Leveraging baseline covariates to analyze small cluster-randomized trials with a rare binary outcome

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Epidemiology

Background:

• Cluster-randomized trials (CRTs) are essential for evaluating interventions but often have limited clusters.
• Individual-level covariate adjustment increases statistical efficiency in individually randomized trials.
• Analytical methods for individual-level covariate adjustment in small CRTs are underexplored.

Purpose of the Study:

• To systematically investigate individual-level covariate adjustment strategies in small CRTs with rare binary outcomes.
• To compare the efficiency of propensity score weighting and multivariable regression for estimating participant-average causal effects.
• To examine the finite-sample performance of variance estimators for quantifying treatment effect uncertainty.

Main Methods:

• Extensive simulations were conducted to evaluate propensity score weighting and multivariable regression.
• Operating characteristics and relative efficiency advantages of each strategy were identified.
• Bias-corrected sandwich variance estimators were assessed for uncertainty quantification.

Main Results:

• Both propensity score weighting and multivariable regression demonstrated potential for improving efficiency in small CRTs.
• Specific scenarios were identified where one adjustment strategy offered a relative efficiency advantage over the other.
• The finite-sample performance of variance estimators was evaluated for reliable uncertainty estimation.

Conclusions:

• Individual-level covariate adjustment strategies, including propensity score weighting and multivariable regression, are valuable for small CRTs.
• These methods can enhance statistical efficiency and provide accurate estimates of participant-average treatment effects.
• Practical recommendations are provided for selecting appropriate adjustment strategies based on study characteristics.