The effectiveness and cost-effectiveness of minimal access surgery amongst people with gastro-oesophageal reflux disease - a UK collaborative study. The REFLUX Trial

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 97/10/03. The contractual start date was in June 2000. The draft report began editorial review in November 2006 and was accepted for publication in April 2008. As the funder, by devising a commissioning brief, the HTA Programme specified the research question and study design. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the referees for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

None known

Copyright statement

© Queen’s Printer and Controller of HMSO 2008. This monograph may be freely reproduced for the purposes of private research and study and may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NCCHTA, Alpha House, Enterprise Road, Southampton Science Park, Chilworth, Southampton SO16 7NS, UK.

2008 Queen’s Printer and Controller of HMSO

Gastro-oesophageal reflux disease

GORD causes some of the most frequently seen symptoms in both primary and secondary care; between 20% and 30% of a ‘Western’ adult population experience heartburn and/or reflux intermittently. 1–3 There is a clinical spectrum. The majority has only mild symptoms and requires little if any medication. A minority has severe reflux and develops overt complications, despite full medical therapy, and requires surgical intervention. Amongst the remainder, control of symptoms requires regular or continuous medical therapy, and it is from this intermediate group of patients with significant disease that most of the treatment costs for the health service arise.

Treatment of GORD includes a range of options, both medical and surgical. The simplest is self-administered antacids with advice to alter lifestyle factors such as dietary modification, smoking cessation and weight reduction. Many will require acid suppression therapy using either histamine receptor antagonists (H₂RAs) or proton pump inhibitors (PPIs). Initial high-dose therapy may be followed by maintenance treatment using these drugs either intermittently or continuously at reduced doses sufficient to suppress symptoms. The role of surgery has traditionally been confined to the treatment of those with severe symptoms not responding to medication in appropriate dosage and medically fit for surgery. There has, however, been a paradigm shift since the introduction of laparoscopic techniques, with surgery suggested as an alternative treatment to long-term medication. The NHS costs of GORD are considerable. The yearly drug budget for H₂RAs is in excess of £200 M and for PPIs it is £300 M. Of this budget, most of this prescribing occurs within the primary care setting. 4,5 Once started on PPIs, the majority of patients with significant GORD remain on long-term treatment,6 and an estimated 4–5 patients (age 18–60) per 10,000 are taking maintenance PPIs for oesophagitis and reflux.

Although PPIs are increasingly assumed to be safe there is a spectrum of short-term symptoms caused by PPIs7 and there are concerns regarding the impact of long-term use through profound acid suppression. PPIs cause hypergastrinaemia, the long-term significance of which is unknown but potentially important. Conditions associated with chronic hypergastrinaemia and low acid levels have been linked to a long-term increased risk of developing gastric cancer. There is some evidence of the formation of gastric carcinoid tumours in patients taking long-term PPIs8 and also of vitamin B₁₂ deficiency. 9 Adenocarcinoma of the lower oesophagus is a complication of long-term GORD,10–13 and the incidence of this highly malignant disease has trebled in Western communities in the last 25 years. Whilst the overall incidence of gastric cancer is falling, adenocarcinoma of the gastro-oesophageal junction is now a common cause of death, especially in men. The reasons for this change are probably multifactorial, but there is a clear relationship between Helicobacter pylori infection with migration to the gastric fundus and acid suppression, whether naturally occurring or induced by drug therapy. 14,15

Laparoscopic fundoplication

Interest in surgery as an alternative to long-term medical therapy for GORD has been considerable since the introduction of the minimal access approach in the early 1990s. The operative method, whether using an open or a laparoscopic approach, involves performing a fundoplication by wrapping the fundus of the stomach around the lower oesophagus to create a high-pressure zone, thus reducing gastro-oesophageal reflux. The wrap created can be either complete (360°) or partial. Many operative variants have been described. The commonest operation is a 1-cm complete wrap fashioned over a large bougie, the so-called ‘short floppy Nissen’. 16,17 The use of a partial fundoplication has a number of potential advantages but several controlled studies have shown broad equivalence between the two approaches. 18,19 For the purpose of this study they are therefore regarded as equivalent. Although fundoplication will produce resolution of reflux symptoms in upwards of 90% of patients, there is continuing debate regarding the risks, side effects and durability of surgical therapy.

Medical management

There is no doubt that PPIs, sometimes combined with prokinetic agents, are the most effective treatment for moderate to severe GORD. For the purpose of this study, medical therapy has been taken to mean long-term therapy with PPIs (or H₂RAs if intolerant to PPIs). Although fundoplication is highly effective for controlling GORD, there has been considerable uncertainty whether exchanging symptoms associated with the best medical management of GORD for those of the side effects of surgery is advantageous for the patient and cost-effective for the health-care provider.

The costs of laparoscopic fundoplication appear to be equivalent to those of 2–3 years of maintenance treatment with PPIs, although it is acknowledged that the costs of PPIs are falling. 20 The costs of surgery are related largely to two factors – the incidence of complications/length of hospital stay and the number of patients requiring long-term medical interventions after surgery.

Rationale for the study design

The study described in this report aimed to clarify the place of laparoscopic fundoplication in the belief that decisions about the management of GORD should be based on unbiased, statistically precise comparisons of alternative policies. All patients in this study fulfilled three criteria: they were on long-term acid suppression with PPIs; they had symptoms that were thought to be adequately controlled; and they were suitable in terms of fitness and co-morbidity for either surgical or continuing medical treatment for their GORD.

The most likely sources of bias were in the ways in which the groups being compared were selected, the ways in which their outcomes were assessed, and how the management was actually delivered. This is the basis for using a pragmatic randomised controlled trial design. Random allocation protects against selection bias. Confining the trial to those with no clear treatment preference limits biased patient-centred assessment of outcome, and pragmatic comparison of alternative policies (with intention to treat analysis) avoids bias introduced by non-compliance. This approach has limitations, however, and for this reason we chose to incorporate two parallel, non-randomised preference groups.

Excluding those with a clear preference for one policy or the other limits extrapolation and generalisation. Study of this group may give insights into the reasons for preference and hence give pointers to patient choices after the study. 21 Furthermore, preference may influence outcome and, if so, this may also help when making treatment decisions. 21,22 A third reason for including the parallel, non-randomised preference groups23 is that the addition of data from the preference groups may reduce imprecision around the estimates from the randomised comparison and this may be particularly useful for rare events, such as complications, that can be confidently ascribed to one or other treatment. (The limitation is that these groups are not derived by random allocation and hence the comparisons are prone to the biases of non-randomised studies.)

The decisions about, and comparisons between, randomised and preference groups require valid measurement of treatment outcome. Although there are a number of quality of life tools available, none was sufficiently specific to assess the spectrum of gastrointestinal symptoms associated with the treatment of GORD, particularly surgery. For this reason the development and validation of a new outcome measure (the REFLUX questionnaire; see Chapter 4) was an essential component of the study. 24

GORD and its management represent a very significant call on NHS resources. Although clinical effectiveness, acceptability and safety will be important determinants of future policy, the issues of cost and resource use may be over-riding. A prospective, multicentre study25 found that the total cost for chronic PPI (omeprazole) therapy over 5 years was less than the cost of an open fundoplication; however, two other studies26,27 found laparoscopic surgery to be less expensive in the long run than daily treatment with 20–40 mg of omeprazole. In one of these studies27 laparoscopic fundoplication became more cost-effective at 1.4 years post procedure. A Canadian Markov model comparing medical management with laparoscopic fundoplication concluded that laparoscopic fundoplication became cost-effective at 3.3 years post operation. 28

A recent UK trial-based economic analysis comparing laparoscopic fundoplication with PPIs using data on 100 GORD patients29 reported that the incremental cost per point improvement in combined gastrointestinal and psychological well-being scores at 12 months for laparoscopic fundoplication versus PPI was £293, and the incremental cost per additional patient returned to a physiologically normal acid score at 3 months was £5515. 29 There are, however, no existing studies in the UK that have compared laparoscopic fundoplication with PPIs using a generic measure of health, such as quality-adjusted life-years (QALYs). Expressing health benefits in terms of QALYs would provide decision-makers with a basis for comparison with other uses of health-care resources in a range of disease areas and specialties.

There is little doubt that PPIs are the most effective pharmacotherapy30 for moderate to severe GORD and, for the purpose of this analysis, medical therapy will be taken to mean long-term therapy with PPIs. Although fundoplication is a highly effective therapy for controlling GORD, the question is whether surgery, which can alleviate GORD symptoms but may have unwanted side effects, is advantageous for the patient and cost-effective for the health-care provider.

This is the reason for the economic evaluation. Policy should be guided by both assessment of the relative cost-effectiveness of alternative policies and assessment of the impact that possible policy changes would have for the NHS and for patients with GORD.

Study design

The study had two complementary components:

1. a multicentre, pragmatic randomised trial (with parallel non-randomised preference groups) comparing a laparoscopic surgery-based policy with a continued medical management policy to assess their relative clinical effectiveness

2. an economic evaluation of laparoscopic surgery for GORD comparing the cost-effectiveness of the two management policies to identify the most efficient provision of future care and describe the resource impact that various policies for fundoplication would have on the NHS.

Patients who consented to participate in the randomised trial were randomly allocated to either laparoscopic surgery or continued medical management. Those patients who had a strong preference for one or other of the two treatment options could be recruited to the preference study. Clinical history at trial entry was recorded on participants’ entry forms (see Appendix 1). Participants completed health status questionnaires at the time of recruitment to the study and then at specified times equivalent to 3 and 12 months after surgery (see Appendix 2).

Approval for this study was obtained from the Scottish Multicentre Research Ethics Committee and the appropriate local research ethics committees.

Clinical centres

Clinical centres were based on local partnerships between surgeons with experience of laparoscopic fundoplication and the gastroenterologists with whom they shared the secondary care of patients with GORD. Centres were eligible if they included:

1. a surgeon who had performed at least 50 laparoscopic fundoplication operations

2. one or more gastroenterologists who agreed to collaborate with the surgeon(s) in the trial.

Study population

Potential participants, who were identified both retrospectively and prospectively, were invited to attend an outpatient appointment (see Appendix 3). The participating clinician reviewed each patient’s symptoms and treatment regimen and assessed eligibility (see Appendix 4).

Eligible patients were those for whom care had been provided by a participating clinician who was uncertain which management policy (surgical or medical) was better. In addition, patients had to have documented evidence of GORD (based on endoscopy and/or manometry/24-hour pH monitoring) as well as symptoms for more than 12 months requiring maintenance PPI therapy for reasonable symptom control. Patients who were intolerant to PPIs and who therefore required H₂RA therapy to control their symptoms were also included. Patients who were morbidly obese [body mass index (BMI) > 40 kg/m²], patients with Barrett’s oesophagus of more than 3 cm or who had evidence of dysplasia, patients who had a para-oesophageal hernia and patients with an oesophageal stricture were all excluded.

If eligibility was confirmed the patient was invited to see the local research nurse who described the trial, giving supplementary information describing the operation (see Appendix 5) and answering any questions or concerns. This process is summarised in Figure 1.

FIGURE 1.

Flow chart describing patient recruitment.

Consent to participate

Health technology policies being compared

Study registration (and treatment allocation when randomised)

The entry procedure distinguished between those who agreed to randomisation and those who agreed to participate in the preference part of the study.

Once a participant had agreed to join the trial the research nurse recorded basic identifying and descriptive information on a standard form (see Appendix 1). A letter was sent to each participant, confirming their participation and whether they were taking part in the randomised or preference component of the trial. At this point the participant was also asked to complete a baseline questionnaire (see Appendix 2).

The treatment allocation for participants in the randomised component of the trial was computer generated in the trial office; it was stratified by centre, with balance in respect of other key prognostic variables – age (18–49 years or 50 + years), sex (male or female) and BMI (≤ 28 or > 29 kg/m²) – by a process of minimisation. Randomisation was organised centrally at the Health Services Research Unit, Aberdeen, and was independent of all clinical collaborators.

Clinical management

The first 146 randomised participants (70 allocated surgery and 76 allocated medical management) were sent details of their allocation at the same time as the baseline questionnaires. This was changed for subsequent participants at the request of the Data Monitoring Committee (DMC; see page 10) such that the allocation was only generated once completed baseline forms had been returned. This was to ensure that there was no possibility that knowledge of the allocation might influence responses to the baseline questionnaire (as well as ensuring that a completed baseline questionnaire would be received from all randomised participants). A summary of the trial procedure pathways is illustrated in Figure 2.

FIGURE 2.

Flow chart showing trial procedures post recruitment.

Participants who were allocated to the surgical arm were invited to a consultation with the collaborating surgeon. During this consultation the surgeon confirmed that there were no contraindications to surgery and discussed the operation in more detail, before arranging an operation date. The intraoperative details were recorded by the surgeon on specially designed study forms (see Appendix 8). All other in-hospital data collection was the responsibility of the local research nurse. In all respects, other than the trial interventions, clinical management was left to the discretion of the clinician responsible for care.

