Propensity score methods for estimating relative risks in cluster randomized trials with low-incidence binary...

Area of Science:

• Biostatistics
• Epidemiology
• Clinical Trials

Background:

• Cluster randomized trials (CRTs) are susceptible to selection bias despite randomization.
• Low incidence binary outcomes in CRTs can pose challenges for classical statistical methods.
• Separation problems can hinder the effectiveness of multivariable regression for bias adjustment.

Purpose of the Study:

• To evaluate propensity score (PS)-based methods for estimating relative risks in CRTs.
• To compare the performance of PS-based methods against multivariable regression for binary outcomes with low incidence.
• To identify the most effective method for correcting selection bias in this specific trial design and outcome setting.

Main Methods:

• Simulation study design.
• Implementation of propensity score (PS) methods: direct adjustment, inverse weighting, and stratification.
• Comparison with classical multivariable regression.
• Focus on binary outcomes with low incidence in cluster randomized trials.

Main Results:

• Direct adjustment on the propensity score (PS) fully corrected bias.
• Direct adjustment on the PS demonstrated superior statistical properties compared to other methods.
• Multivariable regression and other PS-based methods showed limitations in addressing bias for low incidence binary outcomes.

Conclusions:

• Direct adjustment on propensity scores is a reliable method for bias correction in cluster randomized trials with low incidence binary outcomes.
• Propensity score methods, particularly direct adjustment, offer advantages over traditional regression techniques in specific challenging scenarios.
• The findings support the use of direct PS adjustment for robust treatment effect estimation in CRTs.