Variance reduction in randomised trials by inverse probability weighting using the propensity score.

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Epidemiology

Background:

• Covariate adjustment in regression models is standard for increasing precision in individually randomized controlled trials.
• Inverse probability-of-treatment weighting (IPTW) using propensity scores is an alternative adjustment method.
• Accurate variance estimation is crucial for reliable treatment effect assessment.

Purpose of the Study:

• To calculate the large-sample marginal variance of IPTW estimators for continuous and binary outcomes.
• To demonstrate that IPTW adjustment consistently increases the precision of treatment effect estimates.
• To highlight the advantages of IPTW over covariate adjustment, particularly for binary outcomes like risk differences and ratios.

Main Methods:

• Calculation of large-sample marginal variance for IPTW estimators of mean difference (continuous outcomes).
• Calculation of large-sample marginal variance for IPTW estimators of risk difference, risk ratio, and odds ratio (binary outcomes).
• Comparison of IPTW variance with standard analysis of covariance and unadjusted estimators.

Main Results:

• IPTW adjustment demonstrably increases the precision of treatment effect estimates across outcome types.
• For continuous outcomes, IPTW yields the same large-sample marginal variance as analysis of covariance.
• Failure to account for propensity score estimation in variance calculations leads to inaccurate precision estimates.

Conclusions:

• IPTW is a robust method for adjusting treatment effects in randomized trials, always improving precision.
• IPTW circumvents convergence issues common in covariate-adjusted regression models for binary outcomes.
• Correctly accounting for propensity score estimation in variance calculation is essential for valid IPTW analysis.