ABSTRACT: A composite endpoint combining multiple outcomes is frequently used for clinical trials. Conventional time-to-first-event analyses have some limitations as the first event analyzed is often a less important, but more frequent outcome. The win ratio addresses these limitations. It considers the clinical importance order and the relative time sequence of the multiple outcomes. The (unmatched) win ratio compares each patient in the treatment group with every patient in the control group, following the prioritized pairwise comparison process. For any pair, the time to the most important event (e.g. death) determines the “winner” (e.g. the one who died later). If the winner cannot be determined (e.g., both patients remain alive), the second-most important event is considered (e.g., graft loss in transplant studies), and so forth until the winner can be determined; otherwise the pair is tied. The win ratio is the ratio of the total numbers of wins between the two groups.
In this talk, Gaohong will focus on stratified designs and present the stratified win ratio constructed similarly to the Mantel-Haenszel stratified odds ratio. Gaohong will present its statistical performance and evaluate the corresponding test of homogeneity of the stratum-specific win ratios. Gaohong will also discuss the connections of the win ratio to some common measures of treatment effect, as well as win ratio estimands. As the win ratio analysis usually involves many more ties than wins, we have also explored the role of ties. For this purpose, Gaohong will present the win odds, which assigns 50% of ties to the numerator and the denominator. Gaohong will illustrate this statistic and compare it with the win ratio using real data from clinical trials.
For questions and inquiries about this event, please reach out to Evan Wardell at evan.wardell@nyu.edu.