Data collection

Follow-up by postal questionnaire was performed at least twice at participant-specific time intervals after joining the study. This was around 3 and 12 months after surgery or at an equivalent time amongst those who did not have surgery. The latter times were chosen through a process of matching participants in the various groups. Participants received up to two reminder telephone calls or letters to encourage non-responders to return their postal questionnaires. On occasion, and at the convenience of participants, questionnaires were completed over the phone.

All data were sent to the trial office in Aberdeen for processing and staff in Aberdeen worked closely with participants’ local research nurses to secure as complete and accurate data as possible. A random 10% sample of all data was double entered to check accuracy. Extensive range and consistency checks further enhanced the quality of the data.

The principal study outcome measures

The primary outcomes for measuring the differences in effects between medical and surgical treatment were:

• a ‘disease-specific’ measure incorporating assessment of reflux and other gastrointestinal symptoms and the side effects and complications of both therapies (the REFLUX questionnaire was developed specifically for this study as described in Chapter 4)

• NHS costs including treatments, investigations, consultations and other contacts with the health service.

The secondary outcome measures were:

• health-related quality of life (HRQoL) – EuroQol-5 Dimensions (EQ-5D) and Short Form with 36 Items (SF-36)

• patient costs including loss of earnings, reduction in activities, and the costs of prescriptions and travel for health care

• other serious morbidity, such as operative complications

• mortality.

The instruments for collecting this information are shown in Appendix 2.

Sample size

The original aim was to recruit 600 participants to the randomised trial to give 80% power to identify a difference between the two groups of 0.25 of a standard deviation in respect of the disease-specific instrument and other continuous variables such as EQ-5D or SF-36, using a significance level of 5%. Based on the same arguments it was planned that 300 people would be recruited to each arm of the preference study. The cost savings of a surgical policy largely depend on the number of patients managed surgically who no longer require PPI treatment, and a trial with 300 surgically managed patients would have estimated this proportion to within about 5% with 95% statistical confidence.

However, prompted by a lower rate of recruitment than expected, this target was revised in January 2003 in consultation with the DMC and representatives of the HTA programme. It was agreed that a larger benefit (0.3 of a standard deviation) was clinically plausible based on improvements seen after surgery amongst more severely affected people. This was calculated to require 196 in each group to give 80% power (p = 0.05). On this basis it was agreed that recruitment would be extended for an extra year, aiming for this revised sample size.

Statistical analysis

A single principal analysis of the randomised trial was planned when all participants had been followed up for 12 months after surgery (or an equivalent time if managed medically). The primary outcome measure [REFLUX quality of life (QoL) score at 12 months] and secondary outcome measures (REFLUX QoL score at 3 months; SF-36, EQ-5D, REFLUX symptom scores and use of reflux-related drugs at 3 months and 12 months) were analysed using general linear models that always adjusted for the minimisation covariates (age, BMI and sex) and where appropriate (defined by significant at the 5% significance level) also adjusted for baseline score and baseline score by treatment interaction. A secondary, pre-stated, subgroup analysis explored the differential effects of surgeons’ preferred operative procedures on the primary outcome measure. All analyses used 95% confidence intervals.

The primary analysis of the randomised trial was by intention to treat. The intention to treat approach gives the least biased estimate of effectiveness of the two interventions. As a secondary comparison we were also interested in estimating the efficacy of the treatment received. Given that a relatively large proportion of the randomised surgical participants did not receive surgery, we used two approaches to estimate the efficacy of the treatment – a per protocol analysis and an adjusted treatment received analysis. 32 In the per protocol analysis, participants who were randomised to surgery and actually received surgery were compared with participants who were randomised to medication and actually received medication (i.e. the compliers in the surgical group were compared with the compliers in the medical group). In an open trial design the per protocol estimate can have substantial selection bias. One way to estimate the effect when the allocation was complied with while adjusting for possible selection biases is to use a latent variable approach. 33 We used the method of adjusted treatment received as described by Nagelkerke et al. 32 The method used a two-stage least squares approach whereby treatment randomised was regressed onto treatment received and the residuals from that model were used as an independent variable in a second model together with the treatment received to estimate the effects on the various primary and secondary outcome measures.

For the preference group, only the primary outcome was analysed statistically. The analysis compared the preference surgical group with the preference medical group and adjusted for the minimisation factors. As described above, for logistical reasons and to maintain balance between the randomised and preference groups, we capped the number of preference participants at 20 per group per centre. The study design was not therefore a true comprehensive cohort. We did consider modelling differences between the randomised and preference groups; however, it is not universally accepted that formal modelling is appropriate in this context. In this case we knew from the randomised arms that there was a strong interaction with baseline reflux QoL, and in addition we also knew that there was a large difference in QoL between preference arms at baseline (and patient demographics). We therefore decided that formal modelling of the arms would not add much to the comparison given the large confounding between preference groups.

Missing items in the health-related outcome measures were treated as per the instructions for that particular measure. No further imputation for missing values was undertaken.

Data monitoring

In March 2003 an independent DMC met for the first time to review the overall conduct of the trial, patient accrual, data collection and an interim analysis of the data. They considered data available to them up to January 2003. At that time 146 participants had been recruited to the randomised trial, 76 allocated to the randomised medical group and 70 allocated to the randomised surgical group. Of the 177 preference participants, 77 chose the medical group and 100 chose the surgical group. On the basis of the data available to them they requested that the treatment allocation procedure be investigated. This led the DMC to instruct that the entry procedures be amended (as described on page 7) so that participants were only randomised once the trial office had received the baseline questionnaire and all of the other baseline paperwork (see Appendices 2, 4, and 6).

The DMC met on two further occasions (July 2003, January 2004) and were happy with the trial progress and interim analyses and saw no reason to recommend any further changes to the protocol.

A preliminary comparison of the cost-effectiveness of pharmacotherapy and surgery (laparoscopic fundoplication) in the treatment of gastro-oesophageal reflux disease

Background

Early in the study we chose to develop a preliminary economic model. Using the best evidence then available we developed a decision analytic model to provide preliminary estimates of costs and outcomes for medical and surgical management prior to the REFLUX trial reporting. 34,35 This chapter describes the preliminary economic model.

Methods

Description of the model

The model was probabilistic and took the perspective of the UK NHS. Health outcomes were expressed in terms of QALYs with a lifetime horizon. The model related to a 45-year-old patient as this is the peak age of presentation with GORD. 36 There proved to be very little difference between men and women; thus, only the results for males are presented here. Costs and QALYs have been discounted at a rate of 3.5% per annum. 37

The structure of the model can be seen in Figure 3. Two strategies were compared: long-term medical management and immediate laparoscopic surgery for GORD. Medical management was assumed to be prescribed for the remainder of a patient’s lifetime (30 years for a 45-year-old patient). Surgery was assumed to occur immediately following entry into the surgical arm of the model. The model was also split into short-term and long-term elements. The short-term model related to the period immediately following allocation to surgery or medical management. The longer-term element tracked the patient’s progression through a series of states over the remainder of their lifetime. Patients were assumed to stay in a ‘wait’ state before surgery, during which they would have received a maintenance dose of PPIs. The effects of alternative waiting times for surgery were also explored using alternative scenarios (1 month and 1 year) to represent the possible length of delay. In these cases it was still assumed that surgery following relapse would occur immediately, that is there would be no delay. Monthly cycles represented the monthly transition probabilities between states in the model.

FIGURE 3.

Structure of the preliminary economic model.

For patients receiving surgery a small mortality risk is associated with laparoscopic fundoplication (approximately 5 per 10,000 patients)38–47 and this was included in the model. If patients survived surgery the outcomes could be success (cured) or failure (relapse). In addition, patients could relapse from a successful surgery each month. This rate was constant and lasted for only one cycle, during which a patient received a double dose of PPI. A scenario is also presented in which the risk of failure from surgery (and the need for revision) ended at 5 years after initial surgery. Patients could be given a reoperation following surgery failure. If the reoperation failed, surgery was deemed a total failure and patients were considered to have been prescribed long-term medical management with PPIs. For patients offered medical management following initial surgical failure, medical management was deemed a total failure if there was subsequent relapse from medical management, and patients were placed on a double dose of PPIs for the remainder of their lives.

Medical management patients had a risk of relapsing each month. They could be offered surgery or could receive a double dose of PPIs for a cycle, followed by a return to a stable (well) medical management state at a normal dose of PPIs. Patients receiving surgery following relapse on medical management faced the same transition probabilities as surgical patients post surgery. They could also receive one reoperation following surgery failure. Medical management following two operations was deemed a total failure and patients were placed on a double dose of PPIs for the remainder of their lives.

For both surgical and medical management patients there was a monthly risk of all-cause mortality. The age-specific death rate for men aged from 45 to 54 years was obtained from the UK Office of National Statistics (ONS)48 and used to calculate the probability of death from natural causes from one cycle to the next.

Results

Base-case cost-effectiveness

The base-case estimates of costs and QALYs associated with surgery are shown in Table 4. Over a lifetime, medical management (£4890) was estimated to cost less than surgery (£5014) but it was associated with fewer QALYs than surgery: 12.36 compared with 13.04.

TABLE 4 - Cost-effectiveness of surgery versus medical management for the treatment of gastro-oesophageal reflux disease

ICER, incremental cost-effectiveness ratio; QALYs, quality-adjusted life-years.

	Total costs (£)	Total QALYs	ICER
Surgery	5014.17	13.04	£180.61
Medical management	4890.59	12.36

The lifetime ICER for surgery versus medical management is thus £180. Based on this, as long as decision-makers are willing to pay more than £180 for an additional QALY, surgery would be regarded as the more cost-effective treatment option. However, mean costs and QALYs were estimated with uncertainty. Figure 4 shows the potential impact of the uncertainty in mean differences (surgery minus medical management) in costs and QALYs gained between the two groups (i.e. it shows mean costs and QALY differences based on the 1000 simulations). Figure 5 represents this uncertainty in the form of a cost-effectiveness acceptability curve. The probability that surgery is cost-effective at a threshold of cost-effectiveness of £30,000 per QALY is 0.639.

FIGURE 4.

Representation of the uncertainty in differential mean costs and quality-adjusted life-years (QALYs).

FIGURE 5.

Cost acceptability curve for surgery versus medical management.

Expected value of perfect information

The per patient EVPI for adults with GORD is illustrated in Figure 6. At a cost-effectiveness threshold of £30,000, EVPI is substantial at £15,106. At a threshold of £20,000, the EVPI is £10,081. EVPI for groups of parameters showed that all of the value of further research (£11,346 at a threshold of £30,000 for cost-effectiveness) is associated with the quality of life implications of medical or surgical therapies, indicating that this is where future research should focus.

FIGURE 6.

Per patient expected value of perfect information (EVPI).

Discussion

This was the first investigation of the cost-effectiveness of lifelong medical treatment compared with immediate laparoscopic fundoplication for the treatment of GORD. The results of this model suggest that, even when the risk of spontaneous failure of surgery exists for a patient’s lifetime, surgery for GORD is more cost-effective than lifelong management with drugs. However, the true cost-effectiveness of surgery is uncertain and, at a threshold for cost-effectiveness of £30,000 per additional QALY, the value of information surrounding the decision problem is high. The number of people with GORD suitable for surgery is likely to be sizeable and therefore the EVPI of £15,106 at a threshold of £30,000 per QALY implies that the EVPI will exceed the cost of further investigation. This, in turn, suggests that further research will be potentially cost-effective. EVPI analysis on groups of parameters suggested that further research should focus on collecting evidence relating to the HRQoL of patients on medical management and following surgery.

It was necessary to make a number of assumptions in the model. First, in the absence of applicable data, it was necessary to simplify the dosing adjustment used to deal with relapse. In clinical practice a more complicated titration of dose and duration of step-up or step-down dosing would be used. The effect of this would probably be that patients in relapse spend more than 1 month on a higher dose (and at higher cost) and simultaneously experience lower HRQoL for longer than modelled here. At present, given that equal consequences of relapse have been applied to those patients relapsing on medical management or surgery, it is unlikely that applying a more complex relapse dosing structure would have a significant effect on the results of the model.

Second, the costs of surgery only partially capture true cost. Surgery may have unwanted side effects or may spontaneously fail at some point in the future. Treatment of side effects or surgical failure has costs both in monetary and quality of life terms. A common side effect, temporary difficulty with swallowing (dysphagia), has been considered in the model and a probabilised cost loading used to incorporate its treatment. However, no disutility of dysphagia, bloating, flatulence or other unwanted side effects following surgery has been included in the model because of a lack of data and consensus on the magnitude of effect. Related to this is the availability of data for other states in the model. In the absence of other suitable data, the utility values used to reflect the post-surgical state were based on patients measured following laparoscopic cholecystectomy,71 which has some surgical similarities to laparoscopic fundoplication but may not have the same spectrum of postoperative discomfort or complications. The utility value associated with a surgical cure has been taken from UK age-specific population norms. 70 Also, it is unclear to what extent the post-surgical state can be likened to the utility of an average member of the UK population, that is, whether surgery actually generates a cure in utility terms.

