The diagnosis of gastro-oesophageal reflux disease (GERD) remains challenging. An algorithm, facilitated by a questionnaire, may provide a more structured and cost-effective care of patients.
The diagnosis of gastro-oesophageal reflux disease (GERD) remains challenging. An algorithm, facilitated by a questionnaire, may provide a more structured and cost-effective care of patients.
To compare symptom control achieved with empirical therapy for GERD, in an algorithm based on the GerdQ (new structured pathway, NSP), with that of current care after endoscopy (ordinary clinical pathway, OCP).
Patients with symptoms of GERD, but without alarm features, were randomised in an open, parallel-group study and followed for 4–8 weeks. In the NSP, GerdQ score was used as a basis for both diagnosis and a treatment algorithm. Patients with high likelihood of GERD were treated empirically with a PPI whereas patients with low likelihood of GERD received therapy chosen by the clinician. In the OCP, diagnosis and treatment were based on endoscopy or pH-metry findings. The statistical hypothesis was non-inferiority of NSP to OCP.
A total of 147 patients (86.5%) in the NSP and 133 patients (80.1%) in the OCP arm were responders. Overall, NSP was non-inferior to OCP, but not superior (P = 0.14). Patients with high likelihood of GERD had significantly better symptom relief in the NSP (P = 0.03), whereas those with low likelihood of GERD showed a numerical difference in favour of an endoscopy-based approach (OCP). NSP saved 146 € per patient.
A symptom-based approach using GerdQ reduced health care costs without loss in efficacy. Patients with high likelihood GERD benefited from empirical treatment. An algorithm based on GerdQ may provide physicians with a tool for a more structured care of patients (ClinicalTrials.gov NCT00842387).
Gastro-oesophageal reflux disease (GERD) is a widespread disorder with a prevalence of 10–20% among adults in Western countries, when defined as at least weekly symptoms of heartburn and/or regurgitation. As a consequence of the high prevalence of GERD in the general population health care resource utilisation is substantial.[2, 3] Moreover, since symptoms largely affect an employed population, a considerable productivity loss is associated with GERD.
The diagnosis of GERD is difficult and both symptom evaluation and invasive investigations have some obvious limitations. Symptom evaluation by physician interview can often lead to misinterpretation as both symptom burden and symptom localisation can be difficult to determine. Invasive investigations have traditionally been used to differentiate between erosive and non-erosive GERD and to detect complications, but neither oesophagogastroduodenoscopy (OGD) nor 24-h pH-metry has the adequate sensitivity to be accepted as a diagnostic gold standard. The diagnosis is further hampered by poor correlation between symptom load and endoscopic findings.[7, 8] Even though reflux symptoms are strongly associated with the development of oesophageal adenocarcinoma the absolute risk of developing cancer is very low in the typical GERD patient who has no alarm features. Endoscopically, GERD also seems to show minor changes over time with progression from non-erosive to more severe reflux oesophagitis or Barrett's oesophagus in a minority of patients. Current international guidelines therefore recommend symptom-based diagnosis and therapy unless alarm symptoms such as dysphagia, weight loss or haemorrhage mandate prompt endoscopy.[11-14]
Consequently, there is a need for optimising the management of GERD patients by implementing symptom-based management algorithms, preferably facilitated by a patient-completed questionnaire. Several different questionnaires have been developed to facilitate the diagnosis of GERD, but many of them lack proper validation or lack the simplicity required to be an integrated part of routine care.[15-19] The GerdQ is a self-administered six-item questionnaire that was recently developed as a tool to improve and standardise symptom-based diagnosis and evaluation of treatment response in patients with GERD.
We aimed to evaluate the use of GerdQ in an algorithm to diagnose and select medical therapy, as an alternative to routinely performed OGD before initiating therapy, in patients with symptoms of GERD, but with no alarm features.
This was an open, randomised, parallel-group multi-centre study conducted over 4–8 weeks and performed at 18 gastroenterology outpatient clinics in Norway from March to December 2009 (called Symptom-based vs. Endoscopy-based Approach – SVEA study). Before any study procedures were conducted, the study protocol was approved by the Independent Committee for Research Ethics in Western Norway. This study was a part of a European project evaluating GerdQ as a tool for a more structured management of GERD.
