Unstructured treatment interruption of antiretroviral therapy in clinical practice: a systematic review

Authors


Corresponding Author Katharina Kranzer, Department of Clinical Research, Faculty of Infectious and Tropical Disease, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, UK. Tel.: +44 20 7636 8636; E-mail: katharina.kranzer@lshtm.ac.uk

Summary

Objective  To characterize the frequency, reasons, risk factors, and consequences of unstructured anti-retroviral treatment interruptions.

Method  Systematic review.

Results  Seventy studies were included. The median proportion of patients interrupting treatment was 23% for a median duration of 150 days. The most frequently reported reasons for interruptions were drug toxicity, adverse events, and side-effects; studies from developing countries additionally cited treatment costs and pharmacy stock-outs as concerns. Younger age and injecting drug use was a frequently reported risk factor. Other risk factors included CD4 count, socioeconomic variables, and pharmacy stock outs. Treatment interruptions increased the risk of death, opportunistic infections, virologic failure, resistance development, and poor immunological recovery. Proposed interventions to minimize interruptions included counseling, mental health services, services for women, men, and ethnic minorities. One intervention study found that the use of short message service reminders decrease the prevalence of treatment interruption from 19% to 10%. Finally, several studies from Africa stressed the importance of reliable and free access to medication.

Conclusion  Treatment interruptions are common and contribute to worsening patient outcomes. HIV/AIDS programmes should consider assessing their causes and frequency as part of routine monitoring. Future research should focus on evaluating interventions to address the most frequently reported reasons for interruptions.

Abstract

Objectif:  Caractériser la fréquence, les raisons, les facteurs de risque et les conséquences des interruptions non structurées du traitement antirétroviral (ART).

Méthode:  Revue systématique.

Résultats:  70 études ont été incluses. La proportion médiane des patients qui ont interrompu le traitement était de 23% pour une durée médiane de 150 jours. Les raisons d’interruption les plus fréquemment rapportées étaient la toxicité des médicaments, les effets indésirables et les effets secondaires. Les études dans les pays en développement ont en outre cité comme inquiétant, les coûts du traitement et les ruptures de stock de pharmacie. L’âge plus jeune et l’utilisation de drogues injectables étaient des facteurs de risque fréquemment rapportés. D’autres facteurs de risque comprennent le taux de CD4, les variables socio-économiques et les ruptures de stock de pharmacie. Les interruptions de traitement augmentaient le risque de décès, les infections opportunistes, l’échec virologique, le développement de la résistance et une pauvre récupération immunologique. Les interventions proposées pour minimiser les interruptions comprennent des conseils, des services de santé mentale, des services pour les femmes, pour les hommes et pour les minorités ethniques. Une étude d’intervention a constaté que l’utilisation d’un service de courts messages de rappel diminuait la prévalence des interruptions du traitement de 19%à 10%. Enfin, plusieurs études en Afrique ont souligné l’importance d’un accès fiable et gratuit aux médicaments.

Conclusion:  Les interruptions de traitement sont courantes et contribuent à l’aggravation des résultats des patients. Les programmes VIH/SIDA devraient envisager d’évaluer les causes et la fréquence dans le cadre de la surveillance de routine. Les recherches futures devraient mettre l’accent sur l’évaluation des interventions pour s’attaquer aux causes les plus fréquemment rapportées pour les interruptions.

Abstract

Objetivo:  Caracterizar la frecuencia, las razones, los factores de riesgo y las consecuencias de las interrupciones no estructuradas del tratamiento antirretroviral (TAR).

Método:  Revisión sistemática.

Resultados:  Se incluyeron 70 estudios. La mediana de la proporción de pacientes que interrumpían el tratamiento era del 23% para una duración media de 150 días. Las razones más frecuentemente reportadas para las interrupciones eran la toxicidad de la droga, los eventos adversos y los efectos colaterales; los estudios en países en vías de desarrollo reportaban además los costes y la falta de existencias en las farmacias. La interrupción del tratamiento aumentaba el riesgo de muerte, las infecciones oportunistas, el fallo virológico, el desarrollo de resistencias y una mala recuperación inmunológica. Las intervenciones propuestas para minimizar las interrupciones incluían el aconsejamiento, servicios de salud mental, servicios para mujeres, hombres y minorías étnicas. Un estudio de intervención encontró que el uso de un servicio de avisos por SMS disminuyó la prevalencia de interrupción del tratamiento del 19% al 10%. Finalmente, varios estudios en África subrayaron la importancia de un acceso fiable y gratuito a la medicación.

Conclusión:  Las interrupciones en el tratamiento son comunes y contribuyen al empeoramiento de los resultados obtenidos por los pacientes. Los programas para el VIH/SIDA deberían considerar evaluar sus causas y frecuencia como parte de la monitorización rutinaria. Las investigaciones futuras deberían focalizarse en la evaluación de intervenciones con el fin de identificar las razones de interrupción del tratamiento más frecuentemente reportadas.

