Rforce: Random Forests for Composite Endpoints

Area of Science:

• Biostatistics
• Clinical Trials
• Medical Informatics

Background:

• Composite endpoints are crucial for evaluating treatment efficacy in medical research.
• Analyzing only the time to the first event in composite endpoints leads to significant information loss.
• Terminal events pose competing risks and are often overlooked in standard analyses.

Purpose of the Study:

• To address limitations in analyzing composite endpoints, particularly nonlinear covariate effects.
• To introduce a novel statistical method for handling both non-fatal and terminal events within composite endpoints.
• To improve the comprehensive analysis of clinical outcomes in medical research.

Main Methods:

• Development of a novel random forest approach for composite endpoints (Rforce).
• Utilization of generalized estimating equations for tree building within Rforce.
• Incorporation of pseudo-at-risk duration to manage dependent censoring from terminal events.

Main Results:

• Rforce effectively analyzes composite endpoints including non-fatal and terminal events.
• The method accounts for information loss inherent in traditional first-event analyses.
• Simulation studies and real-world data confirmed Rforce's robust performance.

Conclusions:

• Rforce offers an advanced solution for analyzing complex composite endpoints in clinical studies.
• This method enhances the utilization of data by considering all events, not just the first.
• Rforce provides a valuable tool for researchers investigating treatment efficacy with composite outcomes.