Eligible patients were referred for symptoms suggestive of GERD and randomised in equal numbers to follow either (i) the new structured pathway (NSP) with diagnosis and treatment based on the GerdQ score or (ii) the ordinary clinical pathway (OCP) comprising endoscopy, and if necessary and available, pH-metry.
After randomisation and start of treatment patients were scheduled for a visit after 4 weeks of therapy, including an assessment of response to treatment, again based on the GerdQ score. For patients with inadequate symptom control or if experiencing an adverse drug reaction, treatment could be changed or maintained, and patients continued for an additional 4 weeks in the study. Patients with symptom resolution after 4 weeks terminated the study (Figure 1).
GerdQ (Supplementary Table S1) was developed by combining six questions from three different validated patient reported outcome (PRO) questionnaires (GERD impact scale – GIS, gastrointestinal symptom rating scale – GSRS and reflux disease questionnaire – RDQ) as assessed in the Diamond study. The ROC analysis gave a sensitivity of 65% and a specificity of 71% for the diagnosis of GERD when using the cut-off level of 8 in total GerdQ score.
Eligible patients had symptoms suggestive of GERD (heartburn or regurgitation as predominant symptoms), were aged ≥18 years and provided written informed consent. Patients presenting with alarm symptoms such as unintentional weight loss, severe or progressive dysphagia, or GI bleeding were excluded as well as patients who had undergone endoscopy and/or pH-metry during the last year. Pre-endoscopy use of acid-suppressive medications was restricted in accordance with existing local guidelines, usually forbidding continuous use of both PPI and histamine-2-receptorantagonists (H2RA) 2 weeks before OGD, but allowing limited on-demand use of antacids or H2RA.
|NSP n = 174||OCP n = 170|
|Males, n (%)||88 (50.6)||91 (53.5)|
|Age (years), mean (s.d.)||49.4 (15.3)||46.9 (14.6)|
|Currently employed, n (%)||118 (68.2)||115 (68.5)|
|Body mass index, BMI, mean (s.d.)||27.1 (4.4)||26.8 (4.4)|
|Current nicotine use, n (%)||49 (28.2)||38 (22.4)|
|Current alcohol use, n (%)||122 (70.1)||124 (72.9)|
|Units per weeks, median||3||3|
|Time since GERD symptom onset (years)|
|Mean (s.d.)||8.1 (9.3)||9.0 (9.9)|
|Time since GERD diagnosis (years)|
|Mean (s.d.)||1.9 (3.7)||1.8 (5.5)|
|History of GI disease, n (%)|
|History of reflux oesophagitis||6 (3.4)||8 (4.7)|
|History of gastric ulcer||3 (1.7)||2 (1.2)|
|History of duodenal ulcer||5 (2.9)||4 (2.4)|
|Heartburn last 7 days scored by subject, n (%)|
|Never||7 (4.0)||13 (7.7)|
|1 day||19 (10.9)||19 (11.2)|
|2–3 days||61 (35.1)||52 (30.8)|
|4–7 days||87 (50.0)||85 (50.3)|
|GerdQ Total Score (score 0–18)|
|Mean (s.d.)||10.0 (2.7)||9.6 (3.0)|
|GerdQ Impact Score (score 0–6)|
|Mean (s.d.)||2.4 (1.7)||2.3 (1.8)|
|GerdQ subgroups, n (%)|
|Low likelihood of GERD (total score 0–7)||34 (19.5)||43 (25.4)|
|High likelihood and low impact GERD (total score 8–18/impact score 0–3)||91 (52.3)||77 (45.6)|
|High likelihood and high impact GERD (total score 8–18/impact score 4–6)||49 (28.2)||49 (29.0)|
|Reflux oesophagitis, n (%)|
|Los Angeles grade A||81 (47.9)|
|Los Angeles grade B||42 (24.9)|
|Los Angeles grade C||12 (7.1)|
|Los Angeles grade D||2 (1.2)|
|Barretts oesophagus, n (%)||3 (1.8)|
|Hiatus hernia, n (%)||92 (54.4)|
|Strictures, n (%)||5 (3.0)|
A referral list from primary care physicians containing patients with symptoms suggestive of GERD and requesting further evaluation and OGD examination was used for recruitment. Patients were consecutively allocated a randomisation code by the electronic case report form system. Randomisation was done in blocks of four into the two distinct pathways; NSP and OCP.
