensive glycaemic control versus conventional glycaemic control for type 2 diabetes mellitus

Primary outcomes

All-cause mortality

Several trials predefined death from any cause as the primary or secondary outcome (see section 'Description of studies', Appendix 9, and Appendix 10). Twenty-four trials provided information on all-cause mortality and could be included in the analyses. The included trials reported 3440 deaths in 34,325 participants (Analysis 1.1). Most of the deaths were reported by trials with low risk of bias according to sequence generation and allocation concealment (3406 deaths) (ACCORD 2008; ADDITION-Europe 2011; ADVANCE 2008; Araki 2012; Cao 2011; DIGAMI 2 2005; Fantin 2011; Natarajan 2012; Steno-2 2008; UGDP 1975; UKPDS 1998; VADT 2009). Meta-analyses with both the fixed-effect model and random-effects model showed no significant effect of intensive glycaemic control (random RR 1.00, 95% CI 0.92 to 1.08; 34,325 participants, 24 trials; Analysis 1.1; moderate quality of the evidence (GRADE); see Summary of findings table 1). Heterogeneity was moderate (I2 = 16%, P = 0.26). Trial sequential analysis with data from all included trials showed that 33,806 patients of the required diversity-adjusted information size of 46,305 were accrued and no firm evidence for benefit or harm was reached (Figure 4). The cumulative Z-curve crossed the futility boundaries suggesting that a 10% or greater relative risk reduction could be rejected at this point. One trial reported three deaths after the randomisation (Becker 2003), however the report did not describe to which intervention group the participants were randomised or the cause of death. It was therefore not possible to use data on death from this trial. Mortality data from the ACCORD, DIGAMI 2, and Steno-2 trials were for the longest follow-up and were therefore after the intervention stopped (ACCORD 2008; DIGAMI 2 2005; Steno-2 2008). The DIGAMI 2 follow-up data were based on post-trial data from a database (DIGAMI 2 2005).

Figure 4.

Trial sequential for all-cause mortality for all trials. Diversity-adjusted required information size of 46,305 participants calculated on basis of proportion of mortality of 9.6% in the conventional glucose control group, relative risk reduction of 10% in the intensive glycaemic control group, alpha = 5%, beta = 20%, and diversity = 39%. Actually accrued number was 33,806 participants, 73% of required information size. Horisontal green lines illustrate conventional levels of statistical significance (P = 0.05). Blue cumulative Z-curve does not cross trial sequential monitoring boundaries for benefit or harm, but boundaries for futility (inner wedge) are crossed.

Inspection of the funnel plot indicated bias. The funnel plot suggested that smaller trials favouring conventional glycaemic control may be unpublished (Figure 5).

Figure 5.

Funnel plot of comparison: 1 Intensive glycaemic control versus conventional glycaemic control, outcome: 1.1 All-cause mortality.

Repeating the analyses with the trials having an HbA1c target of 7% in the intensive intervention group did not change the results to significant values (random RR 0.69, 95% CI 0.27 to 1.74; I2 = 0%) (Bagg 2001; Guo 2008; Kumamoto 2000; REMBO 2008; Yang 2007). Three of these trials were conducted in Asia and contributed only 17 events in 543 participants. The ADDITION-Europe had an intervention target of HbA1c < 7% but additional glycaemic intervention was initiated when the HbA1c was > 6.5% (ADDITION-Europe 2011).

Sensitivity analysis excluding the largest trial, the ADVANCE trial contributing 11,140 participants, showed that there was still no statistical significance of the effect estimate (random RR 1.02, 95% CI 0.93 to 1.12; I2 = 13%, P = 0.31).

Sensitivity analysis with a continuity correction of zero-event trials did not change the statistical significance of the effect estimate (random RR 1.00, 95% CI 0.94 to 1.07).

Sensitivity analysis excluding the ACCORD trial with documented statistical interaction between intensive glycaemic control and the intensive blood pressure intervention did not change the statistical significance of the effect estimate (random RR 0.96, 95% CI 0.89 to 1.03).

Sensitivity analyses excluding the trials assessing the acute effects of glycaemic control (< 48 hours) and excluding unpublished trials could not be performed due to lack of data.

The subgroup analyses stratifying the trials according to risk of bias, study duration, diagnostic criteria, or funding source did not reveal any significance in effect estimates in the risk of all-cause mortality (Analysis 1.2; Analysis 1.3; Analysis 1.4; Analysis 1.5; Analysis 1.6; Analysis 1.7; Analysis 1.8; Analysis 1.9; Analysis 1.10; Analysis 1.11). Tests of interaction showed no significance between the subgroups. Because of lack of data, we were not able to conduct subgroup analyses on the trials published in languages other than English.

In stratifying for diagnostic criteria we chose to stratify according to whether the diagnostic criteria for T2D were described or not (Analysis 1.11).

Subgroup analyses stratifying the included trials according to the intervention (trials exclusively dealing with glycaemic control in usual care setting, glycaemic control as a part of acute intervention, glycaemic control initiated with surgical intervention, or multimodal intervention in usual care setting) were performed. The trials exclusively dealing with glycaemic control in the usual care setting showed no significant effect of the intervention (random RR 1.01, 95% CI 0.92 to 1.12; 28,354 participants, 12 trials; Analysis 1.12: subgroup 1). Heterogeneity was moderate (I2 = 28%, P = 0.19). Applying trial sequential analysis on all-cause mortality from trials exclusively dealing with glycaemic control in usual care settings showed that no evidence of benefit or harm could be established on all-cause mortality as only 28,149 participants (60%) of the 46,677 participants calculated as the required information size were accrued so far. The cumulative Z-curve just crossed the futility boundaries, but when including zero events trials the futility for a 10% relative risk reduction was clearly established (Figure web 7 and 8; http://www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx). Trials applying intensive glycaemic control as an acute intervention also showed no significant effect (random RR 1.11, 95% CI 0.89 to 1.38; 903 participants, 3 trials; Analysis 1.12: subgroup 2). Heterogeneity was absent (I2 = 0%, P = 1.00). Two-hundred and forty-two of the 246 reported deaths were from the DIGAMI 2 trial. Glycaemic control initiated with surgical intervention showed no statistical significance between the interventions (RR 0.63, 95% CI 0.21 to 1.92; 429 participants, 4 trials; Analysis 1.12: subgroup 3). Heterogeneity was absent (I2 = 0%, P = 0.85). Separate analyses of multimodal intervention in the usual care setting also showed no significant effect (RR 0.87, 95% CI 0.62 to 1.23; 4639 participants, 5 trials; Analysis 1.12: subgroup 4). Heterogeneity was present (I2 = 65%, P = 0.06). A test of interaction between the subgroups did not show any statistical significance.

We performed a meta-analysis of trials with available hazard ratios (HR) for all-cause mortality. The HR from the UKPDS trial was obtained from the meta-analysis of Turnbull et al and was truncated after five years of intervention (Turnbull 2009). Neither the fixed-effect model nor the random-effects model showed significant differences between the interventions (random HR 1.00, 95% CI 0.87 to 1.15; 7 trials; Analysis 1.13). Heterogeneity was moderate (I2= 54%, P = 0.04). A relatively large proportion of the participants from the cumulative meta-analysis of all-cause mortality was included in the meta-analysis of HR (90.2% of participants). The HR from the ADDITION-Europe trial was first estimated within each country with adjustment for clustering within a practice and then combined across countries with a fixed-effect model meta-analysis. All trials included in the analyses, except the ACCORD trial, provided an unadjusted HR (ACCORD 2008). The HR available from the ACCORD trial was adjusted for the following variables: assignment to the blood pressure trial or the lipid trial, assignment to the intensive blood pressure intervention in the blood pressure trial, assignment to receive fibrate in the lipid trial, the seven clinical centre networks, and a previous cardiovascular event. Excluding the ACCORD trial did not show any statistical significance (random HR 0.96, 95% CI 0.83 to 1.10). Separate meta-analysis of the trials exclusively dealing with glycaemic control in the usual care setting did not change the statistical significance of the effect estimate (random HR 1.05, 95% CI 0.90 to 1.20). Heterogeneity was moderate (I2 = 41%, P = 0.17).

In Natarajan et al it was not possible to estimate how many of the randomised participants had unknown mortality status at the end of follow-up. This trial was therefore not included in the available case, best-case, and worst case scenarios. Available case analysis did not result in any significant changes of effect estimates (random RR 1.00, 95% CI 0.92 to 1.09; 33,521 participants, 23 trials; Analysis 1.14). Analysing the missing data as the best-case scenario (assuming that participants with unknown vital status receiving intensive glycaemic control were alive and that all participants receiving the conventional intervention with unknown vital status were dead) showed statistical significance in favour of intensive glycaemic control (random RR 0.82, 95% CI 0.74 to 0.91; P = 0.0001; 34,247 participants, 23 trials; Analysis 1.15). No statistical significance was present for the worst-case scenario (assuming that participants with unknown vital status receiving intensive glycaemic control were dead and all participants with unknown vital status receiving conventional intervention were alive) in the random-effects model, but showed statistical significance in the fixed-effect model (random RR 1.20, 95% CI 0.97 to 1.49; fixed RR 1.23, 95% CI 1.16 to 1.30; P < 0.00001; 34,247 participants, 23 trials; Analysis 1.16).

Meta-regression for all the trials showed no significant association between disease duration and all-cause mortality (P = 0.10). Meta-regression did not detect statistically significant associations between fasting blood glucose at baseline, HbA1c at baseline, or duration of the intervention and all-cause mortality. Meta-regression for the trials exclusively dealing with glycaemic control in the usual care setting, however, showed a positive correlation between fasting blood glucose as well as HbA1c at baseline and the log of the risk ratios (RR) of the intervention groups for all-cause mortality. This indicated that RR may increase when the fasting blood glucose (P = 0.02) as well as HbA1c (P = 0.04) at baseline increase, that is, the higher the baseline fasting blood glucose or HbA1c, the higher risk there will be with intensive glycaemic control versus conventional glycaemic control. Meta-regression for the trials exclusively dealing with glycaemic control in the usual care setting showed no statistically significant association between disease duration and duration of the intervention.

Cardiovascular mortality

Several trials predefined cardiovascular mortality as the primary or secondary outcome (see section 'Description of studies', Appendix 5, Appendix 9, and Appendix 10). A total of 22 trials provided information on cardiovascular mortality and were included in the analyses. A total of 1690 cardiovascular deaths in 34,177 participants were included in the meta-analysis (Analysis 1.17). Data from the ACCORD, DIGAMI 2, and Steno-2 trials were for the longest follow-up and therefore after the intervention period (ACCORD 2008; DIGAMI 2 2005; Steno-2 2008). For the ADDITION-Europe, the number of cardiovascular deaths was calculated based on the number of cardiovascular deaths as the first, second, or third cardiovascular event in the participants (ADDITION-Europe 2011).

Most of the events were provided by trials with low risk of bias according to sequence generation and allocation concealment (1672 deaths) (ACCORD 2008; ADDITION-Europe 2011; ADVANCE 2008; Araki 2012; Cao 2011; DIGAMI 2 2005; Steno-2 2008; UGDP 1975; UKPDS 1998; VADT 2009). Neither the random-effects model nor the fixed-effect model showed a significant difference in effect estimates between intensive glycaemic control and conventional glycaemic control (random RR 1.06, 95% CI 0.94 to 1.21; I2 = 20%, P = 0.23; 34,177 participants, 22 trials; Analysis 1.17; moderate quality of the evidence (GRADE); see Summary of findings table 1). Trial sequential analysis for all included trials showed a lack of firm evidence for a benefit of targeting intensive glycaemic control for the reduction of cardiovascular mortality. Merely 33,658 of the 115,094 required patients are randomised at this point. That is, only 29.2% of the required diversity-adjusted information size to detect or reject a 10% relative risk increase was actually accrued in randomised trials so far (Figure 6). For trials reporting cardiovascular mortality and exclusively dealing with glycaemic control in usual care settings barely 17% of the required information size is accrued so far (Figure web 9; http://www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx).

Figure 6.

Trial sequential for cardiovascular mortality for all trials. Diversity-adjusted required information size of 115,094 participants calculated on basis of proportion of cardiovascular mortality of 4.5% in the conventional glucose control group, relative risk reduction of 10% in the intensive glycaemic control group, alpha = 5%, beta = 20%, and diversity = 45%. Actually accrued number was 33,658 participants, 29% of required information size. Horisontal green lines illustrate conventional levels of statistical significance (P = 0.05). Blue cumulative Z-curve does not cross trial sequential monitoring boundaries for benefit or harm, or reach futility.

Inspection of the funnel plot indicated bias. The funnel plot suggested that smaller trials favouring conventional glycaemic control are unpublished (Figure web 10; http://www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx).

By excluding the ACCORD and VADT trials, heterogeneity fell to 3% (P = 0.41). The ACCORD and VADT trials were the two trials with the lowest HbA1c target values in the intensive intervention groups. The effect estimate did not change to significant values (random RR 1.00, 95% CI 0.88 to 1.12).

Sensitivity analysis excluding the largest trial, the ADVANCE trial contributing 11,140 participants, showed that there was a significant benefit of targeting conventional glycaemic control (random RR 1.16, 95% CI 1.03 to 1.30; P = 0.02; I2 = 0%, P = 0.64).

Sensitivity analysis with a continuity correction of zero-event trials did not change the statistical significance of the effect estimate (random RR 1.05, 95% CI 0.96 to 1.16).

Sensitivity analyses excluding the trials with documented interaction between factorial intervention, the acute effects of glycaemic control (< 48 hours) and excluding unpublished trials could not be performed due to lack of data.

Subgroup analyses stratifying the trials according to risk of bias, study duration, and funding source did not reveal any significant effect estimates for cardiovascular mortality (Analysis 1.18; Analysis 1.19; Analysis 1.20; Analysis 1.21; Analysis 1.22; Analysis 1.23; Analysis 1.24; Analysis 1.25; Analysis 1.26). The subgroup of trials describing how the diagnosis of T2D was established showed statistical significance in favour of conventional glycaemic control (RR 1.17, 95% CI 1.04 to 1.31; P = 0.01; Analysis 1.27: subgroup 1). The direction of effect of the intervention was reversed for the trials not describing how the diagnosis of T2D was described (random RR 0.87, 95% CI 0.74 to 1.03; Analysis 1.27: subgroup 2). The test for subgroup differences between the trials describing how the diagnosis of T2D was described compared to the trials not describing how the diagnosis was established showed statistical significance (P = 0.005). No statistical significance was shown with the test of interaction for the remaining subgroup analyses.

Because of lack of data, we were not able to conduct subgroup analyses on the trials not published in English.

A meta-analysis of the 12 trials investigating the effect of intensive glycaemic control in trials exclusively dealing with glycaemic control in usual care settings showed no significant difference in effect estimates (random RR 1.09, 95% CI 0.92 to 1.29; 28,354 participants, 21 trials; Analysis 1.28: subgroup 1). Heterogeneity was present (I2 = 38%, P = 0.12). Analysing trials assessing glycaemic control as part of an acute intervention showed no significance in the effect estimate (random RR 1.14, 95% CI 0.86 to 1.51; 903 participants, 3 trials; Analysis 1.28: subgroup 2). One hundred and sixty-three of the 167 deaths were reported in the DIGAMI 2 trial (DIGAMI 2 2005). Glycaemic control initiated with surgical intervention could not be meta-analysed as only one trial reported deaths due to cardiovascular disease. Separate analyses of multimodal intervention in the usual care setting showed no significant effect of intensive glycaemic control (RR 0.86, 95% CI 0.47 to 1.56; 4639 participants, 5 trials; Analysis 1.28: subgroup 4). Heterogeneity was present (I2 =57%, P = 0.13). A test of interaction between the subgroups did not show any statistical significance.

No statistical significant difference was present when pooling available HRs (random HR 1.05, 95% CI 0.84 to 1.31; 6 trials; Analysis 1.29). Heterogeneity was substantial (I2 = 64%, P = 0.02). A relatively large proportion of the participants from the cumulative meta-analysis of cardiovascular mortality was included in the meta-analysis of HR (81.9% of participants). All trials included in the analyses, except the ACCORD trial, provided an unadjusted HR (ACCORD 2008). Excluding the ACCORD trial did not change the statistical significance of the effect estimate (random HR 0.98, 95% CI 0.77 to 1.25). Meta-analysis of the trials exclusively dealing with glycaemic control in the usual care setting did not change the statistical significance of the effect estimate (random HR 1.09, 95% CI 0.83 to 1.38).

Available case analyses did not result in any significant differences between the effect estimates (random RR 1.06, 95% CI 0.94 to 1.20; 33,451 participants, 22 trials; Analysis 1.30). When analysing the missing data as a best-case scenario and a worst-case scenario statistical significance was present (best-case: random RR 0.71, 95% CI 0.56 to 0.88; P = 0.003; worst-case: random RR 1.54, 95% CI 1.09 to 2.18; P = 0.01; 34,177 participants, 22 trials; Analysis 1.31; Analysis 1.32).

Meta-regressions for all trials and for the subgroup of trials exclusively dealing with glycaemic control in usual care settings could not detect any statistically significant association between duration of disease at baseline, fasting blood glucose at baseline, HbA1c at baseline, or duration of the intervention and intervention effect on cardiovascular mortality.

Secondary outcomes

Macrovascular complications

We predefined a composite outcome of macrovascular complications as a secondary outcome (non-fatal myocardial infarction, non-fatal ischaemic stroke, non-fatal haemorraghic stroke, amputation of lower extremity, and cardiac or peripheral revascularization). The definition of macrovascular disease as a composite outcome was clearly predefined in 10 trials (ACCORD 2008; ADDITION-Europe 2011; ADVANCE 2008; Araki 2012; DIGAMI 2 2005; Fantin 2011; Steno-2 2008; VA CSDM 1995; VADT 2009; Zhang 2011). The definitions varied among trials (Appendix 5). The ACCORD and ADVANCE trials, which contributed most events, included non-fatal myocardial infarction, non-fatal stroke, and death from cardiovascular causes (ACCORD 2008; ADVANCE 2008). For the ACCORD trial, data for the longest follow-up was not possible to include in the meta-analysis (ACCORD 2008). The number of participants with events after the transition period was reported but not the total number of participants from the intervention period and to the end of follow-up. Therefore, we have only included data from the intervention period (ACCORD 2008). Data from the DIGAMI 2 trial were available for the longest follow-up and therefore after the end of the intervention period (DIGAMI 2 2005). The UKPDS trial assessed diabetes-related complications as a composite outcome, which included both macrovascular and microvascular complications (UKPDS 1998). Unfortunately the number of participants was not reported. The number of participants with the composite macrovascular complications for the UKPDS trial was therefore taken from the meta-analysis in Turnbull et al (Turnbull 2009). Three trials did not predefine assessment of a composite macrovascular outcome but it was possible to extract useable data (Bagg 2001; Becker 2003; Kumamoto 2000). The IDA 2009 trial reported a total of 17 participants who received a new percutaneous coronary intervention, coronary bypass surgery, or had symptoms of angina (IDA 2009). From the publication it could not be concluded which group the participants belonged to. The number of patients with cardiovascular disease from Becker et al was calculated as the number of patients with a history of cardiovascular disease at baseline minus the number of patients with cardiovascular disease at follow-up (Becker 2003). Many of the included trials reported a composite macrovascular outcome together with death due to cardiovascular disease or all-cause mortality (ACCORD 2008; ADDITION-Europe 2011; ADVANCE 2008; DIGAMI 2 2005; Steno-2 2008).

Most of the events were provided by trials with lower risk of bias according to sequence generation and allocation concealment (3344 participants with one or more of the composite macrovascular events out of a total of 3426) (ACCORD 2008; ADDITION-Europe 2011; ADVANCE 2008; Araki 2012; DIGAMI 2 2005; Fantin 2011; Steno-2 2008; UKPDS 1998; VADT 2009). Meta-analysis of data from 14 trials on the composite macrovascular outcome did not reveal any significant difference using the random-effects model but showed statistical significance in the fixed-effect model in favour of intensive glycaemic control (random RR 0.91, 95% CI 0.82 to 1.02; fixed RR 0.93, 95% CI 0.87 to 0.99; P = 0.02; 32,846 participants, 14 trials; Analysis 1.33; lower risk of bias). Heterogeneity was substantial (I2 = 47%, P = 0.03).

Subgroup analysis for the trials exclusively dealing with glycaemic control in usual care settings did not show any statistical significance in the random-effects model but statistical significance in the fixed-effect model (random RR 0.92, 95% CI 0.83 to 1.01; fixed RR 0.93, 95% CI 0.86 to 0.99; P = 0.03; 27,666 participants, 9 trials; Analysis 1.34: subgroup 1). The trials assessing glycaemic control as a part of a multimodal intervention did not show any statistical significance (random RR 0.79, 95% CI 0.54 to 1.16; 4330 participants, 3 trials; Analysis 1.34: subgroup 4). The analysis could not be performed for trials assessing glycaemic control as a part of an acute intervention and trials assessing glycaemic control initiated with surgical intervention due to the lack of data. The test of interaction showed no statistical significance between subgroups.

Non-fatal myocardial infarction

A total of 1422 participants with one or more non-fatal myocardial infarctions were recorded in 30,417 participants (Analysis 1.35). The details on how the diagnosis of myocardial infarction was established varied between trials. Eleven trials provided detailed information on how they defined myocardial infarction (ACCORD 2008; ADDITION-Europe 2011; Araki 2012; DIGAMI 2 2005; Fantin 2011; Melidonis 2000; Steno-2 2008; UGDP 1975; UKPDS 1998; VA CSDM 1995; VADT 2009) (Appendix 5). Eight trials had non-fatal myocardial infarction as part of the primary outcome (ACCORD 2008; ADDITION-Europe 2011; ADVANCE 2008; Araki 2012; Steno-2 2008; UGDP 1975; UKPDS 1998; VADT 2009) (Appendix 10). Data from the long-term follow-up for the ACCORD trial reported 80 participants with a non-fatal myocardial infarction after the transition period in the intensive glycaemic intervention arm, and 87 participants in the conventional glycaemic intervention arm. The total number of participants with one or more myocardial infarctions in the total follow-up period (intervention period + post-transition period) was not reported (ACCORD 2008). Data used in this review were therefore until the end of the intervention period (ACCORD 2008). The ADDITION-Europe did not report the total number of participants in each intervention group with a non-fatal myocardial infarction but only reported the number of participants who had myocardial infarction as a part of the first event of the composite outcome (intensive: 29/1678 participants; conventional: 32/1377 participants) (ADDITION-Europe 2011). Three trials had admission to hospital with acute myocardial infarction or unstable angina as an inclusion criterion (DIGAMI 2 2005; Melidonis 2000; Stefanidis 2003). Because all participants had an acute myocardial infarction, we used the number of re-infarctions when meta-analysing these trials. Data from the DIGAMI 2 trial were to the longest follow-up and therefore after the end of the intervention period (DIGAMI 2 2005).

Most of the events were provided by trials with lower risk of bias according to sequence generation and allocation concealment (1404 participants with non-fatal myocardial infarction out of a total of 1422) (ACCORD 2008; ADVANCE 2008; Araki 2012; DIGAMI 2 2005; Steno-2 2008; UGDP 1975; UKPDS 1998; VADT 2009). There was statistical significance in favour of intensive glycaemic control (random RR 0.87, 95% CI 0.77 to 0.98; 29,244 participants, 13 trials; Analysis 1.35; moderate quality of the evidence (GRADE); see Summary of findings table 1). Heterogeneity might not have been important (I2 = 13%, P = 0.32). Trial sequential analysis showed a lack of firm evidence for benefit of targeting intensive glycaemic control for the reduction of non-fatal myocardial infarction. Only 30,194 patients (38.0%) have been accrued so far of the required diversity-adjusted information size of 79,401 to detect a 10% relative risk reduction of non-fatal myocardial infarction (Figure web 11; http://www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx).

The funnel plot did not raise any suspicion of bias (Analysis 1.35).

Combining the data from the trials with detailed information about the definition of non-fatal myocardial infarction showed statistical significance of the effect estimate (random RR 0.87, 95% CI 0.77 to 0.99; P = 0.03). Heterogeneity was moderate (I2 = 21%, P = 0.25). In a meta-analysis of the trials with non-fatal myocardial infarction as part of their primary outcome also showed statistically significant effect estimates (random RR 0.84, 95% CI 0.73 to 0.96; P = 0.009). Heterogeneity was moderate (I2 = 22%, P = 0.26).

In the ACCORD trial almost all participants in the intensive group and more than half in the conventional group received rosiglitazone. Excluding the ACCORD trial from the analysis, neither the random-effects model nor the fixed effect model showed significant benefit of targeting intensive glycaemic control (random RR 0.90, 95% CI 0.80 to 1.02). In the ADVANCE trial more participants in the intensive intervention arm, compared with the conventional intervention arm, also received rosiglitazone. Sensitivity analysis excluding both the ACCORD and the ADVANCE trials changed the effect estimate to non-significant values (random RR 0.73, 95% CI 0.73 to 1.04).

Subgroup analyses stratifying the trials according to risk of bias, study duration, diagnostic criteria, or funding source did not reveal any significant differences in the tests of interaction (Analysis 1.36; Analysis 1.37; Analysis 1.38; Analysis 1.39).

Subgroup analyses stratifying the trials according to intervention were performed. The trials exclusively dealing with glycaemic control in usual care settings showed significant benefit of targeting intensive glycaemic control (random RR 0.85, 95% CI 0.76 to 0.95; P = 0.004; 28,111 participants, 8 trials; Analysis 1.40: subgroup 1). Heterogeneity was absent (I2 = 0%, P = 0.70). Applying glycaemic control in trials exclusively dealing with glycaemic control in usual care settings showed that 44% of the information size was accrued to detect or reject a 10% relative risk reduction (Figure web 12; http://www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx). When excluding the ACCORD trial from the analysis, the significance of the effect estimate disappeared (random RR 0.88, 95% CI 0.77 to 1.01; I2 = 0%, P = 0.71). Three trials were analysed in the subgroup of glycaemic control as a part of acute intervention. The effect estimate did not reveal any significant effect estimate (random RR 1.15, 95% CI 0.84 to 1.58; 903 participants, 3 trials; Analysis 1.40: subgroup 2). Heterogeneity was absent (I2 = 0%, P = 0.82). Subgroup analysis for the subgroup of trials with glycaemic control initiated with surgical intervention could not be performed as only Fantin et al provided data (Fantin 2011). Multimodal intervention in usual care settings did not show any statistical significance (random RR 0.63, 95% CI 0.22 to 1.83; 1333 participants, 2 trials; Analysis 1.40: subgroup 4). The test of interaction showed no statistical significance between the subgroup analyses. However, qualitative interaction was present between the group of trials exclusively dealing with glycaemic control in the usual care setting (direction of intervention effect in favour of intensive glycaemic control) and the trials dealing with glycaemic control as a part of an acute intervention (direction of effect in favour of conventional glycaemic control).

Available case analyses showed significant benefit of intensive glycaemic control (random RR 0.87, 95% CI 0.77 to 0.97; P = 0.01; 28,173 participants, 14 trials; Analysis 1.41). Analysing the missing data as the best-case scenario and worst-case scenario showed statistical significance of the effect estimates (best-case scenario: random RR 0.34, 95% CI 0.24 to 0.47; P < 0.00001; worst-case scenario: random RR 2.28, 95% CI 1.62 to 3.19; P < 0.00001; 30,417 participants, 14 trials; Analysis 1.42; Analysis 1.43).

Neither the meta-regressions of all trials nor trials exclusively dealing with glycaemic control in usual care settings were able to detect a statistically significant association between duration of disease, fasting blood glucose at baseline, HbA1c at baseline, or duration of intervention and the risk of non-fatal myocardial infarction.

Non-fatal stroke

Seven trials had non-fatal stroke as part of their primary outcome (ACCORD 2008; ADDITION-Europe 2011; ADVANCE 2008; Araki 2012; Steno-2 2008; UKPDS 1998; VADT 2009) (Appendix 5, Appendix 10). Of the 900 non-fatal strokes, 423 were reported from the ADVANCE trial (ADVANCE 2008). Data from the long-term follow-up for the ACCORD trial reported 10 participants with a non-fatal stroke infarction after the transition period in the intensive glycaemic intervention arm and 22 participants in the conventional glycaemic intervention arm. The total number of participants with one or more myocardial infarctions in the total follow-up period (intervention period + post-transition period) was not reported (ACCORD 2008). Data were therefore until the end of the intervention period (ACCORD 2008). The ADDITION-Europe did not report the total number of participants in each intervention group with a non-fatal stroke but only reported the number of participants who had stroke as part of the first event of the composite outcome (intensive: 22/1678 participants; conventional: 19/1377 participants) (ADDITION-Europe 2011). Data from the DIGAMI 2 trial was from the extended follow-up period after the end of the intervention (DIGAMI 2 2005).

Most of the events were provided by trials with low risk of bias according to sequence generation and allocation concealment (867 participants with non-fatal stroke out of a total of 900) (ACCORD 2008; ADVANCE 2008; Araki 2012; DIGAMI 2 2005; Steno-2 2008; UKPDS 1998; VADT 2009). No statistically significant difference was found for the risk of non-fatal stroke between the intervention groups (random RR 0.99, 95% CI 0.84 to 1.18; 30,000 participants, 13 trials; Analysis 1.44; moderate quality of the evidence (GRADE); see Summary of findings table 1). Heterogeneity might not have been important (I2 = 17%, P = 0.28). Originally we planned to report ischaemic and haemorrhagic stroke separately, but all trials except one (Kumamoto 2000) defined and reported both aetiologies for the non-fatal stroke composite. Separate analysis of the trials having non-fatal stroke as a part of their primary outcome did not change the statistical significance of the effect estimate (random RR 0.97, 95% CI 0.79 to 1.19).

In a separate meta-analysis of the trials exclusively dealing with glycaemic control in usual care settings the effect estimate remained non-significant (random RR 1.01, 95% CI 0.87 to 1.16; 27,697 participants, 7 trials; Analysis 1.45: subgroup 1). Heterogeneity was absent (I2 = 0%, P = 0.73). It was not possible to meta-analyse data from trials assessing glycaemic control as part of an acute intervention or glycaemic control initiated with surgical intervention due to lack of data. Glycaemic control as a part of multimodal intervention in the usual care setting did not show statistical significance (random RR 0.68, 95% CI 0.19 to 2.49; 1333 participants, 2 trials; Analysis 1.45: subgroup 4). The test of interaction showed no statistical significance between the subgroup analyses.

