COMMUNICATING UNCERTAINTY IN GENERALIZED COST-EFFECTIVENESS ANALYSES: VISUALIZING OPTIMAL INTERVENTIONS AND STRATEGIES

Purpose: The WHO-Choice methodology in general, and the stochastic league table (SLT) approach in particular, have been developed to represent uncertainty in sectoral (or generalized) CEAs. In such CEAs an overall combined approach is taken by considering all interventions in the (disease-specific) medical field, and applying an explicit budget constraint. However, the resulting tables and cost-effectiveness acceptability curves (CEACs) presenting probabilities of inclusion in the optimal strategy have two shortcomings: (1) the probabilities reflect inclusion of individual interventions and their relation to the actual strategy, i.e. combination of interventions, to be chosen is unclear; and (2) their robustness is unspecified. We present an extension of the SLT method with associated visualization addressing these shortcomings.

Method: Analogous to the non-probabilistic MAXIMIN and MAXIMAX decision rules the sensitivity of the performance of interventions as well as strategies in sectoral CEAs may be judged with respect to worst (or best) possible outcomes, in terms of health effects that can be obtained with a given budget. Therefore, after applying Monte Carlo simulation, we assessed robustness by performing the SLT analysis separately on all samples, and on samples in which the optimal strategy yielded health benefits worse (or better) than expected. Taking a certain percentage of worst (or best) outcomes, the robustness of interventions was visualized in a modified starplot. This plot allows the simultaneous depiction of the strategies that are likely to be optimal as well as the interventions that are likely to be included in these optimal strategies. We constructed a panel of modified starplots to visualize results for a range of budgets and a selection of worst and best health outcomes. The approach was tested on two examples, and for comparison the corresponding CEACs were provided, separately for interventions and strategies.

Result: The method was applicable to the original SLT example and to an additional illustrative example and provided clear and easily interpretable results. Furthermore, the modified starplots allow decision makers to pick out interventions with robust performance as well as the most relevant strategies more quickly than combinations of CEACs.

Conclusion: Our SLT extension and corresponding visualization improves the comprehensibility, and therefore usefulness, of outcomes of SLT for decision-makers. Therefore, we recommend its use whenever a SLT approach is considered.