Machine learning assisted adjustment boosts efficiency of exact inference in randomized controlled trials

Area of Science:

• Biostatistics
• Machine Learning
• Clinical Trials

Background:

• Randomized controlled trials (RCTs) are crucial for evidence-based medicine.
• Traditional statistical methods in RCTs may not fully capture complex covariate-outcome relationships.
• Covariate adjustment is essential for enhancing statistical power in RCTs.

Purpose of the Study:

• To propose a novel inferential procedure for RCTs using machine learning-based covariate adjustment.
• To develop a method that replaces traditional linear models with flexible nonparametric models.
• To improve the robustness and efficiency of statistical inference in RCTs.

Main Methods:

• Developed a machine learning-assisted inferential procedure within Rosenbaum's framework for exact tests.
• Employed nonparametric models to capture nonlinear associations and interactions between covariates and outcomes.
• Validated the method through extensive simulation experiments and a real-world case study.

Main Results:

• The proposed method demonstrated robust control of type I error rates.
• Significant improvements in statistical efficiency were observed for RCTs.
• The method's advantages were confirmed in a practical application, showing its real-world utility.

Conclusions:

• The novel procedure offers a simple, flexible, and robust alternative for RCT inference.
• It is particularly advantageous when complex covariate relationships are anticipated.
• This approach has the potential to reduce sample size and costs in clinical trials, including phase III studies.