Sample Size Reduction by Applying ML Based Causal Inference Methods

Area of Science:

• Biostatistics
• Machine Learning in Clinical Trials
• Clinical Trial Design

Background:

• Randomized Controlled Trials (RCTs) traditionally use ANCOVA.
• Causal machine learning (ML) methods offer advanced analytical approaches.
• PROCOVA, TMLE, DML, and GRF are key causal ML methods for trial analysis.

Purpose of the Study:

• To compare causal ML methods against standard ANCOVA for clinical trial efficiency.
• To evaluate the utility of causal ML with and without historical control data.
• To assess the impact of generative AI in simulating trial data for robust analysis.

Main Methods:

• Comparative analysis of ANCOVA, PROCOVA, TMLE, DML, and GRF.
• Utilized historical placebo data from Phase 3 Ophthalmology studies.
• Employed Generative Adversarial Networks (GANs) to simulate RCT data under various scenarios.

Main Results:

• Causal ML methods increased statistical power without historical data.
• TMLE demonstrated a 21% increase in effective sample size in one scenario.
• PROCOVA improved power and controlled Type 1 error when borrowing historical data, showing robustness.

Conclusions:

• Causal ML methods significantly enhance clinical trial power and efficiency.
• These methods offer advantages over traditional ANCOVA, especially in complex trial designs.
• Generative AI simulation aids in evaluating method performance under diverse conditions.