Eliciting Distributions to Populate Decision Analytic Models

BackgroundElicitation can be used to characterize structural uncertainty within a decision analytic model. This allows the value of acquiring further evidence to resolve these uncertainties to be established.AimThis article demonstrated the use of expert elicitation for this purpose and also compared the elicited results with the results from alternative assumptions previously used to characterize the uncertainties.Materials and MethodsDistributions for two unknown parameters were elicited. These were used within a model developed to assess the cost-effectiveness of infliximab and etanercept for the treatment of active psoriatic arthritis (PsA), compared with palliative care. The experts’ distributions were synthesized using two approaches: linear pooling and random effects meta-analysis. Weighting of experts is also explored.ResultsThe four methods produce broadly similar results, and in each, the choice of optimum strategy is between etanercept and palliative care (incremental cost-effective ratio for etanercept is between £29,021 and £39,259 per costs and quality adjusted life years). Decision uncertainty, at a £30,000 threshold, is high in all of the synthesis models thus generating high values of further research at between £141 and £634 million. In each model, the greatest value of further research was for the short-term effectiveness of treatment (£47–£406 million).DiscussionAlthough the cost-effectiveness results do not differ substantially between the models using the elicited values and the original scenarios, there are some stark contrasts in terms of the values of further research generated.ConclusionElicitation offers a feasible method to generate evidence for the missing information but there are a number of key issues for which further research is required.