PURPOSE: To introduce win ratio analysis for quantifying benefits of an intervention, particularly in the rare disease, and illustrate how the outputs can be used to derive the intervention’s value.

DESCRIPTION: Designing and analyzing rare disease clinical trials is challenging, not only due to limited small sample sizes, but also the multisystemic and heterogenous nature of their manifestations. This necessitates assessments on multiple end points to capture fully the disease impacts and treatment benefits. Treatment developers, regulators and health technology assessors are, thus, faced with how best to address these multiple endpoints, particularly if using standard statistical approaches. Traditional approaches force selecting a primary endpoint, “putting all the eggs in that basket”, thereby relegating other, potentially important endpoints, to secondary testing. These standard methods can fail to achieve statistical significance because of small sample sizes and multiple testing. This impairs translating clinical trial results to value and may lead to important innovations being withheld from patients with few other options.

In this workshop, we will outline the challenges and introduce an innovative solution. The recently proposed win ratio method will be detailed and its application in rare disease discussed. A case study will be used to illustrate the considerations involved in setting up win ratio analyses, including selection of endpoints, definitions for wins and losses, capturing uncertainty in estimates and interpreting results. Outcomes of traditional analysis and win ratio will be compared. The difficulties for valuation posed by the newer approach will also be addressed.

Participants will be asked to pose and discuss issues with the method, particularly in terms of value assessment. How would they handle win ratio results? Would these be acceptable as the core efficacy input to a health technology assessment? Given the “comingling” of endpoints of different nature, how will valuations be done? Are new modeling approaches required?