Abstract

Multi-state survival models extend traditional survival analysis, which typically focuses on a single endpoint, by representing disease progression as a series of states and transitions. Nonetheless, capturing complex relationships between patient characteristics and transitions between disease states remains a methodological challenge. This study therefore aims to assess whether integrating machine learning into multi-state survival models can improve the prediction of chronic disease progression. A simulation study is conducted to compare the performance of conventional multi-state models with machine-learning-enhanced approaches across varying sample sizes, censoring levels, and predictor structures. The study investigates the conditions under which machine learning improves predictive performance relative to conventional multi-state modelling approaches. The findings are expected to contribute to the development of flexible predictive methods for chronic disease progression.