Presentation (CTI)

OBJECTIVES: Healthcare evaluations often use the expected value (EV) of net benefit (NB) to recommend a single best treatment, under the risk neutrality assumption. However, recommending multiple-treatment options may be desirable, and risk-averse decision-makers may wish to consider uncertainty. The loss-adjusted expected value (LaEV) and Grading of Recommendations Assessment, Development and Evaluation (GRADE) approaches have been proposed to account for uncertainty and make multi-treatment recommendations based on a single efficacy outcome. We aimed to extend the EV approach to multi-treatment recommendations, and adapt LaEV and GRADE approaches to NB-based decisions.
METHODS: We took a 2-stage approach to EV multi-treatment decisions. The first stage identifies the optimal treatment. At the second stage, all treatments superior to the reference are compared with the optimal treatment within a minimally important difference (MID) in NB (MID-NB), which is calculated from MID in quality-adjusted life years (MID-Q). LaEV follows a similar approach but penalises EV by the expected loss due to decision-making under uncertainty. GRADE involves multiple stages where treatments are excluded until no treatment exceeds a threshold probability of being superior by MID-NB to at least one other treatment in the recommendation set. We applied and compared these methods using one hypothetical example and three cost-effectiveness models with multiple options, where disease-specific MID-Q is derived from a rapid review.
RESULTS: The LaEV approach recommends the same treatments or fewer than EV, as high uncertainty options are excluded. Depending on circumstances, GRADE may recommend more treatments than EV, potentially excluding the highest-NB treatment. GRADE results can depend on the selection of the reference treatment. All approaches are sensitive to MID-NB, but this remains to be further explored in future research.
CONCLUSIONS: Within the NB framework, the LaEV approach is conservative and reliable for risk-averse decision-makers, who have the flexibility to set a MID-NB threshold that reflects their risk tolerance.