For the symptom-based approach a pre defined algorithm was constructed (Figure 2). This algorithm gave us the possibility to tailor treatment alternatives in accordance with symptom burden ranging from no advice on treatment for the individuals with low likelihood of GERD to esomeprazole 40 mg once daily for the patients with high likelihood of GERD and high symptom impact.
Based on the GerdQ score at baseline patients were subdivided into three different groups each with a pre defined algorithm for treatment:
After 4 weeks all patients attended visit 2, and once more completed GerdQ, and for patients with inadequate response to treatment there was a possibility to maintain or change treatment.
Patients randomised to the OCP arm also completed the GerdQ questionnaire at each visit, but the score was kept blinded to the investigator for the duration of the study (Figure 1). The investigator initiated treatment according to ordinary clinical practice and in line with current rules for reimbursement in Norway, which specifies generic PPI (lansoprazole, omeprazole or pantoprazole in standard doses) in most patients and esomeprazole only for severe oesophagitis, complications such as strictures and metaplasia or symptoms refractory to treatment. After the initial period, in case of inadequate response, the investigator was free to alter the treatment by giving a higher dose, changing to a different PPI or changing the dosage regimen. If endoscopy revealed any finding that suggested it would be inappropriate for the patient to continue in the study, the patient was excluded from further participation. Patients with normal findings on OGD were referred for pH-metry, if clinically indicated and available. Those with both normal endoscopy and pH-metry, as well as those in whom pH-metry could not be undertaken within the timeframe of the study, were also followed up and treated according to recommended clinical practice. OGD were performed according to local routines and the Los Angeles (LA) classification was used for grading of reflux oesophagitis. Patients with normal OGD were referred to 24 h or, if available, 48 h wireless pH-metry. Percent time with pH <4 was registered as total, upright and supine. Total number of refluxes and numbers of reflux episodes longer than 5 min were also recorded. An abnormal pH-metry was defined as pH below 4 for >5.5% of time. During the pH-metry patients were instructed to register bothersome GERD symptoms on the receiver, and a cut-off level of ≥95% in Symptom Association Probability (SAP) indicated a positive test for GERD irrespective of acid exposure. All evaluations of OGD and pH-metry were performed by experienced and board certified endoscopists.
The primary outcome variable was the proportion of patients who were treatment responders based on not more than 1 day (score ≤1) during the last week of any of the symptoms heartburn, regurgitation, over-the counter (OTC) use of H2RA/antacids and sleep disturbance, i.e. the positive predictors of GERD. For nonresponders after 4 weeks the symptom score at week 8 was used in the primary analysis.
The main statistical hypothesis was non-inferiority (NI) of NSP to OCP. The clinically acceptable difference was defined based on clinical judgment taking into consideration the symptomatic response rate for PPIs reported in studies on a similar patient population. Consequently, we assumed 85% response rate in both groups and the NI margin was set to 10%. A 95% confidence interval (CI) was calculated for the difference in proportion of treatment responders and NI would be stated if the left end of this CI was larger than -10%. In case of NI of NSP, a pre specified superiority analysis of NSP over OCP was conducted. Superiority would be stated if the left end of the 95% CI would be larger than 0. Chi-squared statistics were used for the analysis on the primary outcome variable and two-sided P-values were calculated using Fischer's exact test. Sub-analyses were done for GerdQ cut-off level ≥8 (high likelihood of GERD = HL) and <8 (low likelihood of GERD = LL) and furthermore for the high likelihood separated by impact of symptoms (High likelihood and low impact GERD = HLLI and High likelihood and high impact GERD = HLHI). The intention-to-treat (ITT) population included all randomised patients except three patients in the OCP arm who had endoscopic findings indicating disease other than GERD, necessitating other therapy. The per protocol (PP) population consisted of a subset of patients who did not violate any major inclusion or exclusion criteria, and completed all the assigned study visits. Patients with incomplete or missing GerdQ at baseline and visit 2 were excluded from ITT analysis on the primary outcome variable. Statistical analyses were performed using the statistical packages SAS version 8.2 and SPSS version 19.
Assuming that symptom control in both groups was 85%, with 10% difference at worst (for NSP relative to OCP) to consider the two pathways to be equally effective (95% CI limit), the number of patients required with a statistical power of 80% would be 158 patients per group. Considering a 10% dropout rate or lost to follow-up during the treatment period, it was estimated that a sample of 348 subjects was needed.