Amputation of lower extremity

Five trials reported amputation of a lower extremity without further description (Fantin 2011; Kumamoto 2000; Melidonis 2000; Stefanidis 2003; UGDP 1975). The Steno-2 and VA CSDM trials specified that the number of amputations was due to ischaemia, and the VADT specified amputation for ischaemic diabetic gangrene (Steno-2 2008; VA CSDM 1995; VADT 2009). The ADDITION-Europe defined amputation as caused by cardiovascular disease, neuropathic disease, or any other reasons (ADDITION-Europe 2011). UKPDS defined amputation as major limb complications requiring amputation of a digit or any limb for any reason (UKPDS 1998). It was therefore not clear if the trials without further specification of amputation added minor amputation (for example a digit) to the reported number, as the UKPDS has done. Besides, the UKPDS included amputation for any reason, which was not the case for the VADT and Steno-2 trials. Accordingly, an amputation due to infection may not have been reported as part of the outcome for the VADT and Steno-2 but would be for UKPDS. The different definitions of amputation of a lower extremity may explain the dominance of the UKPDS.

It was unfortunately not possible to get reliable data from the two largest included trials (ACCORD 2008; ADVANCE 2008). Thus, it was very likely that amputation of a lower extremity was grossly under-reported.

Most of the events were provided by trials with low risk of bias according to sequence generation and allocation concealment (117 participants with amputation of lower extremity out of a total of 118) (ACCORD 2008; ADVANCE 2008; DIGAMI 2 2005; Steno-2 2008; UGDP 1975; UKPDS 1998; VADT 2009). Meta-analysis showed a significantly reduced risk of amputation of a lower extremity when targeting intensive glycaemic control (random RR 0.65, 95% CI 0.45 to 0.94; P = 0.02; 11,200 participants, 11 trials; Analysis 1.46; low quality of the evidence (GRADE); see Summary of findings table 1). Heterogeneity was absent (I2 = 0%, P = 0.82). Trial sequential analysis for all included trials showed that 6.1% of the diversity-adjusted information size has actually been accrued so far to detect or reject a 10% relative risk reduction (Figure web 13; http://www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx). However, the number of reported amputations was very low in both the intensive and conventional intervention groups (59 in each group). The UKPDS contributed almost half of the reported events (UKPDS 1998).

Stratifying the trials according to intervention could only be done for trials exclusively dealing with glycaemic control in usual care (random RR 0.70, 95% CI 0.45 to 1.09; 6677 participants, 5 trials; Analysis 1.47: subgroup 1) and trials dealing with glycaemic control as a part of a multimodal intervention in usual care (random RR 0.54, 95% CI 0.27 to 1.07; 4330 participants, 3 trials; Analysis 1.47: subgroup 4), which did not show any statistical significance. Heterogeneity was absent (I2 = 0%, P = 0.59; I2 = 0%; P = 0.75, respectively). The test of interaction showed no statistical significance between the subgroup analyses. The trials exclusively dealing with glycaemic control in the usual care setting showed only 4.5% of the diversity-adjusted information size to detect or reject a 10% relative risk reduction (Figure web 14; http://www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx).

Cardiac revascularization

The data included in the outcome of cardiac revascularization were surgical revascularizations (for example, artery bypass grafting). The revascularization procedures in the DIGAMI 2 trial were primarily done as acute thrombolysis and it was not possible to extract the data regarding surgical revascularization (DIGAMI 2 2005). Araki et al reported that 18 participants underwent coronary revascularization but it was not reported how many participants in each intervention group (Araki 2012). Stefanidis et al reported separately the number of participants with T2D and acute cardiovascular events undergoing invasive cardiovascular surgery (Stefanidis 2003). Melidonis et al, who also included participants with acute cardiovascular events and T2D, did not specify whether revascularization was surgical or medical; the number was therefore not included in the analyses (Melidonis 2000). The ADVANCE trial investigators reported coronary revascularization procedures as a part of the total coronary events. It was not possible to obtain the number of cardiac revascularizations as a separate number (ADVANCE 2008). Five trials reported cardial revascularization without further specifications (Fantin 2011; Kumamoto 2000; Stefanidis 2003; VA CSDM 1995; VADT 2009). The Steno-2 trial defined cardiac revascularization as coronary bypass grafting (Steno-2 2008).

Of the 289 participants undergoing a cardiac revascularization procedure, most were reported from the VADT trial (234 procedures) (VADT 2009). The effect estimate was not statistically significant (random RR 0.81, 95% CI 0.65 to 1.01; 3532 participants, 7 trials; Analysis 1.48; lower risk of bias). The I2 was 0% (P = 0.53).

Stratifying the trials according to the intervention, the trials exclusively dealing with glycaemic control in usual care settings showed no statistical significance (random RR 0.85, 95% CI 0.67 to 1.07; 2054 participants, 3 trials; Analysis 1.49: subgroup 1). The I2 was 0% (P = 0.67). Meta-analysis of the trials applying glycaemic control as part of multimodal intervention in the usual care setting showed statistical significance in favour of intensive glycaemic control (random RR 0.50, 95% CI 0.26 to 0.94; P = 0.03; 1333 participants; 2 trials; subgroup 4). There was a lack of data for trials with glycaemic control as part of an acute intervention and glycaemic control initiated with surgical intervention. The test of interaction showed no statistical significance between the subgroup analyses.

Peripheral revascularization

The ADVANCE contributed the majority of data (709 out of 768) (ADVANCE 2008). Unfortunately, the definition of peripheral revascularization was not described in the ADVANCE (ADVANCE 2008). We therefore used the number of peripheral vascular events without exactly knowing what this reporting included. It might be that amputation was reported as part of this outcome. The decision on when to intervene with peripheral revascularization might differ among the trials as well as the study centres within each trial. All trials contributing with data were classified as lower risk of bias according to sequence generation, allocation concealment, and blinding of our subjective outcomes. A meta-analysis for peripheral revascularization did not reveal any significant differences in the effect of intensive versus conventional intervention (random RR 0.93, 95% CI 0.81 to 1.06; 13,547 participants, 8 trials; Analysis 1.50; lower risk of bias). The I2 was 0% (P = 0.66).

When stratifying the trials according to the intervention, it was only possible to meta-analyse the subgroup consisting of trials exclusively dealing with glycaemic control in usual care settings. The effect estimate was not significant (random RR 0.93, 95% CI 0.81 to 1.07; 13,194 participants, 4 trials; Analysis 1.51: subgroup 1). Heterogeneity was absent (I2 = 0%, P = 0.83). Meta-analysis of trials with glycaemic control as part of an acute intervention, glycaemic control initiated with surgical intervention, and multimodal intervention in the usual care setting could not be conducted due to lack of data.

Microvascular complications

We predefined a composite outcome of microvascular complications as a secondary outcome (manifestation and progression of nephropathy, manifestation and progression of retinopathy, and retinal photocoagulation). It was possible to extract useable data from six trials that had predefined a composite microvascular outcome (ACCORD 2008; ADVANCE 2008; Fantin 2011; Steno-2 2008; UKPDS 1998; Zhang 2011). The Kumamoto trial did not report a composite microvascular outcome. On request, the authors gave us information on the total number of microvascular events after 10 years of follow-up (22 in the intensive group and 58 in the control group), but not the number of patients.

The definitions of the reported composite outcome varied between the included trials. In the Steno-2 trial microalbuminuria was an inclusion criterion. The reported composite outcome for microvascular disease was progression in any microvascular outcome during the follow-up period after 13.3 years (Steno-2 2008). This definition included both severe and less severe microvascular complications, for example, onset of neuropathy and mild retinal changes (that is, some outcomes were surrogate markers), which were likely to make a major contribution to the meta-analyses (for example, in the ADVANCE trial the statistical significance of the microvascular outcome was primarily due to reduction of albuminuria observed in the intensively treated group, whereas no significance was present for the severe microvascular complications, for example, death due to renal failure) (ADVANCE 2008). Neither the ADVANCE, ACCORD, nor the UKPDS trials included neuropathy in their composite microvascular outcomes. The ADVANCE, ACCORD, and UKPDS trials included moderate to severe retinal events in their composite microvascular outcome (for example, development of proliferative retinopathy, retinal photocoagulation). The nephropathy component of the composite microvascular outcome of the ADVANCE trial included development of macroalbuminuria, whereas the ACCORD and the UKPDS trials reported renal failure (Appendix 6) (ACCORD 2008; ADVANCE 2008; UKPDS 1998). Zhang et al reported a composite microvascular outcome and major microvascular events, without further description (Zhang 2011). Fantin et al reported through correspondence that no participants developed composite microvascular complications (Fantin 2011).

All trials contributing data on the composite microvascular outcome were of lower risk of bias according to sequence generation and allocation concealment, except for one trial (Zhang 2011). We found benefit of targeting intensive glycaemic control compared with targeting conventional glycaemic control (random RR 0.88, 95% CI 0.82 to 0.95; P = 0.0008; 25,927 participants, 6 trials; Analysis 1.52; lower risk of bias). The I2 was 20% (P = 0.28). Trial sequential analysis for all trials showed firm evidence for a 10% relative risk reduction of the composite outcome of microvascular complications in favour of targeting intensive glycaemic control (Figure web 15; http://www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx). The risk difference (RD) showed a statistical significance with a magnitude of a 1% absolute risk reduction (random RD -0.01, 95% CI -0.03 to -0.00; P = 0.02).

Analysing the trials exclusively dealing with glycaemic control in usual care settings showed statistically significant effect estimates favouring intensive glycaemic control (random RR 0.87, 95% CI 0.78 to 0.97; P = 0.01; 25,697 participants, 4 trials; Analysis 1.53: subgroup 1). The I2 was 40% (P = 0.17). The trials exclusively dealing with glycaemic control in the usual care setting showed that firm evidence for a 10% relative risk reduction was not established (Figure web 16; http://www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx). The RD showed a 1% absolute risk reduction (random RD -0.01, 95% CI -0.02 to -0.00; P = 0.009). It was not possible to include in the meta-analysis glycaemic control as part of an acute intervention and a multimodal intervention in usual care settings due to the lack of data.

Nephropathy

We predefined assessing the manifestation and progression of nephropathy. The definition of nephropathy varied among trials (see 'Definition of microvascular outcomes in study or as reported', Appendix 6). The ACCORD trial assessed nephropathy in different ways (development of microalbuminuria, development of macroalbuminuria, development of renal failure, doubling of serum creatinine, or a decrease of glomerular filtration rate (GFR)). The outcome included in this analysis was the predefined composite renal outcome, which did not include development of microalbuminuria (ACCORD 2008). The ADVANCE trial also reported a composite nephropathy outcome, which was defined similarly to the composite nephropathy outcome in the ACCORD trial but did not include decrease in GFR (ADVANCE 2008). The only trial including death due to renal disease in the nephropathy outcome was the ADVANCE trial. However, the statistical significance for the benefit in nephropathy in the ADVANCE trial was primarily driven by changes in onset of micro- and macroalbuminuria (ADVANCE 2008). The UGDP trial assessed kidney function in three different ways: serum creatinine ≥ 1.5 mg/dL, urine protein ≥ 1 gm/L, and urine protein 2+, which were all reported separately (UGDP 1975). We chose to report on the participants with urine protein > 1 gm/L. This definition might have underestimated the number of participants with nephropathy compared to the other included trials because of the high protein limit. The surrogate marker for nephropathy reported from the UKPDS trial was a two-fold plasma creatinine increase after nine years of follow-up (UKPDS 1998). The VA CSDM trial reported nephropathy as an elevated albumin-creatinine ratio (> 0.30), which was defined as overt nephropathy (VA CSDM 1995). The VADT divided nephropathy into three components that were reported separately. We chose to report on the number of participants with doubling of creatinine levels (VADT 2009). Bagg et al reported the number with nephropathy, defined as macroalbuminuria, based on a single urine assessment at the end of follow-up (Bagg 2001). Fantin et al reported through correspondence that no participants developed or progressed in nephropathy, without further description (Fantin 2011).

The participants of the Kumamoto trial were stratified at inclusion to a primary prevention population and a secondary prevention population (Kumamoto 2000). The primary prevention population only included participants without retinopathy and a urinary albumin excretion less than 30 mg/24 hour. The secondary prevention population had simple retinopathy and urinary albumin excretion less than 300 mg/24 hour. The primary prevention population reported the number of participants who developed nephropathy and the secondary intervention population reported the participants who progressed to nephropathy. The numbers were reported together after 10 years of intervention (Kumamoto 2000). The number for nephropathy therefore included onset of microalbuminuria in the primary prevention population, which was not the case for the other trials reporting nephropathy (Kumamoto 2000). A large proportion of the participants in the Steno-2 trial progressed to nephropathy (defined as albumin excretion > 300 mg/24 hour) after 13.3 years of follow-up. An inclusion criterion for the Steno-2 trial was microalbuminuria (Steno-2 2008).

Most of the events were provided by trials with low risk of bias according to sequence generation, allocation concealment, and blinding of the outcome (6905 participants with nephropathy out of a total of 6979) (ACCORD 2008; ADVANCE 2008; Steno-2 2008; UGDP 1975; UKPDS 1998; VADT 2009). Targeting intensive glycaemic control showed significant reductions in nephropathy (random RR 0.75, 95% CI 0.59 to 0.95; P = 0.02; 28,096 participants, 11 trials; Analysis 1.54; lower risk of bias). Heterogeneity was considerable (I2 = 77%, P < 0.0001), which might be due to the different definitions and populations in the included trials. Trial sequential analyses for all trials showed firm evidence that a 10% relative risk reduction of nephropathy was not established (Figure web 17; http://www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx). Applying RD showed no statistical significance in the random-effects model but statistical significance in the fixed-effect model (random RD -0.01, 95% CI -0.02 to 0.01; fixed RD -0.01, 95% CI -0.02 to -0.00; P = 0.02).

Most of the data using the composite nephropathy outcome were reported from the ACCORD trial, which did not show any difference in the number of participants. The composite nephropathy outcome from the ACCORD trial was the only one which included GFR (ACCORD 2008). When looking at each component of the composite nephropathy outcome separately, all were reduced by intensive glycaemic control but doubling of serum creatinine and decrease in GFR, which contributed the most events. Additional information obtained from a published letter by the authors reported that most of the events by far were due to decreased GFR (Ismail-Beigi 2010). Excluding the ACCORD trial from the analysis did not change the statistical significance. Both the UKPDS and the UGDP included participants with relatively mild metabolic disturbances (early disease stage) and reported few cases of nephropathy compared to the other trials reporting nephropathy. When excluding the UKPDS and the UGDP trials, the effect estimate remained statistically significant (random RR 0.74, 95% CI 0.58 to 0.95; P = 0.02).

The trials exclusively dealing with glycaemic control in usual care settings showed no significant effect estimates (random RR 0.79, 95% CI 0.61 to 1.01; 27,866 participants, 9 trials; Analysis 1.55: subgroup 1). Heterogeneity was substantial (I2 = 75%, P < 0.0001). Trial sequential analyses for all trials, as well as for trials exclusively dealing with glycaemic control in the usual care setting showed that firm evidence for a 10% relative risk reduction of nephropathy was not established (Figure web 18; http://www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx). It was not possible to analyse subgroups of trials with glycaemic control as part of an acute intervention, glycaemic control initiated with surgical intervention, or multimodal intervention in the usual care setting due to lack of data.

End-stage renal disease

We pooled data on hard renal outcomes from eight trials (ACCORD 2008; ADVANCE 2008; Fantin 2011; Kumamoto 2000; Steno-2 2008; UGDP 1975; UKPDS 1998; VADT 2009). The extractable data varied but all reported a measure of severe renal failure (for example, dialysis, death due to renal disease) (see 'Definition of microvascular outcomes in study or as reported', Appendix 6). As end-stage renal disease was not a predefined outcome in the protocol, the authors did not comment on the data. The results for the ADVANCE and ACCORD trials were a part of the reported outcome for nephropathy (except for three deaths due to renal failure in the ACCORD trial). Data extracted from Steno-2 and UGDP trials were the number of participants initiating renal dialysis. The measure from the VADT was exclusively the number of participants who died because of renal failure. Fantin et al reported through correspondence that no participants developed end-stage renal disease, without further description (Fantin 2011). All the trials providing data to the meta-analysis on end-stage renal disease were from trials with adequate sequence generation, allocation concealment, and blinding of our subjective outcomes. Pooling data from all eight trials did not show any statistical significance (random RR 0.87, 95% CI 0.71 to 1.06; I2 = 0%, P = 0.45; 28,145 participants, 8 trials; Analysis 1.56; moderate quality of the evidence (GRADE); see Summary of findings table 1).

Stratifying the trials after intervention, it was only possible to carry out a meta-analysis of the trials exclusively dealing with glycaemic control in usual care settings. The effect estimate remained non-significant (random RR 0.88, 95% CI 0.72 to 1.07; I2 = 0%; 27,915 participants, 6 trials; Analysis 1.57: subgroup 1).

Retinopathy

We collected data on the manifestation and progression of retinopathy from the included trials (see 'Definition of microvascular outcomes in study or as reported', Appendix 6). The ACCORD and ADVANCE trials conducted a substudy investigating the manifestation and progression of retinopathy from the Early Treatment of Diabetic Retinopathy Study (ETDRS) scale (ACCORD 2008; ADVANCE 2008). Both the ACCORD and ADVANCE trials also reported severe retinopathy based on patient history. To make the comparisons more similar to those in the other included trials we reported retinopathy as defined in the substudies, using the surrogate marker the ETDRS scale. The ACCORD Eye reported data from 2856 participants followed up for four years (ACCORD 2008). The substudy of the ADVANCE trial that assessed retinopathy randomised 1602 participants.

The trials using the ETDRS to classify retinopathy reported either a two-step or three-step increase as progression of retinopathy. The primary outcome of the ACCORD Eye consisted of at least three steps in the ETDRS, photocoagulation, or vitrectomy. The article on the ACCORD Eye did not report each component of the composite primary outcome separately, only the composite outcome. In an answer to a letter, the authors of the ACCORD Eye reported each component separately and the number we report was the number of participants with a three-step increase in ETDRS (Rind 2010). The ADVANCE and Kumamoto trials reported progression of retinopathy by a two-step increase in the ETDRS (ADVANCE 2008; Kumamoto 2000). Besides reporting a two-step increase in the ETDRS, the ADVANCE trial also reported the number of participants with a three-step increase. Because a two-step increase was used in most trials to describe progression of retinopathy we used this number. The number reported for the primary prevention population in the Kumamoto trial was the number of participants who developed retinopathy (Kumamoto 2000). In the secondary intervention population, the number reported was the number of participants who progressed from simple retinopathy. For the UKPDS 1998, only data from the participants in the UKPDS 33 were available (UKPDS 1998). The UKPDS 34 reported a lower rate of progression of retinopathy with intensive glycaemic control using metformin after nine years (P = 0.044) compared with conventional control. However, the benefit of intensive glycaemic control with metformin disappeared after 12 years (Appendix 11).

All but three trials reporting retinopathy used the ETDRS scale to report new retinopathy and progression of retinopathy (Fantin 2011; Steno-2 2008; UGDP 1975). The UGDP trial graded fundus photographs according to the Airlie House Classification, which assesses lesions as absent, mild, moderate, or severe. However, the ETDRS scale is designed from the Airlie House Classification. We chose to report the data for mild retinal abnormalities because these might be comparable to the ETDRS grading. The Steno-2 trial graded diabetic retinopathy according to another scale, the EURODIAB (European Community–funded Concerted Action Programme into the Epidemiology and Prevention of Diabetes) six-grade scale, which is designed from the ETDRS scale (Steno-2 2008). Fantin et al reported through correspondence that no participants developed or progressed in retinopathy, without further description (Fantin 2011).

Most of the events were provided by trials with low risk of bias according to sequence generation and allocation concealment (1366 participants with retinopathy out of a total of 1390) (ACCORD 2008; ADVANCE 2008; Steno-2 2008; UGDP 1975; UKPDS 1998; VADT 2009). The risk of retinopathy was significantly reduced (random RR 0.79, 95% CI 0.68 to 0.92; P = 0.002; 10,300 participants, 9 trials; Analysis 1.58; lower risk of bias). Heterogeneity was substantial (I2 = 53%, P = 0.04). Trial sequential analyses for all trials showed that firm evidence for a 10% relative risk reduction of retinopathy was not established (Figure web 19; http://www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx). The absolute risk reduction was 3% applying the random-effects model (random RD -0.03, 95% CI -0.06 to -0.01; P = 0.02). Excluding the trials not using the ETDRS to classify retinopathy, statistical significance of the effect estimates was still present (random RR 0.77, 95% CI 0.64 to 0.92; P = 0.05). Analysing data from trials using a two-step increase of the ETDRS as progression of retinopathy also showed significant effect estimates (random RR 0.80, 95% CI 0.65 to 0.98; P = 0.03). Heterogeneity was substantial (I2 = 61%, P = 0.04). Both the UKPDS and UGDP included participants with mild glycaemic disturbances. Excluding these trials, the effect estimate still showed significant values (random RR 0.74, 95% CI 0.62 to 0.89; P = 0.002). Heterogeneity was substantial (I2 = 56%, P = 0.05).

Subgroup analysis stratifying the trials according to the intervention was only possible for trials exclusively dealing with glycaemic control in usual care settings. The effect estimate was significant (random RR 0.80, 95% CI 0.67 to 0.94; P = 0.008; 10,070 participants, 7 trials; Analysis 1.59: subgroup 1). Heterogeneity was substantial (I2 = 59%, P = 0.02). Trial sequential analyses for trials exclusively dealing with glycaemic control in the usual care setting showed that firm evidence for a 10% relative risk reduction of retinopathy was not established (Figure web 20; http://www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx). The absolute risk reduction was at 3% (random RD -0.03, 95% CI -0.07, to -0.00; P = 0.03; fixed RD -0.03, 95% CI -0.04 to -0.02; P < 0.0001). Exclusion of the UGDP and UKPDS trials indicated a more beneficial effect of intensive glycaemic control (random RR 0.73, 95% CI 0.58 to 0.93; P = 0.01).

Retinal photocoagulation

In the Kumamoto trial, all participants requiring photocoagulation were from the secondary intervention group (Kumamoto 2000). The VADT reported separate data for new retinal photocoagulation and any retinal photocoagulation (VADT 2009). We chose to group the measures together. The data from the ADVANCE trial were taken from the substudy (ADVANCE 2008). The ACCORD trial reported the number for photocoagulation and vitrectomy together in one publication; and the number of participants with retinopathy graded on ETDRS, vitrectomy and retinal photocoagulation in the report of the ACCORD Eye substudy (ACCORD 2008). However, the authors provided separate data on retinal photocoagulation from the participants in the ACCORD Eye substudy in an answer to a letter (Rind 2010). The UKPDS contributed most of the reported events (346 out of 751 events).

Most of the events were provided by trials with low risk of bias according to sequence generation and allocation concealment (736 participants with retinal photocoagulation out of a total of 751) (ACCORD 2008; ADVANCE 2008; Steno-2 2008; UGDP 1975; UKPDS 1998; VADT 2009). Targeting intensive glycaemic control showed significant reductions in retinal photocoagulation (random RR 0.77, 95% CI 0.61 to 0.97; P = 0.03; 11,212 participants, 8 trials; Analysis 1.60; lower risk of bias). Heterogeneity was moderate (I2 = 43%, P = 0.10). Trial sequential analyses for all trials showed that firm evidence for a 10% relative risk reduction of retinal photocoagulation was not established (Figure web 21; http://www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx). The RD using the random-effects model was not statistical significant (random RD -0.01, 95% CI -0.03 to 0.00; P = 0.15), but statistical significance were present in the fixed-effect model (fixed RD -0.02, 95% CI -0.03 to -0.01; P = 0.003).

Stratifying the trials according to the intervention showed no statistical significance of the effect estimate when applying the random-effects model but a statistical significance when applying the fixed-effect model to trials exclusively dealing with glycaemic control in usual care settings (random RR 0.82, 95% CI 0.65 to 1.03; fixed RR 0.82, 95% CI 0.71 to 0.95; P = 0.008; 10,982 participants, 6 trials; Analysis 1.61: subgroup 1). Heterogeneity was present (I2 = 38%, P = 0.15). Trial sequential analyses for trials exclusively dealing with glycaemic control in the usual care setting showed that firm evidence for a 10% relative risk reduction of retinal photocoagulation was not established (Figure web 22; http://www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx).

Adverse events

We divided the reporting of adverse events into the following types: serious adverse events, non-serious adverse events, dropouts due to adverse events (Analysis 1.62), and hypoglycaemia (Analysis 1.67) (see 'Adverse events', Appendix 8). The reporting of serious adverse events was very heterogeneous. The funnel plot showed asymmetry for serious adverse events and dropouts due to adverse events (Analysis 1.62).

Two trials reported non-serious adverse events; one as adverse effects of angiotensin-converting enzyme inhibitor and simvastatin treatment (Steno-2 2008) and the other as treatment emergent adverse events (Blonde 2009). None of the trials reported adverse events as any untoward medical occurrence in participants (ICH 1997). The data were primarily derived from Blonde et al, which had unclear sequence generation, allocation concealment, and blinding of our subjective outcomes. Combining the data from the two trials did not show any statistical significance of the effect estimate (RR 1.18, 95% CI 0.87 to 1.60; 403 participants, 2 trials; Analysis 1.62; lower risk of bias).

Some trials reported cardiovascular complications of T2D as serious adverse events whereas other trials had complications of T2D as an outcome and did not report them as serious adverse events. The reported measure of serious adverse events for the ADVANCE trial was hospitalisation for more than 24 hours for any cause (ADVANCE 2008). More participants in the intensive intervention group were hospitalised for any cause (44.9% versus 42.8%) with some of the excess hospitalisations due to severe hypoglycaemia (1.1% versus 0.7%). The ACCORD reported any non-hypoglycaemic serious adverse event (ACCORD 2008). Apparently, this did not include all serious adverse events according to the ICH definition (the number of participants reported with non-hypoglycaemic serious adverse events were lower than the number of participants who died) (ACCORD 2008). The numbers from Bagg et al and Jaber et al included cardiovascular events and patients undergoing surgery (Bagg 2001). Blonde et al reported serious adverse events without further specification (Blonde 2009). The data for serious adverse events for the UGDP trial were hospitalisations for cardiovascular disease (UGDP 1975). The VADT trial applied a more broad definition: life threatening, death, hospitalisation, disability or incapacity, cancer, or other important events requiring medical intervention and treatment. The number in the meta-analysis was without hypoglycaemic serious adverse events (VADT 2009). The serious adverse event from the Steno 2 trial was for a bleeding ulcer: "One patient in the intensive-therapy group was hospitalised for a bleeding gastric ulcer. Otherwise, no major adverse events were reported" (Steno-2 2008). Six trials had as an inclusion criterion admission to hospital for coronary heart disease, and all participants were hospitalised as part of the inclusion criteria (Cao 2011; DIGAMI 2 2005; Fantin 2011; IDA 2009; Melidonis 2000; Natarajan 2012; Stefanidis 2003). For these trials we reported serious adverse events other than the 'mandatory' hospitalisation.

The reported number of serious adverse events in 'Data and analyses' included hospitalisation (Analysis 1.62). In the 'Adverse events' appendix (Appendix 8) hospitalisation was reported separately. Most of the events were provided by trials with low risk of bias according to sequence generation and allocation concealment (5376 participants with myocardial infarction out of a total of 5412) (ACCORD 2008; ADVANCE 2008; DIGAMI 2 2005; Fantin 2011; Steno-2 2008; UGDP 1975; VADT 2009). The risk of serious adverse events was significantly higher when targeting intensive glycaemic control (random RR 1.06, 95% CI 1.02 to 1.10; P = 0.007; random RD 0.01, 95% CI -0.00 to 0.02; fixed RD 0.01, 95% CI 0.00 to 0.02; P = 0.003; 24,280 participants, 11 trials; Analysis 1.62; lower risk of bias). Heterogeneity was absent (I2 = 0%, P = 0.77). Trial sequential analyses for all trials showed firm evidence for a 10% relative risk increase of serious adverse events in favour of conventional glycaemic control (Figure web 23; http://www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx). The number of serious adverse events was primarily driven by the reported number from the ADVANCE trial (4882 out of 5412). Substracting the number of participants hospitalised due to severe hypoglycaemia in the ADVANCE trial did not change the statistical significance (random RR 1.05, 95% CI 1.01 to 1.09; P = 0.02).

Serious adverse events were stratified according to intervention. The effect estimate for the trials exclusively dealing with glycaemic control in usual care settings showed statistical significance in favour of conventional glycaemic control (random RR 1.06, 95% CI 1.02 to 1.10; P = 0.007; 23,927 participants, 7 trials; Analysis 1.63: subgroup 1). Heterogeneity was absent (I2 = 0%, P = 0.43). Trial sequential analyses for trials exclusively dealing with glycaemic control in the usual care setting showed firm evidence for a 10% relative risk increase of serious adverse events in favour of conventional glycaemic control (Figure web 24; http://www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx). However, subtracting the number of participants hospitalised with severe hypoglycaemia in the ADVANCE trial changed the effect estimate to non-significant values (random RR 1.06, 95% CI 1.0 to 1.13; P = 0.07). Two trials assessing glycaemic control as part of an acute intervention in patients with T2D contributed data (Melidonis 2000; Stefanidis 2003). Both trials were relatively small and only reported a few events. There was no statistical significance of the effect estimate (random RR 0.95, 95% CI 0.41 to 2.18; 123 participants, 2 trials; Analysis 1.63: subgroup 2). The test of interaction between trials exclusively dealing with glycaemic control in usual care settings and trials assessing glycaemic control as part of an acute intervention showed no significance. It was not possible to separately analyse data on glycaemic control initiated with surgical intervention and multimodal intervention in the usual care setting.

No statistically significant difference in the effect estimate was present for dropouts due to adverse events (random RR 1.47, 95% CI 0.84 to 2.56; 12,989 participants, 11 trials; Analysis 1.62). Dropouts due to adverse events also showed no statistical significant difference between the interventions in trials exclusively dealing with glycaemic control in usual care settings (random RR 1.50, 95% CI 0.80 to 2.79; I2 = 0%; 12,636 participants, 7 trials; Analysis 1.64; lower risk of bias).

