Accounting for interactions and complex inter-subject dependency in estimating treatment effect in cluster-randomized...

Area of Science:

• Biostatistics
• Epidemiology
• Clinical Trials

Background:

• Semi-parametric methods are standard for estimating intervention effects in cluster-randomized trials (CRTs).
• Missing outcomes (MAR) and covariate imbalances pose challenges for accurate estimation.
• Existing methods like Inverse Probability Weighting (IPW) and augmented generalized estimating equations (AUG) have limitations, especially with treatment-covariate interactions.

Purpose of the Study:

• To propose a novel augmented generalized estimating equations-inverse probability weighting (AUG-IPW) estimator.
• To address challenges of missing outcomes and covariate imbalance in CRTs, particularly with treatment-covariate interactions.
• To develop a doubly robust (DR) estimator for consistent marginal treatment effect estimation.

Main Methods:

• Developed an AUG-IPW estimator weighting by the inverse probability of being a complete case.
• Allowed for different outcome models in each intervention arm.
• Investigated the estimator's doubly robust (DR) properties, ensuring consistency if either the missing data model or the outcome model is correctly specified.

Main Results:

• The proposed AUG-IPW estimator is doubly robust (DR), providing consistent estimates under weaker model assumptions.
• The method effectively handles missing outcomes at random (MAR) and covariate imbalances.
• The approach remains unbiased even with unmodeled covariate interference, provided it doesn't affect both outcome and missingness simultaneously.

Conclusions:

• The AUG-IPW estimator offers a robust solution for analyzing CRTs with missing data and potential treatment-covariate interactions.
• An R package is available for implementing the proposed method.
• The method's utility is demonstrated through simulations and an application to an HIV risk reduction CRT in South Africa.