Capturing simple and complex time-dependent effects using flexible parametric survival models: A simulation study

Abstract Non-proportional hazards are common within time-to-event data and can be modeled using restricted cubic splines in flexible parametric survival models. This simulation study assesses the ability of these models in capturing non-proportional hazards, and the ability of the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) in selecting degrees of freedom. The simulation results for scenarios with differing complexities showed little bias in the survival and hazard functions for simple scenarios; bias increased in complex scenarios when fewer degrees of freedom were modeled. Neither AIC nor BIC consistently performed better and both generally selected models with little bias.