All Times EDT
Keywords: Benefit-Risk, Bayesian, DEA (Data Envelopment Analysis)
Current approaches to drug benefit-risk assessment do not combine all benefit and risk data in a manner that permits straightforward quantitative comparisons of treatment options while acknowledging the uncertainty in estimation. Number Needed to Treat and Number Needed to Harm are well-established measures in the health care literature, but they only accommodate a single measure of benefit and risk and ignore uncertainty. Multicriteria Decision Analysis has been used to study trade-offs between benefit and risk, and while it can combine all available efficacy and safety data, it does not facilitate statistical comparisons. We propose a Bayesian Data Envelopment Analysis (DEA) framework to assess benefit-risk. DEA is a nonparametric mathematical programming method for evaluating the relative efficiency of entities that use resources to achieve desired outcomes. It is widely used in healthcare to compare hospital quality and efficiency, but has yet to be applied to benefit-risk. The proposed model produces posterior distributions of benefit-risk that characterize the value of competing treatment options designed to improve skin appearance in patients with mild to moderate acne. The approach also identifies specific sources of inefficiency (low efficacy/high risk) that can be addressed by physician-scientists. We investigate and compare model performance via a simulation study.