"Super-covariates": Using predicted control group outcome as a covariate in randomized clinical trials

Area of Science:

• Clinical Trials
• Biostatistics
• Pharmacoeconomics

Background:

• Clinical trial efficacy relies on drug potency and patient outcome variability.
• Prognostic covariates in trial analysis can reduce outcome variation.
• Regression models are standard for primary statistical analysis, incorporating treatment, stratification factors, and baseline outcomes.

Purpose of the Study:

• To introduce a novel "super-covariate" for improving the statistical power of randomized controlled clinical trials.
• To leverage historical patient data to predict control group outcomes, thereby reducing unexplained variability.
• To offer an alternative to Bayesian methods that avoids type I error inflation while enhancing trial efficiency.

Main Methods:

• A prognostic model or ensemble is trained on external historical patient data (not from the current trial).
• The trained model generates a patient-specific prediction of the control group outcome, used as a "super-covariate" in the primary analysis.
• The "super-covariate" is included as a covariate in regression models, distinct from an offset.

Main Results:

• The
• super-covariate
• approach demonstrated efficiency gains in an example involving neovascular age-related macular degeneration.
• This method has the potential to increase the power of clinical trials by reducing unexplained outcome variability.
• The benefit is greater for larger sample sizes compared to Bayesian approaches, with no type I error inflation.

Conclusions:

• The
• super-covariate
• method effectively utilizes historical data to boost clinical trial power.
• Generalizability of prognostic models across diverse patient populations is crucial for consistent reduction of unexplained variability.
• This approach offers a statistically sound method for enhancing the precision and power of clinical trial analyses.