Increasing the efficiency of randomized trial estimates via linear adjustment for a prognostic score

Area of Science:

• Clinical Trials
• Biostatistics
• Health Data Science

Background:

• Estimating causal effects in clinical research relies on randomized experiments.
• Reducing statistical uncertainty is crucial for efficient trial analysis.
• Historical data from registries, prior trials, and health records offer potential for improving trial efficiency.

Purpose of the Study:

• To propose a method for utilizing historical data to enhance the efficiency of clinical trial analyses.
• To improve statistical uncertainty reduction in causal effect estimation without sacrificing type-I error rate control.
• To develop a method that leverages linear covariate adjustment for unbiased efficiency gains.

Main Methods:

• A prognostic model is trained on historical data to predict patient outcomes.
• Treatment effects are estimated using linear regression, adjusting for predicted outcomes (prognostic scores).
• The method's theoretical properties and performance are evaluated through simulations and a reanalysis of an Alzheimer's disease trial.

Main Results:

• The proposed prognostic covariate adjustment method achieves minimum variance under certain conditions.
• It demonstrates greater efficiency than raw covariate adjustment, with gains proportional to the prognostic model's predictive accuracy.
• Simulations and trial reanalysis show significant reductions in mean-squared error and estimated variance.

Conclusions:

• Prognostic covariate adjustment offers an effective way to improve clinical trial efficiency by incorporating historical data.
• The method provides substantial reductions in variance and mean-squared error, leading to potential sample size reductions of 10-30%.
• A simplified formula for asymptotic variance facilitates power calculations, enabling more precise trial planning.