A systematic comparison of recurrent event models for application to composite endpoints

Area of Science:

• Biostatistics
• Clinical Trials
• Survival Analysis

Background:

• Clinical trials often study rare events, necessitating large sample sizes and long follow-up periods.
• Composite endpoints combine multiple event types to improve feasibility but may lose information if only the first event is analyzed.
• Recurrent event models offer an alternative for analyzing composite endpoints, accounting for multiple events per individual.

Purpose of the Study:

• To conduct a simulation-based comparison of recurrent event models when applied to composite endpoints.
• To evaluate the performance of different statistical models under various realistic clinical trial scenarios.
• To provide recommendations for appropriate statistical analysis and interpretation of results for composite endpoints.

Main Methods:

• A simulation study was performed to compare recurrent event models.
• Models investigated include those based on count data and Cox-based approaches (Andersen-Gill, Prentice-Williams-Petersen, Wei-Lin-Weissfeld).
• Comparisons were made across different realistic clinical trial scenarios.

Main Results:

• The Andersen-Gill and Prentice-Williams-Petersen models yielded similar effect estimators across various data scenarios.
• The Wei-Lin-Weissfeld model demonstrated effect estimators that could significantly deviate under common data conditions.
• The study highlights differences in model performance relevant to composite endpoint analysis.

Conclusions:

• The simulation results aid in understanding the advantages and disadvantages of different recurrent event models for composite endpoints.
• Recommendations are provided for selecting an adequate statistical analysis strategy.
• Guidance is offered for the meaningful interpretation of results when using composite endpoints in clinical trials.