A study of the patient weighted while alive estimand for recurrent events with competing risks

Recurrent event endpoints affected by a terminal event (competing risk) are common in randomised controlled trials (RCTs) within the pharmaceutical industry, e.g., in cardiovascular and heart failure outcome trials. A key challenge is interpreting the average number of events over time, as high mortality shortens follow-up and leads to fewer observed recurrent events. Although the problem is fundamentally bivariate, several univariate approaches have been proposed. In this paper, we focus on the patient weighted while alive estimand, which represents the average number of events per year lived. We study non-parametric estimation and regression methods for this estimand, the latter using either pseudo-values or inverse probability of censoring weighted estimating equations. The properties of these methods are assessed through simulation studies, and illustrated with a practical application using data from a type 2 diabetes trial. We find that the patient weighted while alive estimand is less sensitive to differences in terminal event rates than the average number of recurrent events, and that both regression approaches are feasible. While this estimand is a useful addition for analysing recurrent events with competing risks, the bivariate nature of the problem should be acknowledged, and separate analyses of recurrent and terminal events remain informative.