Statistical Analysis of Composite Time-to-Event Outcomes: The Win Ratio and Beyond

This short course provides a survey of key topics covered in the instructor's forthcoming book with Chapman & Hall/CRC. Clinical studies often collect multiple outcomes per patient. These typically include mortality as well as lesser events such as disease progression or hospitalization. As a standard approach, investigators focus on the first event, giving rise to the composite endpoint of progression-free, or more generally, event-free, survival. This paradigm is starting to change with the introduction and popularization of the win ratio (Pocock et al., 2012) and related methods, which allow for meaningful ranking of component events according to their clinical priority (e.g., death over hospitalization) and fuller utilization of outcome data. Search the ClinicalTrials.gov database using the keyword 'win ratio', and you will surely find numerous example trials that specify the win ratio (or a certain variation of it) as the analysis method for the primary or secondary endpoints. The change in practice is backed by an ever-growing statistical literature. Yet the methodological contributions have so far been scattered in assorted journals, different in notation and style and perhaps lacking in a coherent theme. Some of them may even be too technical for the average practitioner to digest. Most unfortunately, computing packages that implement these methods, widely available as them are, have received even less publicity. Many analysts struggle to find, let alone to learn, the right software to analyze their data. In this half-day short course, we provide a systematic treatment to the newly developed methods for composite endpoints, from theory to practice. The topics will range from two-sample testing to estimation, semiparametric regression, and, to a lesser extent, nonparametric regression. The exposition will be put in the context of the recently released guidelines in the ICH-E9 (R1) Addendum, which highlights the importance of estimand construction and sensitivity analysis. Usage of the corresponding R-packages will be demonstrated in real time and on real data examples. To facilitate learning, a Quarto (https://quarto.org) site will be developed, incorporating slides and R code in a consistent style. This will allow the student to follow along effortlessly in both theoretical instruction and hands-on programming.