"Within-Trial" Prognostic Score Adjustment Is Targeted Maximum Likelihood Estimation

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Epidemiology

Background:

• Prognostic covariate adjustment is increasingly used in randomized trials.
• Covariates are often derived from historical data using predictive models.
• Researchers question if adjustment is possible using only trial data.

Purpose of the Study:

• To clarify the statistical nature of within-trial prognostic covariate adjustment.
• To establish the relationship between within-trial adjustment and existing statistical methods.
• To provide guidance on terminology for these analytical approaches.

Main Methods:

• The study clarifies that within-trial prognostic adjustment is a form of targeted maximum likelihood estimation (TMLE).
• It emphasizes that TMLE is a well-established procedure for improving statistical power in trial analyses.
• No new methods were developed; the focus is on conceptual clarification and terminology.

Main Results:

• Within-trial prognostic covariate adjustment is statistically equivalent to targeted maximum likelihood estimation (TMLE).
• TMLE is a recognized method that enhances the power of randomized trial analyses.
• The findings suggest that existing TMLE frameworks adequately address within-trial adjustment needs.

Conclusions:

• Within-trial prognostic score adjustment should be recognized and referred to as targeted maximum likelihood estimation (TMLE).
• There is no need to develop new terminology or methods for this approach.
• Standardizing the terminology to TMLE promotes clarity and leverages existing statistical knowledge.