Introduction

Antiretroviral therapy (ART) has dramatically reduced HIV-associated mortality and morbidity in high- and low-income countries (Palella et al. 1998; Egger et al. 2002; Jahn et al. 2008; Floyd et al. 2010; Mahy et al. 2010). Treatment outcomes reported from cohort studies and clinical trials have improved over time as a result of improved drug efficacy, reduced toxicity, and simplified treatment through reduced pill burden and dosing intervals (Boyd 2009). Despite these improvements, consistent adherence and uninterrupted treatment remain major challenges (Lazo et al. 2007; Byakika-Tusiime et al. 2009; Lima et al. 2009; Glass et al. 2010; Bastard et al. 2011).

Ensuring high levels of adherence is desirable for the treatment of any chronic conditions (Jackevicius et al. 2002; Kopjar et al. 2003; Cramer 2004; Osterberg & Blaschke 2005) but is particularly important for treatment of HIV in resource-limited settings, where less robust regimens are used and an extremely high level of adherence (>95%) is required to prevent the development of drug resistance (Bangsberg et al. 2006). There are many challenges to maintaining such high levels of adherence (Mills et al. 2006; Nachega et al. 2010). Among these, treatment interruptions are an inconsistently reported yet common phenomenon in clinical practice, often occurring as a result of treatment fatigue or in an attempt to minimize side-effects. Common toxicities such as lipodystrophy and metabolic side-effects related to prolonged use of ART may improve when treatment is stopped (Tuldra et al. 2001; Mocroft et al. 2005; Mussini et al. 2005; Calmy et al. 2007). However, the majority of individuals who discontinue treatment only do so temporarily, as they experience a rapid decline in CD4 count and increase in viral load following discontinuation of therapy (Poulton et al. 2003; Skiest et al. 2004; El-Sadr et al. 2006; Sungkanuparph et al. 2007).

The potential for provider-directed, structured treatment interruptions as a way to limit antiretroviral exposure (and therefore both toxicities and costs) was abandoned after randomized trials and cohort studies found an increased risk of opportunistic infection and death (El-Sadr et al. 2006; Mugyenyi et al. 2008; Seminari et al. 2008). Nevertheless, patient-initiated unstructured treatment interruptions are a reality of routine clinical care and have been reported in both developed (Holkmann Olsen et al. 2007) and developing country settings (Kranzer et al. 2010).

To better characterize the frequency, reasons, risk factors, and consequences of unstructured treatment interruptions in routine clinical practice, we conducted a systematic review of available studies reporting on unstructured treatment interruptions.

Methods

Criteria for selection of studies

We aimed to identify studies reporting on unstructured ART treatment interruptions in clinical practice. Unstructured treatment interruption was defined as discontinuation of all ART drugs for any period of time, after which treatment was resumed. We considered that any interruption was undesirable, and thus did not limit our search to specific causes or durations. We excluded studies reporting on structured treatment interruptions, defined as physician-initiated, cyclical interruptions guided by CD4 count or viral load. We also excluded studies only reporting on patients experiencing virologic failure. We included both cross-sectional and cohort studies, but excluded editorials, case studies, case reports, and reviews.

Search strategy

We searched three electronic databases for primary studies: Medline, Embase, and Global Health using the compound search strategy summarized in Table S1 and searched the bibliographies of retrieved articles for additional studies. Our search was limited to studies published and conducted from 1996 (the time when highly active ART became available) until the end of the search period (March 2011). We also searched for conference abstracts from all conferences of the International AIDS Society (April, 1985–July, 2010), and all Conference on Retroviruses and Opportunitistic Infections (January, 1997–February, 2010) and the PEPFAR implementers meeting 2007–2009. No language restriction was applied.

Study selection and data extraction

Studies were entered into an electronic database (EndNote X1) to screen potentially eligible studies by title and abstract according to our pre-defined inclusion and exclusion criteria. Full-length articles of all studies considered eligible upon initial screening were obtained and reviewed for eligibility; conference abstracts were screened first by title, then by full abstract. All reviews were carried out independently, in duplicate. After agreeing on eligibility, we abstracted the following information using a standardized extraction form: definitions of treatment interruption, frequency and duration of interruption, reasons, risk factors, consequences of treatment interruption, and proposed interventions. Whenever required, we attempted to contact study authors for clarification by email.

Finally, we assessed full articles for determinants of methodological quality using a pre-defined assessment framework. The following factors were assessed: definition and objectivity of treatment interruption provided, appropriateness of the statistical analysis. Studies investigating consequences or treatment failure (e.g. mortality or viral rebound) were assessed for adjustment for potential confounding and use of objective outcome measures.

