Modeling subdistribution hazards using a piecewise approach for time-independent covariates

Dae-Jin Lee, IE University

Co-authors: Juan Carlos Espinosa-Moreno, Basque Center for Applied Mathematics and University of the Basque Country; Fernando García-García, Basque Center for Applied Mathematics

Abstract: This work proposes a piecewise modeling (PM) approach to obtain the cumulative incidence function in subdistribution hazard (SH) scenarios for continuous-time competing risks (CR). Our method employs a weighted binomial Generalized Linear Model with a complementary log-log (cloglog) link function. We evaluated its performance experimentally using synthetic data, both complete and right-censored. Compared to the established CR baselines, our PM in some cases outperformed them in recovering the true SH parameters; in other cases, it closely matched their estimates. Moreover, in some scenarios, it outperformed them regarding the integrated Brier score.