Congestive heart failure

In the REMBO trial all participants had heart failure at inclusion, and the reported measure was therefore progression to non-compensated heart failure (REMBO 2008). The number from the DIGAMI 2 trial was from the extension period of the trial and was the number of deaths due to heart failure (DIGAMI 2 2005). There was no statistically significant difference between the interventions (random RR 0.98, 95% CI 0.87 to 1.10; 29,815 participants, 12 trials; Analysis 1.65; lower risk of bias).

Trials dealing exclusively with glycaemic control in usual care settings did not show any significant difference in the effect estimate for congestive heart failure (random RR 1.01, 95% CI 0.87 to 1.17; 27,587 participants, 6 trials; Analysis 1.66: subgroup 1). Glycaemic control as part of an acute intervention showed no statistical significance (random RR 0.68, 95% CI 0.34 to 1.37; 903 participants, 3 trials; Analysis 1.66: subgroup 2). The test of interaction showed no statistical significance between the subgroups. Because of lack of data we could not analyse congestive heart failure in the trials assessing glycaemic control initiated with surgical intervention and multimodal intervention in usual care settings.

Hypoglycaemia

We predefined assessing hypoglycaemia as mild (controlled by patient), moderate (daily activities interrupted but self-managed), or severe (requiring assistance).

The definition of mild hypoglycaemia varied among trials (Appendix 7). The ACCORD trial did not systematically collect the number of mild hypoglycaemic episodes but the participants randomised to intensive glycaemic control had more episodes of mild hypoglycaemia compared with the conventional intervention group (correspondence, Bonds, ACCORD 2008). The participants in the ACCORD trial reported the number of blood sugar levels < 3.9 mmol/L based on a finger stick measure before each visit. The investigators did not report on whether these episodes of hypoglycaemia were mild or severe. The DIGAMI 2 trial reported hypoglycaemia with or without symptoms. We have reported the data on hypoglycaemia with symptoms. The number was only reported for the initial 24 hours. The DIGAMI 2 trial did not report nor define the severity of the observed hypoglycaemia (DIGAMI 2 2005). The definition of a hypoglycaemic blood glucose level was < 3 mmol/L. The number of mild hypoglycaemic episodes reported for the UGDP trial was estimated from participants who changed their prescription one or more times during the follow-up because of reported (suspected or definite) hypoglycaemic episodes. Hypoglycaemia was not graded in the UGDP (UGDP 1975). The number of hypoglycaemic episodes in Stefanidis et el was only reported for the participants who completed the trial (Stefanidis 2003). Araki et al defined and assessed mild hypoglycaemia, but no data were reported (Araki 2012) (Appendix 11). Cooray et al also defined and registered hypoglycaemia, but no results were reported (Cooray 2011) (Appendix 11). For all the trials the blinding of assessment of mild hypoglycaemia was classified as high risk of bias as it was reported by the participants, who were unblinded to the intervention (Figure 3).

The risk of mild hypoglycaemia was significantly higher for participants randomised to targeted intensive glycaemic control (random RR 1.54, 95% CI 1.35 to 1.75; P < 0.00001; 19,411 participants, 15 trials; Analysis 1.67; high risk of bias). Heterogeneity was considerable (I2 = 85%, P < 0.00001). Trial sequential analyses for all trials showed firm evidence for a 10% relative risk increase of mild hypoglycaemia in favour of conventional glycaemic control (Figure web 25; http://www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx).

Analysing of mild hypoglycaemia for the trials exclusively dealing with glycaemic control in the usual care setting showed a statistically significant effect estimate in favour of conventional glycaemic control (random RR 1.58, 95% CI 1.37 to 1.81; P < 0.00001; 17,860 participants, 9 trials; Analysis 1.68). Heterogeneity was considerable (I2 = 89%, P < 0.00001). Trial sequential analyses for trials exclusively dealing with glycaemic control in the usual care setting showed firm evidence for a 10% relative risk increase of mild hypoglycaemia in favour of conventional glycaemic control (Figure web 26; http://www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx). Trials with intensive glycaemic control as part of an acute intervention showed no statistically significant effect estimate in the random-effects model but a statistically significant effect estimate favouring conventional glycaemic control in the fixed-effect model (random RR 2.13, 95% CI 0.83 to 5.50; fixed RR 1.81, 95% CI 1.03 to 3.17; P = 0.04; 903 participants, 3 trials; Analysis 1.68). Heterogeneity was substantial (I2 = 54%, P = 0.11). The test of interaction showed no statistical significance. Due to lack of data, we could not perform separate analyses of trials assessing glycaemic control with a intervention and a multimodal intervention in usual care settings.

Due to a lack of data, we included the moderate hypoglycaemic events in the reporting of mild hypoglycaemic events as this was how the rest of the included trials reported on this outcome.

The definition of severe hypoglycaemia varied among the included trials (Appendix 7). The ACCORD trial reported severe hypoglycaemia in two ways: requiring any assistance, and requiring medical assistance. We have reported the number requiring any assistance as this definition agreed best with the definition used in the other included trials (ACCORD 2008). Nine trials, besides the ACCORD trial, described the assistance of a third person in their definition of serious hypoglycaemia (ADVANCE 2008; Araki 2012; Bagg 2001; Blonde 2009; Fantin 2011; Kumamoto 2000; Steno-2 2008; UKPDS 1998; VA CSDM 1995). Araki et al defined and assessed severe hypoglycaemia, but no data were reported (Araki 2012) (Appendix 11). The VADT trial reported severe hypoglycaemia as a serious adverse event: hypoglycaemia that was life threatening, hospitalisation, disability, death or medical assistance (VADT 2009). Cao et al defined severe hypoglycaemia according to blood glucose levels, where a value of 2.2 mmol/L or less was severe (Cao 2011). Five trials reported severe hypoglycaemia but did not specify it further (IDA 2009; Jaber 1996; Melidonis 2000; Stefanidis 2003; Zhang 2011).

Severe hypoglycaemia was significantly more frequent when targeting intensive glycaemic control (random RR 2.18, 95% CI 1.53 to 3.11; P < 0.0001; random RD 0.03, 95% CI -0.00 to 0.06; fixed RD 0.05, 95% CI 0.05 to 0.06; P < 0.00001; 28,127 participants, 12 trials; Analysis 1.67; high quality of the evidence (GRADE); see Summary of findings table 1). Heterogeneity was substantial (I2 = 66%, P = 0.001). When applying trial sequential analysis to severe hypoglycaemia for all trials a relative risk increase of 30% (number needed to harm = 50) was applied to construct the trial sequential monitoring boundary. The cumulative Z-curve crossed the trial sequential monitoring boundary, indicating that there was firm evidence for a 30% increase in severe hypoglycaemia with intensive glycaemic control (Figure web 27; //www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx). For the meta-analysis of RD, heterogeneity was 97% (P < 0.00001). Calculating the RD from risk ratio manually and thereby giving the trials with zero events no weight in the calculation of the RD showed statistical significance in the random-effects model. Inspection of the funnel plot for severe hypoglycaemia showed asymmetry, suggesting presence of bias in favour of conventional glycaemic control (Analysis 1.67). Separate analysis of the trials providing a specific definition of severe hypoglycaemia did not alter the significance of the effect estimate (random RR 2.10, 95% CI 1.45 to 3.03; P < 0.0001). Heterogeneity was still substantial (I2 = 72%, P = 0.0005).

The subgroup analysis of severe hypoglycaemia for the trials exclusively dealing with glycaemic control in usual care settings showed a significant effect estimate (random RR 2.44, 95% CI 1.78 to 3.35; P < 0.00001; 28,082 participants, 10 trials; Analysis 1.69: subgroup 1). Heterogeneity was substantial (I2 = 62%, P = 0.02). The cumulative Z-curve crossed the trial sequential monitoring boundary, indicating that there was firm evidence for a 30% increase in severe hypoglycaemia with intensive glycaemic control (Figure web 28; http://www.ctu.dk/publications/supplementary-material/hemmingsen_2013.aspx). Separate analysis of the trials providing a specific definition of severe hypoglycaemia did not change the effect estimate (random RR 2.40, 95% CI 1.74 to 3.31; P < 0.00001). Separate analysis of the trials initiating glycaemic control with a surgical intervention did not show statistical significance (random RR 4.67, 95% CI 0.81 to 26.84; 429 participants, 4 trials; Analysis 1.69: subgroup 3). It was not possible to conduct subgroup analyses for the trials with glycaemic control as part of an acute intervention and multimodal intervention in the usual care setting.

Meta-regression with data from all included trials showed a positive correlation (P = 0.02) between the relative risk ratio (RR) for severe hypoglycaemia and the duration of disease at baseline, suggesting a higher RR (higher relative risk increase) for a higher average duration of disease. The risk of severe hypoglycaemia was not dependent on the fasting blood glucose at baseline or HbA1c at baseline. A negative correlation between the RR for severe hypoglycaemia and the duration of the intervention was found suggesting a lower RR (less relative risk increase) with longer duration intervention (P = 0.00002). Meta-regression for the trials exclusively dealing with glycaemic control in usual care settings could only include information from seven trials. A significant positive correlation was found between disease duration at baseline (P = 0.06) and severe hypoglycaemia, indicating RR increases with longer disease duration at baseline. Fasting blood glucose at baseline also showed a statistically significant positive correlation (P = 0.04). A negative correlation was shown for the duration of intervention (P = 0.04) and the risk ratio of severe hypoglycaemia, indicating that the risk ratio decreased the longer the intervention. No statistical significance was present for the correlation of HbA1c at baseline and the RR of severe hypoglycaemia.

Health-related quality of life and assessment of well-being

Eight trials reported health-related quality of life or well-being (ACCORD 2008; ADDITION-Europe 2011; Becker 2003; Jaber 1996; REMBO 2008; UKPDS 1998; VA CSDM 1995; Zhang 2011) (Appendix 13). However, the quality of life data from ADDITION-Europe was only published for two subgroups: the participants embedded in the ADDITION-Leicester and the ADDITION-Netherlands (ADDITION-Leicester; ADDITION-Netherlands). We meta-analysed health-related quality of life as overall health-related quality of life and in two components: physical component of health-related quality of life and mental component of health-related quality of life (well-being). The physical component refers to physical, pain, and self-rated health. The mental health component refers to psychological distress, social and role disability as a result of emotional problems. These two components are based on the structure of the 36-Item Short Form Health Survey (SF-36) (Ware 1998). However, as not all trials assessed quality of life based on the SF-36, we evaluated whether the scales that were applied were applicable to the mental health component or the physical health component of the SF-36 quality of life assessment.

Of the 10,251 participants enrolled in the ACCORD trial, a randomly selected subgroup of 2053 participants were enrolled for assessment of quality of life (ACCORD 2008). The ACCORD trial reported quality of life using four different scales, therein the SF-36 (Appendix 13). We applied the SF-36 physical and mental components in the meta-analyses (ACCORD 2008) (Appendix 13). The results were reported as change across all visits from baseline until 48 months (ACCORD 2008). The participants were also evaluated at 12 and 36 months. The ADDITION-Netherlands trial applied four different questionnaires (Appendix 13), therein the SF-36 questionnaire (ADDITION-Netherlands). The results were the change from baseline to three years of intervention. The ADDITION-Leicester applied two different scales to assess quality of life (ADDITION-Leicester) (Appendix 13). One of these scales was the Short Form-12, which is an abbreviated form of the SF-36. The data were reported as the end of follow-up value after one year of intervention (ADDITION-Leicester). The design effect of the clustering could not be assessed for the quality of life outcome, so the number of participants was not corrected for a potential clustering effect (ADDITION-Leicester; ADDITION-Netherlands). Through correspondence it could not be clarified if the data were corrected for clustering or not. The UKPDS reported quality of life in two different set-ups: in a cross-sectional study and in a longitudinal study (UKPDS 1998). There was no overlap between the patients filling in the specific questionnaire in the cross-sectional study and the longitudinal study, so data for both set-ups were included in the meta-analyses. Quality of life was assessed by a specific questionnaire and the EQ-5D questionnaire in the cross-sectional study, and only by the specific questionnaire in the longitudinal study (Appendix 13). For the cross-sectional study the data for the meta-analysis for overall quality of life were assessed by the EQ-5D scores and the mental quality of life with the score from the mood-stage assessment in the specific questionnaire. As the mood scale stage was going in the opposite direction from the other quality of life scales, the observed change was multiplied by -1. For the longitudinal study the data for the meta-analysis from the physical-related quality of life was the symptoms score from the specific questionnaire and mental-related quality of life was the mood-stage from the specific questionnaire. The data from the longitudinal assessment were change from baseline to after six years (UKPDS 1998). As the scales for the mood stage as well as symptoms were going in the opposite direction to the other quality of life scales included in our meta-analysis, we multiplied the change from baseline by -1. The UKPDS also performed quality of life analyses to assess the influence of complications related to T2D and their influence on quality of life. The conclusion from these analyses was that diabetes-related complications reduce the quality of life (UKPDS 1998). The VA CSDM assessed quality of life by applying the 20 question version of the medical outcome study instrument (VA CSDM 1995) (Appendix 13). The participants were assessed at baseline and after two years of follow-up. For the physical component of the health-related quality of life meta-analysis data for physical function was applied, and for the mental component of health-related quality of life the data from the mental health component of the questionnaire, were applied to the meta-analysis (VA CSDM 1995).

Five trials reported quality of life in a way that made the data inapplicable for the meta-analyses (Becker 2003; Jaber 1996; REMBO 2008; Steno-2 2008; Zhang 2011). The publication from van der Does et al (included in the Becker trial) assessed quality of life (Becker 2003) (Appendix 13). The end of follow-up values were only reported according to the achieved decrease in HbA1c and not according to the groups the participants were randomised into. However, it was reported that the conventional glycaemic control group experienced better quality of life compared with intensive glycaemic control (Becker 2003). Neither Jaber et al nor Zhang et al reported any scores, but just no statistical significant difference between the intervention groups (Jaber 1996; Zhang 2011) (Appendix 11). The Steno-2 trial reported health-related quality of life as quality-adjusted life expectancy (Steno-2 2008). These values were based on simulated implications of the complications in each intervention group. These data were expressed as quality-adjusted life years, which were expressed as a combined economic and quality of life evaluation. This assessment was very different from the other assessments and was therefore not included in the analysis. The REMBO trial applied the 'Minnesota Living With Heart Failure Questionnaire' scale, which is validated in patients with heart failure (Appendix 13). The REMBO trial only reported the physical component of the Minnesota Living With Heart Failure Questionaire scale (REMBO 2008). This scale is not comparable with the SF-36 and the data were therefore not included in the meta-analyses.

The ADVANCE and VADT trials had health-related quality of life for the intervention groups as a predefined outcome (ADVANCE 2008; VADT 2009). The results of these analyses are not yet available.

The percentage of responders included in the analysis of quality of life, as well as the follow-up period, varied among trials. For a more detailed description, please see Appendix 13.

Overall health-related quality of life, measured by the EQ5D did not show any statistical significance (MD 0.00; 95% CI -0.02 to 0.02; 2776 participants, 2 trials; Analysis 1.70)

Mental component of health-related quality of life did not show any statistically significant influence of intensive glycaemic control, neither in the random-effects model nor the fixed-effect model, and neither when meta-analysed with the change from baseline or end of follow-up values (change from baseline: standardized mean difference (SMD) (SMD 0.04, 95% CI -0.09 to 0.18; 906 participants, 3 trials; end of follow-up values: SMD -0.09, 95% CI -0.19 to 0.01; 1742 participants, 4 trials; Analysis 1.71; high risk of bias). Heterogeneity for the meta-analysis of change from baseline score was 5% (P = 0.35), and heterogeneity was absent for the end of follow-up meta-analysis (P = 0.61). Sensitivity analysis excluding the ACCORD trial, which based the change from baseline across all visits from baseline to 48 months, including measurement at 12 and 36 months, did not change the statistical significance of mental-related quality of life. Exclusion of the cluster-randomised data (ADDITION-Leicester; ADDITION-Netherlands) also did not change the statistical significance of the meta-analysis, neither for the change from baseline nor the end of follow-up value.

Physical component of health-related quality of life did not show any statistical significance in either the random-effects model or the fixed-effect model for the meta-analyses of changes from baseline, nor the end of follow-up values (SMD -0.06, 95% CI -0.21 to 0.08; 743 participants, 2 trials; SMD -0.02, 95% CI -0.16 to 0.11; 813 participants, 3 trials; Analysis 1.72; high risk of bias).

Costs of intervention

Costs of the interventions were assessed in only three trials (Kumamoto 2000; Steno-2 2008; UKPDS 1998). The Kumamoto trial and the UKPDS trial analysed the cost as cost-effectiveness analyses with a 3% (Kumamoto) and 3.5% (UKPDS) annual discounting rate. The Kumamoto trial provided data on the costs of 10 years of intervention and treatment of complications per participant. The UKPDS reported the data for the UKPDS 33 and UKPDS 34 separately; as for all other outcomes we combined the data for the intensive intervention groups in UKPDS 33 and UKPDS 34. The costs for the UKPDS were expressed as cost per participant during the trial period of 10 years. There was an incremental cost of intensive blood glucose control with insulin and sulphonylurea compared with conventional glycaemic control. The costs of intensive blood glucose control with metformin were lower compared with conventional blood glucose control (UKPDS 1998). The Kumamoto trial classified the costs into two classes: costs of treatment, and costs of the complications. Costs of treatment were significantly higher for patients in the intensive intervention group compared to the conventional treatment group. The costs of complications were higher in the conventional group. When combining the costs of treatment and complications the costs were reduced in the intensive treatment group during the 10 years. When discounting the costs at 3% the difference was still present but not statistically significant (Kumamoto 2000). The Steno-2 trial found that lifetime direct medical costs were higher for the intensive treatment group compared to the conventional treatment group because of increased pharmacy utilisation and consultations when targeting intensive control. When including the lifetime expenses for treating the complications in the two intervention groups, intensive treatment was less expensive than conventional treatment even though the patients lived longer in the intensive treatment group. However, data were not presented in monetary units (Steno-2 2008). It was not possible to add suitable data from the Steno-2 trial to the meta-analysis as the data were expressed as direct medical lifetime costs combined with quality-adjusted life expectancy, which differed from Kumamoto and UKPDS (Kumamoto 2000; UKPDS 1998). As the Kumamoto trial reported the costs as USD per participant and the UKPDS reported GBP per participants the data were meta-analysed applying standardized mean difference (SMD). In a meta-analysis of the results from the Kumamoto trial and UKPDS, there was no statistically significant difference (SMD 0.05, 95% CI -0.02 to 0.12; 4319 participants, 2 trials; Analysis 1.73; lower risk of bias).

The UKPDS and Steno-2 trials also expressed the costs as quality-adjusted life years, a measure of both increases in life expectancy and quality of life. The Steno-2 trial showed lower costs per quality-adjusted life year when targeting intensive control compared with conventional control. UKPDS also found a reduced cost per quality-adjusted life year for the participants randomised to intensive glycaemic control with metformin compared with conventional glycaemic treatment. However, there was an incremental cost per quality-adjusted life year gained for intensive blood glucose control with insulin and sulphonylurea compared with conventional glycaemic control.

The ADVANCE trial reported the costs of diabetes complications from centres in countries with different income profiles (low, middle, high) (ADVANCE 2008). However, costs according to the intensive versus conventional glycaemic control was not reported.

The ACCORD, ADVANCE, and VADT trials all included cost analysis as a predefined outcome (ACCORD 2008; ADVANCE 2008; VADT 2009). The results are not published yet.

Discussion

Summary of main results

This updated Cochrane review is the first systematic review that includes all randomised trials assessing targeted intensive glycaemic control versus conventional glycaemic control in patients with type 2 diabetes mellitus (T2D) (Hemmingsen 2011). We included data from 28 trials with a total of 34,912 participants, which is 16% more than the previous version of our review (Hemmingsen 2011). Unlike most other systematic reviews, our systematic review has only focused on the randomised allocation of participants to different glycaemic targets and not the achieved glycaemic levels.

Our key findings are that there is no statistically significant difference between the interventions regarding all-cause mortality or cardiovascular mortality (see Summary of findings table 1). Other important findings are that targeting intensive glycaemic control may reduce the risk of non-fatal myocardial infarction, amputation of a lower extremity, nephropathy, retinopathy, and retinal photocoagulation. However, a firm conclusion will have to await further trials for these outcomes as they might be influenced by the risk of bias of the trials and the risk of random errors. Targeting intensive glycaemic control reduces microvascular complications as a composite outcome, The risk of random errors seems low for this outcome since a 10% relative risk reduction was confirmed in the trial sequential analysis. However, all trials reporting this outcome were judged as high risk of bias on one or more components. Targeting intensive glycaemic control increased the risk of serious adverse events as well as mild and severe hypoglycaemia. These effects seem certain as they are not influenced by the risk of bias or random errors.

Subgroup analyses stratifying the trials according to how the intervention was applied showed no statistically significant difference for all-cause mortality or cardiovascular mortality. Targeting intensive glycaemic control in trials exclusively dealing with glycaemic control in usual care settings may reduce the risk of non-fatal myocardial infarction, whereas this was not shown for intensive glycaemic control as part of an acute intervention in patients with T2D. The risk of the composite microvascular complications, retinopathy, as well as retinal photocoagulation might also be reduced when targeting intensive glycaemic control in trials exclusively dealing with glycaemic control in usual care settings but with increased risk of serious adverse events and mild and severe hypoglycaemia.

Our primary outcomes were all-cause mortality and cardiovascular mortality. Neither a random-effects nor fixed-effect model showed any statistically significant effect estimates of all-cause mortality or cardiovascular mortality when analysing all trials together or according to how the intervention was applied. The test of interaction for subgroup differences also did not show any statistical significance. Separate meta-analysis for cardiovascular mortality with intensive glycaemic control initiated with surgical intervention could not be performed due to lack of data. Stratifying the trials according to risk of bias, study duration, diagnostic criteria, or funding source did not give rise to significant effect estimates for all-cause mortality. A test of interaction between any of the subgroups also did not reveal any statistical significance. The same was the case for cardiovascular mortality, however stratifying the trials for the diagnostic criteria showed a significant effect estimate for cardiovascular mortality in favour of conventional glycaemic control. A test of interaction between the subgroups when stratifying the trials according to the diagnostic criteria for T2D showed significance (P = 0.006). However, it should be noted that stratifying trials according to diagnostic criteria excluded the ADVANCE trial since this trial did not specify its criteria for the diagnosis of T2D. The ADVANCE trial was the largest trial included in the present meta-analysis (11,140 participants) with about one third of the total information size, and it did not find any evidence of increased cardiovascular mortality when targeting intensive versus conventional control (ADVANCE 2008). Thus, excluding the ADVANCE trial might substantially increase the weight of other studies in the analysis. For example, the ACCORD trial had about the same sample size as the ADVANCE trial but, unlike the ADVANCE trial, its findings suggested an increased risk of cardiovascular death with targeted intensive versus conventional glycaemic control (ACCORD 2008). Meta-analysis of all available hazard ratio data for the primary outcomes did not show any statistically significant effect estimates. Available case analysis showed no statistically significant effect estimates for all-cause mortality or cardiovascular mortality. Worst-case scenarios showed a significant effect favouring conventional glycaemic control for all-cause and cardiovascular mortality. However, significant effect estimates favouring intensive glycaemic control were also shown in both random-effects and fixed-effect model for the best-case scenarios for all-cause mortality and cardiovascular mortality. This implies that missing outcome data in trials could considerably influence the effect estimates for targeting intensive versus conventional glycaemic control on all-cause and cardiovascular mortality, although the unrealistic assumptions reveal very different effect estimates. However, the direction of such influence is uncertain.

Trial sequential analysis suggested a 10% relative risk reduction could be rejected for all-cause mortality. For cardiovascular mortality trial sequential analysis suggested that more trials are needed before firm evidence for a lack of a 10% relative risk reduction is established. Meta-regression for all trials indicated a possible positive association between disease duration and all-cause mortality (P = 0.10). For the remaining variables explored in the meta-regressions and all-cause mortality, no statistically significant associations were found. Meta-regressions for the subgroup of trials exclusively dealing with glycaemic control in usual care settings showed a positive correlation between fasting blood glucose and HbA1c at baseline and the risk ratio for all-cause mortality. Thus, for trials exclusively dealing with glycaemic control in usual care settings, patients with poorer glycaemic control at baseline (higher fasting blood glucose or HbA1c) might benefit less from targeting intensive versus conventional glycaemic control in terms of all-cause mortality than do patients with better glycaemic control at baseline. Neither meta-regression of all trials nor the subgroup analysis of trials exclusively dealing with glycaemic control in usual care settings showed any significant influence on cardiovascular mortality for the explored variables.

We found a statistically significant influence of the intervention on macrovascular complications assessed as a composite outcome in the fixed-effect model. Separate analysis of trials exclusively dealing with glycaemic control in the usual care setting showed statistically significant effect estimates in the fixed-effect model only. The subgroup of trials with multimodal intervention in the usual care setting did not show any statistical significance, neither in the random-effects model nor the fixed-effect model. Subgroup analyses for trials assessing the effect of targeting intensive glycaemic control as part of an acute intervention or initiated with surgical intervention could not be performed. The reporting of a composite macrovascular outcome varied between trials.

Data for non-fatal myocardial infarction were combined for 14 trials, of which 11 gave a detailed description of how the diagnosis was established. Meta-analysis of all 14 trials revealed a statistically significant effect estimate in favour of intensive glycaemic control. The fact that we were only able to get data on myocardial infarction in about 50% of the trials opens up the risk of outcome reporting bias. When analysing non-fatal myocardial infarction in trials exclusively dealing with glycaemic control in usual care settings, statistically significant effect estimates were still present in favour of intensive glycaemic control. The trials assessing the effect of targeting intensive glycaemic control as part of an acute intervention all reported non-fatal myocardial infarction as re-infarction. There was no significant effect estimate for the trials assessing targeted intensive glycaemic control as part of an acute intervention. The trials assessing the effect of targeting intensive glycaemic control as part of a multimodal intervention in the usual care setting did not show any statistical significance for non-fatal myocardial infarction. Due to the lack of data, subgroup analysis of trials with glycaemic control initiated with surgical intervention could not be performed. The test of interaction between the subgroup according to the setting in which glycaemic control was applied did not show any statistical significance. Subgroup analysis according to study duration, risk of bias, and industry funding showed no statistical significance in the test of interaction. Available case analysis showed a significant effect estimate favouring intensive glycaemic control. A best-case scenario showed significant effect estimates favouring intensive glycaemic control. A worst-case scenario favoured conventional glycaemic control. Trial sequential analysis, however, showed that more trials are needed before there can be firm evidence for a benefit of intensive glycaemic control, or lack of effect. Meta-regressions for all trials and the trials exclusively dealing with glycaemic control in usual care settings showed no significant association between the risk ratio of non-fatal myocardial infarction and the explored variables.

Originally we planned to report stroke according to the aetiology, but unfortunately this was not possible because of the reporting in the included trials. Stratifying the trials according to the intervention was possible for the trials exclusively dealing with glycaemic control in usual care settings and in the trials assessing the effect of targeting intensive glycaemic control as part of a multimodal intervention in the usual care setting, which did not show any statistical significance of the effect estimate. The reported non-fatal strokes were primarily from the ADVANCE trial. The result remained non-significant when analysing only the trials with predefined non-fatal stroke as a primary outcome.

A statistically significant effect estimate in favour of targeting intensive glycaemic control was evident for amputation of a lower extremity. Stratifying the trials according to the intervention the trials exclusively dealing with glycaemic control in usual care settings did not show a significant effect estimate. The trials assessing the effect of targeting intensive glycaemic control as part of a multimodal intervention in usual care settings showed statistical significance of the effect estimate. Differences in the definitions used for this outcome and the indication for amputation might vary within the different sites of a single trial. The data on amputation were primarily reported by the UKPDS. Trial sequential analysis showed that only a minor proportion of the required sample size has been accrued so far.

Cardiac revascularization was not influenced by the intense or conventional intervention. Subgroup analysis for the trials exclusively dealing with glycaemic control in usual care settings did not show any statistically significant effect of the intervention; most of the reported data were from the VADT trial. The trials assessing the effect of targeting intensive glycaemic control as part of a multimodal intervention in the usual care setting showed statistical significance in favour of intensive glycaemic control. The test of interaction between subgroups did not show any statistical significance.

Targeting intensive glycaemic control did not reveal any significant influence on the need for peripheral revascularization. Subgroup analysis of the trials according to how intensive glycaemic control was applied was only possible for the trials exclusively dealing with glycaemic control in usual care settings, which did not show a significant effect estimate. The indication for revascularization procedures might vary within the sites in a single trial and among trials. The ADVANCE trial contributed the most events, which were reported as peripheral vascular events.

The relative risk of microvascular complications as a composite outcome was reduced when targeting intensive glycaemic control. Subgroup analysis stratifying the trials according to the intervention could only be done for trials exclusively dealing with glycaemic control in usual care settings, which also showed a statistically significant effect estimate in favour of targeting intensive glycaemic control. Definitions of the composite microvascular outcome varied among trials. The composite microvascular outcome from the Steno-2 trial included both severe and non-severe microvascular events; whereas the ACCORD, ADVANCE, and UKPDS trials reported more severe microvascular events. Because of the lack of access to data at the patient level, we were not able to analyse only severe events of microvascular disease as a composite outcome (for example, events not defined by changes in surrogate markers of disease, such as albuminuria). Trial sequential analysis suggested that firm evidence was reached for a 10% relative risk reduction when targeting intensive glycaemic control in all trials, but not in the trials exclusively dealing with glycaemic control in the usual care setting. This suggests that still more trials are needed to show conclusively if targeting intensive versus conventional glucose control results in a reduction in a composite of microvascular outcomes (surrogate markers and clinical events) in usual care settings, that is, those most closely resembling everyday clinical practice.

