Methods: A Markov model was developed to assess the cost-utility of Barretts surveillance compared to non-surveillance and to quantify areas of uncertainty. A cohort of 55 year old men were modelled over a time horizon of 20 years using a monthly cycle. In the treatment arm, surveillance was implemented every three months for high grade dysplasia, annually for low-grade dysplasia and every three years for non-dysplastic Barretts. Data was taken where possible, from published sources although analysis was hampered by the lack of good quality evidence. The model was also complicated by the fact that disease progression may remain undetected (especially in the non-surveillance arm) whilst clinical decisions are based on the observed diagnoses, hence the disease process is uncoupled from the treatment management process. This is an interesting methodological challenge common to many screening and surveillance models.
Results: Base case findings suggest that surveillance for Barretts Oesophagus may do more harm than good, conveying fewer quality-adjusted life years at a greater cost. Probabilistic sensitivity analysis reveals only a small chance (11%) that surveillance is cost-effective at a willingness-to-pay threshold of £30,000 ($55,800). The total expected value of perfect information at the population level was calculated as £6.5 million ($12.0 million). Partial value of information analysis identified a few critical areas where decision uncertainty could be substantially reduced by eliminating parameter uncertainty, notably the rate of recurrence of ACO after surgery, the time taken for ACO to become symptomatic, and the utility values assigned to Barretts states of varying levels of dysplasia.
Conclusions: This study highlights the difficulties associated with cost-effectiveness analyses where limited evidence is available but also highlights the value of uncertainty analysis in identifying research priorities.