Finally, because of the focus on those patients currently maintained on medical management, the analysis reported here did not consider management strategies other than medical management or surgery. In many clinical settings lifestyle management advice is being favoured as a first-line option, with medical management or surgery considered only as second-line therapies in patients who do not respond to lifestyle changes. This may limit the applicability of this model in certain settings.

Despite these necessary assumptions, the model presented here represents the first attempt to generate estimates of cost per QALY for surgical and medical management strategies for the treatment of GORD patients in the UK. The results of the model suggest that, on the basis of current evidence, laparoscopic fundoplication may well represent a cost-effective means of treating GORD rather than lifelong medical management. Coupled with the apparent safety of the surgical procedure (in experienced hands), patients and the health service may benefit from increased substitution of surgery for medical management. What this preliminary analysis confirmed was the need for more robust data, especially in respect of HRQoL, and these data were being generated in the REFLUX trial.

The development of a new measure of quality of life in the management of gastro-oesophageal reflux disease: the REFLUX questionnaire

Introduction

Although several GORD-specific or gastrointestinal-specific symptom scales and quality of life scales have been developed,79–87 we found that none captures the experience of patients receiving alternative treatments in sufficient detail for evaluating outcomes in the REFLUX trial. Of particular concern was that these measures do not reflect patients’ experiences of the side effects of surgery for GORD, which include general gastrointestinal symptoms as well as oesophageal reflux itself. 85 A new condition-specific outcome measure was therefore developed for use within the REFLUX trial. The aim of this measure was not only to assess the symptoms of GORD but also the side effects of both medical and surgical treatment for GORD and the effects that these have on HRQoL. There were two requirements for the new measure: it had to measure HRQoL and not merely symptom experience; and its content had to cover the effects of treatment for GORD as well as the symptoms of GORD. This chapter describes the development and assessment of the new measure.

Method

Piloting

The initial version of the questionnaire (with the 31 items) was piloted on a sample of 21 patients from Aberdeen, some of whom had taken part in the interview phase. The questionnaire was posted out to the patients asking them to complete it. At a later date they were interviewed about its readability and acceptability. Specifically, they were asked about whether they had any problems understanding the items, whether the response categories were appropriate for them and whether they thought that anything was missing from the questionnaire. The questionnaire was modified following the feedback from these interviews. At this stage a small number of items (three) were discarded as unsuitable or potentially ambiguous, others were reworded and three items that were not originally included in the initial version of the questionnaire, but were repeatedly mentioned by the patients and felt to be of importance, were added The new version therefore also had 31 items.

Results

Scoring

Generating weights for the reflux quality of life score

All 727 participants with complete baseline data on the REFLUX quality of life items and EQ-5D VAS were included in the analysis. Although coefficients for three of the seven quality of life items were not statistically significant, we kept them in the regression model for completeness. In contrast, we excluded the wind item from the RQLS model as the coefficient consistently showed the wrong sign and was not statistically significant. In effect, the wind item will receive a weight of zero when calculating the final score. The work, physical and social activities item had the largest coefficient and thus had most effect on the EQ-5D VAS, and the sleep item had the smallest coefficient. The final model coefficients used to calculate the RQLS are given in Table 6.

TABLE 6 - Model coefficients used to calculate the REFLUX quality of life score (RQLS)

B, beta; NS, not significant; SE, standard error.

Adj r² = 0.22.

REFLUX quality of life item	B	SE	Significance
Heartburn	–1.346	0.81	NS
Acid reflux	–1.700	0.70	< 0.05
Eating and swallowing	–1.103	0.68	NS
Bowel movements	–1.954	0.61	< 0.01
Sleep	–0.351	0.66	NS
Work, physical and social activities	–2.147	0.84	< 0.05
Constant	89.995	1.51	< 0.001

The coefficients from this model were used as weights for calculating the quality of life score by multiplying the response to each quality of life item (coded from 0 ‘not at all’ to 4 ‘extremely’) by the corresponding weight (i.e. the coefficient from Table 6) and subtracting these values from the constant term as follows:

Raw RQLS = 90 – ( heartburn quality of life × 1.35 ) – ( acid reflux quality of life × 1.70 ) – ( wind quality of life × 0 ) – ( eating quality of life × 1.10 ) – ( bowel movement quality of life × 1.95 ) – ( sleep quality of life × 0.35 ) – ( activities quality of life × 2.15 ) .

The score was then standardised to a scale from 0 (worst quality of life) to 100 (best quality of life) as follows:

Standardised RQLS = ( raw RQLS – 55.6 ) × 2.91.

Figure 7 presents the frequency distribution of quality of life scores for patients at baseline. The mean score was 65.0 with a standard deviation of 24.3.

FIGURE 7.

Distribution of reflux quality of life scores (RQLS).

Generating weights for the reflux symptom scores

The PCA identified five components that accounted for 57% of the variance in the items (Table 7). In general, the component structure reflected the themes identified when the items were developed; however, component 1 grouped together heartburn-like symptoms and sleep disruption into general discomfort (Table 7). The first component after rotation explained 19% of the total variance and included seven items with loadings above 0.4. Component 2 explained 12% of the total variance and included six main items. The remaining three components accounted respectively for 10%, 9% and 8% of the total variance. Component loadings were used to construct a profile of five REFLUX symptom scores to summarise an individual’s symptom experience. In the first instance we suggested the following labels for these components: 1 = general discomfort; 2 = wind and frequency; 3 = nausea and vomiting; 4 = activity limitation; and 5 = constipation and swallowing.

TABLE 7 - Component loadings used to calculate the REFLUX symptom scores

Factor loadings < 0.3 have been suppressed.

Item	Component 1	Component 2	Component 3	Component 4	Component 5
A1: Heartburn	0.674
A2: Discomfort in chest	0.643
B1: Acid reflux	0.654
B2: Vomiting			0.734
B3: Regurgitation			0.556
B4: Nausea			0.541
B5: Urge to be sick			0.709
C1: Flatulence		0.738
C2: Belching		0.553
C3: Feeling bloated		0.568
C4: Stomach gurgling		0.515
D1: Difficulty swallowing					0.338
D2: Eating restricted	0.421
E1: Diarrhoea		0.722
E2: Constipation					0.839
E3: Urgent need to go		0.696
E4: Feeling like bowels not emptied					0.645
F1: Difficulty sleeping lying down	0.777
F2: Difficulty getting to sleep	0.814
F3: Disrupted sleep	0.791
G1: Paid/unpaid work				0.695
G2: Less strenuous activities				0.571
G3: Strenuous activities				0.755
G4: Social activities				0.588

Each symptom score was calculated by multiplying the response to each of the symptom items in that score (coded from 0 ‘every day’ to 4 ‘not at all’) by the corresponding weight (i.e. the component loading for that item from Table 7) and then summing across the items. For the four items in activity limitation we grouped the response codes ‘not applicable’ and ‘no, my symptoms do not affect me’ as 4, and recoded the other categories from 0 ‘I no longer work/perform these activities because of my symptoms’ to 3 ‘my symptoms have affected me but I still work/perform these activities’. Symptom scores were then standardised to a scale from 0 (worst symptom score) to 100 (best symptom score) as follows:

General discomfort = 5.24 × [ ( item A 1 × 0.674 ) + ( item A 2 × 0.643 ) + ( item B 1 × 0.654 ) + ( item D 2 × 0.421 ) + ( item F 1 × 0.777 ) + ( item F 2 × 0.814 ) + ( item F 3 × 0.791 ) ] .

Wind and frequency = 6.59 × [ ( item C 1 × 0.738 ) + ( item C 2 × 0.553 ) + ( item C 3 × 0.568 ) + ( item C 4 × 0.515 ) + ( item E 1 × 0.722 ) + ( item E 3 × 0.696 ) ] .

Nausea and vomiting = 9.84 × [ ( item B 2 × 0.734 ) + ( item B 3 × 0.556 ) + ( item B 4 × 0.541 ) + ( item B 5 × 0.709 ) ] .

Activity limitation = 9.58 × [ ( item G 1 × 0.695 ) + ( item G 2 × 0.571 ) + ( item G 3 × 0.755 ) + ( item G 4 × 0.588 ) ] .

Constipation and swallowing = 13.72 × [ ( item D 1 × 0.338 ) + ( item E 2 × 0.839 ) + ( item E 4 × 0.645 ) ] .

Table 8 presents the mean symptom scores at baseline. There were pronounced ceiling effects for nausea and vomiting, constipation and swallowing, and activity limitations: 26%, 25% and 17% respectively of the sample had a maximum score of 100. In contrast, wind and frequency showed a more normal distribution.

TABLE 8 - Mean REFLUX symptom scores at baseline

Reflux symptom dimension	Mean	SD	Median
General discomfort	59.4	25.6	60.3
Wind and frequency	50.7	22.1	49.6
Nausea and vomiting	81.7	19.6	89.0
Activity limitation	79.2	16.5	81.5
Constipation and swallowing	77.7	20.6	79.6

Both the RQLS and REFLUX symptom scores were calculated only for individuals with complete data. However, there were few missing data. REFLUX scores could be calculated for over 95% of patients at baseline. Missing data rates for symptom items ranged from 1% to 2%, and for quality of life items from 3% to 5%.

Validity

Table 9 presents the relationship (Pearson’s r) between the RQLS and the eight SF-36 dimension scores. Social functioning and bodily pain showed the best relationships with the RQLS, and mental health the worst.

TABLE 9 - Relationship (Pearson’s r) between RQLS and SF-36 dimension scores

SF-36 dimension	RQLS
Physical functioning	0.42
Role limitations – physical	0.49
Bodily pain	0.56
General health perception	0.46
Energy/vitality	0.34
Social functioning	0.59
Role limitations – emotional	0.41
Mental health	0.18

Table 10 presents the proportion of respondents who had a score of 100 (best health) on the SF-36 dimensions as a percentage of those who had a best score of 100 on the RQLS. Whereas 96% of those who had the maximum score on the SF-36 physical functioning dimension also had a score of 100 on the RQLS, only 31% of those who had a score of 100 on the SF-36 bodily pain dimension also had a score of 100 on the RQLS.

TABLE 10 - Percentage (n) of respondents with the maximum reflux quality of life score (RQLS) with the maximum score on the SF-36 dimensions

SF-36 dimension	% (n)
Physical functioning	96 (70)
Role limitations – physical	66 (48)
Bodily pain	31 (23)
General health perception	–
Energy/vitality	–
Social functioning	74 (54)
Role limitations – emotional	97 (71)
Mental health	–

Figure 8 plots the mean RQLS against the SF-36 mental component score (MCS) and physical component score (PCS) grouped into fifths. The mean RQLS increases steadily and significantly between successive PCS groups. There is a similar pattern for MCS groups except that respondents in the highest fifth have a lower mean RQLS than those in the next lower fifth.

FIGURE 8.

Reflux quality of life score (RQLS) by SF-36 mental component summary score and physical component summary score (grouped into fifths).

Sensitivity to change

Participants reported whether they were being prescribed medication at baseline and first follow-up. This information was used to classify them into four groups: those prescribed medication at baseline and follow-up (n = 293); those prescribed medication at baseline but not follow-up (n = 186); those prescribed medication at follow-up but not baseline (n = 3); and those not prescribed medication at all (n = 7). As the last groups are reassuringly small, Figure 9 presents mean change in RQLS (baseline score – follow-up score) for the first two groups.

FIGURE 9.

Change in reflux quality of life score (RQLS) by change in prescribed medication (baseline to follow-up).

A negative score indicates an improvement in quality of life. Although the RQLS improved for both groups (paired t-tests showed significant change), patients whose medication status changed between baseline and follow-up (medication at baseline but not at follow-up) showed a greater improvement in their RQLS than patients whose medication status stayed the same (medication at baseline and follow-up).

Discussion

Background

This chapter describes a study that was conducted in addition to the research activities described in the trial protocol. It is the result of discussion among the trial team in which it was decided that it would be wise to check the validity of a questionnaire measure that was devised specifically for, and used for the first time in the context of, the REFLUX trial. We have called this measure the Beliefs about Surgery questionnaire (BSQ). It has the potential to be further developed as a tool for use by consultants and surgical teams. In the sections below we describe the initial analyses that were carried out to determine the validity of the measure. In the final section we suggest further work that could result in the development of a tool to support communication between consultants and patients with GORD as they discuss treatment preferences and decisions.