Meta-analysis of all trials reporting retinopathy showed that the risk of retinopathy was statistically significantly reduced. Subgroup analysis of trials exclusively dealing with glycaemic control in usual care settings also showed a statistically significant effect estimate. We reported retinopathy graded using a scale, which was the Early Treatment of Diabetic Retinopathy Study scale for most of the trials. By excluding the UGDP and the UKPDS trials, which included only participants with short duration diabetes, from the analysis of trials exclusively dealing with glycaemic control in usual care settings the effect estimate revealed a slightly larger risk reduction. This might indicate a possible benefit on retinopathy of intensive versus conventional glucose control across various stages of disease duration. Heterogeneity was substantial. Trial sequential analysis showed that more trials are needed before firm evidence for a 10% relative risk reduction is established from randomised clinical trials.

A meta-analysis of all trials using both the random-effects and fixed-effect models showed a statistically significant benefit of targeting intensive glycaemic control for retinal photocoagulation. Analysing the trials exclusively dealing with glycaemic control in usual care settings resulted in a significant effect estimate favouring targeting intensive glycaemic control only when applying the fixed-effect model. The indication for retinal photocoagulation may vary between sites in a single clinical trial as well as between the sites of the different included trials. Most of the retinal photocoagulation was reported by a single trial (UKPDS). Trial sequential analysis suggested that more trials are needed before firm evidence of a 10% relative risk reduction is reached.

A statistically significant effect estimate was shown for nephropathy for all trials. Subgroup analysis stratifying the trials according to the intervention was only possible for the trials exclusively dealing with glycaemic control in usual care settings, which was not significant. The reported nephropathy outcome from the ACCORD trial was primarily based on a reduction in glomerular filtration rate (GFR) that was observed in more than half of the participants. The definition of nephropathy varied among trials, from surrogate markers (for example, developing microalbuminuria) to hard clinical outcomes (for example, renal transplantation). Heterogeneity was considerable.

The effect estimate for end-stage renal disease showed no statistical significance. Stratifying trials according to the intervention was only possible for trials exclusively dealing with glycaemic control in usual care settings, which did not reveal statistical significance. Some trials reported end-stage renal disease and death due to renal disease as part of the nephropathy outcome. Some trials provided separate data on end-stage renal disease and nephropathy. The extractable data for end-stage renal disease varied.

The number of participants with adverse events could be included in a meta-analysis of two trials. However, none of the trials reported adverse events as recommended by the International Conference on Harmonisation (ICH 1997). The risk of serious adverse events was significantly increased. This was also the case analysing the trials exclusively dealing with glycaemic control in usual care settings. No significant effect was shown for glycaemic control as part of an acute intervention. The test of interaction between trials exclusively dealing with glycaemic control in usual care settings and the trials assessing glycaemic control as part of an acute intervention showed no statistical significance. Meta-analyses of trials initiating glycaemic control with a surgical intervention and trials with multimodal intervention in usual care setting were not possible due to lack of data. Serious adverse event reporting varied among trials, and some trials reported cardiovascular complications and severe hypoglycaemia as a serious adverse event whereas other did not. More than half of the serious adverse events were from the ADVANCE trial. Notably, the risk of serious adverse events seemed to be increased with intensive versus conventional glycaemic control even after excluding the hypoglycaemic episodes requiring hospitalisation from the serious adverse events in the ADVANCE trial (ADVANCE 2008). Trial sequential analysis showed firm evidence for a 10% relative risk increase in favour of conventional glycaemic control.

No statistically significant effect estimates were present for dropouts due to adverse events or to congestive heart failure.

The risk of mild hypoglycaemia was increased when targeting intensive glycaemic control, assessing all trials together as well as assessing trials exclusively dealing with glycaemic control in usual care settings. Trials with glycaemic control as a part of acute intervention did not show a statistically significant increase in mild hypoglycaemia in the random-effects model, but statistical significance was present in the fixed-effect model. The test of interaction between trials exclusively dealing with glycaemic control in usual care settings and the trials assessing glycaemic control as part of an acute intervention showed no statistical significance. It was not possible to analyse trials initiating glycaemic control with surgical intervention or with a multimodal intervention in the usual care setting separately due to lack of data. Definitions of mild hypoglycaemia varied among trials. The lack of blinding of the participants and the investigators might influence the reporting of mild hypoglycaemia. Heterogeneity was considerable, so the results should be interpreted extremely cautiously. Firm evidence for a 10% relative risk increase with targeting intensive glycaemic control was confirmed in the trial sequential analysis.

Severe hypoglycaemia was significantly more frequent when targeting intensive glycaemic control, both when assessing all trials together and when assessing the trials exclusively dealing with glycaemic control in usual care settings. Meta-analysis of trials initiating glycaemic control with surgical intervention did not show any statistical significance. Meta-analysis of glycaemic control as part of an acute intervention and a multimodal intervention in usual care settings could not be performed due to lack of data. A definition of severe hypoglycaemia was given for most trials providing data on this outcome. The definitions often included assistance from another person, without further specification. The grade of assistance from another person may vary from handling a juice to giving glucagon injections or in hospital admission. The design of the included trials made it impossible to blind the participants, which in turn may bias the reporting of severe hypoglycaemia. Heterogeneity was substantial, which may reflect differences in both the included trials and the definition of severe hypoglycaemia. Trial sequential analysis suggested firm evidence for a 30% relative risk increase when targeting intensive versus conventional glycaemic control. Meta-regression for all trials and the subgroup of trials exclusively dealing with glycaemic control in usual care settings showed a positive correlation between the relative risk of severe hypoglycaemia and the duration of disease, indicating that the relative risk of severe hypoglycaemia with targeted intensive glycaemic control versus conventional glycaemic control increases with longer disease duration. A negative correlation between the relative risk of severe hypoglycaemia and the duration of the intervention was found for all trials and for the subgroup of trials exclusively dealing with glycaemic control in usual care settings, indicating a lower relative risk of severe hypoglycaemia with increased duration of the intervention for targeting intensive glycaemic control versus conventional glycaemic control. Meta-regression for all trials showed no influence of the HbA1c or fasting blood glucose level at baseline on the risk of severe hypoglycaemia. Meta-regression for the trials exclusively dealing with glycaemic control in usual care settings also did not show any statistical significance for HbA1c at baseline, but a positive correlation was found for fasting blood glucose at baseline. Heterogeneity between trials was substantial and the results should be interpreted extremely cautiously.

We assessed health-related quality of life as overall health-related quality of life and in a mental component and a physical component. Different scales were used for assessment. No statistically significant differences were present. Two larger trials (ADVANCE 2008; VADT 2009) had quality of life as a predefined outcome but the results are not yet published.

Cost data from only two trials could be pooled (Kumamoto 2000; UKPDS 1998). Based on these data we could not conclude whether targeting intensive glycaemic control is economically efficient. The results might be specific to the countries in which the trials were undertaken (Japan, United Kingdom) because of differences between the public health systems.

Overall completeness and applicability of evidence

We conducted an extensive search for trials, included publications in all languages, and had no restriction on the outcomes reported in the trials. We have included trials with wide ranges of duration of T2D, duration of the interventions, age, and with different groups according to risk of cardiovascular disease, and different assessments of glycaemic control. Our primary objective was to assess all-cause as well as cardiovascular mortality.

The participants of the included trials represented a very diverse sampling of the population with T2D. The results of our review should therefore be interpreted with caution. The diagnosis of T2D varied among trials, and some trials used a definition of T2D which may have included participants with impaired glucose tolerance. Some of the trials only included participants with newly diagnosed T2D, whereas others included patients with a longer duration of T2D. Moreover, the cardiovascular risk profile may have differed significantly because of differences in the inclusion criteria, for example inclusion of participants with acute cardiovascular events, microvascular disease, or at high risk of cardiovascular disease. However, it should be kept in mind that participants with existing co-morbidities, especially renal or hepatic disease, were excluded from many of the included trials. Detailed information about the participants was presented in most trials. Many of the trials were conducted in Europe or Northern America. Age, body mass index (BMI), glycaemic control, and the participants' diabetes duration were in keeping with what might be expected in clinical practice. Even though we have included a wide range of patients with T2D, and due to potential selection bias, for instance more healthy and motivated patients in a clinical trial, it is difficult to say how typical the participants in each clinical trial may be compared with the wider population with T2D. However, a recently published study compared the eligibility criteria for some of the large randomised clinical trials that are included in our review (ACCORD 2008; ADVANCE 2008; UKPDS 1998; VADT 2009) to the patient characteristics of patients with T2D in Scotland (Saunders 2013). The proportion of people with T2D who met the inclusion criteria for the trials included in our review varied from 11.4% to 50.7% (ACCORD 2008; UKPDS 1998). Therefore, the external validity of our data might be low (Saunders 2013).

The glycaemic targets in the intensive and the conventional treatment groups, as well as the anti-diabetic interventions used to achieve the targets, differed among the trials. Based on the included trials, it is neither possible to estimate the 'optimal' glycaemic intervention target nor the optimal treatment regimen necessary to achieve that target. These were not part of our objectives. Thus, our review cannot provide evidence of superiority or inferiority of specific glucose-lowering regimens or of specific glycaemic targets.

Quality of the evidence

Among the 28 trials included in this analysis only two trials were classified as having low risk of bias (and they provided very little information) and only 14 trials were classified as having lower risk of bias. Stratifying the trials according to lower versus high risk of bias did not influence the effect estimates on our primary outcomes or reveal any significance in test of interaction between subgroups. We are therefore in a typical meta-analytic situation that we have very few data that we can believe in and we do not know if the lack of difference between the effect estimates in trials with lower risk of bias and trials with high risk of bias is due to bias risks in both (most likely) or none (less likely). Therefore, all positive effects should be viewed as potentially caused by or influenced by bias (systematic error overestimating benefits). This risk does not apply to our negative findings on intensive glycaemic control. We were able to assess some of the predefined outcomes in all but three of the included trials. All of the larger included trials described randomisation and allocation adequately. Because of the design of the trials comparing intensive glycaemic control with conventional glycaemic control, it was not feasible to require blinding of investigators and participants. This might have influenced the reporting from both the participants and the investigators. Reporting of hypoglycaemia, in particular, might have been prone to reporting bias. We defined blinding of outcome assessors as adequate blinding. For all trials one or more authors were contacted in order to get supplemental information on baseline data, bias domains, and outcomes. In addition, some authors were asked to confirm issues in the publication (for example, overlap of the control population in the UKPDS trial, no data on quality of life yet available from the ADVANCE trial). A total of 19 trial authors (67%) either just confirmed a question or provided additional data that could be implemented for the risk of bias assessment or the meta-analyses of outcomes. Hence, unpublished data were included for most outcomes.

Only two trials were judged as low risk of bias on all bias components (Fantin 2011; UGDP 1975). These trials contained very few data. Seventeen trials (61%) were judged as low risk of bias according to sequence generation and 12 trials were judged as low risk of bias according to allocation concealment (43%). Inadequate sequence generation and allocation concealment (selection bias) seems to exaggerate beneficial intervention effects (Savovic 2012).

Only one trial had blinding of participants (ADDITION-Europe 2011), and none had blinding of the investigators, so the risk of subjective bias and provider bias is high. The lack of blinding of participants, as well as differences in blood glucose levels between the intervention groups, might influence most outcomes. Therefore, the classification of outcomes into subjective and objective is problematic in several cases. For example, even though outcome assessors are blinded to the intervention for the participants, the ophthalmologist making the decisions about the photocoagulation might be influenced by the glycaemic level of the participants as well as the intervention group. Therefore, it might be that participants with conventional glycaemic control are more likely to undergo photocoagulation, even though the ophthalmologist is unaware of the study assignment, but the decision might be biased by the higher glucose levels. The same applies to other outcomes (for example, higher glucose levels might mean that the physician is unaware of the assignment of the patient but is biased).

Another source of bias is the difference in visit intervals of the participants in each intervention group. Several of the trials had shorter intervals between clinical visits in the intensive intervention group. This could potentially lead to other known and unknown risk factors associated with the intervention being treated differently between the groups. Besides, some outcomes, especially hypoglycaemia, might be simply reported more due to the fact that the participants have the chance to report the events more often.

Our subgroup analyses stratifying the trials according to risk of bias for each bias domain did not show any statistical significance in test of interaction. Although not statistically significant in the test for subgroup differences, the trials classified as high risk of bias almost uniformly gave a more positive effect estimate of the intervention effect in favour of intensive glycaemic control.

A relatively large proportion of the trials received funding from the pharmaceutical industry. When stratifying all-cause and cardiovascular mortality by source of funding, this did not show any statistically significant interaction between subgroups. However, recently a Cochrane review showed that trials sponsored by the industry report more beneficial intervention effects (Lundh 2012).

Certain potential limitations of this review warrant special consideration, one being that we were dealing with a very heterogeneous group of trials. The heterogeneity might to some extent be due to the differences in baseline characteristics of the participants of the included trials (for example, age, diabetes duration). This meta-analysis is limited by a lack of data at the individual level. Therefore, we cannot use individual patient data to assess whether certain characteristics (for example, history of cardiovascular events, degree of HbA1c reduction, duration of disease at baseline) affect the degree of cardiovascular risk. We explored heterogeneity by sensitivity analyses, subgroup analyses, and meta-regressions. Diagnostic criteria and definitions of outcomes differed among the trials and were not always well-defined. The anti-diabetic intervention also varied among the trials. Moreover, the outcomes we assessed were diabetic complications, both macro- and microvascular, that might have different aetiologies. The effects of intensive glycaemic control were assessed in patients with newly diagnosed T2D, participants with T2D and microvascular disease, participants with elevated risk for cardiovascular disease, and participants with T2D combined with an acute coronary event. The variable risk of developing the outcomes that we assessed might have influenced the results. We have tried to take the differences between trials into account by performing sensitivity analyses and subgroup analyses. Many of the included trials were not designed or powered to detect our predefined outcomes, which might have resulted in insufficient data from these trials. However, we included the trials irrespective of the outcomes reported. When prespecifying a certain primary outcome, the outcome might be more systematically and uniformly collected in the trial. We tried in all cases to ask for supplementary information from the authors. However, outcome reporting bias could influence the results of our meta-analysis. Adverse events outcome reporting, in particular, was lacking and varied among the trials. Reporting of insulin use at the end of follow-up and treatment of cardiovascular risk factors was especially low (Appendix 2; Appendix 12).

Not all trials reported the glucose target in the conventional group (for example, ADVANCE). Therefore we cannot discount the possibility that only small differences in targets existed between groups, potentially too small to result in significant differences in clinical outcomes.

Reporting outcomes that were not predefined in the trials gives rise to other concerns beside reporting bias. Both macrovascular and microvascular complications usually evolve over a long time period. It might therefore be that some of the included trials reported on outcomes where the duration of the trials was likely too short to influence the outcome (for example, retinopathy reported from the VA CSDM). Besides, the outcomes reported from the trials were in several cases a combination of surrogate markers and patient-important outcomes. This makes it difficult to assess the clinical importance of the outcome we report (for example, microvascular disease combined with retinal photocoagulation and end-stage renal disease). It therefore could have made sense to divide some of the assessed outcomes into more classes (Ciani 2013; Gluud 2007).

We have not evaluated the glucose-lowering drugs that were used to achieve the glycaemic target. In the included trials a wide range of glucose-lowering drugs were often used to achieve the glycaemic target. The treatment protocols for the prescription of glucose-lowering drugs were not identical for the intensive glycaemic group and the conventional glycaemic group in all trials, for example, gliclazide was prescribed for all participants in the intensive treatment group in the ADVANCE and the REMBO trials (ADVANCE 2008; REMBO 2008). Besides predefined differences in the anti-diabetic treatment, other differences might appear. In the ACCORD and the ADVANCE trials a greater proportion of the participants randomised to intensive glycaemic control received rosiglitazone compared with the conventional therapy group (ACCORD 2008; ADVANCE 2008). We have not taken such differences in anti-diabetic treatments between the intervention groups into account despite the fact that some anti-diabetic interventions are suspected of causing some of our reported outcomes. Therefore, the most suitable way to assess the objective of this review would be if all the included trials only used one glucose-lowering drug in both intervention arms to achieve glycaemic target. This was done to some extent in the DCCT study in patients with type 1 diabetes mellitus (DCCT/EDIC 2005) and in the Kumamoto trial in patients with T2D (Kumamoto 2000). However, not only did the glycaemic target differ between the intervention groups in these trials but so did the insulin regimen (for example number of daily injections) thus limiting the conclusions that can be drawn about the effect of the glycaemic target per se. A trial design that only used insulin would, however, probably not be applicable to current clinical practice for patients with T2D as a large range of glucose-lowering drugs are currently being used.

To assess whether differences in targeted or achieved glycaemic control caused differences in the investigated outcomes, the respective groups would have to be similar for every known and unknown risk factor that influences the outcome. For the glucose target this should be true at baseline; and for the achieved glycaemic control other confounders during follow-up should be controlled for. We included only randomised trials to best protect against differences in baseline variables (and, in fact, also during follow-up) that may influence the outcomes differently between intervention groups. Potential blinding of participants and investigators would also confer some protection against confounding during follow-up. Unfortunately, however, such blinding is probably not possible when investigating glucose targets. On the contrary, there are probably few, if any, possible ways of protecting against confounding influences during follow-up for the effect of the achieved glycaemic control to influence other outcomes, for example, mortality or cardiovascular risk. Short of blinded trials, we therefore believe that our approach of identifying randomised trials with different predefined glycaemic targets between the intervention groups was the best way to assess the question of possible causality between glucose control and clinical outcomes. For our review, some trials assessed multimodal intervention in usual care settings for blood pressure and cholesterol control together with intensive glycaemic control. To take these differences into account, we planned to conduct separate analysis of these trials.

The method for assessing glucose control varied between the included trials. Some trials defined the target glucose values using blood glucose. However, the levels of blood glucose only provide a 'snapshot' of the overall degree of glycaemic control. Most of the included trials expressed glycaemic control and the glycaemic goal in terms of levels of HbA1c, which are determined by the blood glucose levels over several weeks. In spite of differences in the timeline for blood glucose and HbA1c determinations, we chose to include trials irrespective of the way glycaemic control was assessed.

Potential biases in the review process

Despite an extensive search of major diabetes conference abstracts, and correspondence with authors of the included trials and relevant medical companies, we did not retrieve any additional trials. However, our funnel plots for all-cause mortality as well as cardiovascular mortality indicated that smaller trials favouring conventional glycaemic control are unpublished. However, the reason for the funnel plot asymmetry could also be due to the low quality of the smaller trials included in the meta-analyses, which might exaggerate the intervention effects. For serious adverse events, dropouts due to adverse events and severe hypoglycaemia funnel plot asymmetry were present, suggesting trials favouring intensive glycaemic control might be missing.

Some of the included trials are of a relatively small size, which increases the risk of providing a more unrealistic estimate of the intervention effects due to bias (systematic errors) and chance (random errors). We have tried to clarify systematic errors. All authors were contacted for clarification if one of the bias domains was not adequately reported. We divided the analyses for the primary outcomes into high risk of bias trials and low risk of bias trials to reveal any influence of bias on the effect estimates of our primary outcomes. To reduce the risk of random errors we have conducted trial sequential analysis on the primary outcomes and all secondary outcomes which showed significant effect estimates, applying both the random-effects and fixed-effect models.

Heterogeneity among trials was partly caused by differences in included participants among trials, intervention targets, and anti-diabetic agents used. For each outcome we made efforts to explain the cause of the heterogeneity. Moreover, we conducted all analyses using both the random-effects model and fixed-effect model. Due to large heterogeneity, we by default reported the outcomes using the random-effects model, and the fixed-effect model if the results differed. The fixed-effect model assumes that the true intervention effect is the same in every randomised trial, that is, the effect is fixed across trials. On the contrary, the random-effects model allows for the effects being estimated to differ across trials. When the heterogeneity increases, the estimated intervention effect may differ between the random-effects model and the fixed-effect model, and the confidence interval increases in the random-effects model. When there is no heterogeneity (I2 = 0%) the two models tend to give the same result. By adopting the random-effects model we were therefore able to pool a broader population of studies than by only relying on the results of the fixed-effect model. On the other hand, the random-effects model reduces the weight of the large trials, which might be more representative of a true intervention effect.

Agreements and disagreements with other studies or reviews

The oldest trial we retrieved, the UGDP, did not reveal any benefit of intensive glycaemic control compared with conventional glycaemic control (UGDP 1975). The participants in both groups were exclusively treated with insulin. At the time the UGDP was designed, there was no single definition of T2D that had general acceptance. However, the participants of the UGDP were more likely to be diagnosed with impaired glucose tolerance than diabetes according to modern diagnostic criteria. The UKPDS trial was initiated 10 to 15 years later, in 1977 (UKPDS 1998). By using the fasting plasma glucose criterion of 6.0 mmol/L, about 85% of all UKPDS patients would have fulfilled the 1985 WHO criterion for diabetes (fasting plasma glucose above 7.8 mmol/L). The findings of the UKPDS trial were more positive with respect to the effect of intensive versus conventional glucose control on complications of diabetes than the findings of the UGDP trial. Observational data from the UKPDS trial showed a 14% risk reduction of myocardial infarction for each 1% decrease in HbA1c (UKPDS-35 2000). A longer follow-up period, after the completion of the randomised UKPDS trial, revealed a reduction in all-cause mortality, myocardial infarction, and microvascular complications for all participants receiving regimens targeting intensive glycaemic control during the intervention period (UKPDS-80 2008). The participants in both the UGDP and UKPDS represented patients with T2D with relatively mild abnormalities in glucose metabolism. The data from the UGDP trial have only been included in one of another 18 meta-analyses of intensive versus conventional glucose control because of the diagnostic criteria for T2D in the trial (Boussageon 2011). Excluding UGDP from the analyses did not influence the statistical significance of our results. We predefined that the participants in the included trials should be classified as having T2D according to guidelines or the authors' definition. Therefore, the Outcome Reduction with an Initial Glargine Intervention (ORIGIN) trial was not included in this review (ORIGIN 2012) as the investigators, besides including participants with T2D, also included participants with impaired fasting glucose and impaired glucose tolerance. Only 1415 (11.3%) of the 12,537 randomised participants did not have T2D at baseline. However, as the authors of the ORIGIN trial had a more restrictive approach to establish the diagnosis of T2D this trial did not fulfil the inclusion criteria of this review, even though a larger percentage of the participants had T2D than in the UKPDS trial (ORIGIN 2012; UKPDS 1998). Sensitivity analyses including mortality data from the ORIGIN trial did not change the statistical significance of the effect estimates. Sensitivity analyses from the composite microvascular outcome, amputation of a lower extremity, and congestive heart failure did not change the statistical significance of the effect estimates achieved in the main meta-analyses. The other patient-important outcomes reported in the ORIGIN trial could not be included in sensitivity analyses due to the way of reporting (for example, any revascularization, any stroke) (ORIGIN 2012).

The Steno-2 trial reported a benefit of targeting multiple cardiovascular risk factors, including glycaemia in patients with T2D and microalbuminuria (Steno-2 2008). The intensive glucose regimen was combined with aggressively targeting other well-known risk factors of cardiovascular disease. Unfortunately, this trial was not designed to assess the influence of each component of the treatment regimen. It remains uncertain how much of the improvement was caused by intensive glucose control as an isolated target. In addition, the included participants represented a heterogeneous and relatively selected population. A longer follow-up period of the Steno-2 population indicated a possible benefit of intensive intervention for multiple risk factors, including glycaemic control, after the end of the intervention period. Like in the long-term follow-up of the UKPDS, the differences in HbA1c disappeared. The observational post-trial data from both the Steno-2 and the UKPDS trials indicate a long-term benefit of targeted intensive glycaemic control that may or may not be supported in future randomised trials. However, because of incomplete follow-up for some participants in the UKPDS post-trial analysis, and the observational design of the post-trial period, the data should be interpreted cautiously (UKPDS-80 2008).

Randomised clinical trials have shown that lipid- and blood pressure-lowering treatments reduce the prevalence of cardiovascular disease and mortality in patients with T2D (Collins 2003; Haffner 1999; Patel 2007). We could not perform separate analyses of trials assessing multimodal intervention in usual care settings for all-cause mortality and cardiovascular mortality because we only had data from the Steno-2 trial (Steno-2 2008). The benefit in the intensive intervention group that was reported in the Steno-2 trial is probably a combined effect by the aggressive approach to blood pressure control, aspirin use, and lipid lowering rather than the glycaemic control alone (Steno-2 2008). Moreover, the glycaemic targets were identical in the two interventions groups for the last two years of the intervention period.

The DIGAMI 2 trial was conducted exclusively in participants with T2D and acute coronary events (DIGAMI 2 2005). The trial was designed to answer the question of whether an intensive glucose-insulin regimen followed by intensive insulin therapy reduced mortality and cardiovascular morbidity compared with insulin-glucose infusion followed by conventional treatment, or conventional treatment alone. The first DIGAMI trial indicated lower mortality when applying intensive glycaemic control after a myocardial infarction in patients with diabetes (DIGAMI 1996). The DIGAMI 2 was an attempt to replicate and extend the findings of the first DIGAMI trial. In the DIGAMI 2 trial, the level of blood glucose ended up being identical in all treatment groups and the trial had to be stopped early due to slow patient recruitment. Other trials of smaller scale and with shorter follow-up periods were not sufficiently powered to answer the question (Melidonis 2000; Stefanidis 2003). Subgroup analyses did not show any benefit of intensive glycaemic control for the primary outcomes in the trials with glycaemic control as a part of an acute intervention.

Recently, two large trials were conducted to answer the question whether intensive glycaemic control is superior to conventional glycaemic control (ADVANCE 2008; ACCORD 2008). Worries arose as the results from the ACCORD trial in 2008 showed increased all-cause mortality and cardiovascular mortality with intensive glycaemic intervention compared with conventional glycaemic intervention. The increased mortality caused early termination of the ACCORD trial. Explanations for this finding have been sought by the authors of the ACCORD trial but no firm evidence was found. Post hoc analyses of the ACCORD trial suggest that elevated levels of baseline HbA1c (above 8.5%) influence the risk of mortality with intensive glycaemic control compared with conventional glycaemic control (Calles-Escandon 2010). Meta-regression of our data on trials exclusively dealing with glycaemic control in usual care settings showed a positive correlation between HbA1c and fasting blood glucose at baseline and the risk ratio of all-cause mortality. However, we did not find any association between baseline HbA1c and all-cause mortality using the data from all included trials. On the other hand, the ACCORD trial showed a reduction in the risk of non-fatal myocardial infarction when targeting intensive glycaemic control. It might be that the myocardial infarctions in the ACCORD trial were, for some reason, more severe and caused death. The question remains why the ACCORD trial reported increased deaths but reduced risk of non-fatal myocardial infarction. However, this is a very important clinical problem that may be difficult to solve. Recently, data from the follow-up period, after termination of the intensive glycaemic intervention arm, have been published. It was shown that the increased risk of mortality and reduced risk of non-fatal myocardial infarction persisted (ACCORD 2011). The ADVANCE trial did not find any increased mortality in the treatment arm targeting intensive glycaemic control. The reasons for the differences in the mortality results for these trials have been debated. Several differences exist between the population of the ACCORD trial and the ADVANCE trial (a slightly longer duration of T2D and more patients on insulin at baseline in the ACCORD trial), which indicate that the participants of the ACCORD might have a more progressive T2D. Besides, there was a difference in the anti-diabetic drugs prescribed to reach the glycaemic target. A larger proportion of the participants were prescribed glitazones in the ACCORD trial; in the ADVANCE trial all participants in the intensive treatment group received gliclazide. The ORIGIN trial randomised 12,537 participants to intervention with insulin glargine (targeting a fasting blood glucose of 3.5 mmol/L or less) or standard care in a factorial design with n-3 fatty acid or placebo (ORIGIN 2012). There was neutral effect on cardiovascular outcomes and cancers, but increased risk of weight gain and hypoglycaemia with intensive intervention.

The different interventions applied to achieve glycaemic control in the different trials may influence mortality, and it has specifically been debated whether the glitazones increase the risk of myocardial infarction (Nissen 2010; Singh 2007). We conducted a sensitivity analysis on non-fatal myocardial infarction by excluding the trials (ACCORD 2008; ADVANCE 2008) using more glitazones in the intensive intervention group, which changed the statistical significance of the effect estimate in favour of targeting intensive glycaemic control into not being significant, applying both the random-effects and fixed-effect models. As mentioned previously, it was not an objective of this review to assess the effect of the different anti-diabetic interventions used, and it might well be that some of the reported effects of intensive glycaemic control are due to the differences in the anti-diabetic interventions used and not to differences in the glycaemic target (for example, metformin in the UKPDS, gliclazide in the ADVANCE trial). Either way, the non-factorial design of these trials in this respect does not allow a separation of these effects. To ensure comparability between the interventions with different glycaemic targets, the number of anti-diabetic drug combinations should be limited and the treatment algorithm should be identical for both the anti-diabetic interventions as well as for cardiovascular risk factors.

Epidemiological analyses of the data from the ACCORD trial observed that severe hypoglycaemia was associated with increased risk of death irrespective of the intervention group (Bonds 2010). However, severe hypoglycaemia did not explain the increased risk of mortality in the intensive intervention group.

Our results for mortality and macrovascular outcomes in the present and more comprehensive meta-analysis are in accordance with the results of recent meta-analyses (Boussageon 2011; Callaghan 2012; Castagno 2011; Coca 2012; Johnson 2011; Kelly 2009; Ma 2009; Mannucci 2009; Marso 2010; Ray 2009; Selvin 2004; Slinin 2012; Tkac 2009; Turnbull 2009; Wang 2009; Wu 2010; Zhang 2010).

Glycaemic control is a fundamental part of managing T2D. Today, HbA1c is commonly used in daily clinical practice to assess average glycaemia over several months. A recently published retrospective cohort study with data from the 'General Practice Research Database' somewhat unexpectedly showed in 48,000 patients with T2D that both low and high mean values of HbA1c were associated with increased all-cause mortality and macrovascular events (Currie 2010); and that the HbA1c value with the lowest hazard ratio for all-cause mortality was HbA1c 7.5%. The specific reasons for death were not reported. Notably, a recent large-scale cohort study in non-diabetic people demonstrated an association between lower levels of HbA1c and increased mortality (a J-shaped curve), that is with levels of HbA1c usually not considered to have a risk (Selvin 2010). Hence, any potential causal or non-causal relationship between lower levels of HbA1c and mortality might not necessarily be specific to the diabetic state, its treatments, or other associated conditions (for example hypoglycaemia).

