To assess what is known about the effectiveness, safety, affordability, cost-effectiveness and organisational impact of endoscopic surveillance in preventing morbidity and mortality from adenocarcinoma in patients with Barrett's oesophagus. In addition, to identify important areas of uncertainty in current knowledge for these programmes and to identify areas for further research.
Electronic databases up to March 2004. Experts in Barrett's oesophagus from the UK.
A systematic review of the effectiveness of endoscopic surveillance of Barrett's oesophagus was carried out following methodological guidelines. Experts in Barrett's oesophagus from the UK were invited to contribute to a workshop held in London in May 2004 on surveillance of Barrett's oesophagus. Small group discussion, using a modified nominal group technique, identified key areas of uncertainty and ranked them for importance. A Markov model was developed to assess the cost-effectiveness of a surveillance programme for patients with Barrett's oesophagus compared with no surveillance and to quantify important areas of uncertainty. The model estimates incremental cost--utility and expected value of perfect information for an endoscopic surveillance programme compared with no surveillance. A cohort of 1000 55-year-old men with a diagnosis of Barrett's oesophagus was modelled for 20 years. The base case used costs in 2004 and took the perspective of the UK NHS. Estimates of expected value of information were included.
No randomised controlled trials (RCTs) or well-designed non-randomised controlled studies were identified, although two comparative studies and numerous case series were found. Reaching clear conclusions from these studies was impossible owing to lack of RCT evidence. In addition, there was incomplete reporting of data particularly about cause of death, and changes in surveillance practice over time were mentioned but not explained in several studies. Three cost--utility analyses of surveillance of Barrett's oesophagus were identified, of which one was a further development of a previous study by the same group. Both sets of authors used Markov modelling and confined their analysis to 50- or 55-year-old white men with gastro-oesophageal reflux disease (GORD) symptoms. The models were run either for 30 years or to age 75 years. As these models are American, there are almost certainly differences in practice from the UK and possible underlying differences in the epidemiology and natural history of the disease. The costs of the procedures involved are also likely to be very different. The expert workshop identified the following key areas of uncertainty that needed to be addressed: the contribution of risk factors for the progression of Barrett's oesophagus to the development of high-grade dysplasia (HGD) and adenocarcinoma of the oesophagus; possible techniques for use in the general population to identify patients with high risk of adenocarcinoma; effectiveness of treatments for Barrett's oesophagus in altering cancer incidence; how best to identify those at risk in order to target treatment; whether surveillance programmes should take place at all; and whether there are clinical subgroups at higher risk of adenocarcinoma. Our Markov model suggests that the base case scenario of endoscopic surveillance of Barrett's oesophagus at 3-yearly intervals, with low-grade dysplasia surveyed yearly and HGD 3-monthly, does more harm than good when compared with no surveillance. Surveillance produces fewer quality-adjusted life-years (QALYs) for higher cost than no surveillance, therefore it is dominated by no surveillance. The cost per cancer identified approaches pound 45,000 in the surveillance arm and there is no apparent survival advantage owing to high recurrence rates and increased mortality due to more oesophagectomies in this arm. Non-surveillance continues to cost less and result in better quality of life whatever the surveillance intervals for Barrett's oesophagus and dysplastic states and whatever the costs (including none) attached to endoscopy and biopsy as the surveillance test. The probabilistic analyses assess the overall uncertainty in the model. According to this, it is very unlikely that surveillance will be cost-effective even at relatively high levels of willingness to pay. The simulation showed that, in the majority of model runs, non-surveillance continued to cost less and result in better quality of life than surveillance. At the population level (i.e. people with Barrett's oesophagus in England and Wales), a value of pound 6.5 million is placed on acquiring perfect information about surveillance for Barrett's oesophagus using expected value of perfect information (EVPI) analyses, if the surveillance is assumed to be relevant over 10 years. As with the one-way sensitivity analyses, the partial EVPI highlighted recurrence of adenocarcinoma of the oesophagus (ACO) after surgery and time taken for ACO to become symptomatic as particularly important parameters in the model.
The systematic review concludes that there is insufficient evidence available to assess the clinical effectiveness of surveillance programmes of Barrett's oesophagus. There are numerous gaps in the evidence, of which the lack of RCT data is the major one. The expert workshop reflected these gaps in the range of topics raised as important in answering the question of the effectiveness of surveillance. Previous models of cost-effectiveness have most recently shown that surveillance programmes either do more harm than good compared with no surveillance or are unlikely to be cost-effective at usual levels of willingness to pay. Our cost--utility model has shown that, across a range of values for the various parameters that have been chosen to reflect uncertainty in the inputs, it is likely that surveillance programmes do more harm than good -- costing more and conferring lower quality of life than no surveillance. Probabilistic analysis shows that, in most cases, surveillance does more harm and costs more than no surveillance. It is unlikely, but still possible, that surveillance may prove to be cost-effective. The cost-effectiveness acceptability curve, however, shows that surveillance is unlikely to be cost-effective at either the 'usual' level of willingness to pay ( pound 20,000-30,000 per QALY) or at much higher levels. The expected value of perfect information at the population level is pound 6.5 million. Future research should target both the overall effectiveness of surveillance and the individual elements that contribute to a surveillance programme, particularly the performance of the test and the effectiveness of treatment for both Barrett's oesophagus and ACO. In addition, of particular importance is the clarification of the natural history of Barrett's oesophagus.