All Times EDT
The win ratio, win odds, and net benefit are all effect-size measures calculated by comparing every patient in the treatment to every one in the control. Such generalized pairwise comparisons (GPC) allow flexible ranking of the outcomes according to their clinical severity and, for that reason, are increasingly used in the analysis of prioritized composite endpoints consisting of death and nonfatal events. Most GPC-based methods, however, are first and foremost concerned with the construction of the estimator, oftentimes at the cost of a dubious estimand that either depends on the censoring distribution or lacks causal interpretation. This needs to change in the wake of the recently released ICH-E9(R1) Addendum, which places estimand construction at the core of the design and analysis of clinical trials. In this talk, I summarize two broad-based approaches to defining meaningful estimands that naturally entail GPC-like estimators --- a nonparametric one by restricting the time frame of the comparison and a semiparametric one by modeling the time-dependent comparison result. I also discuss additional challenges such as sensitivity analysis and handling of intercurrent events that are not part of outcome.