Introduction

Current health-care policy and practice acknowledge the importance of offering choice across the spectrum of health care to users of the health-care system. 91 It is plausible that people’s choices about treatment will be influenced by their beliefs about the risks and benefits of various treatments, which in turn will be shaped by their experiences or anticipated experiences of treatment processes. Indeed, this link between beliefs (cognitions) and action is represented in Leventhal’s common sense model of self-regulation in the face of a threat to health92 as follows. People appraise a health threat situation with reference to cognitions about the illness and then implement coping procedures to restore their physical or emotional equilibrium. The model specifies the cognitive components of this appraisal process in terms of factors that have become known as the illness representations framework. 93 The dimensions of this framework include beliefs about effective treatment or control of the illness (e.g. ‘taking medication will be effective’; ‘surgery may be more effective than medication’; ‘recovery from surgery could take a long time’). A questionnaire measure about illness representations has been developed and is frequently used to investigate people’s cognitions about illness in the context of Leventhal’s model. 92 This chapter reports the development and validation of a measure relating to beliefs about surgical treatments.

The treatment beliefs component of the Leventhal model has been investigated by Horne,94 who proposed that behaviour relating to treatment (e.g. adherence) is determined by perceptions about treatment rather than, or in addition to, perceptions about illness. There are two broad classes of treatments: those involving professional intervention (e.g. medicine, surgery, therapy) and those involving the adoption of different lifestyle behaviours (e.g. exercise, diet, stress management). The Beliefs about Medicines questionnaire (BMQ)93 developed by Horne and colleagues assesses perceptions about one form of treatment. The BMQ has been validated using a chronic illness sample (n = 524), including people diagnosed with asthma, diabetes, renal disease and psychiatric illness, and cardiac and general medicine inpatients. On the basis of principal components analysis and confirmatory factor analysis, four subscales were identified relating to beliefs about medications specific to the diagnosed condition (‘concerns’ about taking the medication and ‘necessity’ of taking the medication) and beliefs about medication in general (‘harmfulness’ of medication in general and ‘overuse’ of medication in general). The psychometric properties of these scales have been reported by Horne et al. 93 and demonstrate high levels of discriminant validity, criterion-related validity and stability of the factor structure across the different illness groups. This chapter investigates the measure of beliefs about medicine amongst people with GORD and also a parallel measure of beliefs about surgery, in the context of the REFLUX trial.

Because the REFLUX trial involved non-randomised preference groups, it was felt important from the start to include a measure that would investigate the process of patients’ decision-making about their treatment choices. Thus, this trial provided the opportunity to answer three questions relating to beliefs about treatment. First, would baseline measures provide support for the validity of the BMQ for individuals in a chronic illness group that was different from the groups investigated in the original validation study, namely people suffering GORD? Second, if participants were asked to answer questions relating to beliefs about surgery (in the form of a BSQ), would their answers suggest that these beliefs relate to professional interventions in general (i.e. would the dimensions of the BSQ converge with dimensions of the BMQ) or would distinguishable factors emerge? The answer to this question could be important when treatment options include both medical and surgical interventions. Third, would data from the BMQ and the BSQ, administered at baseline, provide evidence of criterion-related validity? Such evidence would be provided if the profile of scores on the BMQ and BSQ distinguished between the surgery group and the medication group in the preference groups (the ‘criterion’) but not in the randomised groups.

Methods

Results

Of the people recruited to the trial, 329 (92.12%) in the randomised groups and 419 (91.48%) in the preference groups completed the baseline questionnaire. Data from these 748 participants were analysed in this validation study.

Distributions of scores on the BMQ and BSQ items were generally acceptable. Skewness was greater than 1 for only two variables: ‘My health, at present, depends on my medicines’ (sk = 1.48); ‘I would be willing to have an uncomfortable test’ (sk = 1.09). Kurtosis was greater than 1 for six variables: ‘My health at present depends on my medicines’ (ku = 2.20, modal value = 1); ‘Natural remedies are safer than medicines’ (ku = 1.48, modal value = 3); ‘Medicines do more harm than good’ (ku = 1.81, modal value = 4); ‘I would be willing to have an uncomfortable test’ (ku = 1.74, modal value = 2); ‘Surgery does more harm than good’ (ku = 1.10, modal value = 4); ‘Doctors are too quick to suggest surgery’ (ku = 1.40, modal value = 4).

Exploratory principal components analysis on BMQ/BSQ items

A PCA (with oblimin rotation) was conducted on all BMQ and BSQ items together. Based on the structure of the BMQ for this sample, the combined BMQ/BSQ was expected to comprise up to six discriminable factors (corresponding to the two subscales for each of general medicine, specific medicine and general surgery). The scree plot (Figure 10) suggested that it was appropriate to extract five factors, which together accounted for 50.95% of the variance in the scores. Using a cut-off of 0.4 for item inclusion, only one item in the item pool did not load on to a factor and none of the 26 items had diffuse loading. Table 13 presents the item loadings reported by Horne et al. 93 and the item loadings derived from the REFLUX data.

FIGURE 10.

Scree plot indicating that a five-factor solution would be appropriate.

TABLE 13 - Item loadings for the BMQ (four-factor solution) reported by Horne et al.94 and the BMQ/BSQ (five-factor solution) derived from the reflux baseline data (cut-off for inclusion 0.3)

Square brackets indicate original BMQ wording.

Items	Loadings
	Horne et al.94 (n = 524)				REFLUX baseline (n = 739)
	F1: Specific concerns	F2: Specific necessity	F3: General harm	F4: General overuse	F1: General medicine overuse	F2: Specific medicine necessity	F3: Surgery overuse/harm	F4: Specific medicine concerns	F5: Surgery concerns
Having to take medicines [this medicine]a worries me	0.80							0.81
I sometimes worry about becoming too dependent on my medicines	0.78							0.76
I sometimes worry about the long-term effects of my medicines	0.76							0.80
My medicines disrupt my life	0.60							0.60
My life would be impossible without my medicines		0.81				0.84
My health, at present, depends on my medicines		0.76				0.82
Without my medicines I would be very ill		0.74				0.79
My health, in the future, will depend on my medicines		0.70				0.79
My medicines protect me from becoming worse		0.65				0.59
If doctors had more time they would prescribe fewer medicines			0.81		0.74
Doctors place too much trust in medicines			0.75		0.80
Doctors use too many medicines			0.71		0.75
Natural remedies are safer than medicines			0.47	0.45	0.50
Most medicines are addictive				0.71	0.48
Medicines do more harm than good				0.67	0.61
All medicines are poisons				0.58	0.51
My medicines are a mystery to me				0.55	(0.36)
People who take medicines should stop their treatment every now and again				0.51	0.47
Doctors rely on surgery too much							0.75
Surgery does more harm than good							0.74
Doctors place too much trust in surgery							0.73
Doctors are too quick to suggest surgery							0.71
I would be willing to have an uncomfortable test							–0.44
I worry about the risks of surgery							0.40		0.64
Surgery can result in new health problems									0.51
Surgery should only be taken as a last resort							0.40		0.51
Eigenvalue	3.38	2.92	1.60	1.44	4.56	3.57	2.31	1.59	1.22
Percentage variance explained	18.8	16.2	8.9	8.0	17.54	13.73	8.87	6.10	4.70

In the solution for the combined BMQ/BSQ, beliefs about medicines in general again formed one factor; the two factors relating to beliefs about medicines specific to the reflux condition mapped perfectly on to the solution reported by Horne et al. ,93 and beliefs about surgery also corresponded exactly to the pattern that was expected, based on the findings of Horne et al. 93 Beliefs about surgery appeared to be clearly discriminable from beliefs about medicines, as all between-factor correlations were less than 0.3 (Table 14).

TABLE 14 - Component correlation matrix for the BMQ/BSQ five-factor solution

Factor	F1	F2	F3	F4	F5
F1: Medication in general: overuse/harm	–
F2: Specific medication: necessity	0.014	–
F3: Surgery in general: overuse/harm	0.274	–0.053	–
F4: Specific medication: concerns	0.296	0.160	0.035	–
F5: Surgery in general: concerns	–0.048	–0.051	0.119	–0.116	–

Discussion

Baseline measures in the REFLUX trial provided support for the validity of the BMQ for individuals suffering from GORD. However, the two general medicine scales (labelled harm and overuse) merged in this factor solution. It is possible that specific characteristics of the sample may explain this. For example, GORD is a condition for which medication is taken symptomatically whereas the original validation study was conducted with people experiencing chronic illnesses in which medications are taken continuously. This could have increased the tendency of the GORD sample to discriminate between items relating to medications specific to the illness and correspondingly decreased the tendency to discriminate between the items in the general medicine scales. Furthermore, all of the current sample were trial participants and their involvement in the recruitment and informed consent processes of the trial may have made the GORD-specific items more salient and therefore more discriminable than the items about medicines in general.

Importantly, when participants were asked to answer questions relating to beliefs about surgery, their answers yielded factors that were discriminable from those relating to beliefs about medications, suggesting that these participants held distinctive patterns of beliefs about these two kinds of treatment, rather than about professional interventions in general. Furthermore, data from the BMQ and the BSQ provided evidence of criterion-related validity of the BMQ and the BSQ in that the profile of scores on the BMQ and the BSQ distinguished between the surgery group and the medication group in the preference arm of the trial but not in the randomised arm. In other words, knowing nothing about the participants other than their BMQ/BSQ scores allowed a reasonably good prediction of their treatment choices.

Beliefs about treatment have previously been investigated in relation to adherence to medication regimens but little research in this area has explored the issue of patient choices about treatment. The addition of a measure of beliefs about surgery to the existing measure of beliefs about medication provided the opportunity to explore such decisions. In addition, it may be that a patient’s score on the BSQ can provide important clinically relevant information. It is possible that people with less negative beliefs about surgery experience less anxiety associated with the surgery and, as there is evidence that anxiety is associated with poorer clinical outcomes post surgery,96 this could be important information for surgeons to be aware of. Furthermore, it may be that knowing the scores on these two questionnaires could help clinicians to counsel patients who have higher levels of concern or to help patients make better decisions about their treatment. It could be helpful to explore these possibilities in future research.

In terms of the common sense self-regulation model more generally, there is ample evidence that people’s illness perceptions influence their emotional and behavioural responses to an illness threat. 92 Perceptions and beliefs about how an illness may be controlled – including the treatment options of medication and surgery – are potentially important factors that may link with other behaviours such as altered lifestyle. The development of additional measures that relate to other ways of controlling the symptoms of chronic illness could be useful in identifying people’s preferred ways of coping with illness.

In conclusion, the perceptions of people with chronic illness about potential treatments can be measured validly and reliably. Core elements of the factor structure of the BMQ (in particular the distinction between specific and general classes of beliefs) were replicated in this study. Furthermore, responses to the BSQ indicate that beliefs about surgery form a distinct pattern of treatment representations and there is no redundancy between these two scales. Used together, the two measures can significantly distinguish between groups of individuals who choose one form of treatment over the other.

Implications for future research

As proposed above, there is potentially important follow-up work to be carried out in this field. Some possibilities that could be followed up using the REFLUX data set are:

1. Can the correct classification rate be improved further by the addition of other patient characteristics, for example severity of symptoms at baseline, sociodemographic factors, co-morbidities?

2. Can the recruitment of participants into the randomised groups of the trial be predicted by using a discriminant function analysis as described above to classify those who chose (a) surgery, (b) continued medication, and (c) to be randomised?

3. Do underlying treatment beliefs modify treatment effects? If so, can subgroups be identified who are more likely or less likely to respond well to alternative approaches to treatment, such as surgery or medical management?

4. Can the length of the questionnaire be reduced without reducing its predictive power, for example by using item response theory97 to identify the discriminating items?

5. Are these treatment beliefs stable or do they change over time or as a function of changes in symptom severity?

6. Could the questionnaire be adapted for use as a communication tool by consultants and surgical teams?

In conclusion, this work has thus generated a number of possibilities for continued work in this field. It appears that the BSQ is a valid instrument that has a number of potential applications in surgical practice and research.

Recruitment to the trial

Participants were recruited in 21 clinical centres, all within the UK (Table 16). Recruitment to the trial was open from March 2001 until the end of June 2004, although not all centres enrolled over the total period because of the staggered introduction of centres and early closure for logistical reasons in a few places.

Initial recruitment was limited to two centres (Aberdeen Royal Infirmary; St Mary’s Hospital, London) and these acted as pilot centres whilst systems for recruitment were developed. Roll-out of the trial to other centres started after 6 months with gradual extension to all remaining centres over the following 3 years. Figure 11 shows the total recruitment: the dotted line was the expected rate of recruitment over the last 16 months of the trial based on earlier recruitment. As can be seen, it proved difficult to sustain recruitment to the randomised component, although there was evidence for an increase in recruitment towards the time of recruitment closing.

FIGURE 11.

Actual versus expected recruitment rates: (a) randomised component; (b) preference component.

A total of 357 participants were recruited to the randomised component, with 178 allocated to surgery and 179 allocated to medical management. In total, 453 participants agreed to join the preference component, 261 choosing surgery and 192 choosing medical management. Table 16 shows recruitment by centre. Around 20% of the randomised participants were enrolled in Aberdeen; no centre contributed more than 11% of participants in the preference component.