The beneficial effect of targeting intensive glycaemic control on the composite microvascular outcome in our review may be in accordance with results from both randomised clinical trials and observational studies (ADVANCE 2008; Ohkubo 1995; UKPDS 1998; UKPDS-35 2000). However, it has to be kept in mind that the composite microvascular outcome is composed of less severe events (for example, retinal photocoagulation) and critical events (for example, death due to renal failure). In most of the trials, the majority of participants who experience the microvascular composite outcome will have one of the less severe events. Therefore, the effect estimate might be weighted toward that of the less important outcomes. Observational data from the UKPDS showed a 37% risk reduction of microvascular complications for each 1% decrease in HbA1c (UKPDS-35 2000). The ADVANCE trial found a 14% relative risk reduction of major microvascular events when targeting intensive glycaemic control (ADVANCE 2008). The UKPDS 33 showed a 25% risk reduction in microvascular outcomes when targeting intensive glycaemic control (UKPDS 1998). In our review, we observed a relative risk reduction of 11% to 12% for the composite microvascular outcome, and a 1% absolute risk reduction in favour of intensive glycaemic control for all included trials. For the trials exclusively dealing with glycaemic control in usual care settings a relative risk reduction of 11% to 13% was found, and a 1% absolute risk reduction in favour of targeting intensive glycaemic control.

The greatest number of events for the nephropathy outcome was due to a relatively high event rate of the surrogate marker for renal disease (for example, microalbuminuria) compared with hard renal outcomes. This also explains the lack of statistical significance for the meta-analysis exclusively focusing on the patient-important renal outcome end-stage renal disease. The nephropathy outcome can be assessed in several ways, which is done in another systematic review (Coca 2012). Coca et al also did not find any statistical significance of the patient-important clinical outcomes, only a statistical significance for surrogate markers (Coca 2012). The Kumamoto trial showed a pronounced reduction in the incidence of nephropathy in both the primary prevention cohort (11.5% versus 43.5%) and the secondary intervention cohort (16% versus 40%) when targeting intensive glycaemic control (Kumamoto 2000). The ADVANCE trial showed a 21% relative risk reduction in nephropathy when targeting intensive glycaemic control, whereas this could not be shown in the ACCORD and VADT trials (ACCORD 2008; ADVANCE 2008; VADT 2009). We found a 25% relative risk reduction for nephropathy for all included trials in favour of intensive glycaemic control when the random-effects model was applied and a 4% relative risk reduction in the fixed-effect model. However, we found no statistically significant effect in the meta-analysis of the group of trials exclusively dealing with glycaemic control in the usual care setting. Another limitation of the meta-analysis of nephropathy is the pooling of trials with very diverse definitions of nephropathy, including both surrogate and hard clinical outcomes. The risk of end-stage renal disease did not significantly differ between the two intervention groups of the included trials.

We found a 21% relative risk reduction in retinopathy in favour of intensive glycaemic control in the meta-analysis of all included trials. The absolute risk reduction was 3%. The subgroup of trials exclusively dealing with glycaemic control in usual care settings also showed a 19% to 20% relative risk reduction, and a 3% absolute risk reduction. The UKPDS 33 showed a 29% relative risk reduction for retinal photocoagulation when targeting intensive glycaemic control (UKPDS 1998). Retinal photocoagulation showed a 23% relative risk reduction in favour of intensive glycaemic control in our meta-analysis. The absolute risk reduction was 1% in the fixed-effect model. The group of trials exclusively dealing with glycaemic control in usual care settings only showed a 18% relative risk reduction and was only statistical significant in the fixed-effect model.

We report both microvascular disease with surrogate markers (for example, retinopathy initiation and progression expressed on a scale) and hard clinical outcomes (for example, end-stage renal disease). Microvascular data from the ACCORD trial and the UKPDS indicate that the beneficial effects of intensive glycaemic glucose control on microvascular disease take more than five years to emerge, and the benefits on microvascular disease achieved by intensive glycaemic control could be less pronounced for patients with advanced T2D (ACCORD) compared with patients with new onset T2D (UKPDS) (ACCORD 2008; UKPDS 1998). On the other hand, the meta-analysis for retinopathy indicated that patients with more advanced stages of T2D (ACCORD) might benefit more from intensive glycaemic control compared with newly diagnosed patients with T2D (UKPDS) (ACCORD 2008; UKPDS 1998). Most of the recent meta-analyses have not included microvascular disease as an outcome (Castagno 2011; Johnson 2011; Kelly 2009; Mannucci 2009; Marso 2010; Ray 2009; Selvin 2004; Slinin 2012; Tkac 2009; Turnbull 2009; Wu 2010; Zhang 2010). However, Ma et al analysed the included trials according to the HbA1c target in the intensive intervention group and assessed microvascular disease including nephropathy, retinopathy, and neuropathy. For the trials with a HbA1c target less than 7% Ma et al found no significant reduction in the risk of microvascular disease with strict glycaemic control. For trials with a HbA1 target level of 7% to 7.9% in the intensive intervention group a significant reduction was found for nephropathy and retinopathy in favour of intensive glycaemic control (Ma 2009). Wang et al included trials without predefined differences in glycaemic target (for example, 'Prospective Pioglitazone Clinical Trial in Macrovascular Events' (PROactive) and 'Rosiglitazone Evaluated for Cardiac Outcomes and Regulation of Glycaemia in Diabetes' (RECORD)) and showed a 26% reduction in the odds for microvascular events when targeting intensive glycaemic control (Wang 2009). Boussageon et al also included data from trials without predefined differences in glycaemic targets (Boussageon 2011). Coca et al included seven trials investigating the effect of intensive versus conventional glycaemic control on participants with T2D (Coca 2012). Five renal outcomes were assessed separately, microalbuminuria, macroalbuminuria, doubling of the serum creatinine level, end-stage renal disease, and death from renal disease. This systematic review found a statistically significant relative risk reduction of microalbuminuria as well as macroalbuminuria, whereas no statistical significance could be found for the remaining renal outcomes (Coca 2012).

We identified both mild and severe hypoglycaemia as an adverse effect strongly associated with intensive glucose control, which is in accordance with established knowledge and other meta-analyses (Kelly 2009; Ma 2009; Mannucci 2009; Ray 2009; Turnbull 2009; Zhang 2010). We did not have access to in-trial data at the patient level, and therefore we could not investigate whether there was any correlation between severe hypoglycaemic events and the risk of sudden unexpected death. For the same reason, we were not able to investigate the effect of pre-existing cardiovascular disease on the outcomes. Meta-regression for all trials and the subgroup of trials exclusively dealing with glycaemic control in usual care showed a positive correlation between disease duration and the risk ratio for severe hypoglycaemia. An explanation for the increased risk of severe hypoglycaemia with time might be that the glucagon response to hypoglycaemia decreases with the longer duration of diabetes alongside the reduction in endogenous insulin secretion (Cryer 2008). On the other hand, we also found a negative correlation between the risk ratio of severe hypoglycaemia and the duration of the intervention, which could imply that the patients and clinicians become more familiar with the treatment over time, for example, with the prevention of adverse events.

We did not find any statistically significant effect on health-related quality of life; neither the mental nor the physical component. It could be assumed that targeting intensive glycaemic control might affect quality of life negatively as a consequence of the increased patient burden due to the number of finger pricks, insulin injections, as well as an increased risk of hypoglycaemia. The lack of decrement in well-being in the intensive intervention arm may be related to a number of processes. In several trials there was increased access to providers, including both clinic visits and telephone contact, in the intensive intervention group. This may have supported the patients' well-being. Likewise, the lack of significant effect of intensive versus conventional glycaemic control on a number of patient-relevant outcomes, for example, stroke, could have resulted in less improvement in well-being. Results of large-scale randomised clinical trials are not published yet (ADVANCE 2008; VADT 2009).

The economic burden on society due to diabetes is increasing (ADA 2013a). In our meta-analysis of we did not find any statistical significance.

The American Diabetes Association published in January 2013 a guideline recommending an HbA1c goal of less than 7% to reduce microvascular complications (ADA 2013). Treatment targets of HbA1c at 7% in the intensive glucose-lowering group have only been used in five trials, involving 543 participants (Bagg 2001; Guo 2008; Kumamoto 2000; REMBO 2008; Yang 2007). However, only three of these exclusively assessed the effects of glycaemic control and only one of these trials had a duration of more than one year (Kumamoto 2000). The ADDITION trial with 3057 participants with T2D initially stipulated a target HbA1c of < 7.0% in the intensive group, however, local protocols and later reports from the study support that an HbA1c target of 6.5% could also be used (ADDITION-Europe 2011). Besides, most of the included trials had sparse data on the number of participants achieving the glycaemic target at the end of follow-up; and, when reported, the proportion of participants achieving the glycaemic target was relatively low. The American Diabetes Association, however, recommends an individualised approach with less strict glycaemic control in patients with longer duration of diabetes and increased risk of hypoglycaemia (ADA 2013). Current evidence from randomised clinical trials is insufficient to conclude if individualisation of glycaemic targets is superior to either intensive glycaemic control or conventional glycaemic control in patients with T2D.

Authors' conclusions

Implications for practice

There is insufficient evidence to demonstrate whether targeting intensive glycaemic control influences all-cause or cardiovascular mortality. Intensive glycaemic control may reduce the occurrence of some patient-important outcomes such as non-fatal myocardial infarction, lower extremity amputation, and microvascular disease. However, these effects are uncertain and may be due to risks of bias and risks of random errors. Targeting intensive glycaemic control compared with conventional glycaemic control increases the risk of severe adverse events and both mild and severe hypoglycaemia. We did not find any influence of the interventions on health-related quality of life. It has to be kept in mind that conventional glycaemic control is not synonymous with no glycaemic control but is just less stringent than the glycaemic intervention applied in the intensive glycaemic regimes. The data presented in this review do not seem to clearly favour targeting intensive glycaemic control in patients with T2D but rather highlight the insufficient evidence available to guide the choice of the glycaemic intervention strategy. Our review therefore underscores the need to collect more information from randomised clinical trials before evidence-based general recommendations can be made regarding the choice between conventional glycaemic control and intensive glycaemic control in these patients.

Implications for research

For safety purposes, and with the aim of identifying the general optimal glycaemic target, it would be preferable to have more randomised clinical trials assessing cardiovascular disease and mortality in patients with T2D, for example, in younger patients with T2D without complications and older patients with complications. Considering the combined evidence on the influence of intensive glycaemic control on mortality, a 10% relative risk reduction or more of all-cause mortality seems unlikely. Therefore very large randomised clinical trials with the ability to detect or reject less than a 10% relative risk reduction are warranted. We suggest that more uniform treatment regimens should be used in the interventions arms. We also suggest a more uniform and rigorous reporting of outcomes in upcoming trials to ease the comparisons between different glycaemic intervention targets. Future trials ought to be designed according to SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials) and reported according to the CONSORT (CONsolidated Standards of Reporting Trials) statement (Chan 2013; Schulz 2010).

Acknowledgements

Christina Hemmingsen helped with selection and quality assessment of trials, and provided input to data analyses.

The authors would like to thank Karla Bergerhoff, the Trials Search Co-ordinator of the Cochrane Metabolic and Endocrine Disorders Group, and Sarah Klingenberg, the Trials Search Co-ordinator of the Cochrane Hepato-Biliary Group, for their assistance in developing the search strategy. Thanks to Dimitrinka Nikolova from the Cochrane Hepato-Biliary Group for advice during the writing process and for translating and extracting data from a Russian article. Thanks to Xia Yun for extracting data from a Chinese article. Thanks to Britta Tendal, The Nordic Cochrane Centre, for statistical advice on meta-analysing quality of life.

The authors would like to thank Warwick Bagg, the DIGAMI 2 study group, Peter Gaede and Oluf Borbye Petersen, Genell Knatterud, John F Service, Alexander Stefanidis, Carlos Abraira, Thomas Moritz, Camilla Hage, Denise Bonds, Beatriz Schaan Atsushi Araki, Tomohiro Shinozaki, Satoshi Iimuro, and Hideki Ito for answering our request for information on trials.

Data and analyses

Download statistical data

Comparison 1. Intensive glycaemic control versus conventional glycaemic control
Outcome or subgroup titleNo. of studiesNo. of participantsStatistical methodEffect size
1 All-cause mortality2434325Risk Ratio (M-H, Random, 95% CI)1.00 [0.92, 1.08]
2 All-cause mortality; stratified according to risk of bias2434325Risk Ratio (M-H, Fixed, 95% CI)1.00 [0.94, 1.07]
2.1 Lower risk of bias1333344Risk Ratio (M-H, Fixed, 95% CI)1.01 [0.94, 1.07]
2.2 High risk of bias11981Risk Ratio (M-H, Fixed, 95% CI)0.72 [0.37, 1.37]
3 All-cause mortality; stratified according to sequence generation2434325Risk Ratio (M-H, Random, 95% CI)1.00 [0.92, 1.08]
3.1 Adequate sequence generation1733707Risk Ratio (M-H, Random, 95% CI)1.00 [0.92, 1.09]
3.2 Unclear/inadequate sequence generation7618Risk Ratio (M-H, Random, 95% CI)0.74 [0.36, 1.50]
4 All-cause mortality; stratified according to allocation concealment2434325Risk Ratio (M-H, Random, 95% CI)1.00 [0.92, 1.08]
4.1 Adequate allocation concealment1733707Risk Ratio (M-H, Random, 95% CI)1.00 [0.92, 1.09]
4.2 Unclear/inadequate allocation concealment7618Risk Ratio (M-H, Random, 95% CI)0.74 [0.36, 1.50]
5 All-cause mortality; stratified according to blinding (study level)2434325Risk Ratio (M-H, Random, 95% CI)1.00 [0.92, 1.08]
5.1 Adequate blinding1733635Risk Ratio (M-H, Random, 95% CI)1.00 [0.92, 1.09]
5.2 Unclear/inadequate blinding7690Risk Ratio (M-H, Random, 95% CI)0.64 [0.27, 1.49]
6 All-cause mortality; stratified according to outcome data (outcome level)2434325Risk Ratio (M-H, Random, 95% CI)1.00 [0.92, 1.08]
6.1 Complete outcome data1811497Risk Ratio (M-H, Random, 95% CI)0.96 [0.88, 1.06]
6.2 Incomplete outcome data622828Risk Ratio (M-H, Random, 95% CI)1.06 [0.87, 1.29]
7 All-cause mortality; stratified according to outcome reporting bias (study level)2434325Risk Ratio (M-H, Random, 95% CI)1.00 [0.92, 1.08]
7.1 Adequate outcome reporting1333256Risk Ratio (M-H, Random, 95% CI)1.00 [0.91, 1.10]
7.2 Unclear/inadequate outcome reporting111069Risk Ratio (M-H, Random, 95% CI)0.65 [0.33, 1.30]
8 All-cause mortality; stratified according to academic bias2434325Risk Ratio (M-H, Random, 95% CI)1.00 [0.92, 1.08]
8.1 Low risk of academic bias2233470Risk Ratio (M-H, Random, 95% CI)0.98 [0.90, 1.08]
8.2 High risk of academic bias2855Risk Ratio (M-H, Random, 95% CI)1.11 [0.89, 1.38]
9 All-cause mortality; stratified according to source of funding2434325Risk Ratio (M-H, Fixed, 95% CI)1.00 [0.94, 1.07]
9.1 Industry funded1532206Risk Ratio (M-H, Fixed, 95% CI)1.00 [0.94, 1.07]
9.2 No industry funding92119Risk Ratio (M-H, Fixed, 95% CI)1.03 [0.81, 1.32]
10 All-cause mortality; stratified according to study duration2434325Risk Ratio (M-H, Fixed, 95% CI)1.00 [0.94, 1.07]
10.1 Long duration (> 2.0 years)1233275Risk Ratio (M-H, Fixed, 95% CI)1.01 [0.94, 1.07]
10.2 Short duration (≤ 2 years)121050Risk Ratio (M-H, Fixed, 95% CI)0.78 [0.36, 1.68]
11 All-cause mortality; stratified according to diagnostic criteria2434325Risk Ratio (M-H, Random, 95% CI)1.00 [0.92, 1.08]
11.1 Diagnostic criteria for T2D described1922892Risk Ratio (M-H, Random, 95% CI)1.01 [0.92, 1.12]
11.2 Diagnostic criteria for T2D not described511433Risk Ratio (M-H, Random, 95% CI)0.93 [0.83, 1.05]
12 All-cause mortality; stratified according to intervention2434325Risk Ratio (M-H, Random, 95% CI)1.00 [0.92, 1.08]
12.1 Exclusively dealing with glycaemic control in usual care setting1228354Risk Ratio (M-H, Random, 95% CI)1.01 [0.92, 1.12]
12.2 Glycaemic control as a part of acute intervention3903Risk Ratio (M-H, Random, 95% CI)1.11 [0.89, 1.38]
12.3 Glycaemic control initiated with surgical intervention4429Risk Ratio (M-H, Random, 95% CI)0.63 [0.21, 1.92]
12.4 Multimodal intervention in usual care setting54639Risk Ratio (M-H, Random, 95% CI)0.87 [0.62, 1.23]
13 All-cause mortality; hazard ratio7 Hazard Ratio (Random, 95% CI)1.00 [0.87, 1.15]
14 All-cause mortality; available case2333521Risk Ratio (M-H, Random, 95% CI)1.00 [0.92, 1.09]
15 All-cause mortality; best-case scenario2334247Risk Ratio (M-H, Random, 95% CI)0.82 [0.74, 0.91]
16 All-cause mortality; worst-case scenario2334247Risk Ratio (M-H, Random, 95% CI)1.20 [0.97, 1.49]
17 Cardiovascular mortality2234177Risk Ratio (M-H, Random, 95% CI)1.06 [0.94, 1.21]
18 Cardiovascular mortality; stratified according to risk of bias2234177Risk Ratio (M-H, Random, 95% CI)1.07 [0.94, 1.22]
18.1 Lower risk of bias1133196Risk Ratio (M-H, Random, 95% CI)1.08 [0.92, 1.26]
18.2 High risk of bias11981Risk Ratio (M-H, Random, 95% CI)0.74 [0.29, 1.93]
19 Cardiovascular mortality; stratified according to sequence generation2234177Risk Ratio (M-H, Random, 95% CI)1.06 [0.94, 1.21]
19.1 Adequate sequence generation1533559Risk Ratio (M-H, Random, 95% CI)1.07 [0.93, 1.23]
19.2 Unclear/inadequate sequence generation7618Risk Ratio (M-H, Random, 95% CI)0.67 [0.22, 2.00]
20 Cardiovascular mortality; stratified according to allocation concealment2234177Risk Ratio (M-H, Random, 95% CI)1.06 [0.94, 1.21]
20.1 Adequate allocation concealment1533559Risk Ratio (M-H, Random, 95% CI)1.07 [0.93, 1.23]
20.2 Unclear/inadequate allocation concealment7618Risk Ratio (M-H, Random, 95% CI)0.67 [0.22, 2.00]
21 Cardiovascular mortality; stratified according to blinding (study level)2234177Risk Ratio (M-H, Random, 95% CI)1.06 [0.94, 1.21]
21.1 Adequate blinding1533487Risk Ratio (M-H, Random, 95% CI)1.07 [0.93, 1.23]
21.2 Unclear/inadequate blinding7690Risk Ratio (M-H, Random, 95% CI)0.54 [0.14, 2.03]
22 Cardiovascular mortality; stratified according to outcome data (outcome level)2234177Risk Ratio (M-H, Random, 95% CI)1.06 [0.94, 1.21]
22.1 Adequate outcome data1711427Risk Ratio (M-H, Random, 95% CI)1.10 [0.96, 1.27]
22.2 Incomplete outcome data522750Risk Ratio (M-H, Random, 95% CI)1.08 [0.75, 1.54]
23 Cardiovascular mortality; stratified according to outcome reporting bias (study level)2234177Risk Ratio (M-H, Random, 95% CI)1.06 [0.94, 1.21]
23.1 Adequate outcome reporting1233186Risk Ratio (M-H, Random, 95% CI)1.07 [0.92, 1.24]
23.2 Unclear/inadequate outcome reporting10991Risk Ratio (M-H, Random, 95% CI)0.64 [0.21, 1.93]
24 Cardiovascular mortality; stratified according to academic bias2234177Risk Ratio (M-H, Random, 95% CI)1.06 [0.94, 1.21]
24.1 Low risk of academic bias2033322Risk Ratio (M-H, Random, 95% CI)1.05 [0.90, 1.23]
24.2 High risk of academic bias2855Risk Ratio (M-H, Random, 95% CI)1.14 [0.86, 1.51]
25 Cardiovascular mortality; stratified according to source of funding2234177Risk Ratio (M-H, Random, 95% CI)1.06 [0.94, 1.21]
25.1 Industry funding1332047Risk Ratio (M-H, Random, 95% CI)1.06 [0.90, 1.26]
25.2 No industry funding92130Risk Ratio (M-H, Random, 95% CI)1.02 [0.70, 1.50]
26 Cardiovascular mortality; stratified according to study duration2234177Risk Ratio (M-H, Random, 95% CI)1.06 [0.94, 1.21]
26.1 Long duration (> 2 years)1233275Risk Ratio (M-H, Random, 95% CI)1.06 [0.91, 1.24]
26.2 Short duration (≤ 2 years)10902Risk Ratio (M-H, Random, 95% CI)0.83 [0.25, 2.71]
27 Cardiovascular mortality; stratified according to diagnostic criteria2234177Risk Ratio (M-H, Random, 95% CI)1.06 [0.94, 1.21]
27.1 Diagnostic criteria for T2D described1822822Risk Ratio (M-H, Random, 95% CI)1.17 [1.04, 1.31]
27.2 Diagnostic criteria for T2D not described411355Risk Ratio (M-H, Random, 95% CI)0.87 [0.74, 1.03]
28 Cardiovascular mortality; stratified according to intervention2234177Risk Ratio (M-H, Random, 95% CI)1.06 [0.94, 1.21]
28.1 Exclusively dealing with glycaemic control in usual care setting1228354Risk Ratio (M-H, Random, 95% CI)1.09 [0.92, 1.29]
28.2 Glycaemic control as a part of acute intervention3903Risk Ratio (M-H, Random, 95% CI)1.14 [0.86, 1.51]
28.3 Glycaemic control initiated with surgical intervention2281Risk Ratio (M-H, Random, 95% CI)0.95 [0.14, 6.57]
28.4 Multimodal intervention in usual care setting54639Risk Ratio (M-H, Random, 95% CI)0.86 [0.47, 1.56]
29 Cardiovascular mortality; hazard ratio6 Hazard Ratio (Random, 95% CI)1.05 [0.84, 1.31]
30 Cardiovascular mortality; available case2233451Risk Ratio (M-H, Random, 95% CI)1.06 [0.94, 1.20]
31 Cardiovascular mortality; best-case scenario2234177Risk Ratio (M-H, Random, 95% CI)0.71 [0.56, 0.88]
32 Cardiovascular mortality; worst-case scenario2234177Risk Ratio (M-H, Random, 95% CI)1.54 [1.09, 2.18]
33 Macrovascular complications1432846Risk Ratio (M-H, Random, 95% CI)0.91 [0.82, 1.02]
34 Macrovascular complications; stratified according to intervention1432846Risk Ratio (M-H, Random, 95% CI)0.91 [0.82, 1.02]
34.1 Exclusively dealing with glycaemic control in usual care setting927666Risk Ratio (M-H, Random, 95% CI)0.92 [0.83, 1.01]
34.2 Glycaemic control as a part of acute intervention1780Risk Ratio (M-H, Random, 95% CI)1.09 [0.91, 1.32]
34.3 Glycaemic control initiated with surgical intervention170Risk Ratio (M-H, Random, 95% CI)1.67 [0.43, 6.45]
34.4 Multimodal intervention in usual care setting34330Risk Ratio (M-H, Random, 95% CI)0.79 [0.54, 1.16]
35 Non-fatal myocardial infarction1430417Risk Ratio (M-H, Random, 95% CI)0.87 [0.77, 0.98]
36 Non-fatal myocardial infarction; stratified according to study duration1430417Risk Ratio (M-H, Random, 95% CI)0.87 [0.77, 0.98]
36.1 Long duration (> 2 years)1030181Risk Ratio (M-H, Random, 95% CI)0.87 [0.76, 1.00]
36.2 Short duration ( ≤ 2 years)4236Risk Ratio (M-H, Random, 95% CI)0.81 [0.23, 2.78]
37 Non-fatal myocardial infarction; stratified according to risk of bias1430417Risk Ratio (M-H, Random, 95% CI)0.87 [0.77, 0.98]
37.1 Lower risk of bias929988Risk Ratio (M-H, Random, 95% CI)0.87 [0.75, 1.01]
37.2 High risk of bias5429Risk Ratio (M-H, Random, 95% CI)0.82 [0.34, 1.99]
38 Non-fatal myocardial infarction; stratified according to source of funding1430417Risk Ratio (M-H, Random, 95% CI)0.87 [0.77, 0.98]
38.1 Industry funded828594Risk Ratio (M-H, Random, 95% CI)0.86 [0.73, 1.00]
38.2 No industry funding61823Risk Ratio (M-H, Random, 95% CI)1.00 [0.67, 1.49]
39 Non-fatal myocardial infarction; stratified according to diagnostic criteria1319277Risk Ratio (M-H, Random, 95% CI)0.84 [0.74, 0.96]
39.1 Diagnostic criteria for T2D described1319277Risk Ratio (M-H, Random, 95% CI)0.84 [0.74, 0.96]
40 Non-fatal myocardial infarction; stratified according to intervention1430417Risk Ratio (M-H, Random, 95% CI)0.87 [0.77, 0.98]
40.1 Exclusively dealing with glycaemic control in usual care setting828111Risk Ratio (M-H, Random, 95% CI)0.85 [0.76, 0.95]
40.2 Glycaemic control as a part of acute intervention3903Risk Ratio (M-H, Random, 95% CI)1.15 [0.84, 1.58]
40.3 Glycaemic control initiated with surgical intervention170Risk Ratio (M-H, Random, 95% CI)0.0 [0.0, 0.0]
40.4 Multimodal intervention in usual care setting21333Risk Ratio (M-H, Random, 95% CI)0.63 [0.22, 1.83]
41 Non-fatal myocardial infarction; available case1428103Risk Ratio (M-H, Random, 95% CI)0.87 [0.77, 0.97]
42 Non-fatal myocardial infarction; best-case scenario1430417Risk Ratio (M-H, Random, 95% CI)0.34 [0.24, 0.47]
43 Non-fatal myocardial infarction; worst-case scenario1430417Risk Ratio (M-H, Random, 95% CI)2.28 [1.62, 3.19]
44 Non-fatal stroke1330003Risk Ratio (M-H, Random, 95% CI)1.00 [0.84, 1.19]
45 Non-fatal stroke; stratified according to intervention1330000Risk Ratio (M-H, Random, 95% CI)0.99 [0.84, 1.18]
45.1 Exclusively dealing with glycaemic control in usual care setting727697Risk Ratio (M-H, Random, 95% CI)1.01 [0.87, 1.16]
45.2 Glycaemic control as a part of acute intervention3900Risk Ratio (M-H, Random, 95% CI)1.19 [0.62, 2.30]
45.3 Glycaemic control initiated with surgical intervention170Risk Ratio (M-H, Random, 95% CI)0.0 [0.0, 0.0]
45.4 Multimodal intervention in usual care setting21333Risk Ratio (M-H, Random, 95% CI)0.68 [0.19, 2.49]
46 Amputation of lower extremity1111200Risk Ratio (M-H, Random, 95% CI)0.65 [0.45, 0.94]
47 Amputation of lower extremity; stratified according to intervention1111200Risk Ratio (M-H, Random, 95% CI)0.65 [0.45, 0.94]
47.1 Exclusively dealing with glycaemic control in usual care setting56677Risk Ratio (M-H, Random, 95% CI)0.70 [0.45, 1.09]
47.2 Glycaemic control as a part of acute intervention2123Risk Ratio (M-H, Random, 95% CI)0.0 [0.0, 0.0]
47.3 Glycaemic control initiated with surgical intervention170Risk Ratio (M-H, Random, 95% CI)0.0 [0.0, 0.0]
47.4 Multimodal intervention in usual care setting34330Risk Ratio (M-H, Random, 95% CI)0.54 [0.27, 1.07]
48 Cardiac revascularization73532Risk Ratio (M-H, Random, 95% CI)0.81 [0.65, 1.01]
49 Cardiac revascularization; stratified according to intervention73532Risk Ratio (M-H, Random, 95% CI)0.81 [0.65, 1.01]
49.1 Exclusively dealing with glycaemic control in usual care setting32054Risk Ratio (M-H, Random, 95% CI)0.85 [0.67, 1.07]
49.2 Glycaemic control as a part of acute intervention175Risk Ratio (M-H, Random, 95% CI)2.17 [0.21, 22.89]
49.3 Glycaemic control initiated with surgical intervention170Risk Ratio (M-H, Random, 95% CI)2.0 [0.19, 21.06]
49.4 Multimodal intervention in usual care setting21333Risk Ratio (M-H, Random, 95% CI)0.50 [0.26, 0.94]
50 Peripheral revascularization813547Risk Ratio (M-H, Random, 95% CI)0.93 [0.81, 1.06]
51 Peripheral revascularization; stratified according to intervention813547Risk Ratio (M-H, Random, 95% CI)0.93 [0.81, 1.06]
51.1 Exclusively dealing with glycaemic control in usual care setting413194Risk Ratio (M-H, Random, 95% CI)0.93 [0.81, 1.07]
51.2 Glycaemic control as a part of acute intervention2123Risk Ratio (M-H, Random, 95% CI)0.0 [0.0, 0.0]
51.3 Glycaemic control initiated with surgical intervention170Risk Ratio (M-H, Random, 95% CI)0.0 [0.0, 0.0]
51.4 Multimodal intervention in usual care setting1160Risk Ratio (M-H, Random, 95% CI)0.6 [0.23, 1.57]
52 Microvascular complications625927Risk Ratio (M-H, Random, 95% CI)0.88 [0.82, 0.95]
53 Microvascular complications; stratified according to intervention625927Risk Ratio (M-H, Random, 95% CI)0.88 [0.82, 0.95]
53.1 Exclusively dealing with glycaemic control in usual care setting425697Risk Ratio (M-H, Random, 95% CI)0.87 [0.78, 0.97]
53.2 Glycaemic control as a part of acute intervention00Risk Ratio (M-H, Random, 95% CI)0.0 [0.0, 0.0]
53.3 Glycaemic control initiated with surgical intervention170Risk Ratio (M-H, Random, 95% CI)0.0 [0.0, 0.0]
53.4 Multimodal intervention in usual care setting1160Risk Ratio (M-H, Random, 95% CI)0.89 [0.80, 1.00]
54 Nephropathy1128096Risk Ratio (M-H, Random, 95% CI)0.75 [0.59, 0.95]
55 Nephropathy; stratified according to intervention1128096Risk Ratio (M-H, Random, 95% CI)0.75 [0.59, 0.95]
55.1 Exclusively dealing with glycaemic control in usual care setting927866Risk Ratio (M-H, Random, 95% CI)0.79 [0.61, 1.01]
55.2 Glycaemic control as a part of acute intervention00Risk Ratio (M-H, Random, 95% CI)0.0 [0.0, 0.0]
55.3 Glycaemic control initiated with surgical intervention170Risk Ratio (M-H, Random, 95% CI)0.0 [0.0, 0.0]
55.4 Multimodal intervention in usual care setting1160Risk Ratio (M-H, Random, 95% CI)0.54 [0.35, 0.85]
56 End-stage renal disease828145Risk Ratio (M-H, Random, 95% CI)0.87 [0.71, 1.06]
57 End-stage renal disease; stratified according to intervention828145Risk Ratio (M-H, Random, 95% CI)0.87 [0.71, 1.06]
57.1 Exclusively dealing with glycaemic control in usual care setting627915Risk Ratio (M-H, Random, 95% CI)0.88 [0.72, 1.07]
57.2 Glycaemic control as a part of acute intervention00Risk Ratio (M-H, Random, 95% CI)0.0 [0.0, 0.0]
57.3 Glycaemic control initiated with surgical intervention170Risk Ratio (M-H, Random, 95% CI)0.0 [0.0, 0.0]
57.4 Multimodal intervention in usual care setting1160Risk Ratio (M-H, Random, 95% CI)0.17 [0.02, 1.35]
58 Retinopathy910300Risk Ratio (M-H, Random, 95% CI)0.79 [0.68, 0.92]
59 Retinopathy; stratified according to intervention910300Risk Ratio (M-H, Random, 95% CI)0.79 [0.69, 0.92]
59.1 Exclusively dealing with glycaemic control in usual care setting710070Risk Ratio (M-H, Random, 95% CI)0.80 [0.67, 0.94]
59.2 Glycaemic control as a part of acute intervention00Risk Ratio (M-H, Random, 95% CI)0.0 [0.0, 0.0]
59.3 Glycaemic control initiated with surgical intervention170Risk Ratio (M-H, Random, 95% CI)0.0 [0.0, 0.0]
59.4 Multimodal intervention in usual care setting1160Risk Ratio (M-H, Random, 95% CI)0.76 [0.58, 0.99]
60 Retinal photocoagulation811212Risk Ratio (M-H, Random, 95% CI)0.77 [0.61, 0.97]
61 Retinal photocoagulation; stratified according to intervention811212Risk Ratio (M-H, Random, 95% CI)0.77 [0.61, 0.97]
61.1 Exclusively dealing with glycaemic control in usual care setting610982Risk Ratio (M-H, Random, 95% CI)0.82 [0.65, 1.03]
61.2 Glycaemic control as a part of acute intervention00Risk Ratio (M-H, Random, 95% CI)0.0 [0.0, 0.0]
61.3 Glycaemic control initiated with surgical intervention170Risk Ratio (M-H, Random, 95% CI)0.0 [0.0, 0.0]
61.4 Multimodal intervention in usual care setting1160Risk Ratio (M-H, Random, 95% CI)0.52 [0.29, 0.91]
62 Adverse events1337672Risk Ratio (M-H, Random, 95% CI)1.06 [1.02, 1.11]
62.1 Adverse events2403Risk Ratio (M-H, Random, 95% CI)1.18 [0.87, 1.60]
62.2 Serious adverse events1124280Risk Ratio (M-H, Random, 95% CI)1.06 [1.02, 1.10]
62.3 Drop-outs due to adverse events1112989Risk Ratio (M-H, Random, 95% CI)1.47 [0.84, 2.56]
63 Serious adverse events; stratified according to intervention1124280Risk Ratio (M-H, Random, 95% CI)1.06 [1.02, 1.10]
63.1 Exclusively dealing with glycaemic control in usual care setting723927Risk Ratio (M-H, Random, 95% CI)1.06 [1.02, 1.10]
63.2 Glycaemic control as a part of acute intervention2123Risk Ratio (M-H, Random, 95% CI)0.95 [0.41, 2.18]
63.3 Glycaemic control initiated with surgical intervention170Risk Ratio (M-H, Random, 95% CI)1.0 [0.07, 15.36]
63.4 Multifactorial intervention in usual care setting1160Risk Ratio (M-H, Random, 95% CI)3.0 [0.12, 72.56]
64 Drop-outs due to adverse events; stratified according to intervention1112989Risk Ratio (M-H, Random, 95% CI)1.47 [0.84, 2.56]
64.1 Exclusively dealing with glycaemic control in usual care setting712636Risk Ratio (M-H, Random, 95% CI)1.50 [0.80, 2.79]
64.2 Glycaemic control as a part of acute intervention2123Risk Ratio (M-H, Random, 95% CI)1.35 [0.39, 4.65]
64.3 Glycaemic control initiated with surgical intervention170Risk Ratio (M-H, Random, 95% CI)0.0 [0.0, 0.0]
64.4 Multimodal intervention in usual care setting1160Risk Ratio (M-H, Random, 95% CI)0.0 [0.0, 0.0]
65 Congestive heart failure1229815Risk Ratio (M-H, Random, 95% CI)0.98 [0.87, 1.10]
66 Congestive heart failure; stratified after intervention1129733Risk Ratio (M-H, Random, 95% CI)0.99 [0.88, 1.12]
66.1 Exclusively dealing with glycaemic control in usual care setting627587Risk Ratio (M-H, Random, 95% CI)1.01 [0.87, 1.17]
66.2 Glycaemic control as a part of acute intervention3903Risk Ratio (M-H, Random, 95% CI)0.68 [0.34, 1.37]
66.3 Glycaemic control initiated with surgical intervention170Risk Ratio (M-H, Random, 95% CI)0.0 [0.0, 0.0]
66.4 Multimodal intervention in usual care setting11173Risk Ratio (M-H, Random, 95% CI)0.70 [0.30, 1.62]
67 Hypoglycaemia1948205Risk Ratio (M-H, Random, 95% CI)1.80 [1.51, 2.14]
67.1 Mild hypoglycaemia1519411Risk Ratio (M-H, Random, 95% CI)1.54 [1.35, 1.75]
67.2 Severe hypoglycaemia1728794Risk Ratio (M-H, Random, 95% CI)2.18 [1.53, 3.11]
68 Mild hypoglycaemia; stratified according to intervention1519411Risk Ratio (M-H, Random, 95% CI)1.54 [1.35, 1.75]
68.1 Exclusively dealing with glycaemic control in usual care setting918200Risk Ratio (M-H, Random, 95% CI)1.58 [1.37, 1.81]
68.2 Glycaemic control as a part of acute intervention3903Risk Ratio (M-H, Random, 95% CI)2.13 [0.83, 5.50]
68.3 Glycaemic control initiated with operation2148Risk Ratio (M-H, Random, 95% CI)35.00 [2.19, 560.18]
68.4 Multimodal intervention in usual care setting1160Risk Ratio (M-H, Random, 95% CI)1.14 [0.95, 1.37]
69 Severe hypoglycaemia; stratified according to intervention1728794Risk Ratio (M-H, Random, 95% CI)2.18 [1.53, 3.11]
69.1 Exclusively dealing with glycaemic control in usual care setting1028082Risk Ratio (M-H, Random, 95% CI)2.44 [1.78, 3.35]
69.2 Glycaemic control as a part of acute intervention2123Risk Ratio (M-H, Random, 95% CI)7.00 [0.38, 128.61]
69.3 Glycaemic control initiated with surgical intervention4429Risk Ratio (M-H, Random, 95% CI)4.67 [0.81, 26.84]
69.4 Multimodal intervention in usual care setting1160Risk Ratio (M-H, Random, 95% CI)0.71 [0.34, 1.51]
70 Health-related quality of life; EQ5D22776Mean Difference (IV, Random, 95% CI)0.0 [-0.02, 0.02]
71 Quality of life: mental component52648Std. Mean Difference (IV, Random, 95% CI)-0.04 [-0.13, 0.04]
71.1 Change in score from baseline3906Std. Mean Difference (IV, Random, 95% CI)0.04 [-0.09, 0.18]
71.2 End of follow-up value41742Std. Mean Difference (IV, Random, 95% CI)-0.09 [-0.19, 0.01]
72 Quality of life: physical component41556Std. Mean Difference (IV, Random, 95% CI)-0.04 [-0.14, 0.06]
72.1 Change in score from baseline2743Std. Mean Difference (IV, Random, 95% CI)-0.06 [-0.21, 0.08]
72.2 End of follow-up value3813Std. Mean Difference (IV, Random, 95% CI)-0.02 [-0.16, 0.11]
73 Cost of intervention24319Std. Mean Difference (IV, Random, 95% CI)0.05 [-0.02, 0.12]
Analysis 1.1.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 1 All-cause mortality.

