In the report of the first phase of the REFLUX trial¹ we concluded that, among the sorts of patients recruited to the trial, laparoscopic fundoplication ‘significantly increases general and reflux related QoL measures, at least up to 12 months after surgery’. There was, however, considerable uncertainty about cost-effectiveness, largely because the follow-up period was so short. Varying plausible assumptions about the longer-term effects of surgery, particularly in terms of QALYs gained and costs of medication, led to markedly differing results. This was the basis for this second phase of the trial, in which follow-up has been extended out to a time equivalent to 5 years after surgery.

The trial has a pragmatic design and compared two policies for managing GORD, rather than directly comparing surgery with PPI therapy. This is the basis for the primary analyses being based on the ITT principle as this directly compares the policies. The first policy can be characterised as relatively early surgery for most eligible patients but with the option to take medication if considered helpful, irrespective of whether or not surgery had been performed. The second policy can be described as medical management as appropriate with ‘delayed’ surgery in selected cases. Hence, we have not made an assumption that those taking medication after surgery are ‘failures’. In our view, although surgery may have improved symptoms, the addition of PPIs may give further improvement and hence should be considered to be a component of both policies.

In contrast to the other large randomised trial (the LOTUS trial,⁴⁸ discussed in Chapter 4), whose primary outcome was ‘treatment failure’, we chose patient-reported outcome measures as our primary and main secondary outcome measures. The advantage is that they provide a ‘common currency’ across the two trial policies and do not depend on clinical judgements (as ‘treatment failures’ do). There is, however, a concern that completion of the patient-reported outcome questionnaires may be influenced by the nature of the management received. We had a reminder of this in the early stages of our trial. The DMC noticed an imbalance in baseline scores of the first few patients randomised, but not in other descriptive characteristics. It seemed that this might have been due to completion of the form after the allocation was known (although it could still have been due to chance); once it was made a requirement that the form had to be filled in before the allocation was known, however, this discrepancy disappeared. We believe that a strength of the long-tem follow-up as reported here is that, as the time from the differentiating event (surgery or no surgery) gets increasingly long, the possibility of such reporting bias becomes remote. Protection was also provided by the partially randomised patient preference design: the randomised component was limited to patients who were uncertain which treatment to choose while those who had strong views were enrolled into the preference groups.

We designed the trial with the aim of making the management policies as similar as possible to normal NHS care. So, for example, a large number of centres were involved (both teaching and non-teaching hospitals); recruitment was based on gastroenterologist–upper gastrointestinal surgeon partnerships; surgeons chose the type of fundoplication and other aspects of the procedure; after optimisation of medical management in secondary care, all subsequent medical care was in general practice; there was no requirement for extra tests or hospital visits; and simple entry criteria identified people with chronic troublesome GORD symptoms that required anti-reflux medication for reasonable control suitable for either policy (average age 46 years). The results should, therefore, be easily generalisable to standard NHS care.

The one area in which we think the trial did not ‘mimic’ usual care is in the relatively low proportion of those allocated surgery who actually had surgery (62%; see Table 10). There are reasons for thinking that the unusual circumstances of a randomised trial comparing medical management with surgery were partly responsible for the large proportion who did not have surgery. We think the rate (84%) in the preference group is likely to be more indicative of ‘normal’ acceptance rates. For this reason we undertook secondary adjusted treatment received analyses aimed to compensate for this. These analyses are likely to give a better estimate of differential effects in usual care, but because they depart from the randomised groups and hence may be prone to bias they should be treated with appropriate caution. We also explored this issue through post hoc analyses stratified by whether or not those allocated surgery actually had fundoplication. This showed (see Figure 18) that those who had surgery had lower baseline REFLUX scores (worse symptoms) than those who did not have an operation, but that, following surgery, their scores were consistently higher that those who did not have surgery.

Despite our best attempts to retain the cohort of participants there has been some attrition over the course of the follow-up period. The response rate of 69% at 5 years can be considered satisfactory in a study of this type and is similar to the rate in the LOTUS trial (67%).⁴⁸ The rate in the REFLUX trial reflects the decision among some participants to withdraw, but with high levels of return among those remaining. Responders did differ from non-responders but we used analysis techniques to make the most of the available data (repeated measures and imputation), and the responders in the two randomised groups were generally reassuringly similar in respect of baseline characteristics.

The new results provide clear evidence of a sustained greater improvement in GORD-related QoL in the group randomised to surgery. The results also suggest sustained benefit in respect of generic health-related measures of QoL, although the differences attenuate over time and are not statistically significant at 5 years. In these respects the REFLUX trial is in line with the results of the other three randomised trials that have compared laparoscopic surgery with medical management. The worse the symptoms at entry (the lower the score at baseline), the greater are the benefits of surgery.