The cost analyses were performed from a societal perspective, i.e. including both direct medical costs, patient out-of-pocket costs and indirect costs resulting from work time lost. Direct medical cost was estimated based on expenditure for diagnostic procedures, medication and unscheduled health care visits (Supplementary Table S2). Data on indirect costs related to absence from work, time required for visits and travelling distance (for both patient and any accompanying person) were collected at visit 1 and used as a standard for all visits. We assumed that all Adverse Drug Reactions resulted in a specialist consultation. The year of the cost analysis was 2010.
If the primary statistical analysis rejected NI, a cost-utility analysis was planned, otherwise a cost-minimisation analysis would be conducted. Cost-minimisation is used when two interventional pathways are of equal efficacy and tolerability, in which case the intervention with the lowest cost is favourable from an economic point of view.
The first analysis performed looked at the costs generated within trial only, however, in clinical practice the duration of therapy is longer. Furthermore, it could be argued that a number of patients in the NSP arm – if not responding to treatment – would later be referred to endoscopy or pH-metry. To investigate the cost implications of these aspects we conducted a sensitivity analysis, in which:
In this study we registered all serious adverse events (SAE) and also any other events believed to be causally related to the study drugs, called the Adverse Drug Reactions (ADR). Moreover, malignancies and organ damage encountered during the study were also recorded.
Of the 347 patients randomised into the study (NSP, n = 174; OCP, n = 173) 83.0% reported heartburn ≥2 days during the week prior to randomisation. Based on the total and impact GerdQ scores, 22% of all patients had low likelihood of GERD, 49% had high likelihood and low impact GERD, and 29% had high likelihood and high impact of GERD. In the OCP arm 81% of the patients had reflux oesophagitis and 9 of 16 patients that underwent pH-metry had an abnormal test. 87% (n = 67) of the high likelihood low impact and 94% (n = 46) of the high likelihood high impact OCP patients had endoscopy or pH-metry confirmed GERD. Three OCP patients (1.8%) had endoscopically suspected oesophageal metaplasia. Demographic data and GERD baseline characteristics show that the groups were well balanced on all variables (Table 1). Patient flow chart and analysis disposition are shown in Figure 3.
Most patients used over-the-counter (OTC) H2RA and antacids (42%) or had no treatment (38%) during the 4 weeks before entering the study, whereas 20% used a PPI. Overall, at the first visit 92% were prescribed a PPI, 4% a H2RA and 4% had no treatment. There were markedly more patients in the NSP arm (28%) compared with OCP arm (10%), who were prescribed esomeprazole as directed by the predefined NSP algorithm (Supplementary Table S3). A total of 65 patients continued the study to visit 3 (week 8) and 51 of 65 patients (78%) made a change in medication, the majority of patients changed from a generic PPI to esomeprazole (supplementary Table S4).
Overall 147 patients (86.5%) in NSP were treatment responders compared to 133 (80.1%) in the OCP arm (Figure 4). The left-sided CI did not overlap 10%. Thus, NSP was non-inferior to OCP, and a superiority analysis of NSP over OCP was done. Superiority was not established (P = 0.143 for ITT analysis and P = 0.068 for PP analysis).
A sub-analysis was conducted dividing the patients into high likelihood (GerdQ total score 8–18) and low likelihood (GerdQ total score 0–7). For the high likelihood patients there were significantly more responders to therapy in the NSP arm compared to the OCP (P = 0.032). For the low likelihood GERD patients there were more responders to therapy in the OCP arm compared to the NSP.
Sub-analysis dividing the high likelihood GERD patients into patients with low impact (impact score 0–3) and high impact (impact score 4–6) of symptoms of GERD showed a larger numerical difference in favour of NSP with increasing impact of GERD (Figure 4).
The public health service cost in the 8 weeks within-trial analysis was 199 € in NSP, and 345 € in OCP resulting in a net cost-saving of 146 € in favour of NSP (Table 2). The public health service costs in the extrapolated 6-month sensitivity analysis was 278 € for NSP and 380 € in OCP resulting in a cost-saving of 102 € in favour of NSP (Table 3). The NSP strategy also reduced patient cost and indirect cost compared to OCP.