Analysis 1.2.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 2 All-cause mortality; stratified according to risk of bias.

Analysis 1.3.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 3 All-cause mortality; stratified according to sequence generation.

Analysis 1.4.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 4 All-cause mortality; stratified according to allocation concealment.

Analysis 1.5.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 5 All-cause mortality; stratified according to blinding (study level).

Analysis 1.6.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 6 All-cause mortality; stratified according to outcome data (outcome level).

Analysis 1.7.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 7 All-cause mortality; stratified according to outcome reporting bias (study level).

Analysis 1.8.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 8 All-cause mortality; stratified according to academic bias.

Analysis 1.9.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 9 All-cause mortality; stratified according to source of funding.

Analysis 1.10.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 10 All-cause mortality; stratified according to study duration.

Analysis 1.11.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 11 All-cause mortality; stratified according to diagnostic criteria.

Analysis 1.12.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 12 All-cause mortality; stratified according to intervention.

Analysis 1.13.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 13 All-cause mortality; hazard ratio.

Analysis 1.14.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 14 All-cause mortality; available case.

Analysis 1.15.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 15 All-cause mortality; best-case scenario.

Analysis 1.16.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 16 All-cause mortality; worst-case scenario.

Analysis 1.17.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 17 Cardiovascular mortality.

Analysis 1.18.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 18 Cardiovascular mortality; stratified according to risk of bias.

Analysis 1.19.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 19 Cardiovascular mortality; stratified according to sequence generation.

Analysis 1.20.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 20 Cardiovascular mortality; stratified according to allocation concealment.

Analysis 1.21.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 21 Cardiovascular mortality; stratified according to blinding (study level).

Analysis 1.22.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 22 Cardiovascular mortality; stratified according to outcome data (outcome level).

Analysis 1.23.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 23 Cardiovascular mortality; stratified according to outcome reporting bias (study level).

Analysis 1.24.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 24 Cardiovascular mortality; stratified according to academic bias.

Analysis 1.25.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 25 Cardiovascular mortality; stratified according to source of funding.

Analysis 1.26.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 26 Cardiovascular mortality; stratified according to study duration.

Analysis 1.27.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 27 Cardiovascular mortality; stratified according to diagnostic criteria.

Analysis 1.28.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 28 Cardiovascular mortality; stratified according to intervention.

Analysis 1.29.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 29 Cardiovascular mortality; hazard ratio.

Analysis 1.30.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 30 Cardiovascular mortality; available case.

Analysis 1.31.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 31 Cardiovascular mortality; best-case scenario.

Analysis 1.32.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 32 Cardiovascular mortality; worst-case scenario.

Analysis 1.33.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 33 Macrovascular complications.

Analysis 1.34.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 34 Macrovascular complications; stratified according to intervention.

Analysis 1.35.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 35 Non-fatal myocardial infarction.

Analysis 1.36.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 36 Non-fatal myocardial infarction; stratified according to study duration.

Analysis 1.37.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 37 Non-fatal myocardial infarction; stratified according to risk of bias.

Analysis 1.38.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 38 Non-fatal myocardial infarction; stratified according to source of funding.

Analysis 1.39.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 39 Non-fatal myocardial infarction; stratified according to diagnostic criteria.

Analysis 1.40.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 40 Non-fatal myocardial infarction; stratified according to intervention.

Analysis 1.41.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 41 Non-fatal myocardial infarction; available case.

Analysis 1.42.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 42 Non-fatal myocardial infarction; best-case scenario.

Analysis 1.43.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 43 Non-fatal myocardial infarction; worst-case scenario.

Analysis 1.44.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 44 Non-fatal stroke.

Analysis 1.45.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 45 Non-fatal stroke; stratified according to intervention.

Analysis 1.46.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 46 Amputation of lower extremity.

Analysis 1.47.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 47 Amputation of lower extremity; stratified according to intervention.

Analysis 1.48.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 48 Cardiac revascularization.

Analysis 1.49.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 49 Cardiac revascularization; stratified according to intervention.

Analysis 1.50.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 50 Peripheral revascularization.

Analysis 1.51.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 51 Peripheral revascularization; stratified according to intervention.

Analysis 1.52.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 52 Microvascular complications.

Analysis 1.53.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 53 Microvascular complications; stratified according to intervention.

Analysis 1.54.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 54 Nephropathy.

Analysis 1.55.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 55 Nephropathy; stratified according to intervention.

Analysis 1.56.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 56 End-stage renal disease.

Analysis 1.57.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 57 End-stage renal disease; stratified according to intervention.

Analysis 1.58.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 58 Retinopathy.

Analysis 1.59.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 59 Retinopathy; stratified according to intervention.

Analysis 1.60.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 60 Retinal photocoagulation.

Analysis 1.61.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 61 Retinal photocoagulation; stratified according to intervention.

Analysis 1.62.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 62 Adverse events.

Analysis 1.63.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 63 Serious adverse events; stratified according to intervention.

Analysis 1.64.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 64 Drop-outs due to adverse events; stratified according to intervention.

Analysis 1.65.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 65 Congestive heart failure.

Analysis 1.66.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 66 Congestive heart failure; stratified after intervention.

Analysis 1.67.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 67 Hypoglycaemia.

Analysis 1.68.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 68 Mild hypoglycaemia; stratified according to intervention.

Analysis 1.69.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 69 Severe hypoglycaemia; stratified according to intervention.

Analysis 1.70.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 70 Health-related quality of life; EQ5D.

Analysis 1.71.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 71 Quality of life: mental component.

Analysis 1.72.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 72 Quality of life: physical component.

Analysis 1.73.

Comparison 1 Intensive glycaemic control versus conventional glycaemic control, Outcome 73 Cost of intervention.

Appendices

Appendix 1. Search strategies

Search terms and databases

Unless otherwise stated, search terms are free text terms.

Abbreviations:

'$': stands for any character; '?': substitutes one or no character; adj: adjacent (i.e. number of words within range of search term); exp: exploded MeSH; MeSH: medical subject heading (MEDLINE medical index term); pt: publication type; sh: MeSH; tw: text word.

The Cochrane Library
1. MeSH descriptor Diabetes mellitus, type 2explode all trees
2. MeSH descriptor Insulin resistanceexplode all trees
3. ((impaired in All Text and glucosein All Text and toleranc* in All Text) or (glucosein All Text and intoleranc* in All Text) or (insulin*in All Text and resistanc* in All Text) )
4. (obes* in All Text near/6 diabet*in All Text)
5. (MODY in All Text or NIDDMin All Text or TDM2 in All Text)
6. ((non in All Text and insulin*in All Text and depend* in All Text) or (noninsulin*in All Text and depend* in All Text) or (nonin All Text and insulindepend* in All Text) or noninsulindepend*in All Text)
7. (typ* in All Text and (2in All Text near/6 diabet* in All Text))
8. (typ* in All Text and (IIin All Text near/6 diabet* in All Text))
9. (non in All Text and (keto*in All Text near/6 diabet* in All Text))
10. (nonketo* in All Text near/6 diabet*in All Text)
11. (adult* in All Text near/6 diabet*in All Text)
12. (matur* in All Text near/6 diabet*in All Text)
13. (late in All Text near/6 diabet*in All Text)
14. (slow in All Text near/6 diabet*in All Text)
15. (stabl* in All Text near/6 diabet*in All Text)
16. (insulin* in All Text and (defic*in All Text near/6 diabet* in All Text)
17. (plurimetabolic in All Text and syndrom*in All Text)
18. (pluri in All Text and metabolicin All Text and syndrom* in All Text)
19. (#1 or #2 or #3or #4 or #5 or #6 or #7or #8 or #9 or #10)
20. (#11 or #12 or #13or #14 or #15 or #16 or #17or #18)
21. (#19 or #20)
22. MeSH descriptor Diabetes insipidusexplode all trees
23. (diabet* in All Text and insipidusin All Text)
24. (#22 or #23)
25. (#21 and not #24)
26. MeSH descriptor Blood glucoseexplode all trees
27. MeSH descriptor Hyperglycemiaexplode all trees
28. MeSH descriptor Hemoglobin A, glycosylatedexplode all trees
29. ((blood in All Text and glucos*in All Text) or hyperglycaemi* in All Text or hyperglycemi*in All Text or (haemoglobin* in All Text and Ain All Text) or (hemoglobin* in All Text and Ain All Text))
30. (HbA1C in All Text or (Hbin All Text and A in All Text) or (HbA in All Text and 1c in All Text) or HbA in All Text or A1Cs in All Text)
31. (glycosylated in All Text near/6 haemoglobin*in All Text)
32. (glycosylated in All Text near/6 hemoglobin*in All Text)
33. (glucos* in All Text near/3 management*in All Text)
34. (#26 or #27 or #28or #29 or #30 or #31 or #32or #33)
35. (#25 or #34)
36. (intensi* in All Text near/3 control*in All Text)
37. (intensi* in All Text near/3 therap*in All Text)
38. (intensi* in All Text near/3 treatment*in All Text)
39. (intensi* in All Text near/3 intervention*in All Text)
40. (intensi* in All Text near/3 management*in All Text)
41. (conventional* in All Text near/3 control*in All Text)
42. (conventional* in All Text near/3 therap*in All Text)
43. (conventional* in All Text near/3 treatment*in All Text)
44. (conventional* in All Text near/3 intervention*in All Text)
45. (conventional in All Text near/3 management*in All Text)
46. (regular in All Text near/3 control*in All Text)
47. (regular in All Text near/3 therap*in All Text)
48. (regular in All Text near/3 treatment*in All Text)
49. (regular in All Text near/3 intervention*in All Text)
50. (regular in All Text near/3 management*in All Text)
51. (usual in All Text near/3 control*in All Text)
52. (usual in All Text near/3 therap*in All Text)
53. (usual in All Text near/3 treatmentin All Text)
54. (usual in All Text near/3 intervention*in All Text)
55. (usual in All Text near/3 management*in All Text)
56. (routin* in All Text near/3 control*in All Text)
57. (routin* in All Text near/3 therap*in All Text)
58. (routin* in All Text near/3 treatment*in All Text)
59. (routin* in All Text near/3 intervention*in All Text)
60. (routin* in All Text near/3 management*in All Text)
61. (tight in All Text near/3 control*in All Text)
62. (tight in All Text near/3 therap*in All Text)
63. (tight in All Text near/3 treatment*in All Text)
64. (tight in All Text near/3 intervention*in All Text)
65. (tight in All Text near/3 management*in All Text)
66. (#36 or #37 or #38or #39 or #40 or #41 or #42or #43 or #44 or #45 or #46or #47 or #48 or #49 or #50or #51 or #52 or #53 or #54or #55 or #56 or #57 or #58 or #59 or #60 or #61 or #62 or #63 or #64 or #65)
67. (#35 and #66)
MEDLINE
1. exp Blood Glucose/
2. exp Hyperglycemia/
3. exp Hemoglobin A, Glycosylated/
4. (blood glucos$ or hyperglyc?emi$ or h?emoglobin$ A).ab,ti.
5. (HbA1C or Hb A or HbA 1c or HbA or A1Cs).ab,ti,ot.
6. (glycosylated adj6 h?emoglobin$).ab,ti.
7. (glucos$ adj3 management$).ab,ti.
8. or/1-7
9. exp Diabetes Mellitus, Type 2/
10. exp Diabetes Complications/
11. (MODY or NIDDM or T2DM).tw,ot.
12. (non insulin$ depend$ or noninsulin$ depend$ or noninsulin?depend$ or non insulin?depend).tw,ot.
13. ((typ$ 2 or typ$ II) adj3 diabet$).tw,ot.
14. ((keto?resist$ or non?keto$) adj6 diabet$).tw,ot.
15. (((late or adult$ or matur$ or slow or stabl$) adj3 onset) and diabet$).ab,ti.
16. or/9-15
17. exp Diabetes Insipidus/
18. diabet$ insipidus.tw,ot.
19. 17 or 18
20. 16 not 19
21. 8 or 20
22. ((intensi$ or conventional$ or regular or tight or usual or routin$  or or standard) adj3 (control$ or therap$ or treatment or intervention$ or management$)).ab,ti.
23. 21 and 22
24. randomized controlled trial.pt.
25. controlled clinical trial.pt.
26. randomi?ed.ab,ti.
27. placebo$.ab,ti.
28. drug therapy.fs.
29. randomly.ab,ti.
30. trial$.ab,ti.
31. group$.ab,ti.
32. or/24-31
33. Meta-analysis.pt.
34. exp Technology Assessment, Biomedical/
35. exp Meta-analysis/
36. exp Meta-analysis as topic/
37. hta.tw,ot.
38. (health technology adj6 assessment$).tw,ot.
39. (meta analy$ or metaanaly$ or meta?analy$).tw,ot.
40. ((review$ or search$) adj10 (literature$ or medical database$ or medline or pubmed or embase or cochrane or cinahl or psycinfo or psyclit or healthstar or biosis or current content$ or systemat$)).tw,ot.
41. or/33-40
42. (comment or editorial or historical-article).pt.
43. 41 not 42
44. 32 or 43
45. 23 and 44
46. (animals not (animals and humans)).sh.
47. 45 not 46
EMBASE
1. exp Diabetes Mellitus, Type 2/
2. exp Insulin Resistance/
3. impaired glucose toleranc$.ab,ti,ot.
4. glucose intoleranc$.ab,ti,ot.
5. insulin$ resistanc$.ab,ti,ot.
6. (obes$ adj diabet$).ab,ti,ot.
7. (MODY or NIDDM or TDM2).ab,ti,ot.
8. (non insulin$ depend$ or noninsulin depend$ or noninsulin?depend$ or non insulin?depend$).ab,ti,ot.
9. ((typ$ 2 or typ$ II) adj diabet$).ab,ti,ot.
10. (diabet$ adj (typ$ 2 or typ$ II)).ab,ti,ot.
11. ((keto?resist$ or non?keto$) adj diabet$).ab,ti,ot.
12. ((adult$ or matur$ or late or slow or stabl$) adj diabet$).ab,ti,ot.
13. (insulin$ defic$ adj relativ$).ab,ti,ot.
14. pluri?metabolic$ syndrom$.ab,ti,ot.
15. or/1-14
16. exp Diabetes Insipidus/
17. diabet$ insipidus.ab,ti,ot.
18. 16 or 17
19. 15 not 18
20. exp Glucose Blood Level/
21. exp Hyperglycemia/
22. exp Glycosylated Hemoglobin/
23. (blood glucos$ or hyperglyc?emi$ or h?emoglobin$ A).ab,ti,ot.
24. (HbA1C or Hb A or HbA 1c or HbA or A1Cs).ab,ti,ot.
25. (glycosylated adj6 h?emoglobin$).ab,ti,ot.
26. (glucos$ adj3 management$).ab,ti,ot.
27. or/20-25
28. 19 or 27
29. ((intensiv$ or conventional$ or regular or tight or usual or routin$) adj3 (control$ or therap$ or treatment or intervention$ or management$)).ab,ti,ot.
30. 28 and 29
31. Randomized Controlled Trial/
32. exp Controlled Clinical Trial/
33. randomi?ed.ab,ti.
34. placebo$.ab,ti.
35. exp Drug Therapy/
36. randomly.ab,ti.
37. trial$.ab,ti.
38. group$.ab,ti.
39. or/31-38
40. exp meta analysis/
41. (metaanaly$ or meta analy$ or meta?analy$).ab,ti,ot.
42. ((review$ or search$) adj10 (literature$ or medical database$ or medline or pubmed or embase or cochrane or cinahl or psycinfo or psyclit or healthstar or biosis or current content$ or systematic$)).ab,ti,ot.
43. exp Literature/
44. exp Biomedical Technology Assessment/
45. hta.tw,ot.
46. (health technology adj6 assessment$).tw,ot.
47. or/40-46
48. (comment or editorial or historical-article).pt.
49. 47 not 48
50. 39 or 49
51. 30 and 50
52. limit 51 to human
LILACS
1. (Blood Glucose or Hyperglycemia or hemoglobin A, glycosylated or Diabetes mellitus) [Subject descriptor]
and
2. (control$ or management) [Palavras]
and 
3. (random$ or placebo$ or trial or group$) [Palavras]
CINAHL
1. MM "Blood Glucose"
2. MM "Glycemic Control"
3. MM "Hyperglycemia+"
4. MM "Hemoglobin A, Glycosylated"
5. TI (blood glucos* OR hyperglyc?emi* OR h?emoglobin A) or AB (blood glucos* OR hyperglyc?emi* OR h?emoglobin A)
6. TI (HbA1C or Hb A or HbA 1c or HbA or A1Cs) or AB (HbA1C or Hb A or HbA 1c or HbA or A1Cs)
7. TI glycosylated N6 h?emoglobin* or AB glycosylated N6 h?emoglobin*
8. TI glucos* N3 management* or AB glucos* N3 management*
9. #1 or #2 or #3 or #4 or #5 or #6 or #7 or #8
10. MM "Diabetes Mellitus, Non-Insulin-Dependent"
11. TX Diabetes Complications
12. TX MODY or NIDDM or T2DM
13. TX non insulin* depend* or noninsulin* depend* or noninsulin?depend* or non insulin?depend
14. TX diabet* AND (typ* 2 or typ* II)
15. TX diabet* AND (keto*resist* or non*keto*)
16. TI (onset AND (late or adult* or matur* or slow or stabl*)) and TI diabet*
17. AB (onset N3 (late or adult* or matur* or slow or stabl*)) and AB diabet*
18. #10 or #11 or #12 or #13 or #14 or #15 or #16 or #17
19. MM "Diabetes Insipidus"
20. TX diabet* insipidus
21. #19 or #20
22. #18 NOT #21
23. #9 or #22
24. TI (control* AND (intensi* or tight or conventional* or regular or usual or routin* or standard)) or AB (control* N3 (intensi* or tight or conventional* or regular or usual or routin* or standard))
25. TI (therap* AND (intensi* or tight or conventional* or regular or usual or routin* or standard)) or AB (therap* N3 (intensi* or tight or conventional* or regular or usual or routin* or standard))
26. TI (treatment* N3 (intensi* or tight or conventional* or regular or usual or routin* or standard)) or AB (treatment* N3 (intensi* or tight or conventional* or regular or usual or routin* or standard))
27. TI (intervention* N3 (intensi* or tight or conventional* or regular or usual or routin* or standard)) or AB (intervention* N3 (intensi* or tight or conventional* or regular or usual or routin* or standard))
28. TI ( management* N3 (intensi* or tight or conventional* or regular or usual or routin* or standard)) or AB (management* N3 (intensi* or tight or conventional* or regular or usual or routin* or standard))
29. #24 or #25 or #26 or #27 or #28
30. #23 and #29
31. TX random* OR blind* OR placebo* OR group*
32. TX animal* NOT (animal* AND human*)
33. #31 NOT #32
34. #30 and #33
Science Citation Index Expanded
1. TS=(blood glucos* or glyc?emic* control or hyperglyc?emi* or h?emoglobin* A)
2. TS=(HbA1C or Hb A or HbA 1c or HbA or A1Cs)
3. TS=(glycosylated SAME h?emoglobin*)
4. TS=(glucos* SAME management*)
5. #4 OR #3 OR #2 OR #1
6. TS=(MODY or NIDDM or T2DM)
7. TS=(non insulin* depend* or noninsulin* depend* or noninsulin?depend* or non insulin?depend*)
8. TS=(diabet* SAME (typ* 2 or typ* II))
9. TS=(diabet* SAME (keto*resist* or non*keto*))
10. TS=((onset SAME (late or adult* or matur* or slow or stabl*)) and diabet*)
11. #10 OR #9 OR #8 OR #7 OR #6
12. #11 NOT TS=(diabet* insipidus)
13. #12 OR #5
14. TS=((intensi* or tight or conventional* or regular or usual or routin* or standard) SAME (control* or therap* or treatment* or intervention* or management*))
15. #14 AND #13
16. TS=(random* OR blind* OR placebo* OR group*)
17. TS=(animal* NOT (animal* AND human*))
18. #16 NOT #17
19. #18 AND #15

Appendix 2. Interventions in trials

Characteristic

Study ID

Intervention(s) and
control(s)
Number of units of insulin/day
[mean (SD)]
Number of units of
insulin/day/kg body weight [mean (SD)]
Monotherapy used Combinationtherapy used
ACCORD 2008I: targeting intensive glycaemic control--yesyes
C: targeting conventional glycaemic control--yesyes
ADDITION-Europe 2011I: targeting intensive glycaemic control--yesyes
C: targeting conventional glycaemic control--yesyes
ADVANCE 2008I: targeting intensive glycaemic control37.3 (28.4)0.5 (0.3)yesyes
C: targeting conventional glycaemic control39.8 (27.2)0.5 (0.3)yesyes
Araki 2012I: targeting intensive glycaemic control--yesyes
C: targeting conventional glycaemic control--yesyes
Bagg 2001I: targeting intensive glycaemic control-0.8 (0.1)yesyes
C: targeting conventional glycaemic control-0.4 (0.1)yesyes
Becker 2003I: targeting intensive glycaemic control--yesyes
C: targeting conventional glycaemic control--yesyes
Blonde 2009I: targeting intensive glycaemic control-0.6noyes
C: targeting conventional glycaemic control-0.5noyes
Cao 2011I: targeting intensive glycaemic control--yesyes
C: targeting conventional glycaemic control--yesyes
Cooray 2011I: targeting intensive glycaemic control--yesyes
C: targeting conventional glycaemic control--yesyes
DIGAMI 2 2005I: targeting intensive glycaemic control--yesyes
C: targeting conventional glycaemic control--yesyes
Fantin 2011I: targeting intensive glycaemic controlMedian: 47 IQR: 29-127-yesno
C: targeting conventional glycaemic controlMedian: 0 IQR: 0-5-yesno
Guo 2008I: targeting intensive glycaemic control--yesyes
C: targeting conventional glycaemic control--yesyes
IDA 2009I: targeting intensive glycaemic control--yesyes
C: targeting conventional glycaemic control--yesyes
Jaber 1996a I: targeting intensive glycaemic control--yesno
C: targeting conventional glycaemic control--yesyes
Kumamoto 2000I: targeting intensive glycaemic control--yesyes
C: targeting conventional glycaemic control--yesyes
Lu 2010I: targeting intensive glycaemic control--yesyes
C: targeting conventional glycaemic control--yesyes
Melidonis 2000I: targeting intensive glycaemic control--yesno
C: targeting conventional glycaemic control--yesyes
Natarajan 2012I: targeting intensive glycaemic control34.7 (23.6)-yesyes
C: targeting conventional glycaemic control--yesyes
REMBO 2008I: targeting intensive glycaemic control--yesyes
C: targeting conventional glycaemic control--yesyes
Service 1983b I: targeting intensive glycaemic control--yesyes?
C: targeting conventional glycaemic control--yesyes?
Stefanidis 2003I: targeting intensive glycaemic control38 (10)-yesyes
C: targeting conventional glycaemic control--noyes
Steno-2 2008c I: targeting intensive glycaemic control75 (57)0.7 (0.5)yesyes
C: targeting conventional glycaemic control75 (61)0.8 (0.5)yesyes
UGDP 1975d I: targeting intensive glycaemic control47.0 (38.0)-yesno
C: targeting conventional glycaemic control13.9 (1.7)-yesno
UKPDS 1998e I: targeting intensive glycaemic controlMedian: 36 (22.2)-yesyes
C: targeting conventional glycaemic control--yesyes
VA CSDM 1995f I: targeting intensive glycaemic control97.5 (26)1.0yesyes
C: targeting conventional glycaemic control57.5 (44.2)0.6yesno
VADT 2009g I: targeting intensive glycaemic controlMedian: 56 (48.1)0.5yesyes
C: targeting conventional glycaemic controlMedian: 45 (40.7)0.5yesyes
Yang 2007I: targeting intensive glycaemic control--yes-
C: targeting conventional glycaemic control--yesyes
Zhang 2011I: targeting intensive glycaemic control--yesyes
C: targeting conventional glycaemic control--yesyes

Footnotes

"-" denotes not reported

aIt is not explicit in the text whether monotherapy was used only or combination therapy in the intensive treatment group.

bIt is not explicit in the text whether monotherapy was used only or if combination therapy also was allowed in the intensive and conventional treatment group.

cAll numbers are from the end of the intervention period (7.8 years of follow-up). The doses of insulin did not have a normal distribution..

dStandard deviations calculated from standard errors.

eStandard deviations for number of units of insulin/day calculated from interquartile ranges. Data are from the UKPDS 33.

fThe number of units of insulin/day is estimated from figure after 24 months of follow-up. The standard deviations for insulin doses are calculated from standards errors.

gData from insulin doses are medians. Standard deviations calculated from interquartile ranges. Data on insulin doses only available after 4 years of follow-up.