Results

Characteristics of included studies

The study selection process is summarized in Figure 1. Our initial search yielded 813 potentially relevant publication and 577 potentially relevant conference abstracts, from which 47 publications and 23 abstracts were considered eligible for inclusion. Three studies considered potentially eligible were excluded because it was unclear whether patients restarted treatment (i.e. interruption) or not (i.e. discontinuation); authors were contacted but did not provide clarification (Berenguer et al. 2004; Braitstein et al. 2007; Ayuo et al. 2008). Sixteen studies were from Africa, 14 from North America, two from Australia, one from South America, two from Eastern Europe, three from Asia, and 32 from Europe. The majority of studies (63) reported results of treatment interruptions in adults from the general population; of the remainder, two studies were among children, one was among adolescents, one was among injecting drug users, one was among men who have sex with men, one was among recurrent prisoners, and one was among women. We judged the methodological quality of studies included as full-length articles to be moderate: a third of studies (15/47) provided a definition and objectivity of treatment interruption; almost all (46/47) used an appropriate statistical analysis approach, and where appropriate the majority (23/25) adjusted for confounders and used an objective outcome measure (29/30).

Figure 1.

 Study selection process.

Definition of treatment interruption and measurement

We found substantial variation and uncertainty in the definition of treatment interruption applied by the individual studies. Twenty-eight did not define the duration of treatment interruption, while of the 42 studies that did specify a definition, duration ranged from 24 h to 1 year (Figure 2). Two cross-sectional studies investigating self-reported treatment interruptions defined interruption as discontinuation of all drugs for more than 24–48 h in the 4 weeks preceding the survey (Glass et al. 2006; Marcellin et al. 2008). Two studies investigating short interruptions defined a maximum duration of treatment discontinuation of 1 month (Oyugi et al. 2007) and 3 months (Taffe et al. 2002).

Figure 2.

 Definition of treatment interruption and their frequencies.

The methods used to determine treatment interruptions varied: self-report (21/70), electronic medication monitoring (4/70) data, prospectively collected by clinicians (7/70), information extracted from clinical records (7/70), pharmacy prescriptions in combination with clinical records (3/70), pharmacy prescriptions only (2/70), combination of data collected by clinicians and/or self-report and/or prescriptions (4/70). Twenty-two studies did not describe the method used to identify treatment interruptions.

Frequency and duration of treatment interruption

Forty-two studies reported frequencies of treatment interruptions, either as proportions (35), rates (1), or proportions and rates (3) of interruption, or as rates or proportions of discontinuation and resumption (3) (Table 1). The proportion of treatment interruptions ranged from 5.8% [adults in Switzerland (Glass et al. 2006)] to 83.1% [recurrent prisoners in the USA (Pai et al. 2009)]; the median proportion of patients interrupting treatment was 23.1% (IQR 15.0–48.0). Rates of treatment interruptions ranged from 2.0 per 100 person-years in the United Kingdom (Bansi et al. 2008), to 6.0 in the Euro-SIDA study (Holkmann Olsen et al. 2007). Eleven studies reported on the mean or median duration of treatment interruptions, with durations ranging from 11.5 days (Oyugi et al. 2007) to 18 months (Holkmann Olsen et al. 2007) (median 150 days). Treatment interruptions were frequently reported as recurrent events, with up to three interruptions per person reported in South Africa (Kranzer et al. 2010) and Senegal (Uhagaze et al. 2006), five in Switzerland (Taffe et al. 2002), six in the EuroSIDA study (Holkmann Olsen et al. 2007), and an average of two in Uganda (Oyugi et al. 2007).