|New structured pathway (n = 174)||Ordinary clinical pathway (n = 170)|
|Item||Public costs||Patient costs||Indirect costs||Total costs||Item||Public costs||Patient costs||Indirect costs||Total costs|
|Visit 1 (specialist consultation + GerdQ)||77||73||101||250||Visit 1 (specialist consultation + OGD/pH-metry)||233||67||119||419|
|Visit 2 (specialist consultation + GerdQ)||76||72||100||248||Visit 2 (specialist consultation)||75||71||97||242|
|Visit 3 (specialist consultation + GerdQ)||15||15||20||50||Visit 3 (specialist consultation)||14||13||4||31|
|Medication 0–4 weeks||9||5||14||Medication 0–4 weeks||6||4||10|
|Medication 4–8 weeks||4||3||7||Medication 4–8 weeks||2||1||4|
|Medication total||13||8||21||Medication total||9||5||14|
|Unscheduled visits||3||4||7||13||Unscheduled visits||1||2||3||6|
|Adverse drug reaction||15||14||18||48||Adverse drug reaction||14||46||57||116|
|New structured pathway||Ordinary clinical pathway|
|Public costs (€)||Public costs (€)|
|Baseline from trial first 4 weeks||199||Baseline from trial first 4 weeks||345|
|Endoscopy/pH-metry in nonrespondersa||32||Endoscopy/pH-metry in nonrespondersa||0|
|Additional 5 months PPI treatmentb||47||Additional 5 months PPI treatmenta||34|
|Total costs||278||Total costs||380|
Four SAEs were reported during the study, but none was judged to be causally related to treatment. A total of 14 ADRs were reported, 11 ADRs occurred in the NSP and 3 in the OCP and were dominated by abdominal complaints in line with previously reported data. No malignancies were reported during the course of the study.
Upper GI symptoms is one of the commonest reasons for contacting the healthcare service. In clinical practice an algorithm, based on the GerdQ score, could therefore facilitate the diagnosis and choice of treatment and improve resource utilisation. Based on this argumentation, this study compared, in patients without alarm features, the structured use of GerdQ vs. current clinical practice comprising the gastroenterologist's interview together with OGD and 24-h pH-metry. With this algorithm, we assumed that patients with low likelihood of GERD could benefit from having endoscopy undertaken for correct diagnosis and therapy. For patients with high likelihood of GERD symptom guided therapy with PPI should be the first option. We also hypothesised that patients with high impact GERD could benefit from more potent acid-inhibition from start (esomeprazole 40 mg) providing longer duration of acid-inhibition to overcome reflux episodes related to late evening meal and reflux-related sleep disturbances.
The overall primary efficacy analysis showed a numerical improvement in symptom response rate for NSP (86.5%) over OCP (80.1%), but results were within the NI margin. Analysis on patients with high and low likelihood of GERD, respectively, indicated that patients with low likelihood of GERD (total score 0–7) benefited from the endoscopy approach, whereas patients with high likelihood GERD (total score 8–18) did best with a symptom-based approach, the latter reaching statistical significance when analysed for superiority (P = 0.032). From the clinical viewpoint this contributed to a validation of the algorithm. We also assessed the sub-group of high likelihood GERD with high and low impact symptoms and this analysis showed a numerically larger difference in favour of NSP with increasing impact of symptom burden, although this difference did not reach statistical significance.
Although the two diagnostic pathways overall were equally effective in relieving symptoms the cost analysis showed a 146 € saving in health care related costs per patient in the within-trial analysis (8 weeks) and 102 € saving in cost per patient in the extrapolated 6 months analysis, all in favour of the NSP pathway.
The patients in this study represents a selected population where patients with upper GI symptoms of functional or atypical nature as well as mild reflux symptoms likely to be self-managed, have been excluded by the primary care physician. Patients were relatively young (mean age 48 years), had a relatively long symptom history (mean 8.5 years) and 83% of patients had 2 days or more during the last week experienced heartburn symptoms. The Norwegian reimbursement scheme necessitates objectively verified disease (OGD or pH-metry) for reimbursement of GERD medication. Consequently, OGD units need to be meticulous in their instructions to stop pre-OGD medications. These are all plausible explanations for the high proportion of OCP patients (81%) who had OGD-verified reflux oesophagitis reported in this study. For the generalisability of the findings in this study it must be taken into consideration that a true primary care population seeking health care for upper GI symptoms might results in more patients with functional complaints and a higher proportion of non-erosive reflux disease. It is therefore plausible that a similar trial conducted in a true primary care population would result in lower symptom response rates, but probably so in both treatment groups.
