Fig. 4.

Predictive performance of estimated cut-points when (k₁, k₂, a) equals (− 8/5, 8/3, 1/2). a Simulation results when P_c = 0%; (b) Simulation results when P_c = 20%; (c) Simulation results when P_c = 50%. Four discretization methods are used to find optimal cut-points of simulated continuous variables. The continuous variables are transformed into categorical variables and then fitted in univariate Cox models. The boxplots present predictive performance of Cox models in term of c-index, CPE, IBS, R_PM² and R_D²