Regression methods for cost-effectiveness analysis with different censoring times or terminating events for survival...

Cost-effectiveness analysis (CEA) is crucial for evaluating new medical treatments. In many studies, both costs and effectiveness are censored. While standard survival analysis methods suit survival time, they cannot be directly applied to cumulative outcomes (e.g., costs or quality-adjusted lifetime) due to induced informative censoring. Additional challenges arise when costs and effectiveness have different terminating events or censoring times. Our motivating examples are two cardiovascular trials: MADIT-CRT and MADIT-II. In MADIT-CRT, effectiveness was defined as heart failure-free survival, while costs accumulated until death, leading to different terminating events for costs and effectiveness. Furthermore, subgroup identification for patient heterogeneity was also of interest. In MADIT-II, early stopping of cost collection for some patients complicated CEA. These examples highlight the need for methods handling different censoring structures while allowing covariate adjustment to improve efficiency and address imperfect randomization. Although some CEA methods have been proposed for different terminating events or censoring times, none provide covariate adjustment. We propose regression-based methods for CEA with different terminating events or censoring times, focusing on estimating the incremental cost-effectiveness ratio and the incremental net benefit. By incorporating covariates within a regression framework, our methods enable covariate adjustment and subgroup identification. Simulation studies demonstrate good finite-sample performance, and applications to the two motivating examples show that our approach provides practical tools for CEA under complex censoring mechanisms.