SD: standard deviation

ACCORD: Action to Control Cardiovascular Risk in Diabetes Study; ADDITION: Anglo-Danish-Dutch study of Intensive Treatment In PeOple with screeN
detected diabetes in primary care; ADVANCE: Action in Diabetes and Vascular disease – PreterAx and DiamicroN MR Controlled Evaluation; DIGAMI: Diabetes and Insulin-Glucose Infusion in Acute Myocardial Infarction; IDA: Insulin Diabetes Angioplasty; REMBO: Rational Effective Multicomponent Therapy in the Struggle Against DiaBetes Mellitus in Patients With COngestve Heart Failure; UGDP: University Group Diabetes Program; UKPDS: United Kingdom Prospective Diabetes Study; VACSDM: Veterans Affairs Cooperative Study in Type 2 Diabetes Mellitus; VADT: Veterans Affairs Diabetes Trial

Appendix 3. Baseline characteristics (I)

Characteristic

Study ID

Intervention(s) and
control(s)
Duration of intervention
(duration of follow-up)
Participating
population
CountrySettingDuration of disease
[mean/range years (SD), or as reported]
ACCORD 2008I: targeting intensive glycaemic controlMean (median) 3.5 (3.4) yr
(median 5.0 yr)
Type 2 diabetes mellitus and high risk of cardiovascular diseaseCanada
USA
Outpatients10 (median)
C: targeting conventional glycaemic control10 (median)
ADDITION-Europe 2011I: targeting intensive glycaemic control5 yr (5.3 (1.6) yr)Type 2 diabetes mellitus, screen detectedUnited Kingdom
Denmark
the Netherlands
Genral practiceAll newly diagnosed
C: targeting conventional glycaemic controlAll newly diagnosed
ADVANCE 2008I: targeting intensive glycaemic control5 yr (median 5 yr)Type 2 diabetes mellitus and elevated risk of cardiovascular disease20 countries from
Asia, Australia,
Europe, North America
Outpatients7.9 (6.3)
C: targeting conventional glycaemic control8.0 (6.4)
Araki 2012I: targeting intensive glycaemic control6 years (6 years)Type 2 diabetes mellitusJapanOutpatients16.7
C: targeting conventional glycaemic control18.0
Bagg 2001I: targeting intensive glycaemic control20 weeks (20 weeks)Type 2 diabetes mellitusNew ZealandOutpatients7.9 (4.5)
C: targeting conventional glycaemic control5.9 (3.2)
Becker 2003I: targeting intensive glycaemic controlMean 22 months (mean 22 months)Type 2 diabetes mellitusNetherlandsOutpatients and general practitioners3.4
C: targeting conventional glycaemic control3.2
Blonde 2009I: targeting intensive glycaemic control20 weeks (20 weeks)Type 2 diabetes mellitusUSAOutpatients7.9
C: targeting conventional glycaemic control20 weeks (20 weeks)9.0
Cao 2011I: targeting intensive glycaemic control28 days (28 days)Type 2 diabetes mellitus and undergoing gastrectomy for gastric cancerJapanHospital5.5 (median)
C: targeting conventional glycaemic control28 days (28 days)6.0 (median)
Cooray 2011I: targeting intensive glycaemic control2 months (2 months)Type 2 diabetes mellitusSwedenOutpatients8.6 (range: 1-22)
C: targeting conventional glycaemic control2 months (2 months)11.2 (range: 2-28)
DIGAMI 2 2005I: targeting intensive glycaemic controlMedian 2.1 yr; IQR 1.03-3.00 yr (2.1; data from epidemiological database: 4.1 yr)Type 2 diabetes mellitus with suspect acute myocardial infarctionSweden, Finland, Norway, Denmark, The Netherlands, and the UKHospital (coronary care units)7.9 (8.2)
C: targeting conventional glycaemic control8.3 (8.3)
Fantin 2011I: targeting intensive glycaemic control24 hours (6 months)Type 2 diabetes mellitusBrazilHospital9.1 (7.8)
C: targeting conventional glycaemic control13.3 (12)
Guo 2008I: targeting intensive glycaemic control6 months (6 months)Type 2 diabetes mellitusChinaOutpatientsLess than 1 year
C: targeting conventional glycaemic controlLess than 1 year
IDA 2009I: targeting intensive glycaemic control6 months and 3 weeks; attempts were made to optimise glycaemic control during three weeks preceding the percutaneous coronary intervention in the intensive intervention group (6 months and 3 weeks)Type 2 diabetes mellitus and undergoing percutaneous coronary interventionSwedenHospital and outpatient6.4 (5.8)
C: targeting conventional glycaemic control6.5 (7.4)
Jaber 1996I: targeting intensive glycaemic control4 months (4 months)Type 2 diabetes mellitusUSAOutpatients6.8 (6.5)
C: targeting conventional glycaemic control6.2 (4.8)
Kumamoto 2000I: targeting intensive glycaemic control10 years (10 years)Type 2 diabetes mellitusJapanOutpatients8.6 (5.4)
C: targeting conventional glycaemic control8.5 (5.2)
Lu 2010I: targeting intensive glycaemic control12 weeks (12 weeks)Type 2 diabetes mellitus with microalbuminuriaChinaOutpatients8.0 (4.0)
C: targeting conventional glycaemic control8.3 (4.7)
Melidonis 2000I: targeting intensive glycaemic control6 days (6 days)Type 2 diabetes mellitus with acute coronary eventGreeceHospital (coronary care unit)10.5 (4.4)
C: targeting conventional glycaemic control12.4 (range: 3-38)
Natarajan 2012I: targeting intensive glycaemic control6 months (12 months)Type 2 diabetes mellitus and undergoing catheter-based revascularizationCanadaOutpatients9.7 (10.4)
C: targeting conventional glycaemic control7.9 (11.7)
REMBO 2008I: targeting intensive glycaemic control12 months (12 months)Type 2 diabetes mellitus and heart insuffienceyRussiaOutpatients5.0 (7.4)
C: targeting conventional glycaemic control6.0 (8.5)
Service 1983I: targeting intensive glycaemic control1.5 yr (1.5 yr)Type 2 diabetes mellitusUSAOutpatients0.1
C: targeting conventional glycaemic control2.0 yr (2.0 yr)0.8
Stefanidis 2003I: targeting intensive glycaemic control72 hours (72 hours)Type 2 diabetes mellitus with acute coronary eventGreeceHospital16 (7)
C: targeting conventional glycaemic control15 (9)
Steno-2 2008I: targeting intensive glycaemic control7.8 yr (13.3 yr)Type 2 diabetes mellitus and microalbuminuriaDenmarkOutpatients5.5 (5.0)
C: targeting conventional glycaemic control6.0 (4.4)
UGDP 1975I: targeting intensive glycaemic controlMean 12 years; range 10-14.5 years (mean 12 years)Type 2 diabetes mellitus, newly diagnosedUSAOutpatientsAll patients were diagnosed with type 2 diabetes within 12 months prior to enrolment in the trial
C: targeting conventional glycaemic control
UKPDS 1998I: targeting intensive glycaemic controlMedian of 10.0 years; IQR 7.7-12.4. For the participants taking part in the UKPDS 34, the median was 10.7 years (the median follow-up for endpoint analyses was 10.0 years (IQR 7.7-12.4). For UKPDS 34 the median follow-up was 10.7 years)Type 2 diabetes mellitus, newly diagnosedUnited KingdomOutpatientsAll participants were newly diagnosed with type 2 diabetes mellitus
C: targeting conventional glycaemic control
VA CSDM 1995I: targeting intensive glycaemic control27 months (27 months)Type 2 diabetes mellitusUSAOutpatients8.0 (3.6)
C: targeting conventional glycaemic control7.7 (4.3)
VADT 2009I: targeting intensive glycaemic control5.6 yr (median 5.6 yr, up to 7.5 yr)Type 2 diabetes mellitusUSAOutpatients11.5 (8.0)
C: targeting conventional glycaemic control11.5 (7.0)
Yang 2007I: targeting intensive glycaemic control1 yr (1 yr)Type 2 diabetes mellitusChinaOutpatients1
C: targeting conventional glycaemic control1
Zhang 2011I: targeting intensive glycaemic control5 yr (5 yr)Type 2 diabetes mellitus and one or more risk factors for cardiovascular diseaseChinaOutpatients7.9 (5.0)
C: targeting conventional glycaemic control8.9 (4.7)

Footnotes

"-" denotes not reported

SD: standard deviation

ACCORD: Action to Control Cardiovascular Risk in Diabetes Study; ADDITION: Anglo-Danish-Dutch study of Intensive Treatment In PeOple with screeN detected diabetes in primary care; ADVANCE: Action in Diabetes and Vascular disease – PreterAx and DiamicroN MR Controlled Evaluation; DIGAMI: Diabetes and Insulin-Glucose Infusion in Acute Myocardial Infarction; IDA: Insulin Diabetes Angioplasty; REMBO: Rational Effective Multicomponent Therapy in the Struggle Against DiaBetes Mellitus in Patients With COngestve Heart Failure; UGDP: University Group Diabetes Program; UKPDS: United Kingdom Prospective Diabetes Study; VACSDM: Veterans Affairs Cooperative Study in Type 2 Diabetes Mellitus; VADT: Veterans Affairs Diabetes Trial

Appendix 4. Baseline characteristics (II)

Characteristic

Study ID

Intervention(s) and
control(s)
Sex
[N female/ male]
Age
[mean/range
years (SD),
or as reported]
Fasting
blood glucose [mean mmol/L (SD)]
HbA1c
[mean %
(SD)]
BMI
[mean kg/m2 (SD)]
Co-morbidities
[N]
ACCORD 2008I: targeting intensive glycaemic control1985/314362.2 (6.8)9.7 (3.1)8.3 (1.1)32.2 (5.5)Cardiovascular disease: 1826
C: targeting conventional glycaemic control1967/315662.2 (6.8)9.8 (3.1)8.3 (1.1)32.2 (5.5)Cardiovascular disease: 1783
ADDITION-Europe 2011I: targeting intensive glycaemic control697/98160.3 (6.9)-7.0 (1.6)31.6 (5.6)History of myocardial infarction: 109
C: targeting conventional glycaemic control589/79060.2 (6.8)-7.0 (1.5)31.6 (5.6)History of myocardial infarction: 79
ADVANCE 2008I: targeting intensive glycaemic control2376/319566 (6)8.5 (2.8)7.5 (1.7)28 (5)Cardiovascular disease: 1794
C: targeting conventional glycaemic control2357/321266 (6)8.5 (2.8)7.5 (1.6)28 (5)Cardiovascular disease: 1796
Araki 2012I: targeting intensive glycaemic control314/27171.99.3 (2.7)8.4 (0.8)24.0 (3.9)Ischaemic heart disease: 87
C: targeting conventional glycaemic control316/27271.79.4 (2.9)8.5 (0.9)24.3 (7.3)Ischaemic heart disease: 96
Bagg 2001  I: targeting intensive glycaemic control12/957.2 (7.4)13.7 (0.6)10.8 (0.2)31.9 (1.1)Cardiovascular disease: 2
C: targeting conventional glycaemic control12/1054.5 (9.2)13.2 (0.6)10.5 (0.2)29.4 (1.1)Cardiovascular disease: 2
Becker 2003I: targeting intensive glycaemic control50/5663.3 (8.4)9.4 (2.8)-28.0 (4.8)Cardiovascular disease: 21
C: targeting conventional glycaemic control60/4863.3 (8.3)9.7 (3.3)-29.1 (4.3)Cardiovascular disease: 23
Blonde 2009I: targeting intensive glycaemic control54/6856.69.18.033.0-
C: targeting conventional glycaemic control44/7857.29.17.933.6-
Cao 2011I: targeting intensive glycaemic control64/2858.2 (6.3)6.8 (0.6)7.5 (0.7)21.1 (2.0)Coronary artery disease: 2
C: targeting conventional glycaemic control57/3059.4 (7.3)7.0 (0.7)7.3 (0.6)22.2 (2.6)Coronary artery disease: 1
Cooray 2011I: targeting intensive glycaemic controlOnly reported for both groups togetherOnly reported for
both groups together: 61 (range: 51-69)
-8.0 (range: 5.8-12.9)30 (range: 23-39)-
C: targeting conventional glycaemic controlOnly reported for both groups togetherOnly reported for
both groups together 61 (range: 51-69)
-8.0 (range: 5.1-10.0)29 (range: 23-37)-
DIGAMI 2 2005I: targeting intensive glycaemic control156/31868.1 (11.4)12.8 (4.5)7.2 (1.7)28.3 (4.9)Cardiovascular disease: 474
C: targeting conventional glycaemic control97/20968.4 (11.2)12.9 (4.6)7.3 (1.7)28.4 (4.4)Cardiovascular disease: 306
Fantin 2011I: targeting intensive glycaemic control14/2160.3 (10)         7.8 (3.7)8.0 (1.8)30 (5.6)Previous myocardial infarction: 12
C: targeting conventional glycaemic control14/2160.7 (10.3)       8.7 (4.1)7.5 (1.6)29.5 (5.4)Previous myocardial infarction: 14
Guo 2008I: targeting intensive glycaemic control-49.3 (8.8)8.2 (2.6)7.1 (1.9)25.7 (3.1)Cardiovascular disease: -
C: targeting conventional glycaemic control-48.3 (8.7)9.0 (2.5)7.7 (2.5)25.3 (4.1)Cardiovascular disease: -
all:92/128 
IDA 2009I: targeting intensive glycaemic control10/2966 (9.6)7.0 (1.9)6.5 (1.4)-Cardiovascular disease: 51
C: targeting conventional glycaemic control10/3362 (6.7)7.3 (1.6)6.5 (1.3)-Cardiovascular disease: 51
Jaber 1996I: targeting intensive glycaemic control12/559 (12)11.1 (4.0)11.5 (2.9)34 (7)Cardiovascular disease: -
C: targeting conventional glycaemic control15/765 (12)12.7 (4.7)12.2 (3.5)33 (7)Cardiovascular disease: -
Kumamoto 2000I: targeting intensive glycaemic control27/2848.2 (11)9.4 (1.8)9.4 (1.6)20.5 (2.1)Cardiovascular disease: 0
C: targeting conventional glycaemic control29/2650.9 (14)9.0 (1.9)8.9 (1.4)20.4 (2.6)Cardiovascular disease: 0
Lu 2010I: targeting intensive glycaemic control7/1457.5 (11.0)9.2 (4.3)8.8 (4.4)24.5 (3.6)Cardiovascular disease: -
C: targeting conventional glycaemic control6/1461.5 (10.4)9.4 (3.8)9.1 (2.6)24.2 (4.2)Cardiovascular disease: -
Melidonis 2000I: targeting intensive glycaemic control11/1366.6 (6.7)13.2 (3.6)7.6 (0.6)26.3 (4.0)Cardiovascular disease: 24
C: targeting conventional glycaemic control8/1666.5 (9.6)13.9 (3.9)7.9 (0.8)27.4 (5.0)Cardiovascular disease: 24
Natarajan 2012I: targeting intensive glycaemic control6/3958.7 (9.5)9.3 (3.8)8.0 (1.2)30.9 (5.9)Previous myocardial infarction: 7
C: targeting conventional glycaemic control14/2862.9 (10)8.4 (2.4)7.5 (1.2)31.9 (6.3)Previous myocardial infarction: 13
REMBO 2008I: targeting intensive glycaemic control10/3164 (11.9)6.5 (1.3)7.1 (1.2)31.6 (5.0)Cardiovascular disease: 41
C: targeting conventional glycaemic control14/2664 (7.4)6.6 (1.9)7.2 (1.4)30.1 (4.4)Cardiovascular disease: 40
Service 1983I: targeting intensive glycaemic control3/5449.911.4-Cardiovascular disease: -
C: targeting conventional glycaemic control3/7567.711.4-Cardiovascular disease: -
Stefanidis 2003I: targeting intensive glycaemic control14/2266 (11)15.4 (5.2)8.0 (1.0)28 (3.1)Cardiovascular disease: 36
C: targeting conventional glycaemic control18/2168 (9)14.8 (5.6)8.2 (1.2)27.5 (3.2)Cardiovascular disease: 39
Steno-2 2008I: targeting intensive glycaemic control17/6354.9 (7.2)10.1 (3.1)8.4 (1.6)29.7 (3.8)Cardiovascular disease: 18
C: targeting conventional glycaemic control24/5655.2 (7.2)10.5 (3.0)8.8 (1.7)29.9 (4.9)Cardiovascular disease: 21
UGDP 1975I: targeting intensive glycaemic control158/46-7.8--Cardiovascular disease: 7
C: targeting conventional glycaemic control153/57-7.9--Cardiovascular disease: 16
all:52.7 (11.2) 
UKPDS 1998I: targeting intensive glycaemic control1260/181153.2 (8.6)8.1 (1.9)7.1 (1.5)27.5 (5.1)Cardiovascular disease: -
C: targeting conventional glycaemic control433/70553.4 (8.6)8.0 (2.0)7.1 (1.4)27.8 (5.5)Cardiovascular disease: -
all:Cardiovascular disease: 77
VA CSDM 1995I: targeting intensive glycaemic control0/7560.4 (6.4)11.4 (0.4)9.3 (0.2)30.7 (4.4)Cardiovascular disease: 31
C: targeting conventional glycaemic control0/7859.9 (6.7)12.4 (0.4)9.5 (0.2)31.3 (5.5)Cardiovascular disease: 27
VADT 2009I: targeting intensive glycaemic control26/86660.5 (9.0)10.8 (4.0)9.4 (2.0)31.3 (3.0)Cardiovascular disease: 355
C: targeting conventional glycaemic control26/87360.3 (9.0)11.0 (3.7)9.4 (2.0)31.2 (4.0)Cardiovascular disease: 368
Yang 2007I: targeting intensive glycaemic control-50 (8)7.2 (1.7)7.4 (1.7)26 (3.4)Cardiovascular disease: -
C: targeting conventional glycaemic control-53 (9)7.3 (1.9)6.9 (1.2)25.6 (3.5)Cardiovascular disease: -
Zhang 2011I: targeting intensive glycaemic control16/3267.8 (5.3)7.8 (2.1)7.4 (1.3)24.2 (2.5)Cardiovascular disease: 9
C: targeting conventional glycaemic control19/3066.6 (4.8)7.8 (2.4)7.4 (1.3)24.9 (2.6)Cardiovascular disease: 8

Footnotes

"-" denotes not reported

BMI: body mass index; HbA1c: glycosylated haemoglobin A1c; SD: standard deviation

ACCORD: Action to Control Cardiovascular Risk in Diabetes Study; ADDITION: Anglo-Danish-Dutch study of Intensive Treatment In PeOple with screeN detected diabetes in primary care; ADVANCE: Action in Diabetes and Vascular disease – PreterAx and DiamicroN MR Controlled Evaluation; DIGAMI: Diabetes and Insulin-Glucose Infusion in Acute Myocardial Infarction; IDA: Insulin Diabetes Angioplasty; REMBO: Rational Effective Multicomponent Therapy in the Struggle Against DiaBetes Mellitus in Patients With COngestve Heart Failure; UGDP: University Group Diabetes Program; UKPDS: United Kingdom Prospective Diabetes Study; VACSDM: Veterans Affairs Cooperative Study in Type 2 Diabetes Mellitus; VADT: Veterans Affairs Diabetes Trial

Appendix 5. Definition of mortality and cardiovascular outcomes in study or as reported

Characteristic

Study ID

Cardiovascular mortalityMacrovascular
complications (composite outcome)
Non-fatal myocardial infarctionNon-fatal strokeAmputation of lower extremityCardial revascularizationPeripheral revascularization
ACCORD 2008Unexpected death and death due to myocardial infarction, congestive heart failure, after invasive cardiovascular interventions, arrhythmia, stroke, cardiovascular causes after non-cardiovascular surgery, other cardiovascular diseases (e.g., pulmonary emboli or abdominal aortic rupture), and presumed cardiovascular death (every component described in details in study protocol p 87-88)Non-fatal myocardial infarction or non-fatal stroke or death from cardiovascular causesProlonged ischaemic symptoms > 20 minutes and or raised cardiac enzymes (Troponin T or
I and/or serum CK-MB), included Q-wave myocardial infarction, non Q-wave myocardial infarction, silent myocardial infarction, probable non Q-wave myocardial infarction, myocardial infarction after coronary bypass graft surgery, myocardial infarction after cardiovascular invasive interventions, and myocardial infarction after non-cardiovascular surgery
Included ischaemic stroke, primary intracerebral haemorrhage, subarachnoid haemorrhage, stroke of unknown aetiology, non-fatal stroke after cardiovascular invasive interventions, and non-fatal stroke after non-cardiovascular surgery (more details of each component, see study protocol p 89)Limb amputation: including partial or digit amputation due to vascular disease (a part of cardiovascular revascularization procedures)A part of cardiovascular revascularization procedures:
1. Percutaneous transluminal coronary angioplasty (balloon);
2. Percutaneous transluminal coronary angioplasty with stent;
3. coronary-artery bypass grafting
A part of cardiovascular revascularization procedures: Peripheral angioplasty with or without stent and peripheral vascular surgery (including aortic aneurysm repair)
ADDITION-Europe 2011

Acute myocardial infarction
Death from a new (within 30 days) confirmed acute myocardial
infarction, confirmed
• In hospital by appropriate biochemistry, electrocardiographies or imaging tests
or
• By autopsy findings showing a recent myocardial infarction or a
recent occluding coronary thrombus, whether or not the patient
was in hospital
Heart failure death
Death form worsening of heart failure, even if the terminal event was
an arrhythmia. Death will generally be preceded by persistent or
frequently recurrent New York Heart Association class IV symptoms an escalating need for supportive therapy and often by evidence of organ failure (e.g. renal).
Patients with cardiogenic shock or pulmonary oedema resistant for therapy should be included in this definition
Cardiac arrhythmic death
Documented sudden onset of arrhythmias directly leading to death or sudden unexpected death without evidence of acute myocardial
infarction, dissecting aneurysm, cardiovascular, or other attributable cause.
Cerebrovascular (stroke) death
Death occurring within 30 days from the onset of symptoms thought
to be due to a cerebrovascular event, including athero/thrombotic
infarction, embolism or haemorrhage, assuming no other more relevant intervening event. In the absence of other obvious causes for the sudden onset of neurological signs and symptoms the endpoint committee should presume a vascular cause. Death due to
subarachnoid or subdural haemorrhage should be included in this
category.
or
• By autopsy findings showing a recent cerebrovascular event,
including athero/thrombotic infarction, embolism or haemorrhage
whether or not the patient was in hospital.

Cardiovascular procedure-related death
Death related to an invasive cardiovascular procedure or surgery and
occurring within 30 days from the onset of the event.
Other cardiovascular death
Death from a cardiovascular cause not compatible with the previous
categories will be included in this category

Any of cardiovascular death,

myocardial infarction, stroke, revascularisation and amputation

Myocardial infarction (non-fatal)
Definition: The term myocardial infarction should be used when there is evidence of
myocardial necrosis in a clinical setting consistent with myocardial ischaemia. Under
these conditions any one of the following criteria meets the diagnosis for myocardial
infarction:
• Detection of rise and/or fall of cardiac biomarkers (preferably troponin) with at
least one value above the 99th percentile of the upper reference limit
together with evidence of myocardial ischaemia with at least one of the
following:
• Symptoms of ischaemia;
• electrocardiograph changes indicative of new ischaemia (new ST-T
changes or new left bundle branch block);
• Development of pathological Q waves in the electrocardiograph;
• Imaging evidence of new loss of viable myocardium or
new regional wall motion abnormality
The diagnosis of stroke requires evidence of a neurological deficit, usually localised,
lasting 24 hours or more, usually confirmed by diagnostic testing (e.g. Computer Tomografi-scan).
The clinical characteristics of stroke include the sudden onset of a neurological deficit
typically manifested as:
• Depression of state of consciousness
• Disturbance of vision
• Paresis of paralysis of one or more extremities
• Sensory impairment
• Speech impairment
• Central cranial nerve dysfunction
• Memory defect
• Ataxia
• Movement disorder
Amputation caused by cardiovascular disease,
amputation caused by neuropathic disease and amputation caused by a mixture of the above. Amputation caused by other reasons (e.g. traumatic)

Coronary artery bypass graft surgery

Percutaneous coronary interventions (including coronary artery bypass graft surgery) or coronary stent procedure

Attempt at Percutaneous coronary interventions/coronary artery bypass graft surgery: uses if it is attempted but the procedure did not succeed

for instance because the catheter could not pass the plaque or the plaque could not be reached for technical reasons

At lower extremities:

  • Bypass surgery

  • Percutaneous transluminal angioplasty (PTA) or stent placement

  • Attempt at PTA: use if PTA was attempted but not successful

At neck:

  • Bypass surgery

  • Thromboendarterectomy / thrombectomy (thrombus removing (surgical))

ADVANCE 2008Death from cardiovascular causesDeath from cardiovascular causes, non-fatal myocardial infarction, or non-fatal strokeNon-fatal myocardial infarctionNon-fatal strokeNDA part of total cardiovascular disease events (major coronary events, silent myocardial infarction, coronary revascularization, or hospital admission for unstable angina)Peripheral vascular events
Araki 2012NDMacroangiopathy: Ischemic heart disease, stroke and peripheral vascular disease

A. Diagnostic electrocardiogram at the time of the event

B. Ischaemic cardiac pain and diagnostic enzyme profile

C. Ischaemic cardiac pain and equivocal enzymes and equivocal electrocardiogram

D. A routine electrocardiogram diagnostic for myocardial infarction while the previous electrocardiogram was not

Stroke was defined as clinical signs of a focal neurological

deficit with rapid onset persisting < 24 hours confirmed by

either brain computed tomography or magnetic resonance

imaging

Diabetic footCoronary revascularizationND
Bagg 2001aNDNon-fatal stroke, unstable anginaNon-fatal myocardial infarctionNon-fatal strokeNDNDND
Becker 2003bNDMyocardial infarction, angina pectoris, stroke, transient ischaemic attack, and intermittent claudicationNDNDNDNDND
Blonde 2009NDNDNDNDNDNDND
Cao 2011Cerebrovascular accident, pulmonary embolism and acute cardiac failureNDNDNDNDNDND
Cooray 2011NDNDNDNDNDNDND
DIGAMI 2 2005cSudden cardiovascular deaths were those that occurred within 24 hour following onset of symptoms and without any other obvious reason for the fatal outcome. Deaths were labelled as cardiovascular or non-cardiovascular, and those without any obvious non-cardiovascular cause were considered cardiovascularDeath, reinfarction, or stroke

Myocardial infarction was diagnosed according to the joint recommendations of the European Society of Cardiology and the American College of Cardiology.