By 5 years, 24 (13%) of the participants randomly allocated to medical management had undergone anti-reflux surgery. Exploratory analyses (see Figure 19) showed that, as a group, these 24 had low REFLUX questionnaire scores (worse symptoms) at trial entry, which subsequently improved markedly after surgery. Hence, this group is at least a contributory factor to the narrowing of differences between the randomised groups over time (see, in particular, Figures 3 and 17) and a reason for thinking that the ITT-based analyses comparing the two management policies are likely to underestimate the effects of surgery.

The follow-up has clarified the rates of longer-term use of PPI medication in both policies. In the randomised medical group, 87% were taking medication at 1 year, falling gradually to 82% at 5 years (see Figure 2). The equivalent figures in the randomised surgery group were 36% at 1 year (15% among those who had surgery) and 41% (26%) at 5 years. This was in response to a question that, to avoid problems with recall, asked just about the preceding 2 weeks (rather than the full year), and we have assumed that the 2 weeks are typical of the previous year. We know, however, that medication use is sometimes dynamic – that patients stop and start. This is apparent in Table 13, for example: among those in the medical group who were not taking medication at the end of the first year, 13 (68%) of the 19 respondents reported that they were taking PPIs at 5 years.

Short-term complications of surgery were described in more detail in the first report of this trial. However, the REFLUX trial is consistent with the other three trials in this respect, with small numbers having associated visceral injuries, postoperative problems and dilatation of the wrap. The longer-term follow-up has now clarified the likelihood of further surgery following a fundoplication. Overall, 4% (n = 16) of the total 364 in the study who had fundoplication had a subsequent reflux-related operation, of whom two had a further (i.e. third) operation. Reoperation was most often conversion to a different type of wrap or a reconstruction of the same wrap. There were only two cases of reversal of the fundoplication and neither was in the randomised comparison. In total, 3% (n = 12) of those who had fundoplication required surgical treatment for a complication directly related to the original surgery, including oesophageal dilatation (n = 4) and repair of incisional hernia (n = 3). As described in Chapter 4, although it is not possible to extract exactly comparable data, these results are broadly in line with those of the other trials.

Where the REFLUX trial results do differ from the results of the other trials, especially the LOTUS trial, is in the likelihood and extent of adverse symptoms associated with fundoplication. Dysphagia, flatulence and bloating, and inability to vomit despite wanting to have all been reported to be problematical after fundoplication. However, in the REFLUX trial, the patterns of difficulty swallowing, flatulence and wanting to vomit but being physically unable to do so were similar in the two randomised groups (see Table 15), with no statistically significant differences.

The economic analysis of the 5-year data from the REFLUX trial had two phases. First, a within-trial 5-year cost-effectiveness analysis was undertaken; this was followed by an exploration of the need to develop a longer-term model. Differences in mean costs and mean QALYs at 5 years were used to derive an estimate of relative cost-effectiveness. The base-case approach used multiple imputation (principally because of the extent of missing data), an ITT analysis and adjustment for baseline QALYs. As described in Chapter 5, complete-case and PP analyses were also undertaken, as were a range of structural, scenario and probabilistic sensitivity analyses. Costs were estimated from a health-care perspective and consideration was limited to randomised trial participants. Costs for each participant were calculated by multiplying their use of health-care resources by associated unit costs and were discounted at an annual rate of 3.5%. HRQoL was calculated from serial EQ-5D measurements. The mean (SD) costs in the first year were £2501 (£1698) in the surgical group compared with £560 (£1007) in the medical group; in each subsequent year the mean costs were around £175 higher in the medical group. The estimated incremental mean cost of the surgical policy was £1518 (95% CI £1006 to £2029) with incremental mean QALYs of 0.2160 (95% CI 0.0205 to 0.4115), giving an ICER of £7028. The probability of the surgical policy being the more cost-effective was 0.93 at a threshold of £20,000 per QALY and 0.96 at a threshold of £30,000 per QALY. The complete-case analysis gave similar results and the conclusions were robust to plausible changes in assumptions, the only exception being when surgery-allocated patients with missing data were assumed to experience lower HRQoL than other patients. A regression-based mixed-model approach was then used to explore the robustness of the findings and to gauge the likelihood that the current strong evidence for cost-effectiveness might be reversed over subsequent years. The regression-based model gave very similar results to the base-case imputation approach. Given the trends in both costs and benefits, it was concluded that it was highly unlikely that the cost-effectiveness of surgery would be reversed when extrapolated beyond 5 years.

Thus, this second phase of the REFLUX trial has accomplished what it set out to do. After 5 years' follow-up, a policy of relatively early laparoscopic fundoplication among patients for whom reasonable control of GORD symptoms requires long-term medication and for whom both surgery and medical management are suitable continues to provide better relief of GORD symptoms with associated better QoL. Although surgery carries risks, complications were rare. And despite being initially more costly, a surgical policy was found to be highly likely to be cost-effective for such patients at conventional threshold costs per QALY.