Table 1.   Frequency of treatment interruptions
AuthorStudy populationCountryTime periodStudy descriptionMeasure of TIDefinition of TINProportion of TI (%)Length of TI (median, mean)TI rate per 100 PYRate or proportion of stopping treatmentRate or proportion of treatment resumption
Adeyemi and Olaogun (2006)AdultsNigeria2005Cross-sectional studySelf-report56022.0
Ahonkhai et al. (2011)AdultsSouth Africa2004–2008Prospective cohort studyUnknown11 39711.0
Ammassari et al. (2004)AdultsItalyCross-sectional studySelf-report11615.0
Bansi et al. (2008)AdultsUK1996–2005Prospective cohort studyUnknown>2 weeks12 97721.74.4 months2.0
Boileau et al. (2008)AdultsBurkina Faso, Mali2005Cross-sectional studySelf-report60622.3
Compostella et al. (2005)AdultsItalyCross-sectional studySelf-report11956.3
Das-Douglas et al. (2009)Homeless and marginally housedUSA2006Cross-sectional studySelf-report>48 h12511.2
Ekstrand et al. (2010)AdultsIndiaProspective cohort studySelf-report>48 h55220.0
Ekstrand et al. (2008)AdultsIndiaProspective cohortSelf-report>48 h22948.0
Ekstrand et al. (2008)AdultsIndiaCross-sectional studySelf-report>48 h9331.0
Gandhi et al. (2004)WomenUSAProspective cohortSelf-report>48 h12027.2
Glass et al. (2006)AdultsSwitzerland2003Cross-sectional studySelf-report>24 h in the 4 weeks pre-survey36075.8
Grierson et al. (2005)AdultsAustralia2003Cross-sectional national surveySelf-report105947.087 days
Grierson et al. (2004)AdultsAustralia2001/2002Cross-sectional national surveySelf-report64071.7
Holkmann Olsen et al. (2007)AdultsEurope (EuroSIDA)Until September 2005Prospective cohort studyStart and Stop date of each ARV recorded by clinician>3 months381123.118 months6.0
Kaptue et al. (2002)AdultsCameroonProspective cohort studyUnknown5030.0
Kaufmann et al. (2011)AdultsSwitzerland1996–2008Prospective cohort studyRecorded by clinician>1 month249151.09 months
Kavasery et al. (2009)Injecting drug usersUSAUntil July 2005Prospective cohort studySelf-report>6 months33577.612 months
Knobel et al. (2009)AdultsSpainUntil July 2007Prospective cohort studyComputer assisted pharmacy dispensing system and self-report>3 days54042.8
Knobel et al. (2002)AdultsSpain1998/1999Cross-sectional survey with self-reported TISelf-report with validation of a subset>2 days300415.0
Kouanfack et al. (2008)AdultsCameroon2006/2007Cross-sectional surveyUnknown4279.6
Kranzer et al. (2010)AdultsSouth Africa2004–2009Prospective cohort studyPharmacy record and clinical records>30 days1154228 days12.8/100 PY21.4/100 PY
Lazar et al. (2010)AdolescentsRomaniaCross-sectional surveySelf-report9651.60
Li et al. (2005)Homosexual menUSAUntil March 2002Prospective cohort studySelf-report68710.5 – 1997
5.2 – 1999
7.7 – 2001
61 days
Marcellin et al. (2008)AdultsCameroon2006/2007Cross-sectional national surveySelf-report>2 days in the 4 weeks pre-study53312.8
Martsinovskaya et al. (2010)AdultsUkraiine2008Cross-sectional surveyUnknown313322.0
Mbanya (2003)Adults, self-payingCameroonProspective cohort studyUnknown508.0
Mocroft et al. (2001)AdultsUKUntil December 1998Prospective cohort studyStart and Stop date of each ARV recorded by clinician5567 months26.0%56.1%
Moore et al. (2009)Adults, outpatientsBritish Columbia2000–2006Prospective cohort studyRecorded by clinician>3 months170737.7
Murri et al. (2002)AdultsItaly2001Cross-sectional surveySelf-report8026.0
Murri et al. (2009)AdultsItaly2006Cross-sectional surveySelf-report35924.7
Nacher et al. (2006)Adults, hospital based cohortFrench Guiana1992–2003Prospective cohort studyUnknown>1 year12134.3
Oyugi et al. (2007)Adults, self-payingUganda2002–2004Prospective cohort studyElectronic medication monitor, self-report, pill count>48 h ≤30 days9765.011.5 days
Pasquet et al. (2010)AdultsIvory Coast2006–2008Prospective cohort studyClinical records>1 month155453.4
Pai et al. (2009)Recurrent prisonersUSA1996–2005Prospective cohort studyDispensing pharmacy and community providerNot taking antiretroviral therapy while outside of jail46783.1
Protopopescu et al. (2010)AdultsFranceProspective cohort studyClinical records>60 days83211.5109 days2.9
Saitoh et al. (2008)ChildrenUSA2000–2004Prospective cohort studyUnknown>3 months40516 months17.8%66.6%
Taffe et al. (2002)AdultsSwitzerlandUntil May 2001Prospective cohort studySelf-report>1 month <3 months472027.5
Touloumi et al. (2006)AdultsEurope, Cascade studyUntil August 2003Prospective cohort studyUnknown>2 weeks155119.3
Uhagaze et al. (2006)AdultsSenegal2004–2005Cross-sectional surveyUnknown6027.0150 days
Wenkel et al. (2006)Adults, user feesNigeriaJune 2005Cross-sectional national survey with self-reported TISelf-report12272.0189 days
Zhang et al. (2010)Adultthe NetherlandsUntil February 2008Prospective cohort studyStart and Stop date of each ARV recorded by clinicianAny duration332115.43.1 months

Reasons for treatment interruption

Twenty-two studies, 18 from developed countries and four from Africa, investigated reasons for treatment interruptions (Table 2). Toxicity, adverse events, and side effects were the most frequently reported reasons, with between 6% (Saitoh et al. 2008) and 80% of patients reporting these reasons (Chen et al. 2002). Other reasons included pill burden (Moore et al. 2009), intercurrent illness (Wolf et al. 2005), patient’s decision (Krentz et al. 2003; Sommet et al. 2003; Gibb et al. 2004; Pavie et al. 2005; Saitoh et al. 2008; Moore et al. 2009), treatment fatigue (Saitoh et al. 2008), social and psychiatric issues (Uhagaze et al. 2006; Saitoh et al. 2008), perceived lack of benefits (Tarwater et al. 2003; Gibb et al. 2004) and physician’s decision (Wolf et al. 2005) because of drug interactions, surgery, or other reasons. A study from Australia found that 38% of patients interrupted treatment for solely clinical reasons and 29% for solely lifestyle reasons (Grierson et al. 2004). Costs were the main reason for treatment interruptions (>60%) in two studies from Nigeria (Adeyemi & Olaogun 2006; Wenkel et al. 2006). Pharmacy stock outs and poor access to drugs were reported in three of the four studies from developing countries (Adeyemi & Olaogun 2006; Wenkel et al. 2006; Pasquet et al. 2010).