A reinfarction was defined as a new event > 72 hour from the index infarction

Stroke was defined as unequivocal signs of focal or global neurological deficit of sudden onset and a duration of > 24 hour that were judged to be of vascular originNDNot possible to divide the revascularizations into thrombolysis or invasive surgical interventionND
Fantin 2011NDMajor cardiovascular eventsMyocardial infarction was defined according to the criteria of the American College of CardiologyNon-fatal strokeAmputation of lower extremityCardial revascularizationPeripheral revascularization
Guo 2008NDNDNDNDNDNDND
IDA 2009dNDNew percutaneous coronary intervention, coronary bypass surgery, anginaNDNDNDNDND
Jaber 1996NDNDNDNDNDNDND
Kumamoto 2000Sudden death (probably myocardial infarction) and death due to cerebral vascular diseaseCardiovascular events (angina pectoris or myocardial infarction), cerebrovascular events (stroke), and peripheral vascular events (intermittent claudication, gangrene, or amputation)Non-fatal myocardial infarctionNon-fatal ischaemic stroke, non-fatal haemorraghic strokeAmputation of lower extremityCardial revascularizationPeripherial revascularization
Lu 2010NDNDNDNDNDNDND
Melidonis 2000eNDND

The diagnosis of AMI required fulfilment of at least two of the following criteria: 1. Anginal chest pain of at least 30 min duration;
2. development of new Q waves in 2 of the 12 electrocardiogram leads;
3. serum levels of creatine phosphokinase and creatine phosphokinase-MB fraction to more than twice the upper limit of normal 10-16 hour after the onset of symptoms.

Reinfarction reported

Non-fatal strokeAmputation of lower extremityCardial revascularizationPeripheral revascularization
Natarajan 2012NDNDNDNDNDNDND
REMBO 2008Stroke, heart failureNDNDNDNDNDND
Service 1983NDNDNDNDNDNDND
Stefanidis 2003fReported as death due to myocardial infarctionNDReinfarction reportedNon-fatal strokeAmputation of lower extremityCardial revascularizationPeripheral revascularization
Steno-2 2008Death from cardiovascular causesDeath from cardiovascular causes, non-fatal myocardial infarction, coronary-artery bypass grafting, percutaneous coronary intervention, non-fatal stroke, amputation as a result of ischaemia, or vascular surgery for peripheral atherosclerotic artery diseaseWHO criteriaWHO criteriaAmputation because of ischaemiaCoronary-artery bypass graftingSurgical interventions for peripheral atherosclerotic artery disease
UGDP 1975Death due to:
Sudden death; defined as a death occurring within three hours of the onset of symptoms in an otherwise clinically stable patient and in a manner consistent with a cardiovascular event.
Myocardial infarction; this diagnosis was made from electrocardiogram changes and changes in serum enzymes observed during the terminal course of illness, or if the events leading to death were clinically compatible with the diagnosis and autopsy findings provided evidence that an myocardial infarction was the principal cause of death.
Other heart disease, included deaths due to congestive heart failure, valvular heart disease, atherosclerotic heart disease, and hypertensive heart disease.
Extracardiac Vascular Disease: cerebral vascular disease, pulmonary embolism, and peripheral vascular
NDPatients hospitalised with a diagnosis of non-fatal myocardial infarction or changes from a less severe finding for Q/QS and T patterns on the baseline electrocardiograph to a more severe finding for these abnormalities on a follow-up electrocardiographNDAmputation of all or part of lower limbNDND
UKPDS 1998gFatal myocardial infarction, fatal stroke, death from peripheral vascular disease, and sudden deathIs not reported separately in trial. Is reported as a part of the aggregate outcome; any diabetes-related endpointWHO clinical criteria with associated electrocardiogram/enzyme changes or new pathological Q wave (ICD 9 Code 410)Major strokes with symptoms that persisted for more than one month (ICD 430 to 434.9 and 436)Major limb complications- requiring amputation of digit or limb for any reason (ICD codes 5.845 to 5.848)NDND
VA CSDM 1995Cardiovascular death is classified as sudden death, coronary heart disease, cerebrovascular attack, or other cardiovascular causes (pulmonary embolism, cardiomyopathy, etc)Myocardial infarction, stroke, congestive heart failure, amputation for gangrene, new angina and/or coronary artery disease, coronary artery by-pass graft, percutaneous transluminal coronary angioplasty, ischaemic ulcer, transient ischaemic attack, new intermittent claudicationMyocardial infarctions are classified by the CER-Lab using the Minnesota code. Patients with suspected acute myocardial infarction, treated with thrombolytic therapy or with acute coronary angioplasty (within 24 hour of the onset of symptoms), who do not meet the electrocardiogram criteria, also are countedNon-fatal strokeLimb ulcers or amputation were computed end points only if diagnosed as ischaemicCoronary revascularizationPeripheral revascularization
VADT 2009hIn appendix listed as death caused by:
myocardial infarction, congestive heart failure, coronary revascularization, stroke, cerebral revascularization, complications of occlusions, peripheral revascularization, sudden death, and pulmonary embolism
Acute myocardial infarction, death from cardiovascular disease, stroke, congestive heart failure, amputation from peripheral vascular disease, surgical intervention for coronary or peripheral vascular disease, and critical limb ischaemiaQ wave in 2 consecutive leads or a new R-wave in V1 of at least 50% accompanied with motion abnormality in MUGA scan or echocardiography;
or ST depression over 1 mm or new T-wave in 2 consecutive leads with injury changes in creatine phosphokinase over 2 times and elevated CK-MB or troponins
Non-haemorrhagic stroke: sudden onset of focused symptoms over 24 hours;
intracranial haemorraghic stroke: with meningeal symptoms in the absence of focal signs, and bloody spinal fluid with increased pressure;
embolic stroke: rapid onset, localized symptoms, presence of embolic condition
Amputation for ischaemic diabetic gangreneCoronary revascularizationPeripheral revascularization
Yang 2007NDNDNDNDNDNDND
Zhang 2011Death due to cardiovascular diseaseMajor macrovascular eventsNDNDNDNDND

Footnotes

a"One suffered a brainstem cerebrovascular accident after 2 weeks, one developed unstable angina. One further patient in IC developed exertional angina during the study but was able to complete the study."
bFor macrovascular complications, definition of previously cardiovascular disease was used.
cAn inclusion criterion was previously myocardial infarction. Therefore we have recorded the number from reinfarction as myocardial infarction.
dMacrovascular disease was not clearly defined in text.
eThe number of myocardial infarction reported in analysis is the number of reinfarction. Reinfarction was not defined in trial.
fThe number of myocardial infarction reported is the number of reinfarction. Reinfarction was not defined in trial.
gDefinition of major cardiovascular events from Turnbull et al. (Turnbull 2009): cardiovascular death or non-fatal stroke or non-fatal myocardial infarction, stroke (fatal or non-fatal), myocardial infarction (fatal or non-fatal) and heart failure resulting in hospitalisation or death.
hMyocardial infarction, stroke, coronary revascularization, and peripheral revascularization are listed in appendix in the same table as death due to cardiovascular disease. Therefore we assume the number of these events is the non-fatal events.

ACCORD: Action to Control Cardiovascular Risk in Diabetes Study; ADDITION: Anglo-Danish-Dutch study of Intensive Treatment In PeOple with screeN detected diabetes in primary care; ADVANCE: Action in Diabetes and Vascular disease – PreterAx and DiamicroN MR Controlled Evaluation; CK-MB: Creatine kinase isozyme component MB; DIGAMI: Diabetes and Insulin-Glucose Infusion in Acute Myocardial Infarction; ICD: International Classification of Diseases; IDA: Insulin Diabetes Angioplasty; MUGA scan: multiple-gated acquisition scan; ND: not defined; REMBO: Rational Effective Multicomponent Therapy in the Struggle Against DiaBetes Mellitus in Patients With COngestve Heart Failure; UGDP: University Group Diabetes Program; UKPDS: United Kingdom Prospective Diabetes Study; VACSDM: Veterans Affairs Cooperative Study in Type 2 Diabetes Mellitus; VADT: Veterans Affairs Diabetes Trial; WHO: World Health Organisation

Appendix 6. Definition of microvascular outcomes in study or as reported

Characteristic

Study ID

Microvascular complications (composite outcome)NephropathyEnd-stage renal diseaseRetinopathyRetinal photocoagulation
ACCORD 2008Fatal or non-fatal renal failure (initiation of dialysis or end-stage renal disease, renal transplantation, or rise of serum creatinine
> 291.7 μmol/L) or retinal photocoagulation or vitrectomy for diabetic retinopathy
Composite nephropathy outcome: Doubling of serum creatinine or a 20 mL/min/1.73m2 or decrease in estimated glomerular filtration rate, development of macroalbuminuria (albumin/creatinine ratio > 300 mg albumin per gram creatinine in random urine sample), development of renal failure (renal transplantation or initiation of dialysis or a rise in serum creatinine > 3.3 mg/dL in the absence of an acute reversible causeDevelopment of renal failure as defined by renal transplantation or initiation of
dialysis or a rise in serum creatinine > 3.3 mg/dL in the absence of an acute
reversible cause. Death due to renal failure
Progression of diabetic retinopathy of at least 3 stages the Early Treatment of Diabetic Retinopathy Study scalePhotocoagulation
ADDITION-Europe 2011NDNDNDNDND
ADVANCE 2008New or worsening nephropathy or retinopathy (development of proliferative retinopathy, macular edema or diabetes-related blindness, or the use of retinal photocoagulation therapy)Development of macroalbuminuria, defined as a urinary albumin:creatinine ratio of more than 300 μg of albumin per milligram of creatinine (33.9 mg per millimoles), or doubling of the serum creatinine level to at least 200 μmol/L, the need for renal-replacement therapy, or death due to renal diseaseRenal-replacement
therapy or death from renal causes
Progression of ≥2 steps in Early Treatment of Diabetic Retinopathy Study classification with laser coagulation therapy during follow-up as the final step in Early Treatment of Diabetic Retinopathy Study classification, including both incidence and progression
of retinopathy
Laser coagulation therapy
Araki 2012Microangiopathy (retinopathy, nephropathy, and neuropathy)

Mean urinary albumin-to creatinine

ratio (mg/mg creatinine) in two or three

successive urinalyses was used to classify diabetic nephropathy

as no nephropathy (albumin to creatinine ratio < 30), microalbuminuria (30≦ albumin-to creatinine ratio <300) or persistent proteinuria (albumin-to creatinine

ratio ≧ 300 or urinary protein ≧ 30 mg/dL)

ND

Retinopathy status was assessed by the Japanese Diabetes Complication Study method and classified into five stages: stage 0: no retinopathy; stage 1: dot

haemorrhages, haemorrhages or hard exudates; stage 2: soft exudates; stage 3: Intraretinal microvascular abnormalities or venous deformities; stage 4: neovascularization, preretinal proliferative tissues, vitreous haemorrhages or retinal detachment

ND
Bagg 2001aNDMacroalbuminuriaNDNDND
Becker 2003NDNDNDNDND
Blonde 2009NDNDNDNDND
Cao 2011NDNDNDNDND
Cooray 2011NDNDNDNDND
DIGAMI 2 2005NDNDNDNDND
Fantin 2011Microvascular complications (composite outcome)NephropathyEnd-stage renal diseaseRetinopathyRetinal photocoagulation
Guo 2008NDNDNDNDND
IDA 2009NDNDNDNDND
Jaber 1996NDNDNDNDND
Kumamoto 2000ND

The patients with nephropathy were divided into three stages depending on their urinary albumin excretion: normoalbuminuria (< 30 mg/24 hour), microalbuminuria (30-300 mg/24 hour), or albuminuria (> 300 mg/24 hour).

Reported for the primary prevention population as participants developing nephropathy. Reported for the secondary intervention population as participants progressing to nephropathy

End-stage renal diseaseThe degree of retinopathy for each patient was determined by the two eye examiners using the modified Early Treatment of Diabetic Retinopathy Study classification with a scale of 19 stages. The development and progression of retinopathy were defined as a change of at least two steps up from stage 1 in the primary prevention population and as a change of two or more steps up from stages 2 to 5 in the secondary intervention populationRetinal photocoagulation
Lu 2010NDWHO 1999 criteriaNDNDND
Melidonis 2000NDNDNDNDND
Natarajan 2012NDNDNDNDND
REMBO 2008NDNDNDNDND
Service 1983NDNDNDNDND
Stefanidis 2003NDNDNDNDND
Steno-2 2008Progression of microvascular complications (incident diabetic
nephropathy or the development or progression of
diabetic retinopathy or neuropathy)
Nephropathy was defined as median albumin excretion rates greater than 300 mg/24 hour in at least one of the two-yearly examinationsEnd-stage renal disease requiring dialysisDiabetic retinopathy was graded according to the six-level grading scale of the European Community - funded Concerted Action Programme into the Epidemiology and Prevention of DiabetesND
UGDP 1975bNDUrine protein ≥ 1 gm/LRenal dialysisMild retinal abnormalities: hard exudates, soft exudates, and/or haemorrhages or microaneurysmsND
UKPDS 1998Retinopathy requiring photocoagulation, vitreous haemorrhage, and or fatal or non-fatal renal failureTwo-fold plasma-creatinine increaseRenal failure dialysis and/or plasma creatinine > 250 ųmol/L not ascribable to any acute intercurrent illness. Death from renal diseaseRetinopathy was defined as one microaneurysm or more in one eye or worse retinopathy, and progression of retinopathy as a two-step change in Early Treatment of Diabetic Retinopathy Study gradeRetinal photocoagulation
VA CSDM 1995NDOvert nephropathy was defined as an albumin:creatinine ratio > 0.30Serum creatinine > 265 ųM (without a reversible cause), and/or need for dialysis or kidney transplantSeven-field fundus photograph and ophthalmological examination. The first two photographic end points is the presence of at least 3 counts of microaneurysms for the two eyes, and the second is the worsening of retinopathy as defined by a progression of two or more levels in the final Early Treatment of Diabetic Retinopathy Study scaleND
VADT 2009Retinopathy, nephropathy, and neuropathySevere nephropathy was defined as a doubling of the serum creatinine level, a creatinine level of more than 3 mg per decilitre (265 μmol/L), or a glomerular filtration rate of less than 15 ml per minuteDeath due to renal failureThe 23-point Early Treatment Diabetic Retinopathy Study grading scale was used to define progression to new proliferative diabetic retinopathy. The progression of retinopathy was defined as a 2-point increase on the scaleND
Yang 2007NDNDNDNDND
Zhang 2011Major microvascular eventsNewly diagnosed with microalbuminuriaNDNDND

Footnotes

aThe urine assessment at the end of the follow-up period was a single albumin creatinine ratio.
bThe authors have not defined nephropathy in the articles, but we have chosen to report urine protein ≥1 gram/L as nephropathy.

ACCORD: Action to Control Cardiovascular Risk in Diabetes Study; ADVANCE: Action in Diabetes and Vascular disease – PreterAx and DiamicroN MR Controlled Evaluation; DIGAMI: Diabetes and Insulin-Glucose Infusion in Acute Myocardial Infarction; IDA: Insulin Diabetes Angioplasty; ND: not defined; REMBO: Rational Effective Multicomponent Therapy in the Struggle Against DiaBetes Mellitus in Patients With COngestve Heart Failure; UGDP: University Group Diabetes Program; UKPDS: United Kingdom Prospective Diabetes Study; VACSDM: Veterans Affairs Cooperative Study in Type 2 Diabetes Mellitus; VADT: Veterans Affairs Diabetes Trial; WHO: World Health Organisation

Appendix 7. Definition of hypoglycaemia in study or as reported

Characteristic

Study ID

Hypoglycaemia (when not further specified)Mild hypoglycaemiaModerate hypoglycaemiaSevere hypoglycaemia
ACCORD 2008Hypoglycaemia specified after severeness in trialMild hypoglycaemia is defined as self-reported transient symptoms, such as light headedness, tremor, shaking, sweating, tingling, blurry vision, trouble concentrating, and so on, that resolve after self-treatment with the ingestion of simple carbohydratesNDSevere hypoglycaemia is defined as hypoglycaemia with documented blood glucose < 2.8 mmol/L (50 mg/dL) or symptoms that promptly resolve with oral carbohydrate, intravenous glucose, or glucagon that require the assistance of medical or paramedical personnel
ADDITION-Europe 2011NDNDNDND
ADVANCE 2008Hypoglycaemia was defined as a blood glucose level of less than 2.8 mmol/L (50 mg/dL) or the presence of typical symptoms and signs of hypoglycaemia without other apparent causeMinor hypoglycaemiaNDPatients with transient dysfunction of the central nervous system who were unable to treat themselves (requiring help from another person) were considered to have severe hypoglycaemia
Araki 2012NDMild hypoglycaemia episodes included the appearance of or recovery from hypoglycaemic symptomsNDSevere hypoglycaemia episodes were defined as coma, convulsion or incapacity of the patient sufficient to require the assistance of another person
Bagg 2001Hypoglycaemic episodes were defined as any capillary glucose record < 4 mmol/L, or symptoms of hypoglycaemia relieved by treatment expected to raise the level of blood glucose in the absence of a capillary glucose testMild hypoglycaemiaNDSevere hypoglycaemia was defined as the presence of impaired consciousness requiring the help of another person, coma or seizure and the presence of low blood glucose
Becker 2003NDNDNDND
Blonde 2009Specified according to severeness in trialSelf measured plasma glucose (SMPG) < 3.1 mmol/L and not requiring third party assistanceNDHypoglycaemia requiring third part assistance
Cao 2011NDNDNDBlood glucose level of 2.2 mmol/L or less.
Cooray 2011Blood glucose levels <3 mmol/LNDNDND
DIGAMI 2 2005Hypoglycaemia was defined as a blood glucose < 3.0 mmol/L and was recorded as with or without symptomsNDNDND
Fantin 2011

Defined as blood glucose lower than 70 mg/dl,

irrespective of the presence or absence of symptoms

Defined as blood glucose lower than 70 mg/dl,

irrespective of the presence or absence of symptoms

Moderate hypoglycaemia

Defined as a blood glucose level lower than 40 mg/dl or less than 70 mg/dl in an unable or unwilling patient, because of

neuro glycopenia, to take carbohydrate orally

Guo 2008NDNDNDND
IDA 2009NDNDNDSevere hypoglycaemic episodes
Jaber 1996NDMild to moderate: classic autonomic symptoms, recognised by the patients and successfully self-treatedNDND
Kumamoto 2000Hypoglycaemia specified after severeness in trialMild hypoglycaemia was defined as an
event with symptoms consistent with hypoglycaemia (sweating, palpitations, hunger, or blurred vision) in which the patient did not require the assistance of another person and which was associated with a blood glucose level < 50 mg/dL by self-monitoring
NDSevere hypoglycaemia was defined as an event with symptoms consistent with hypoglycaemia in which the patient required the assistance of another person and which was associated with a blood glucose level < 50 mg/dL and a prompt recovery after intravenous glucose loading
Lu 2010NDNDNDND
Melidonis 2000NDMild hypoglycaemic episodesModerate hypoglycaemic episodesSevere hypoglycaemic episodes
Natarajan 2012Hypoglycaemic eventsNDNDND
REMBO 2008NDNDNDND
Service 1983NDNDNDND
Stefanidis 2003NDMild hypoglycaemic episodesModerate hypoglycaemic episodesSevere hypoglycaemic episodes
Steno-2 2008NDMinor episode of symptomatic hypoglycaemiaNDMajor hypoglycaemic event that impaired consciousness and required help from another person
UGDP 1975Suspected or observed period of hypoglycaemia (Fasting values below 50 mg/100mL)NDNDND
UKPDS 1998aHypoglycaemia specified after severeness in trialHypoglycaemic episodes were defined as minor if the patient was able to treat the symptoms unaidedNDHypoglycaemia requiring third-party help or
medical intervention
VA CSDM 1995Hypoglycaemia specified after severeness in trialMild hypoglycaemia is defined as serum glucose < 2.8 mmol/L, with or without symptoms, or consistent symptoms (sweating, palpitations, blurred vision etc.) relieved by treatments that raise blood glucoseSymptoms that caused substantial discomfort and interfered with normal activity but that did not meet the criteria for either mild or severe hypoglycaemiaComa, seizure, or impaired consciousness requiring assistance
VADT 2009Hypoglycaemia specified after severeness in trialRelieved by food or sugar intakeNDDefined with as a serious adverse event i.e., life threatening, death, hospitalisation, disability or incapacity, cancer or other important event requiring medical intervention/treatment
Yang 2007NDNDNDND
Zhang 2011NDOnset of hypoglycaemic episodesNDSevere hypoglycaemic episodes

Footnotes

aThe number of hypoglycaemia is on intention-to-treat percentages from UKPDS 33 and UKPDS 34. We assumed the number being reported is the number of patients with at least one episode of hypoglycaemia over 10 years of follow-up.

ACCORD: Action to Control Cardiovascular Risk in Diabetes Study; ADVANCE: Action in Diabetes and Vascular disease – PreterAx and DiamicroN MR Controlled Evaluation; DIGAMI: Diabetes and Insulin-Glucose Infusion in Acute Myocardial Infarction; IDA: Insulin Diabetes Angioplasty; ND: not defined; REMBO: Rational Effective Multicomponent Therapy in the Struggle Against DiaBetes Mellitus in Patients With COngestve Heart Failure; UGDP: University Group Diabetes Program; UKPDS: United Kingdom Prospective Diabetes Study; VACSDM: Veterans Affairs Cooperative Study in Type 2 Diabetes Mellitus; VADT: Veterans Affairs Diabetes Trial

Appendix 8. Adverse events

Characteristic

Study ID

Intervention(s) and
control(s)
Non-serious adverse events
[N (%)]
Serious adverse events
[N (%)]
Hospitalisation
[N (%)]
Left study due to adverse events
[N (%)]
Congestive heart failure
[N (%)]
ACCORD 2008I: targeting intensive glycaemic control-113/5128 (2.2)part of left reported11/5128 (0.2)152/5128 (3.0)
C: targeting conventional glycaemic control-82/5123 (1.6)10/5123 (1.2)124/5123 (2.4)
all:-195/10,251 (1.9)21/10,251 (0.2)276/10,251 (2.7)
ADDITION-Europe 2011I: targeting intensive glycaemic control-----
C: targeting conventional glycaemic control-----
all:-----
ADVANCE 2008I: targeting intensive glycaemic control--2501/5571 (44.9)-220/5571 (3.9)
C: targeting conventional glycaemic control--2381/5569 (42.8)-231/5569 (4.1)
all:--4882/11,140 (43.8)-451/11,140 (4.0)
Araki 2012I: targeting intensive glycaemic control-----
C: targeting conventional glycaemic control-----
all:-----
Bagg 2001I: targeting intensive glycaemic control-4/21 (19.0)4/21 (19.0)4/21 (19.0)0/21 (0.0)
C: targeting conventional glycaemic control-0/22 (0.0)0/22 (0.0)0/22 (0.0)0/22 (0.0)
all:-4/43 (9.3)4/43 (9.3)4/43 (9.3)0/43 (0.0)
Becker 2003I: targeting intensive glycaemic control-----
C: targeting conventional glycaemic control-----
all:-----
Blonde 2009I: targeting intensive glycaemic control51/121 (42.1)5/121 (4.1)-5/121 (4.1)-
C: targeting conventional glycaemic control44/122 (36.1)4/122 (3.3)-3/122 (2.5)-
all:95/243 (39.1)9/143 (6.3)-8/243 (3.3)-
Cao 2011I: targeting intensive glycaemic control--92/92 (100.0)--
 C: targeting conventional glycaemic control--87/87 (100.0)--
 all:--179/179 (100.0)--
Cooray 2011I: targeting intensive glycaemic control-----
C: targeting conventional glycaemic control-----
all:-----
DIGAMI 2 2005I: targeting intensive glycaemic control--474/474 (100.0)-9/474 (1.9)
C: targeting conventional glycaemic control--306/306 (100.0)-9/306 (2.9)
all:--780/780 (100.0)-18/780 (2.3)
Fantin 2011I: targeting intensive glycaemic control-1/35 (2.8)35/35 (100.0)0/35 (0.0)0/35 (0.0)
C: targeting conventional glycaemic control-1/35 (2.8)35/35 (100.0)0/35 (0.0)0/35 (0.0)
all:-2/70 (2.8)70/70 (100.0)0/70 (0.0)0/70 (0.0)
Guo 2008I: targeting intensive glycaemic control-----
C: targeting conventional glycaemic control-----
all:-----
IDA 2009I: targeting intensive glycaemic control--51/51 (100.0)--
C: targeting conventional glycaemic control--51/51 (100.0)--
all:--102/102 (100.0)--
Jaber 1996a I: targeting intensive glycaemic control-2/23 (8.7)1/23 (4.3)1/23 (4.3)-
C: targeting conventional glycaemic control-2/22 (9.1)2/22 (9.0)0/22 (0.0)-
all:-4/45 (8.9)3/45 (6.7)1/43 (2.3)-
Kumamoto 2000I: targeting intensive glycaemic control-----
C: targeting conventional glycaemic control-----
all:-----
Lu 2010I: targeting intensive glycaemic control-----
C: targeting conventional glycaemic control-----
all:-----
Melidonis 2000I: targeting intensive glycaemic control-6/24 (25.0)24/24 (100.0)0/24 (0.0)4/24 (16.7)
C: targeting conventional glycaemic control-4/24 (16.7)24/24 (100.0)0/24 (0.0)5/24 (20.8)
all:-10/48 (20.8)48/48 (100.0)0/48 (0.0)9/48 (18.8)
Natarajan 2012I: targeting intensive glycaemic control--36/36 (100.0)--
C: targeting conventional glycaemic control--42/42 (100.0)--
all:--78/78 (100.0)--
REMBO 2008I: targeting intensive glycaemic control----14/41 (34.1)
C: targeting conventional glycaemic control----19/40 (47.5)
all:----33/81 (40.7)
Service 1983b I: targeting intensive glycaemic control-----
C: targeting conventional glycaemic control-----
all:-----
Stefanidis 2003I: targeting intensive glycaemic control-6/36 (16.7)36/36 (100.0)5/36 (13.9)1/36 (2.8)
C: targeting conventional glycaemic control-6/39 (15.4)39/39 (100.0)4/39 (10.3)2/39 (5.1)
all:-12/75 (16.0)75/75 (100.0)9/75 (12.0)3/75 (4.0)
Steno-2 2008c I: targeting intensive glycaemic control7/80 (8.8)1/80 (1.3)1/80 (1.3)0/80 (0.0)-
C: targeting conventional glycaemic control5/80 (6.3)0/80 (0.0)0/80 (0.0)0/80 (0.0)-
all:12/160 (7.5)1/160 (0.6)1/160 (0.6) (same patient as left)0/160 (0.0)-
UGDP 1975d I: targeting intensive glycaemic control--14/204 (6.9)--
C: targeting conventional glycaemic control--13/210 (6.2)--
all:--27/414 (6.5)--
UKPDS 1998e I: targeting intensive glycaemic control----91/3071 (3.0)
C: targeting conventional glycaemic control----36/1138 (3.2)
all:----127/4209 (3.0)
VA CSDM 1995f I: targeting intensive glycaemic control---2/75 (2.7)1/75 (1.3)
C: targeting conventional glycaemic control---0/78 (0.0)4/78 (5.1)
all:---2/153 (1.3)5/153 (3.3)
VADT 2009g I: targeting intensive glycaemic control-139/892 (15.6)-7/892 (0.8)76/892 (8.5)
C: targeting conventional glycaemic control-130/899 (14.5)-3/899 (0.3)82/899 (9.1)
all:-269/1791 (15.0)-10/1791 (0.6)158/1791 (8.8)
Yang 2007I: targeting intensive glycaemic control-----
C: targeting conventional glycaemic control-----
all:-----
Zhang 2011I: targeting intensive glycaemic control-----
C: targeting conventional glycaemic control-----
all:-----

Footnotes

"-" denotes not reported

ACCORD: Action to Control Cardiovascular Risk in Diabetes Study; ADDITION: Anglo-Danish-Dutch study of Intensive Treatment In PeOple with screeN
detected diabetes in primary care; ADVANCE: Action in Diabetes and Vascular disease – PreterAx and DiamicroN MR Controlled Evaluation; DIGAMI: Diabetes and Insulin-Glucose Infusion in Acute Myocardial Infarction; ICD: International Classification of Diseases; IDA: Insulin Diabetes Angioplasty; REMBO: Rational Effective Multicomponent Therapy in the Struggle Against DiaBetes Mellitus in Patients With COngestve Heart Failure; UGDP: University Group Diabetes Program; UKPDS: United Kingdom Prospective Diabetes Study; VACSDM: Veterans Affairs Cooperative Study in Type 2 Diabetes Mellitus; VADT: Veterans Affairs Diabetes Trial

Appendix 9. Matrix of trial outcomes (publications)

Characteristic

Study ID

 Endpoint reported in publicationEndpoint not reported in publicationTime of measurement

ACCORD 2008

 

 

Review's primary outcomes
All-cause mortalityX 5 yr
Cardiovasular mortalityX 5 yr
Review's secondary outcomes
Serious adverse eventsX 3.5 yr
Congestive heart failureX 3.5 yr
Cost(s) of treatment XN/A
Health-related quality of lifeX 4 yr
HypoglycaemiaX 3.5 yr
Macrovascular complicationsX 3.5 yr
Microvascular complicationsX 3.5 yr
  Endpoint reported in publicationEndpoint not reported in publicationTime of measurement
ADDITION-Europe 2011 Review's primary outcomes
All-cause mortalityX 5.3 yr
Cardiovasular mortalityX 5.3 yr
Review's secondary outcomes
Serious adverse events XN/A
Congestive heart failure XN/A
Cost(s) of treatment XN/A
Health-related quality of life XN/A
Hypoglycaemia XN/A
Macrovascular complicationsX 5.3 yr
Microvascular complications XN/A
  Endpoint reported in publicationEndpoint not reported in publicationTime of measurement
ADVANCE 2008 Review's primary outcomes
All-cause mortalityX 5 years
Cardiovasular mortalityX 5 years
Review's secondary outcomes
Serious adverse eventsX 5 years
Congestive heart failureX 5 years
Cost(s) of treatment XN/A
Health-related quality of life