A1429
Title: Modeling the restricted mean survival time as a function of time horizons with pseudo-value regression trees
Authors: Alina Schenk - University of Bonn (Germany) [presenting]
Matthias Schmid - University of Bonn (Germany)
Abstract: The restricted mean survival time (RMST) has become an increasingly important estimand in time-to-event studies. Defined as the restricted area under the survival function over $[0, \tau]$, the RMST represents the average event-free survival time up to the time horizon $\tau$. In practice, directly modeling the RMST conditional on covariates is particularly useful for assessing the impact of a treatment or exposure on the expected lifetime. However, most direct modeling approaches for the RMST focus on a single fixed time horizon $\tau > 0$. Choosing an appropriate value for $\tau$ can be challenging and has been widely discussed in the literature. The purpose is to introduce an alternative approach to modeling the RMST as a function of $\tau$ using pseudo-value regression trees (PRT). PRT are characterized by a multivariate regression tree built on a pseudo-value outcome and by successively fitting a set of regularized additive models to the data in the nodes of the tree using gradient boosting. Like previously published approaches, PRT models RMST values at various time horizons simultaneously and incorporates time-varying covariate effects. A simulation study and a real-world application are presented to demonstrate the properties of the proposed method.