Analysis populations

Throughout the analyses presented later in this chapter the participants in the randomised component are kept separate from those in the preference component (other than for rare surgical events). Primary analyses of the comparisons between surgical and medical management in both of these components are based on the allocated management at trial entry, that is, they are based on the intention to treat (ITT) principle. This sustains the integrity of the randomisation in particular. However, as described later in this chapter, a sizeable minority of participants did not actually receive their allocated management. To allow exploration of the impact (‘blunting effect’) that this might have on any observed differences, secondary analyses based on those who actually received their allocated management – per protocol (PP) analyses – were also undertaken and are presented alongside the ITT analyses.

The number of participants in each of the four main analysis populations is shown in Table 17. All 357 who joined the randomised component are in the randomised intention to treat (RITT) population whereas only the 280 within this group who actually received their allocated management are in the randomised per protocol (RPP) population. Similarly, all 453 participants who joined the preference component are in the preference intention to treat (PITT) population, and the 407 of these who were managed as originally chosen are in the preference per protocol (PPP) population.

Trial conduct

The derivation of the main study groups and their progress through the trial is shown in Figure 12. This is in the form of a CONSORT (Consolidated Standards of Reporting Trials) flow diagram. In total, 1078 patients were considered for trial entry. Of these, 200 were found not to meet one or more of the eligibility criteria. Of the 68 patients eligible for the study but not recruited, 51 declined to participate, 6 were subsequently deemed inappropriate for the study by the surgeon responsible for care, and the remaining 11 were missed.

FIGURE 12.

CONSORT diagram.

In total, 357 participants were recruited to the randomised component, with 178 randomly allocated to surgery and 179 to best medical management. A further 453 patients who wished to have one or other of the alternative approaches to management agreed to join the preference component – 261 to the surgical group and 192 to the medical management group.

In the early stages of the trial a few participants failed to return baseline questionnaires. After the first meeting of the Data Monitoring Committee, procedures were changed to prevent this, such that formal entry to the study (and random allocation if appropriate) occurred only after full baseline questionnaires had been received. The 1-year follow-up questionnaires were received from approximately 90% of the study participants. There were no substantive differences in response rates between the groups.

Three participants died before the 1-year follow-up was reached, two in the preference surgery group and one in the randomised medical group. None of these participants actually had surgery.

Description of the groups at trial entry

Surgical management

Table 21 gives details of the surgical management of those randomly allocated or in the preference for surgery group. For 47 allocated surgery there was subsequently a definite decision not to have surgery. For 25 of these, this was a clinical decision, most commonly the surgeon deciding that surgery was not appropriate. Most of the others changed their minds about having surgery for a variety of work- or home-related reasons, because of worries about the risks of surgery, because of a wish to avoid the preoperative tests, or because their symptoms had improved. A further 20 withdrew for uncertain reasons. There is no doubt, however, that a number of these participants suffered long delays before being formally offered surgery, and this was an important factor in their eventual decision to choose not to have surgery after all. The trial was conducted at a time when there was great pressure on surgical services in the NHS, with long delays for elective surgery for non-life-threatening benign conditions being common. Indeed, the average time between trial entry and surgery in the trial was 8–9 months (see Table 23).

In total, 111 (62.4%) of those randomised to surgery and 218 (83.5%) of the preference participants actually received surgery. Amongst the randomised participants, about 50% had a total wrap and 50% a partial wrap fundoplication. A total wrap was, however, the predominant procedure in the preference group (72.8%). The difference between the randomised and preference group fundoplication procedures was a reflection of the surgeon’s preferred procedure and not any systematic surgeon bias between a surgeon’s randomised and preference participants. This is illustrated in Figure 13, which shows that, within a given centre, the surgeon(s) performed the same procedures on their randomised and preference patients. The majority of operations were performed by a consultant and took around 2 hours to complete.

FIGURE 13.

Type of fundoplication performed by centre.

Intra- and postoperative surgical outcomes

Table 22 shows the intra- and postoperative surgical outcomes in the randomised and preference surgical participants who actually had a fundoplication. Two (0.6%) participants out of the total of 329 participants who had surgery required conversion to an open procedure (95% CI 0.2%–2.2%), and 8 (2.4%) had a visceral injury (95% CI 1.2%–4.7%). One participant had a blood transfusion. Three were admitted to a high dependency unit, but none to an intensive care unit. Nearly all were discharged to their homes after a median length of stay of 2 days. Three participants (0.9%) required a reoperation (95% CI 0.3%–2.6%) – all in the preference group – and three had dilatation of an oesophageal stricture or food disimpaction within 12 months of their initial surgery.

Follow-up at the time equivalent to 3 months after surgery

Follow-up at the time equivalent to 12 months after surgery

Health status

The health status measures at the 12-month follow-up are shown in Table 28. Within the randomised trial (RITT groups) there were still substantial differences across all measures (of the order of magnitude of one-third or one-half of a standard deviation of the score), with the surgery group having better scores than the medical group. The differences were larger when only the per protocol participants were considered (RPP groups). Details of statistical testing of the health status scores can be found in the next section of this chapter. For the reflux symptoms, although there were improvements across all symptom groups for surgical participants, the largest improvement in symptom score was for the general discomfort dimension. A detailed description of the responses to each symptom question is given in the REFLUX questionnaire (see Appendix 2). These improvements were also reflected in the SF-36 scores where the biggest differences were observed in the general health and bodily pain dimensions.

TABLE 28 - Follow-up at the time equivalent to 12 months after surgery – health status

ITT, intention to treat; PP, per protocol; QoL, quality of life; VAS, visual analogue scale.

	Randomised participants				Preference participants
	Surgical		Medical		Surgical		Medical
	ITT (n = 178)	PP (n = 111)	ITT (n = 179)	PP (n = 169)	ITT (n = 261)	PP (n = 218)	ITT (n = 192)	PP (n = 189)
REFLUX QoL, mean (SD)	84.6 (17.9)	88.3 (15.6)	73.4 (23.3)	73.1 (23.7)	83.3 (20.7)	86.0 (17.9)	79.2 (19.2)	79.4 (19.0)
Reflux symptom score, mean (SD)
General discomfort symptom score	84.7 (17.5)	90.2 (14.0)	67.4 (25.8)	66.7 (25.8)	85.0 (19.4)	87.7 (16.5)	73.9 (20.7)	74.0 (20.8)
Wind and frequency symptom score	56.7 (21.0)	56.9 (21.7)	52.6 (23.3)	52.7 (23.5)	56.9 (22.5)	57.5 (22.1)	61.4 (21.9)	61.5 (22.0)
Nausea and vomiting symptom score	91.9 (14.4)	94.7 (11.8)	84.0 (18.6)	83.3 (18.8)	91.1 (16.5)	93.3(13.8)	88.6 (15.4)	88.9 (14.4)
Activity limitation symptom score	90.7 (12.8)	93.3 (11.5)	82.2 (19.2)	81.6 (19.4)	90.8 (16.8)	92.4 (14.8)	87.3 (14.7)	87.4 (14.8)
Constipation and swallowing symptom score	79.3 (19.1)	80.2 (19.6)	74.5 (22.8)	75.2 (22.3)	78.5 (20.2)	79.1 (19.7)	83.6 (17.6)	83.8 (17.4)
SF-36 scores, mean (SD)
Norm-based physical functioning	48.9 (10.3)	49.6 (10.3)	47.2 (11.0)	47.2 (10.9)	49.7 (10.8)	50.3 (10.5)	47.4 (10.5)	47.4 (10.6)
Norm-based role physical	46.7 (11.4)	47.4 (11.3)	45.8 (11.8)	46.0 (11.7)	49.0 (11.2)	49.6 (10.5)	46.8 (10.7)	46.8 (10.7)
Norm-based bodily pain	47.7 (10.4)	48.5 (10.7)	44.5 (10.9)	44.5 (10.9)	49.1 (11.3)	49.9 (11.1)	47.4 (9.9)	47.4 (10.0)
Norm-based general health	45.2 (11.1)	46.2 (11.8)	40.7 (11.2)	40.5 (11.1)	46.4 (10.8)	47.2 (10.6)	42.3 (10.1)	42.3 (10.1)
Norm-based vitality	46.9 (11.5)	47.6 (11.6)	44.2 (11.9)	44.4 (11.7)	47.3 (12.0)	48.0 (11.7)	45.1 (10.3)	45.2 (10.3)
Norm-based social functioning	46.9 (11.6)	47.8 (11.7)	45.2 (12.2)	45.4 (12.1)	46.9 (12.5)	47.8 (12.1)	46.6 (10.6)	46.6 (10.6)
Norm-based role emotional	46.4 (13.5)	47.2 (12.9)	44.2 (14.4)	44.4 (14.2)	47.3 (13.3)	48.1 (12.7)	46.2 (12.0)	46.1 (12.0)
Norm-based mental health	47.2 (11.7)	48.5 (11.6)	46.4 (12.1)	46.5 (12.2)	46.9 (12.0)	47.4 (12.0)	46.5 (10.9)	46.6 (10.9)
EQ-5D, mean (SD)	0.75 (0.25)	0.78 (0.23)	0.71 (0.27)	0.71 (0.27)	0.79 (0.26)	0.80 (0.25)	0.74 (0.24)	0.74 (0.24)
EQ-5D VAS, mean (SD)	74.3 (18.0)	75.9 (17.8)	69.3 (20.1)	69.2 (20.0)	75.6 (16.7)	76.5 (16.1)	71.5 (18.1)	71.7 (17.8)

For preference participants the health status measure scores tended to favour the surgical group. However, the differences between the preference groups were less marked than the differences between the randomised groups, mainly because the preference medical group had better scores than the randomised medical group.

Graphical displays of the changes in REFLUX QoL scores and EQ-5D scores for all study groups are displayed in Figures 14 and 15 respectively.

FIGURE 14.

Reflux quality of life (QoL) scores at baseline and follow-up. Scores ranged from 0 to 100; the higher the score the better the patient felt.

FIGURE 15.

EQ-5D scores at baseline and follow-up. Scores ranged from 100 (perfect health) to 0 (equivalent to death).

Three participants died, one in the randomised medical group (road traffic accident) and two in the preference surgical group, neither of whom had surgery (alcoholic liver disease and cause unknown).

Statistical analyses

Discussion

The trial provides strong evidence of improvement in GORD symptoms following laparoscopic fundoplication as judged by the REFLUX quality of life score and its constituent domains. There were large differences between the randomised groups in these respects at 3 months post surgery, which were broadly sustained 9 months later. Also, scores in the preference surgical group were somewhat higher than those in the preference medical group despite starting from much lower baseline levels. The estimated sizes of differences varied depending on the assumptions being made. However, significant differences were observed even in the most conservative of the three main analyses – that based on intention to treat – where about one-third of those randomised to surgery did not actually receive it. Similar differences were also seen in most of the other measures of health status. There is, however, some evidence of a narrowing of the differences when the 3-month and 12-month follow-up results are compared. This was most marked for the EQ-5D, in which the surgical values had decreased and the medical values had increased somewhat (most easily seen in Figure 15).

We anticipated that this would be a difficult trial to deliver and so it proved. Trials comparing strikingly different interventions (such as surgery versus medical management) are often a challenge to recruit to. The explanation of such a trial needs to encompass a range of considerations and it is not unusual for some people, both clinicians and patients, to have strong views on the alternative procedures. As expected, many potential participants did have preferences for one approach or the other, and it was partly because we anticipated this that we included preference groups alongside the randomised core of the study. By enrolling surgeon/gastroenterologist pairs who were uncertain about the place of minimal access surgery in this context, we aimed to avoid clinician preferences. However, the differential recruitment to the preference groups in the clinical centres, in part reflecting which clinician actually first saw a potential participant, showed that there were differences in clinical perspective. This became a problem within the randomised comparison on the (relatively few) occasions in which a patient recruited by one clinician was deemed unsuitable for surgery by another clinician in the same centre.

To make the study more attractive to potential participants, those allocated medical management underwent a review of their medication to ‘optimise this’, rather than just carrying on with their existing regimen. This may be the reason why the types of PPI taken at follow-up differed from those at the time of trial entry (predominantly lansoprazole at entry, but omeprazole or lansoprazole equally at follow-up).

People suitable for the trial were not easy to identify. Most patients on long-term PPI treatment are managed in general practice, often through a repeat prescription system. We used a combination of three approaches: retrospective case note review to identify potentially eligible patients who had been seen in a participating hospital; prospectively, especially through endoscopy clinics; and (in selected centres) advertisements to the general public. All potentially eligible people had to be assessed clinically and they were only then formally approached about the trial. This was extra work over and above normal clinical duties, often through specially established monthly clinics. As described in Chapter 2, the numbers enrolled in individual centres tended to be small, reflecting all of these constraints. In the event, we found that those who agreed to join the randomised trial had characteristics mid-way between those of the two preference groups.