A previous study compared a symptom-based approach, using empirical PPI treatment, with endoscopic based approach, with PPI treatment according to endoscopic findings. An equal response rate between the two pathways was established, but the empirical treatment approach gave a cost-saving in direct medical costs of 38.72 Euros per patient over 24 weeks. This study did not use a questionnaire to facilitate the symptom-based approach and all patients received the same treatment (esomeprazole).
The fear of missing severe underlying disease and malignancies has been an argument for pre-treatment endoscopy. Results from this short-term study showed that only 1.8% (n = 3) of the OCP patients had endoscopically suspected oesophageal metaplasia (ESOM) at baseline and no malignancies were encountered during the course of the study. In a previous Norwegian primary care study (n = 594) over 6-months only 1.2% were diagnosed with ESOM, and again no cancers were encountered. Moreover, for patients with confirmed Barrett's oesophagus, the risk of developing oesophageal adenocarcinoma has been shown to be lower than previously estimated.
We postulate that when facilitated by GerdQ the responsibility for diagnosis and initial therapy of patients with GERD, without indication for OGD, could confidently be transferred to primary care. The GerdQ would not only assist with the diagnosis, but would also provide the primary care physician with a tool for a more structured follow-up of GERD patients, to evaluate treatment response and select patients for referral to secondary care. GerdQ would also secure that the physician does not underestimate symptom burden or overestimate the effect of medical treatment, as has been previously described. The results from this study have shown that GerdQ can serve as a tool to identify patients who have a high likelihood of GERD and are in need of anti-secretory medication. Likewise, for the low likelihood group of patients, GerdQ can identify those in need of OGD such as patients with atypical reflux symptoms, functional dyspepsia and other differential diagnoses. The limitations of this study are its short duration of follow-up and the lack of a post study endoscopy for patients randomised to the symptom-based approach. Selection of patients for open-access endoscopy units also needs to be in mind when interpreting the results.
A high proportion of patients will eventually undergo OGD if reflux symptoms, irrespective of treatment or not, persist over time. Hence, an even longer time perspective will reduce the cost difference between the pathways. Still, a symptom-based approach may lead to fewer upper GI endoscopies in a life-time perspective taking into account the proportion of patients with spontaneous relapse. What is not accounted for in our model is that a requirement for a pre-treatment OGD approach creates long waiting lists for referral and thereby prolongs the time before proper care of patients can be given, leading to sub-optimal treatment, possible deterioration in patient well-being and anxiety and loss of productivity during waiting time.
A validated and patient-centered questionnaire, such as GerdQ, may be a better gatekeeper than endoscopy for directing the appropriate diagnostic work-up and for tailoring of subsequent medical treatment, thereby contributing to better use of healthcare resources.
For GERD patients without alarm symptoms, a symptom-based approach facilitated by a questionnaire (GerdQ) reduces health care costs without any loss in clinical efficacy. An algorithm based on GerdQ supports the fact that the typical GERD patient should be taken care of in the primary care setting. Furthermore, the algorithm may also help primary care physicians to identify patients with low likelihood of GERD who may benefit more from an endoscopy and to provide the typical GERD patient with better treatment and follow-up.
Declaration of personal interests: We thank Lars Langkilde for helping with the health economics analysis and Hans Wedel for providing statistical help. The 18 participating sites in Norway: J. Hatlebakk, Bergen, B. Moum, Oslo, O. Sandstad, Oslo, T. Sandanger, Asker, G. Noraberg, Arendal, J. Matre, Kristiansand, C. Bang, Nesttun, J. Takvam, Tønsberg, R. Breckan, Bodø, V. Høeg, Tynset, J. Langtind, Orkanger, A. Wilskow, Mosjøen, U. Fjøsne, Levanger, O. Lange, Molde, V. Glazkov, Haugesund, E Melsom, Kristiansund, D.A. Hoff, Ålesund, J.A Sparby, Kongsvinger. Christian Jonasson and Knut Robert Andersen are former employees of AstraZeneca. Bjørn Moum or his institution has received grants, speaker fees or consultancy fees from Abbott, Ferring, MSD and Pharmacosmos for activities not related to submitted work. Christen Bang declares no competing interests. Jan Hatlebakk's institution has received a research grant from AstraZeneca. Declaration of funding interests: This study was funded in full by AstraZeneca. The study sponsor was involved in the study concept, design and analysis, but not in manuscript writing or submission.