Table 2.   Reasons for treatment interruption
AuthorCountryTime periodStudy descriptionMeasure of TIDefinition of TINReasons
  1. *Children.

Adeyemi and Olaogun (2006)Nigeria2005Cross-sectional studySelf-report123Cost (69%), side effects (22%), missing of clinic days (12%), poor access to drug (urban 52%, rural 87%)
Bedimo et al. (2006)USA1996–2001Prospective cohort studyUnknown>180 days71Complete viral suppression (1%), treatment failure (4%), non-adherence and adverse events (94%)
Chen et al. (2002)USAProspective cohort studyClinical records>30 days75Side effects (80%), new opportunistic infection (1%), virologic failure (12%), non-adherence (7%), financial (15%)
Gibb et al. (2004)*UK, Ireland1999–2002Prospective cohort study Clinical records>4 weeks71Poor adherence (23%), parent or child request (24%), adverse drug reactions (9%), perceived lack of virologic and immunologic benefits (21%)
Gonzalez et al. (2003)SpainProspective cohort studyUnknown64Drug-related adverse events (55%), patient or physician decision (45%),
Grierson et al. (2004)Australia2001/2002Cross-sectional national surveySelf-report263Solely clinical reasons (38%), both/neither lifestyle and clinical reasons (33%), solely lifestyle reason (29%)
Krentz et al. (2003)Canada1999–2002Prospective cohort studyUnknown>2 months50Virologic failure and a drug resistance (41%), adverse effects or toxicity (36%), patient decision (14%)
Landman et al. (2003)France1998–2002Retrospective cohort studyUnknown>2 months80Patient’s request (19%), lipodystrophy (21%), other drug toxicity (23%), pregnancy or post-partum (11%), high CD4 count (20%), early therapy (6%)
Lazar et al. (2010)RomaniaCross-sectional survey with self-reported TISelf-report50Neglect (59%), boredom (14%), the wish that other do not know that one is ill (10%), lack of medication (10%)
Moore et al. (2009)Canada2000–2006Prospective cohort studyRecorded by clinician>3 months74Medication associated adverse event (7%), pill burden (2%), interaction with methadone (0.3%), pregnancy (0.2%), patient-initiated (2%), treatment failure (0.3%), unknown (88%)
Munoz-Moreno et al. (2010)Spain2006–2008Cross-sectional studyHIV database, clinical records>15 days27Structured TI (42%), toxicity (22%), individual decision (36.%)
Murri et al. (2002)Italy2001Cross-sectional surveySelf-report23Side effects (43%) – particularly vomiting and gastrointestinal symptoms, other reasons included being bored of therapy and being in holiday
Pasquet et al. (2010)Ivory Coast2006–2008Prospective cohort studyClinical records830Drug stock outs (9%), travel/funeral/adverse events/traditional medicine/inability to pay (12%), not recorded (79%)
Pavie et al. (2005)France1999–2003Retrospective chart reviewUnknown30Patient initiated (50%), side effects (50%)
Saitoh et al. (2008)USA2000–2004Prospective cohort studyUnknown>3 months72Medical fatigue (69%), toxicity (14%), adverse events (6%), social and behavior issues (6%), social issues (11%), behavior issues (7%), psychiatric disease (3%)
Sanchez et al. (2007)SpainProspective cohort studyPharmacy prescriptions>4 weeks20Toxicity (65%)
Sommet et al. (2003)France1998–2001Prospective cohort studyUnknown>30 days163Virologic failure (43%), side effects (33%), patient initiated (24%)
Uhagaze et al. (2006)Senegal2004–2005Cross-sectional surveyUnknown42Fear of side effects (72%), having forgotten to take the drugs (26%), the illness (33%), falling asleep (15%), depression (17%)
Tarwater et al. (2003)USAProspective cohort studyClinical records, clinician, self-report105Perceived lack of an indication for therapy on the part of the clinician (44%), drug toxicity (15%), non-adherence (14%), performance of resistance testing (15%), failure (8%)
Van Valkengoed et al. (2003)EuropeProspective cohort studyUnknown>7 days201Toxicity (43%), patient’s decision (29%)
Wolf et al. (2005)Germany1999–2002Prospective frequency matched cohort studyRecorded by clinician>2 weeks133Toxicity and/or side effects (39%), physician’s decision or recommendation (20%), intercurrent illnesses (5%), other reasons (3%)
Wenkel et al. (2006)Nigeria2006Cross-sectional surveySelf-report88Financial constraints (61%), ARVs out of stock (14%), side effects (6%), others (19%)

