Use of the Win Ratio for Analysis of Stroke Trials: Description, Illustration, and Planned Use in the Second European Carotid Surgery Trial (ECST-2)

Abstract

Randomized trials in stroke often focus on outcomes beyond a single clinical event. Trials of stroke prevention commonly use composite outcomes that include multiple components (eg, death, stroke, or myocardial infarction). A major limitation is that all events count equally but may differ markedly in terms of clinical severity. Trials in acute stroke often use ordinal outcomes or scale scores. Limitations include the requirement for statistical assumptions and the difficulty of handling the competing risk of death. We introduce the win ratio as an alternative method. It works by placing components of a composite into a hierarchy, whereby clinically more important outcomes take priority over less important ones. We illustrate how it works using data from 2 major stroke trials: the ICSS (International Carotid Stenting Study, a trial in stroke prevention) and the MR CLEAN (Multicenter Randomized Clinical Trial of Endovascular Treatment for Acute Ischemic Stroke in the Netherlands). Potential benefits of the win ratio approach include the possibility to (1) emphasize the clinically most important outcomes, (2) combine components of different outcome types into a composite (eg, a mixture of time-to-event, continuous, and categorical), and (3) naturally handle the competing risk of death in analyses of quantitative outcomes. The win ratio will be used in the upcoming analysis of the ECST-2 (Second European Carotid Surgery Trial), which has a hierarchical primary outcome of (1) time to perioperative death, fatal stroke, or fatal myocardial infarction (most important); (2) time to nonfatal stroke; (3) time to nonfatal myocardial infarction (excluding silent infarcts); and (4) new silent cerebral infarct on brain imaging (least important). The win ratio provides a useful clinically relevant method for analyzing trial outcomes. It has some advantages over conventional methods, and we recommend its wider application in future stroke trials.