|
Abstract:
|
For composite outcomes whose components can be prioritized on clinical importance, following generalized pairwise comparisons (GPC), patients in the two groups can be compared using importance order to produce wins and subsequently win proportions. Because the win ratio (ratio of win proportions), the net benefit (difference in win proportions) and the win odds (odds of win proportions) are derived using the same win proportions and they test the same hypothesis of equal win probabilities in the two groups, we refer to them as win statistics. In this talk, we will focus on the censoring issues of time-to-event outcomes. We will first present impact of follow-up time. If the treatment has a long-term benefit from a more important but less frequent endpoint (eg, death), the win statistics can show this benefit by following patients longer. Then, we will show that inverse-probability-of-censoring weighting (IPCW) can correct the win statistics from censoring bias, and IPCW-adjusted and CovIPCW (IPCW with baseline and/or time-dependent covariates that can predict dependent censoring) adjusted win statistics are unbiased estimators of treatment effect in the presence of censoring.
|