Risk factors for treatment interruption

Sixteen studies (12 from developed countries) reported on risk factors for treatment interruption. The most commonly reported risk factors were younger age (Mocroft et al. 2001; Gandhi et al. 2004; Li et al. 2005; Nacher et al. 2006; Holkmann Olsen et al. 2007; Moore et al. 2009; Kranzer et al.2010) and injecting drug use (Taffe et al. 2002; Compostella et al. 2005; Touloumi et al. 2006; Kavasery et al. 2009; Moore et al. 2009) (Table 3). The effect of gender and CD4 count on treatment interruption was inconsistent across studies: a high CD4 count (baseline or current) was associated with interruptions in some studies (Taffe et al. 2002; Touloumi et al. 2006; Holkmann Olsen et al. 2007; Moore et al. 2009; Kranzer et al. 2010) while others reported an association between low CD4 count and treatment interruptions (Li et al. 2005; Touloumi et al. 2006; Kavasery et al. 2009). Socioeconomic variables such as employment, income, education, and being homeless were also identified as risk factors for interruption in some studies (Taffe et al. 2002; Oyugi et al. 2007; Marcellin et al. 2008; Das-Douglas et al. 2009; Kavasery et al. 2009). One study reported that the odds of treatment interruption among homeless and marginally housed patients was six times higher if their health care plan included consumer cost-sharing (Das-Douglas et al. 2009). Finally, a study from Cameroon reported that pharmacy stock shortages were identified as a major risk factor for treatment interruption (Marcellin et al. 2008).

Table 3.   Risk factors for treatment interruptions
AuthorStudy populationCountryTime periodStudy descriptionMeasure of TIDefinition of TINRisk factors for TI
  1. *Associated with discontinuation (not TI).

Compostella et al. (2005)AdultsItalyCross-sectional studySelf-report119Older age
Injecting drug use
Time lag between HIV diagnosis and treatment initiation
Anxiety related to therapy
Subjective antiretroviral therapy (ART) intolerance
Experience of more than four regimens
Das-Douglas et al. (2009)Homeless and marginally housedUSA2006Cross-sectional studySelf-report>48 h125Consumer cost-sharing
Emergency department visits in the past year
Being homeless
Depression
Gandhi et al. (2004)WomenUSAProspective cohort studySelf-report>48 h120Younger age
Reduced adherence
Alcohol use
Higher viral load
Holkmann Olsen et al. (2007)AdultsEurope (EuroSIDA)Until September 2005Prospective cohort studyStart and Stop date of each ARV recorded by clinician>3 months3811Higher current log viral load
Higher current CD4 count
Women
Younger age
Kavasery et al. (2009)Injecting drug usersUSAUntil July 2005Prospective cohort studySelf-report>6 months335Younger age
Lower CD4 count
Higher HIV RNA level
Daily injecting drug use
Unemployment
ART initiation in later calendar years
Using crack and alcohol
Kranzer et al. (2010)AdultsSouth Africa2004–2009Prospective cohort studyPharmacy record and clinical records 0>30 days1154Men*
Higher baseline CD4 count*
Shorter time on ART*
ART initiation in later calendar years*
Li et al. (2005)Homosexual menUSAUntil March 2002Prospective cohort study,Self-report687Younger age
Black race
Lower CD4 count
Higher HIV RNA level
Shorter time on ART
Not taking 3TC
Marcellin et al. (2008)AdultsCameroon2006/2007Cross-sectional national surveySelf-report>2 days in the 4 weeks preceding the study533Men
Low educational level
Low monthly household income
Treatment with 3TC
Binge drinking
Number of symptoms
Pharmacy stock shortages
Mocroft et al. (2001)AdultsUKUntil end of 1998Prospective cohort studyStart and Stop date of each ARV recorded by clinician556Younger age*
Men*
Higher viral load*
Moore et al. (2009)Adults, outpatientsBritish Columbia2000–2006Prospective cohort studyRecorded by clinician>3 months1707History of IDU
Higher baseline CD4 count
Hepatitis C pos
Women
Younger age
No AIDS diagnosis at baseline
Less experienced physician
Murri et al. (2009)AdultsItaly2006Cross-sectional surveySelf-report359Suboptimal adherence
Higher viral load
Smokers
NNRTIs
Nacher et al. (2006)Adults, hospital based cohortFrench Guiana1992–2003Prospective cohort studyUnknown>1 year1213Younger age
Initial CD4 count >500 cells/μl
Oyugi et al. (2007)Adults, self-payingUganda2002–2004Prospective cohort studyElectronic medication monitor, self-report, pill count>48 h
≤30 days
97Financial difficulties
Protopopescu et al. (2010)AdultsFranceProspective cohort studyClinical records>60 days832Good patient-provide relationship
No social support from their main partner
No prior history of viral rebound
Fewer HIV-related clinical events
Taffe et al. (2002)AdultsSwitzerlandUntil May 2001Prospective cohort studySelf-report>1 month
<3 months
4720High baseline viral load
High baseline CD4 count
Injecting drug use
Low education
Touloumi et al. (2006)AdultsEurope, Cascade studyUntil August 2003Prospective cohort studyUnknown>2 weeks1551Women
Injecting drug use
High baseline viral load
High baseline CD4 count
Low current CD4 count

Consequences of treatment interruption

Thirty-eight studies reported on various consequences of treatment interruption, comprising mortality, opportunistic infections, immunological and virologic changes, the development of resistance mutations, neurocognitive impairment, and decreased health-related quality of life.