What we did not predict were the long waiting times for surgery in many centres. This was due to ambiguity about the responsibilities of participating hospitals in terms of the extra treatment costs of surgery. The intention had been that surgical slots would be pre-booked for the trial and that participants randomised to surgery would take the next available of these slots. In the event, emergencies such as cancer cases were given precedence, sometimes with repeated postponements of REFLUX trial patients. Anecdotally, long delays were an important factor in the decision of some of those participants allocated surgery ultimately not to have surgery. Delays became intractable in a few centres to the extent that special subvention funds were eventually found to allow the operations to be performed without any impact on normal clinical services. The availability of such funds to all centres from the start of the trial would almost certainly have overcome much of the waiting list problem.

In retrospect, given the long waits for surgery, it might have been better following enrolment to delay random allocation until there was a definite operation appointment. However, the likely impact would have been significant uncertainty amongst those enrolled about what they had agreed to, and greatly reduced numbers actually randomised (with some operations still postponed).

The standard rule in most trials is to time follow-up from randomisation. This was not appropriate in this trial because of the variable time between randomisation and surgery, exacerbated by the waiting list problem. The protocol specified follow-up at times equivalent to 3 and 12 months after surgery. It was important to have follow-up in the medical groups at equivalent times. We arranged this by pairing surgical and medical participants such that follow-up was linked and at (about) the same time after randomisation. The success of this manoeuvre can be assessed in Tables 23 and 26.

The large number of participants who did not get the management that they had been allocated to did have an impact on the results. For example, only 20% (10/50) of those allocated surgery who were taking reflux-related drugs at the 3-month follow-up had actually had surgery. As discussed earlier in this chapter, we have gone to some lengths to explore the likely impact of this non-adherence to the trial allocation. One way is through per protocol analyses limited to those randomised who received their allocated management. The second way is through an adjusted approach as a way of attempting to circumvent the likely selection bias of per protocol analyses. In this study the direction of effects was so clear, irrespective of the way that the analysis was performed, that the main issue became the size of effects. These did vary substantially (see, for example, Table 29) and this could be very important when policy decisions are being made, for example in the context of an economic evaluation. The approach that we took to address this in our economic analyses is described in the next two chapters.

One reason why we elected to have parallel non-randomised preference groups was to get more experience of the two forms of management. This particularly applied to surgery. Complications amongst the 319 participants who actually had fundoplication were rare (Table 22). Two operations were converted to open procedures, there were six visceral injuries and two pneumothoraces, and there were three admissions to a high dependency unit with no admissions to an intensive care unit. Patients stayed in hospital for a median of only 2 days. Three had reoperations and three had operations related to oesophageal stenosis.

As discussed in the following two chapters, an important measure of outcome is the proportion of patients continuing to take reflux-related drugs, especially after surgery. Rates did go up somewhat between the 3-month and 12-month assessment. Our rate of 11–14% at 12 months is higher than that in some other studies although estimates do vary both above and below this. Funding for this project was for follow-up to the time equivalent to 12 months after surgery. We have, however, instituted further annual follow-ups using similar questionnaires to those used at 3 and 12 months. Further follow-up will be important for assessing whether the benefits of surgery are sustained or whether differences in health status further narrow over time. We expect to report this after 5 years of follow-up are available for all participants.

The next two chapters on the economic evaluation reflect the position that we are in currently, having only 1 year of follow-up data, while recognising that long-term lifetime effects are likely to determine whether laparoscopic fundoplication is cost-effective. First, the within-trial cost-effectiveness analyses reported in Chapter 7 are developed within an economic framework. Then, an economic model is used to explore and extrapolate cost-effectiveness over a longer-term perspective.

Introduction

This chapter presents the within-trial cost-effectiveness analysis comparing laparoscopic fundoplication with medical management. Mean costs and health outcomes per patient are evaluated over 1 year using data from the REFLUX trial. The analysis is conducted from the perspective of the health and social care services. Costs are at 2006 prices and include the use of reflux-related health-care resources. Costs and outcomes are not discounted for this 1-year analysis.

Methods

Results

Cost-effectiveness

The estimated mean ICER is around £19,000 per QALY using intention to treat (Table 34). Bootstrap simulations were undertaken to estimate the uncertainty around the treatment decision. At a cost-effectiveness threshold ICER of £20,000 per QALY, surgery has a probability of 46% of being cost-effective, and at a threshold of £30,000 per QALY surgery is 86% likely to be cost-effective (Figure 16). This indicates that there is considerable uncertainty about whether surgery is cost-effective using the REFLUX trial data.

FIGURE 16.

Cost-effectiveness plane for laparoscopic surgery versus medical management using an intention to treat analysis. This figure shows the difference in mean cost and difference in mean quality-adjusted life-years (QALYs) per patient in 1000 bootstrap simulations of the data.

Discussion

Cost-effectiveness analysis is intended to inform two separate decisions. First, which strategy (medical management or laparoscopic fundoplication) is most cost-effective for the management of patients with reflux who are stable on medication from the perspective of the NHS and, second, what value there is in acquiring further information about these strategies.

This chapter presented the expected differences in costs and health outcomes between laparoscopic surgery and medical management over 1 year using the REFLUX trial data only. Surgery was on average more effective (in terms of QALYs gained over 1 year) but more costly. The ICER of £19,000 suggests that laparoscopic fundoplication might be cost-effective given that the threshold value in England and Wales is between £20,000 and £30,000. 102 However, there is still considerable uncertainty about this result (probability that surgery is cost-effective between 46% and 86%). The main limitation, however, is that the within-trial analysis ignores events, costs and health benefits that accrue after 1 year. The benefits of surgery are likely to be experienced by patients over the longer term,104 and the costs of medical management, even with widespread use of generics, are considerable when continued over a patient’s lifetime. Conversely, although no revisions of laparoscopic fundoplication were observed in the randomised surgical group over 1 year in this trial, the procedure may fail in the longer term. 105

More generally, new trials have to be placed in the context of existing evidence. Other randomised trials, and observational studies, have evaluated these strategies in the United Kingdom and elsewhere. A modelling framework can be used to extrapolate events, costs and health outcomes over a longer time horizon and to synthesise data from different sources to evaluate cost-effectiveness. 106 A modelling framework can also inform decisions about whether, and with what purpose, further research is needed. Value of information analysis can help to identify which variables contribute most to the overall uncertainty in the treatment decision and to quantify the benefits that would arise from having further information about these parameters. 107 To address these questions, a revised version of the decision model described in Chapter 3, updated with the evidence from the first year of the REFLUX trial, is described in the next chapter of this report.

Introduction

The REFLUX trial compared a strategy of laparoscopic surgery with one of continued medical management for patients with reasonable symptom control on anti-reflux medications. Data are now available from the clinical trial for at least 1 year after surgery for all patients. However, as discussed in the preceding chapter, as reflux is a chronic disease, an analysis that considers only events, costs and health benefits accruing over 1 year is too restricted. To accurately determine cost-effectiveness, and the value of conducting further research, a modelling approach is required that extrapolates costs and health benefits over an appropriate time horizon and allows the synthesis of evidence from different sources. This chapter describes a long-term decision-analytic model including evidence from the REFLUX trial and other sources.

The model presented here differs somewhat from the preliminary model described in Chapter 6. The preliminary model required evidence of the underlying disease process, that is, knowledge of whether treatment failure was temporary or permanent. Further treatment, such as dose adjustment, withdrawal of medication or revision of surgery, was then carried out conditional on the type of failure that occurred. However, in the reports from clinical trials, including the REFLUX trial, the follow-up points are infrequent and/or the underlying disease is rarely observed or described consistently. On the other hand, the treatments administered during follow-up are usually well recorded in these reports. Therefore, the model has been revised to define treatment failure in terms of change in treatment rather than return of symptoms.

Methods

Model structure

In the model patients follow a strategy of either immediate laparoscopic surgery or continuation of medical management (without an option of surgery following failure of medical management). In principle, immediate open surgery is feasible in this patient group but it is not considered here because it is widely considered to have been superseded by laparoscopic surgery, although conversion to open surgery is an option when laparoscopic surgery fails. Patients are assumed to be male and aged 45 when entering the model, which is the median age and commoner gender of patients in the REFLUX trial. Costs and health benefits are discounted at 3.5% per year and the price year is 2006.

The model structure is represented by Figure 17. As mentioned above, we define treatment failure following surgery as a change of treatment. Two options are considered for patients who fail surgery: patients may return to the use of anti-reflux medication or they may undergo a revision of surgery.

FIGURE 17.

Model structure diagram.

It is assumed that patients randomised to receive surgery who are found to be using medication at the end of a clinical trial are doing so to control reflux symptoms or symptoms related to surgery. It is further assumed that such patients will incur costs of medication indefinitely.

Although, in practice, patients who fail surgery may recommence anti-reflux drugs followed by revision of surgery, data on sequences of therapies were not available and so these events are treated as mutually exclusive competing risks in the model. Patients are assumed to have the same prognosis following revision as surgical patients who were not reoperated on.

To estimate the rates of return to medical management and revision of surgery, all of the available studies, whether randomised or not, are treated as observational data. The rates of failures could simply be estimated as the total number of events divided by the total patient years of exposure (Tables 38 and 39), which would estimate the rate of return to medical management as 4.8 per 100 patient-years. This estimate ignores any between-study heterogeneity, which might arise from patient selection, definition of outcomes, study design, surgical technique or other sources. To assess the assumption of homogeneity we also estimated the rate using a random effects Poisson regression using the statistical package WinBUGS^® (see Appendix 11 for code). 113 We also explored whether any observed factors (length of follow-up, study design) might explain some of the heterogeneity, but these variables were not found to be statistically significant and were omitted from the final model.

TABLE 38 - Surgical patients requiring medical management: results of laparoscopic surgery arms of randomised trials or observational studies

Study	Number of subjects	Years of follow-up	Exposure (person-years)	Number of failures	Rate of failure per person-year	Proportion of failures at end of study (%)
Mahon et al., 2005108	109	1	109	2	0.018	1.8
Booth et al., 200242	179	4	716	19	0.027	10.6
Bammer et al., 200157	171	6.3	1094	24	0.022	14.0
Contini et al., 200238	103	1	103	5	0.049	4.9
Pessaux et al., 200245	1470	3	4410	60	0.014	4.1
Papasaras et al., 2005109	289	2	578	150	0.260	51.9
Granderath et al., 2002110	27	4	108	2	0.019	7.4
Dassinger et al., 2004111	52	5	260	11	0.042	21.2
Bloomston et al., 2003112	100	1	100	19	0.190	19.0
Bloomston et al., 2003112	84	4	336	31	0.092	36.9
Vidal et al., 2006105	124	4.3	533.2	10	0.019	8.1
Madan and Minocha, 2006104	100	3	300	80	0.267	80.0
Laine et al., 199767	18	1	18	0	0.000	0.0
Reflux trial, 2006	104	1	104	14	0.135	13.5
All studies			8769	427	0.049

TABLE 39 - Surgical patients requiring re-operation during follow-up: results of single arms of randomised trials or observational studies

Study	Number of subjects	Exposure (years)	Number of failures	Rate of failure
Mahon et al., 2005108	50	50	2	0.040
Booth et al., 200242	179	716	11	0.015
Bammer et al., 200157	171	1094	5	0.005
Contini et al., 200238	103	103	0	0.000
Pessaux et al., 200245	1470	4410	35	0.008
Laine et al., 199767	18	18	0	0.000
Reflux trial, 2006	104	104	0	0.000
All studies		6495	53	0.008

A state of treatment failure for patients following medical management is not defined because there is no feasible alternative treatment, that is, in this model (unlike the preliminary model), surgery is not an option for patients following medical management. The estimates of mean HRQoL after successful surgery and after medical management, and the standard errors, are those observed in the randomised REFLUX trial at 1 year as there are no other randomised trials comparing surgery with medication that have used a preference-based utility instrument (see Chapter 6 for details of HRQoL data collected in the trial). HRQoL following treatment failure is estimated by the mean EQ-5D of all surgical patients (preference or randomised to surgery) who returned to medical management or required revision of surgery by 1 year. The base-case assumption in the model is that, although HRQoL decreases with age at the same rate as that in the age- and sex-matched general population, proportionate differences in HRQoL between health states are maintained over time.

To account for the decline in HRQoL with age, the HRQoL for each outcome observed at the end of the trial was compared with the average HRQoL of the general population aged from 45 to 55 years. 70 It was assumed that this proportionate decrement of HRQoL was constant as the cohort aged (Table 40). The age- and sex-stratified rate of death was taken from life tables,115 assuming that this patient group has a similar life expectancy to the UK general population after surgery. There is a small risk of operative mortality, estimated in a literature review as 4 deaths in 4000 procedures (see Chapter 3).

TABLE 40 - Mean (SE) HRQoL parameters used in the model

GORD, gastro-oesphageal reflux disease; HRQoL, health-related quality of life; SE, standard error.

