Comparative Review of Modern Competing Risk Methods in High-dimensional Settings

Competing risk analysis accounts for multiple mutually exclusive events, improving risk estimation over traditional survival analysis. Despite methodological advancements, a comprehensive comparison of competing risk methods, especially in high-dimensional settings, remains limited. This study evaluates penalized regression (LASSO, SCAD, MCP), boosting (CoxBoost, CB), random forest (RF), and deep learning (DeepHit, DH) methods for competing risk analysis through extensive simulations, assessing variable selection, estimation accuracy, discrimination, and calibration under diverse data conditions. Our results show that CB achieves the best variable selection, estimation stability, and discriminative ability, particularly in high-dimensional settings. while MCP and SCAD provide superior calibration in $n>p$ scenarios. RF and DH capture nonlinear effects but exhibit instability, with RF showing high false discovery rates and DH suffering from poor calibration. Further, we compare the flexibility of these methods through the analysis of a melanoma gene expression data with survival information. This study provides practical guidelines for selecting competing risk models to ensure robust and interpretable analysis in high-dimensional settings and outlines important directions for future research.