Consistent with the findings of structured interruption studies, unstructured treatment interruptions were commonly associated with a higher risk of death and opportunistic infection and a lower probability of increased CD4 cell counts (Hogg et al. 2002; Taffe et al. 2002; Schrooten et al. 2004; Holkmann Olsen et al. 2007; Pai et al. 2009; Zhang et al. 2010; Kaufmann et al. 2011). Furthermore, a high prevalence of neurocognitive impairment (Munoz-Moreno et al.2010) and lower health-related quality of life (Krentz et al. 2003) were reported in individuals interrupting therapy.

All studies investigating CD4 and viral load response during treatment interruption reported a substantial drop of CD4 count and increase in viral load compared with pre-interruption levels (Gonzalez et al. 2003; Sommet et al. 2003; Tarwater et al. 2003; Gibb et al. 2004; Achenbach et al. 2005; Burton et al. 2005; Giard et al. 2005; Pavie et al. 2005; Wolf et al. 2005; Bedimo et al. 2006; Hull et al. 2006; Sanchez et al. 2007; Saitoh et al. 2008; Mussini et al. 2009; Sarmati et al. 2010). The influence of nadir CD4 counts, CD4 counts, and viral load levels prior to treatment interruption on CD4 decay was inconsistent, with some studies reporting an effect (Gonzalez et al. 2003; Wolf et al. 2005; Hull et al. 2006; Mussini et al. 2009) while others reported no effect (Saitoh et al. 2008).

CD4 counts rose after resumption of therapy (Chen et al. 2002; Sommet et al. 2003; Gibb et al. 2004; Giard et al. 2005; Wolf et al. 2005; Sanchez et al. 2007; Touloumi et al. 2008; Mussini et al. 2009). The increase was biphasic with a steeper slope in the first months after re-initiation of therapy (Touloumi et al. 2008; Mussini et al. 2009). However, CD4 recovery was incomplete: in studies reporting CD4 recovery, the proportion of patients experiencing an increase in CD4 counts to levels before treatment interruption at 24 months ranged from 28% to 69% (Chen et al. 2002; Giard et al. 2005). One study that investigated the effect of treatment interruption in a prison setting found that patients with continuous ART treatment gained on average 0.67 CD4 cells per months compared with intermittently treated patients who lost cells at an average of 0.93 CD4 cells per month (Pai et al. 2009).

The majority of studies reported that patients experienced virologic suppression once treatment was restarted (Chen et al. 2002; Yozviak et al. 2002; Gibb et al. 2004; Wolf et al. 2005; Touloumi et al. 2008; Mussini et al. 2009). However, treatment interruptions were associated with an increased risk of rebound and virologic failure in developed and developing countries (Murri et al. 2002; Parienti et al. 2004, 2008; Spacek et al. 2006; Laher et al. 2007; Oyugi et al. 2007; Bansi et al. 2008; Boileau et al. 2008; Kouanfack et al. 2008; Knobel et al. 2009; Datay et al. 2010; Ekstrand et al. 2010). A study from Spain differentiated treatment interruptions because of patients’ choice and adherence difficulties or physician’s advice for toxicity, severe side effects, or intercurrent illness. After adjusting for drug regimen and adherence level, the risk of a detectable viral load (>500 copied/ml) or death was 3.62 for the former and 1.36 for the latter, compared with continuous treatment (Knobel et al. 2009). A study among adults receiving boosted protease inhibitors (PI) reported that average adherence predicted viral suppression, whereas treatment interruption did not in multivariate analysis (Parienti et al. 2010).

Four studies investigated the development of resistance mutations (Parienti et al. 2004; Spacek et al. 2006; Oyugi et al. 2007; Sanchez et al. 2007). In a study from France, interrupting treatment more than once was significantly associated with the development of resistance to the non-nucleoside-reverse-transcriptase inhibitors (NNRTI) class (hazard ratio 22.5, 95% CI 2.8–180.3) (Parienti et al. 2004). Among 19 treatment interrupters in Spain, nine had mutations in the reverse transcriptase gene and 17 had polymorphism in the protease gene, with L63P being the most commonly found (Sanchez et al. 2007). In Uganda, none of the patients with continuous treatment had evidence of resistance mutations, but 13% of patients with a history of treatment interruption had resistance mutations: all of them had mutations conferring nevirapine resistance, five had mutations conferring lamivudine resistance, and three had mutations conferring stavudine resistance (Oyugi et al. 2007). Another study from Uganda showed resistance to NNRTI class in 26 of 36 patients with detectable viral load with the most common mutation being K103N. Twenty-three of the 36 patients had the M184V/I mutation and three had genotypic resistance to PIs (Spacek et al. 2006).