Parameter	Mean	SE	Source
HRQoL following medical management	0.711	0.018	Reflux trial EQ-5D in randomised medical arm at 1 year
Additional HRQoL following successful laparoscopic surgery compared with medical management	0.071	0.028	Reflux trial EQ-5D in randomised surgery arm at 1 year (off drugs)
HRQoL following unsuccessful laparoscopic surgery (on medication)	0.686	0.048	Reflux trial EQ-5D in all patients who failed surgery at 1 year
Average HRQoL during year if undergoing re-intervention	0.686	0.048	As for unsuccessful surgery
HRQoL for general population aged 45–55: men; women	0.84; 0.85		Kind et al., 199971
HRQoL for general population aged 55–65: men; women	0.78; 0.81		Kind et al.,199971
HRQoL for general population aged 65–75: men; women	0.78; 0.78		Kind et al., 199971
HRQoL for general population aged 75+: men; women	0.75; 0.71		Kind et al., 199971
Prevalence of GORD in population aged 18–60	0.0045		McDougall et al., 1996;30 Trimble et al., 1995114

During the first year of follow-up, 35% of patients require an outpatient visit and 35% a day-case or hospital admission following surgery compared with 15% who require an outpatient visit and 14% a day-case or hospital admission following medical management (see Table 32; Chapter 7). The Nordic GORD study62 found that only a small proportion of patients required endoscopy after 12 months in either group, and here it is assumed that no further hospital admissions or outpatient visits are needed after 1 year other than revisions of surgery.

The model time horizon is a patient’s lifetime. However, there are significant sources of uncertainty surrounding several model parameters given that the main source of data in the model, the REFLUX trial, has reported only 1 year of follow-up. To provide a point of reference, the model analysis starts from a set of assumptions that are similar to those used in the within-trial cost-effectiveness analysis presented in Chapter 7, which assumed that there were no differences in cost or health benefits beyond 1 year. This is unlikely to be the case and so a series of scenarios that relax these assumptions is explored, described in Box 1.

BOX 1 - Assumptions used in the within-trial analysis and alternative scenarios explored using a series of sensitivity analyses (see Chapter 3 for data sources for assumptions)

Scenario number	Assumption
	Duration of cost of medication	Duration of relative health benefit of surgery	Annual rate of conversion from surgery to medical	Annual rate of reoperation
11 (within-trial analysis)	1 year	1 year	13% convert year 1, 0.0% thereafter	0.0%
15 (temporary QoL advantage)	Lifetime	5 years	13% per year up to year 2, 4.9% thereafter	0.0% year 1, 0.8% thereafter
16 (low rate of surgical failure)	Lifetime	Lifetime	4.9%	0.8%
17 (base-case)	Lifetime	Lifetime	13% per year up to year 2, 4.9% thereafter	0.0% year 1, 0.8% thereafter
18 (very high rate of surgical failure)	Lifetime	Lifetime	13% per year	0.8%
19 (high rate of surgical failure)	Lifetime	Lifetime	8% per year	0.8%

One way of proceeding with this analysis might be to vary the time horizon over which the model is run, from 1 year up to a lifetime, in a series of scenarios; however, this would be naïve. Reflux is a chronic disease and, therefore, the only reasonable analysis is over a lifetime. The role of scenario analysis is to explore different assumptions about HRQoL, costs and clinical events over this time horizon. The sources for the alternative assumptions are presented in Chapter 3.

Results

Value of information analysis

The value of conducting additional research that, in principle, would reduce parameter uncertainty can be estimated using value of information analysis. The expected value of perfect information (EVPI) is the amount that a decision-maker should be willing to pay to eliminate all uncertainty that arises because of imprecision in the parameters of the model. The partial EVPI represents the amount a decision-maker should be willing to pay to eliminate all uncertainty in individual parameters or a subset of parameters, given the uncertainties elsewhere in the model.

To illustrate this we estimated the EVPI and partial EVPI for five sets of parameters: (1) the HRQoL of patients who fail surgery; (2) the estimates of HRQoL for all other model states; (3) the estimates of the annual rates of failure of surgery; (4) the estimates of unit costs used in the model; and (5) the rate of return to medical management post surgery together with the HRQoL of patients who fail. The analysis requires an estimate of the percentage of the population who would be eligible for surgery if it were cost-effective. A Spanish population survey (both sexes, ages 40–79 years) found that 287 out of 2500 (11%) interviewees used anti-reflux drugs, and 119 (4.8%) were stable (not having had reflux symptoms in the past year), although 43 (1.7%) acknowledged taking anti-reflux drugs to prevent symptoms. 118 This might be considered a conservative estimate of patients who could be considered for surgery. If we assume that about one-half of these might be excluded because of age, preference or co-morbidity, then prevalence is estimated at 1% of this population, equivalent to about 160,000 people in the UK. 115 Figure 18 shows an estimate of population EVPI at a range of values of the threshold ICER. EVPI in this case is increasing with the threshold ICER because at higher values of the ICER we are more willing to pay for the health benefits associated with surgery (and therefore more certain that surgery is the correct decision), but the consequences of a wrong decision are also greater (in terms of loss of health and wasted resources) and we are willing to pay more to avoid the possibility of these losses. Because the population is large, the model indicates that the EVPI is £300 million at a threshold ICER of £30,000, indicating that we would be willing to pay up to this to eliminate all of the uncertainty in the decision.

FIGURE 18.

Expected value of perfect information (EVPI) assuming a population of 160,000 patients in England and Wales, and partial EVPI for three sets of parameters: (1) all HRQoL parameters; (2) HRQoL after surgery failure; and (3) failure rates for surgery.

Figure 18 also shows the partial EVPI for selected sets of parameters. Partial EVPI is greatest for the rate of return to medical management post surgery together with the HRQoL of patients who fail. This indicates that almost all of the variation affecting the treatment decision is due to the interaction of these two parameters. Relatively little information is available on the HRQoL of patients who fail surgery; this could be captured in a longer follow-up but does not necessarily require a randomised trial. Figure 18 shows relatively little value of information in other parameters. However, this analysis does not on its own capture all of the uncertainty in the decision for two reasons. First, we have used mean estimates of HRQoL collected in a short-term trial to extrapolate over the longer term, without adjusting standard errors to take account of this additional uncertainty. There is, therefore, additional uncertainty over long-term differences in HRQoL, which is not captured in this value of information analysis. Second, we have assumed that the pooled rates of failure from observational studies of between 1 and 6 years are generalisable to our population, and that these rates will continue over the long term. We have attempted to represent this uncertainty as a series of scenario analyses. Taken together, this implies value in a continuing long-term follow-up to the randomised trial. The question remains, however, over how long this follow-up should optimally be.

Discussion

This chapter has presented a decision-analytic model comparing laparoscopic surgery with medical management, using data from the REFLUX trial and other sources to estimate cost-effectiveness over a lifetime. The results of this model are similar to those of the preliminary model presented in Chapter 6, which indicated that surgery was cost-effective but with a high degree of uncertainty. Other authors have examined the cost-effectiveness of laparoscopic surgery versus medical management. Cookson et al. 29 found that laparoscopic surgery broke even compared with medical management after 8 years and was cost saving thereafter. Romagnuolo et al. 28 evaluated cost-effectiveness over 5 years in a Canadian setting, in which both surgery and medical therapy is on average more expensive (generic formulations were not used in that model) than that found in the REFLUX trial and by Cookson in a UK setting. They concluded that there was little difference in HRQoL between the treatments and that surgery broke even relative to medical management after 3 years. Arguedas et al. 119 evaluated the strategies in a US setting with costs similar to Canada, assuming a relatively higher rate of symptom recurrence and failure of surgery, and relatively lower differences in HRQoL between the treatments, and concluded that medical therapy dominated surgery using a 10-year time horizon.

Although surgery seems likely to be cost-effective in terms of expected (mean) costs and health effects, there remains considerable uncertainty about this conclusion. Balances between risks and benefits and between costs and health gain will depend on patient characteristics such as age, the presence of serious co-morbidity and the severity of GORD symptoms. Furthermore, there are a number of practical issues to consider before the NHS could consider offering surgery to a wider range of patients who are currently stable on medical management. In particular, surgical capacity and availability of trained surgeons are potential barriers to implementation and should be addressed.

We have estimated the value of reducing some of the model uncertainty in the analysis of EVPI and partial EVPI, and through a series of scenario analyses. These have indicated that continued follow-up of the randomised trial would be valuable, particularly to obtain more information on HRQoL following surgery failure and the long-term difference in HRQoL between strategies. Further research to obtain more information on the long-term HRQoL and prognosis of patients would be valuable.

Introduction

This chapter reports an application of a discrete choice experiment (DCE) to measure patient preferences for treatment options for GORD. DCEs are increasingly recognised as an important method in health services research for measuring the strength of patients’ preferences (utility) for treatments and methods of delivery of care. 120

The aims of this work were to identify the strength of the trial participants’ preferences for the different treatments and outcomes of GORD; to investigate whether these preferences differ between the different arms of the trial; and finally to identify whether the mean benefits associated with each treatment vary. It should be noted that the utilities produced by the DCE reflect the preferences of people with GORD for the treatment and outcomes of GORD. As such, they are different from the utilities used to generate QALYs, which were based on the responses to the EQ-5D questionnaire and reflect the public preferences for the outcomes following treatment of GORD. It is these latter utilities that are arguably most useful for priority setting within the NHS. 102

In the following section a brief description of the DCE approach is provided. This is followed by the methods used to achieve the aims stated above and the subsequent results. Finally, a brief discussion is presented outlining the strengths and limitations of the approach and the implications of the findings.

Methods

The study was performed in two stages: a methodological stage to develop the questionnaire and an applied stage to derive the utility estimates for the different treatments and outcomes for GORD.

Eliciting preferences

Once the scenarios to be presented to patients were identified, preferences for these scenarios were obtained by using a forced choice approach. An example of the choices presented to participants is shown in Figure 19; respondents were asked which option they would choose, ‘A’ or ‘B’.

FIGURE 19.

Example of discrete choice experiment question presented to trial participants.

One important issue in preference elicitation is whose preferences should be elicited. Patients with the experience of both disease and treatment were considered appropriate for this study and therefore the completed questionnaire was sent to all active participants in the REFLUX trial during August 2006.

Results

Of the 705 questionnaires sent out, 441 (63%) were returned; 17(3%) were returned uncompleted and 424 (60%) were fully or partially completed. Of these 424 questionnaires, 87 (21%) were from the randomised surgical group, 103 (24%) were from the randomised medical group, and 109 (26%) and 125 (29%) were from the preference medical and preference surgical groups respectively.

Discussion

The aim of this chapter was to use a DCE to explore the strength of preference for the treatment and outcomes of GORD. This approach has been used to measure preferences of GORD patients previously128,129 but this earlier work sought to establish willingness to pay for complete symptom relief of GORD and for diagnostic uncertainty. The DCE reported in this chapter was different in that it attempted to explore preferences for the outcomes of treatment (e.g. troublesome symptoms and serious complications) and preferences for the process by which these outcomes were obtained.

The results of the DCE indicate that the most important single dimension is serious complications, followed by a 10% change in the chance of having surgery or receiving lifelong medication. Suffering troublesome symptoms most days was less important, although the unit of analysis was a 1% chance of this event occurring. There was no evidence that respondents placed any importance on suffering troublesome symptoms once a week in comparison with no symptoms.

The group that was associated with the highest utility relative to a best case situation was the preference surgical group, and the group that was associated with the lowest utility was the randomised surgical group. If the effect of serious complications is removed from the consideration of utility, then the preference groups are associated with higher levels of utility relative to the best case than the randomised groups. Furthermore, the surgical group is associated with higher utility than the medical group.

The exclusion of serious complications from the consideration of utility might be considered contentious. However, an analysis was conducted to explore whether the preferences for the continuous variables (risk of serious complications, risk of surgery and risk of receiving lifelong medication) in the econometric analysis were linear. The results of this analysis indicated that, although utility fell as risk increased, it fell at a decreasing rate for both the risk of serious complications and surgery (there was no evidence of this effect for lifelong medication). The implication of this is that it is possible that there is little or no difference in the loss of utility caused by serious complications between groups. Research is required to further investigate how this non-linearity in preferences might be most appropriately modelled, as the quadratic function would result in implausible utility estimates for higher risks of serious complications than were considered in the DCE questionnaire.

The results of the analysis presented in this chapter also provide some insight into the importance of people’s preferences for treatment with respect to utility. For example, people who have a preference for medicine but who actually undergo surgery experience almost twice the loss of utility (1.059 or –0.557 + –0.502) as those people in the randomised arm who receive surgery (–0.557) for a 1% increase in the risk of surgery. Similarly, people who preferred surgery and received surgery lost less utility (–0.008 or –0.557 + 0.549). This result indicates the importance of patient choice when decisions are made about which type of treatment to provide.

