Targeted estimation and inference for the sample average treatment effect in trials with and without pair-matching

Area of Science:

• Biostatistics
• Epidemiology
• Clinical Trials

Background:

• Cluster randomized trials often lack a clearly defined target population, making population average treatment effect (PATE) interpretation challenging.
• Logistical constraints frequently dictate study unit selection, deviating from simple random sampling.
• The sample average treatment effect (SATE) offers a more interpretable measure for the specific study units involved.

Purpose of the Study:

• To propose and evaluate targeted maximum likelihood estimation (TMLE) for the sample average treatment effect (SATE) in cluster randomized trials.
• To assess the asymptotic and finite sample properties of TMLE for SATE estimation.
• To provide a conservative variance estimator for TMLE-based SATE inference.

Main Methods:

• Application of targeted maximum likelihood estimation (TMLE) for SATE in cluster randomized trials.
• Analysis of both pair-matched and unmatched trial designs.
• Asymptotic and finite sample simulations to evaluate TMLE performance.

Main Results:

• TMLE provides an interpretable and efficient method for estimating the sample average treatment effect (SATE).
• Simulations demonstrate potential gains in precision and statistical power when focusing on SATE.
• A conservative variance estimator is proposed for robust inference.

Conclusions:

• TMLE is a viable and advantageous method for estimating SATE in cluster randomized trials.
• The proposed methodology enhances the analysis of treatment effects when the target population is ill-defined.
• This approach is applicable to complex trial designs, including pair-matched studies like the SEARCH trial.