Score test for association between recurrent events and a terminal event

Area of Science:

• Biostatistics
• Clinical Trials
• Survival Analysis

Background:

• Recurrent event analysis typically assumes independent censoring, which may be violated when a terminal event (e.g., death) is present.
• Existing joint models for recurrent and terminal events are complex, hindering clinical application and assessment of model appropriateness.
• The dependence between censoring and random effects in the presence of a terminal event is a critical, often overlooked, assumption.

Purpose of the Study:

• To propose and evaluate simple, efficient statistical methods for testing the association between terminal events and recurrent events.
• To assess the performance of these methods through simulation studies and evaluate their sensitivity to model misspecification.
• To demonstrate the practical utility of the proposed methods using a real-world clinical dataset.

Main Methods:

• Development of novel statistical tests for the association between recurrent and terminal events.
• Simulation studies to assess the power and Type I error rates of the proposed methods.
• Sensitivity analysis to evaluate the robustness of the methods to model misspecification.
• Application of the methods to a dataset of T-cell lymphoma patients with repeated skin tumor observations.

Main Results:

• The proposed methods provide a simple and efficient way to test for association between terminal and recurrent events.
• Simulation results demonstrate good performance of the methods under various scenarios.
• The methods show reasonable robustness to certain types of model misspecification.
• The analysis of the T-cell lymphoma data provides insights into the relationship between tumor recurrence and patient survival.

Conclusions:

• The developed methods offer a practical approach for clinicians and researchers to assess the relationship between terminal and recurrent events.
• These simple tests can help determine if more complex joint models are necessary for accurate statistical analysis.
• The findings contribute to a better understanding of event dependencies in longitudinal clinical data, particularly in oncology.