Interventions

We only identified one intervention study. This randomized controlled trial from Kenya showed that short message service reminders either daily or weekly reduced the prevalence of treatment interruptions exceeding 48 h from 19% to 10% (P = 0.03) (Pop-Eleches et al. 2011).

Six studies investigating risk factors associated with treatment interruptions discussed possible interventions. Studies from developed countries suggested appropriate counseling on the consequences of drug discontinuation, encouragement of optimal adherence, offering of mental health services, addressing addictions, and providing services specifically engaging women and ethnic minorities (Li et al. 2005; Moore et al. 2009; Murri et al. 2009). Studies from Uganda and Cameroon emphasized the importance of steady and reliable access to medication, as well as free access to ART and possibly food supply programs (Oyugi et al. 2007; Marcellin et al. 2008). A study from South Africa concluded that interventions should be targeted at men and during the first 6 months on ART (Kranzer et al. 2010).

When patients were asked to give at least one suggestion how to improve adherence and reduce treatment interruptions: 46% suggested reduction in daily doses, 28% more detailed information about therapy, 27% more attention to side effects, 20% more time dedicated to adherence-related issues, 19% supervised treatment interruptions, and 16% psychological help (Ammassari et al. 2004).

Conclusions

Recent research has highlighted the importance of non-adherence to and defaulting from antiretroviral care in contributing to poor program outcomes (Garcia De Olalla et al. 2002; Nieuwkerk & Oort 2005; Mills et al. 2006; Maggiolo et al. 2007; Rosen et al. 2007; Brinkhof et al. 2009). Our review highlights that unstructured treatment interruptions, while far less frequently reported, are an important phenomenon both in developed and in developing countries and may result in excess mortality and opportunistic infections, increased risk of virologic failure, and poor immunological recovery.

Medication-taking behavior is characterized by adherence which is defined as ‘extent to which a patient acts in accordance with the prescribed interval, and dose of a dosing regimen’ and persistence defined as ‘the duration of time from initiation to discontinuation of treatment’ (Cramer et al. 2008). Persistence emphasizes the concept of continuous therapy and is influenced by both defaulting from antiretroviral care and treatment interruption’ (Bae et al. 2011). Adherence and persistence are both important for optimal treatment outcomes, but their impact may vary dependent on the type of regimen prescribed and the duration and frequency of treatment interruptions.

We found that the characterization of treatment interruption in the literature to date is confused by heterogeneous definitions. A quarter of studies provided no definition, while for those that did definitions varied from more than 24 h to more than 1 year of discontinuation of treatment. Only half of studies reported on median duration of interruption. Similar problems with regard to uniformity of definitions have been encountered in studies investigating loss to follow-up where definitions ranged from 1 to 6 months late for a scheduled consultation or medication pick-up (Rosen et al. 2007). In addition, the method of determination of treatment interruption varied considerably: over a quarter of studies using self-report, while a similar number did not specify the method used to identify treatment interruptions.

The reported causes of treatment interruption are multi-dimensional and context-specific. However, research to date has largely assessed risk factors and reasons for treatment interruption, few in developing country settings. Studies from developing countries highlighted pharmacy stock outs and costs as important factors for treatment interruptions. While several interventions have been proposed, only one has been formally assessed.

Data synthesis is a desirable goal for systematic reviews. However, in view of the substantial degree of heterogeneity between studies with regard to definitions of treatment interruption and methods used to identify treatment interruptions, we decided against providing a data synthesis. In addition, because treatment interruptions depend on duration of ART, incidence would be a more informative measure, but few studies provided incidence estimates. Another limitation of our review, reflecting a limitation of the published evidence, is that only four studies investigated the association between treatment interruption and genotypic resistance. The sample size of these studies was small. One of these studies relied on self-report to identify treatment interruptions. Larger studies using objective measures of treatment interruptions are needed to confirm the association between treatment interruption and genotypic resistance. Finally, although our search strategy was extensive, yielding a high number of studies, we cannot exclude the possibility that our search strategy may not have captured all reports of treatment interruption.

Our study highlights several directions for future research and practice. First, reporting on treatment interruptions should be encouraged, both to improve the quality of program outcome reports, and support better characterization and quantification of the problem. Second, more uniform reporting of treatment interruption should be encouraged to support comparability across studies, as has been proposed for treatment defaulting. The range of proposed interventions in the literature does not reflect the range of causes reported, with a notable absence of attention on some of the most frequently reported drivers of treatment interruption, including drug toxicity, adverse events, and side effects. This suggests that a first step to minimizing treatment interruptions in many settings is simply to provide better care to patients. Finally, intervention studies should be planned to determine the effectiveness of approaches to minimize treatment interruption and encourage treatment resumption.

In conclusion, treatment interruptions are common both in developed and in developing countries and are associated with increased morbidity, mortality, and possibly genotypic resistance. Future research should focus on evaluating interventions to address the most frequently reported reasons for interruptions to support patients in a way that maximizes the chances of continuous and effective treatment.

Acknowledgement

KK is funded by the Welcome Trust, London, United Kingdom.

Ancillary