Some of the limitations of the analysis reported in this chapter have already been described but one further limitation relates to how the information derived by the DCE could have been used in the economic model reported earlier. It was not possible, nor was it planned, for these two ‘economic’ elements to be integrated. Indeed, methods to integrate DCEs into a trial remain relatively undeveloped. However, future work should consider how a DCE and an economic model conducted as part of a trial analysis can be developed in an integrated fashion. It is likely that this will involve the attributes and levels of the DCE being reflected in the model structure, with the values of attribute levels being produced by the model and fed into the estimation of utilities as part of the DCE analysis. Any attempt to integrate these approaches would be facilitated by the use of a common continuous measure, such as willingness to pay, so that all dimensions could be valued in terms of this numeraire.

The methods used to analyse and present the results of the DCE have limitations. One of the main limitations is the limited handling of uncertainty in the analysis. In economic studies it is expected that an extensive sensitivity analysis would be conducted to assess how robust the conclusions are. Increasingly, as exemplified by the economic evaluation presented in Chapter 7, it is becoming expected that a probabilistic sensitivity analysis will be used to develop credible intervals around mean estimates. Although sensitivity analysis has been performed as part of the work reported in this chapter, probabilistic sensitivity analysis has not been conducted. Probabilistic sensitivity analysis would also help overcome a further limitation of the DCE. When analysing the DCE, we followed the common econometric convention of combining levels of dimensions when there was no evidence of a statistically significant difference and of dropping parameters from an analysis when the coefficients were not statistically significant. There is some debate about how appropriate this approach is as it reduces the information available to decision-makers. However, with probabilistic sensitivity analysis a full model, including both statistically significant and insignificant coefficients, can be used to develop both mean utility scores and credible intervals. Therefore, further work might focus on conducting a probabilistic sensitivity analysis.

DCEs use hypothetical questions and, as such, they have been criticised. This is because it is unclear whether people would pick these scenarios if they were faced with these choices in real life. Nevertheless, the respondents to the DCE all had experience of GORD and its treatment (either medical, surgical or both); hence, it was hoped that the respondents would be able to consider the choices and trade-offs involved in each choice question.

A final concern relates to the number of choice questions to present to potential respondents. The greater the number of dimensions and levels that are considered relevant the greater the number of possible scenarios that individuals could potentially be presented with. Experimental design techniques were used to reduce the number of scenarios that were presented to individuals while still allowing for utilities to be inferred for all possible scenarios. However, even after the use of these techniques it was felt that the number of questions to be presented (n = 16) was too great. As a consequence, the questions were randomly split into two questionnaires, each containing eight questions. It was hoped that this would increase the completion rate of the questionnaire, although it did have the effect of reducing our ability to detect important differences in preferences. Overall, the completion rate achieved was quite high for a DCE questionnaire (which are thought to be cognitively demanding on respondents) and this may be attributed to the relatively short length of the questionnaire.

Conclusions

The results of the DCE presented in this chapter complement the evidence reported in earlier chapters. The results also aid the interpretation of the clinical evidence by indicating the importance placed on type of treatment and the ability of a treatment to resolve symptoms. The most important single dimension is serious complications, followed by changes in surgery, lifelong medication and troublesome symptoms most days. There was no statistically significant evidence that respondents placed any importance on suffering troublesome symptoms once a week in comparison with no symptoms. Relative to a best case situation the trial arm associated with the highest mean utility was the preference surgical group and that associated with the lowest mean utility was the randomised surgical group. The utility associated with surgery is dependent upon the risk of serious complications, which was assumed to be greater than that for lifelong medical treatment. If the effect of serious complications is removed from the consideration of utility, the preference arms are associated with higher levels of utility than the randomised groups. Furthermore, the surgical arms are associated with higher utility than the medical arms. Thus, the results of the analysis indicate the importance of quantifying the risk of serious complications and of considering patient choice when decisions are made about which types of treatment to provide and the type of treatment to recommend.

Additional further research is also indicated. Part of this research should focus on how approaches such as DCEs can be made more useful to trials-based research. A more specific research need is to consider how best to describe the imprecision surrounding the mean estimates of utilities that are generated.

Implications for health care and recommendations for research

The advent of less invasive fundoplication performed laparoscopically opened up new possibilities for the management of people with chronic symptoms of GORD. Good results obtained amongst people whose symptoms were not satisfactorily controlled by medical management raised questions about the place of relatively early surgery in people with GORD whose symptom control from long-term medical management was reasonably acceptable. Would surgery be more effective than continuing medical management? Would surgery be sufficiently safe? And would widening the use of laparoscopic fundoplication to such patients be cost-effective? These are the principal questions addressed in this study.

The study had two main components: a pragmatic randomised controlled trial to assess clinical effectiveness and an economic evaluation to explore cost-effectiveness and the wider implications for efficient health-care provision.

The trial provided clear evidence of effectiveness in respect of reflux-related quality of life. Even though the number of participants in the trial was not as large as originally intended, the sizes of differences observed in the condition-specific reflux quality of life measure were so large that they were highly statistically significant. As with other disease-specific measures, the magnitude of these differences is hard to conceptualise. However, broadly similar differences were also observed in most components of the more accessible generic health status measures, SF-36 and EQ-5D.

As described in Chapter 6, clear differences were observed even though as many as one-third of those allocated surgery did not have fundoplication. Extra analyses explored how much of a blunting effect this might have had on the results and, arguably, these adjusted analyses provide better estimates of the true effects of surgery in this type of population as it might be used in normal clinical practice.

Current follow-up is to the equivalent of 12 months after surgery. In comparison with the results obtained at 3 months there were sustained better scores but with some evidence of attenuation of the differences. For example, the number taking reflux-related drugs after surgery went up from around 9% to 14%. Narrowing of differences was most marked for the EQ-5D health status measure. It is possible that some of the ‘improvement’ is due to a placebo effect of surgery and one explanation of any attenuation of difference is that the placebo effect has diminished over time. This could be clarified by further follow-up to find out if differences are sustained or whether there is more narrowing of the differences.

In addition to the randomised groups the trial also had two preference groups, which aid interpretation of the randomised trial results. As a group, the preference surgical participants had the lowest baseline REFLUX scores (worst symptoms) and the preference medical group the highest (with the randomised groups between them (see Figure 14). After surgery the preference surgical group had scores that rose to the level of the preference medical group and by 12 months they were the better of the two groups. The preference groups give an indication of likely behaviour if surgery were to become more freely available. In addition to having the least well-controlled symptoms at baseline, the preference surgical group had been on medication longer and were less concerned about possible adverse effects of surgery (described in Chapter 5).

The preference groups also add extra information about clinical events, in particular rare serious adverse events. Taken at face value, laparoscopic fundoplication appears to be a relatively safe procedure; however, even the experience of all of the 329 participants who had surgery is too little to provide sufficiently precise estimates of uncommon events. So, questions still remain about the extent of possible adverse effects of surgery and their frequency.

The within-trial (i.e. up to 12 months of follow-up) cost-effectiveness analysis related the extra mean costs associated with the surgical policy with the increase in mean QALYs that followed surgery to generate an incremental cost-effectiveness ratio. This was around £19,000 when the intention to treat approach to analysis was used. Taking into account uncertainties around the various estimates, it was calculated that the chances that the surgical policy would be cost-effective at a threshold of £20,000 per QALY was 46%. When a per protocol approach was used, the incremental cost per QALY increased to around £23,000, with a probability that this would be cost-effective at a threshold of £20,000 of only 19%. These results indicate considerable uncertainty at thresholds that are currently commonly applied to costs per QALY.

The within-trial analyses have significant limitations, however, as discussed in Chapters 7 and 8. The most important is that they ignore events, costs and benefits that accrue after 1 year. It is likely that surgery will continue to bestow benefits after 1 year, although there could also be relapse of symptoms, and medical management may require lifelong medication with significant costs. For this reason, the REFLUX trial data were synthesised with other data to develop an extended cost-effectiveness model. This explored a number of possible scenarios. Assuming that the benefits of surgery persist throughout a lifetime, that without surgery mediation use would continue for a lifetime, that there would be a 4.8% annual rate of additional uptake of medication in the surgery group, and that there would be an annual 0.8% reoperation rate led to an estimated incremental cost per QALY of around only £2000, with a 74% probability of surgery being considered cost-effective at a threshold of £20,000 per QALY. Applying other plausible assumptions, however, gave a range of incremental costs per QALY of between £1000 and £44,000, again indicative of wide uncertainty. The factors most contributing to the uncertainty were the projected HRQoL parameters and the long-term rate of uptake of medication following surgery.

The DCE was performed to provide an alternative way of assessing the weights that people with GORD place on their outcome and treatment. The results were broadly in line with the other economic evaluation in this project, based on the EQ-5D. The DCE did show, however, that respondents put considerable weight on avoiding rare but serious risks. The economic analysis found that these risks have little impact on QALYs on average and that the uncertainty in the clinical results about their incidence does not affect the treatment decision at the population level, all other things being equal. Nevertheless, the DCE highlights that these risks may be important when patients choose whether to accept surgery if it is offered.

Currently available evidence from the REFLUX trial indicates that surgery could be cost-effective at the thresholds (£20,000–£30,000) currently applied by the National Institute for Health and Clinical Excellence (NICE) but with considerable uncertainty. The extended model suggests that the true cost-effectiveness, when lifetime costs and benefits are taken into account, is likely to be more favourable. But this, too, is prone to major uncertainty.

Questions also remain about the generalisability of the study’s results. The economic model was based on a 45-year-old man, whereas many people receiving PPIs for GORD are older than this and can have significant co-morbidities.

The most urgent need for further research, therefore, is to acquire improved estimates of longer-term benefits and costs. This could be accomplished relatively easily by continuing annual follow-up in the REFLUX trial, and indeed arrangements for this have been put in place. Funds have recently been awarded by the HTA Programme to support follow-up to 5 years after surgery. Our analyses of cost-effectiveness will then be updated to take these results and other changes (such as in the costs of PPIs) into account. In the meantime it may be worth exploring whether there are other longer-standing non-randomised cohorts that could be useful in this respect. Perceptions of the risks of rare adverse events may play a major role in decision-making about surgery. Such cohorts could also be useful for getting more precise estimates of uncommon events associated with both surgical and medical management.

Contributions of authors

Adrian Grant (Director, HSR trialist) was the principal grant applicant and contributed to the development of the trial protocol and the preparation of the report and was responsible overall for the conduct of the trial.

Samantha Wileman (Trial Co-ordinator, HSR trialist) contributed to the development of the trial design, was responsible for the day-to-day management of the trial, monitored data collection and assisted in the preparation of the report.

Craig Ramsay (Senior Research Fellow, Health Statistics) contributed to the grant application and the trial design and conducted the statistical analysis.

Mark Sculpher (Senior Health Economist, Health Economics) was responsible for the economic evaluation section of the grant application and protocol, and Laura Bojke (Research Fellow, Health Economics) and David Epstein (Research Fellow, Health Economics) conducted the analysis of the economic models for the report.

Sue Macran (Research Fellow, Health Outcomes) led the development of the REFLUX outcome measure.

Luke Vale (Senior Research Fellow, Health Economics) and Mary Kilonzo (Research Fellow, Health Economics) conducted the discrete choice experiment (DCE) and assisted in the preparation of the DCE for the report.

Jill Francis (Senior Research Fellow, Health Psychology) conducted the analysis of the belief questionnaires and wrote the report for this part of the study.

Zygmunt Krukowski (Surgeon, Gastroenterology) and Ashley Mowat, Robert C Heading and Mark Thursz (Physicians, Gastroenterology) advised on clinical aspects of the trial design and the conduct of the trial and commented on the draft report.

Ian Russell (Director, HSR, Health Outcomes) contributed to the development of the trial design.

Marion Campbell (Programme Director, HSR statistician/trialist) contributed to the development of the trial design, commented on all aspects of the conduct of the trial and contributed to the preparation of the report.

Other members of the Trial Steering Group were as follows (those marked with an asterisk were independent of the trial): Wendy Atkin* (Chair), John Bancewicz, Garry Barton (1999–2002), Ara Darzi, Janusz Jankowski*, Richard Lilford*, Iain Martin (1997–2000).

Members of the independent Data Monitoring Committee were as follows: Jon Nicholl, Chris Hawkey, Iain MacIntyre.

The authors wish to thank the following researchers for their assistance in nurse co-ordination and patient recruitment and follow-up: Maureen GC Gillan, Marie Cameron, Christiane Pflanz-Sinclair; Sharon McCann who was involved in the piloting of the practical arrangements of this trial; Allan Walker for database and programming support; and Janice Cruden and Pauline Garden for their magnificent secretarial support and data management.

Members of the REFLUX trial group responsible for recruitment in the clinical centres are shown in the table below.

Finally, the authors are indebted to the referees for reading the report and for the quality of their comments.

The Health Services Research Unit is funded by the Chief Scientist Office of the Scottish Government Health Directorates.

Disclaimers

The views expressed in this publication are those of the authors and not necessarily those of the HTA Programme or the Department of Health.