Covariate-constrained randomization in cluster randomized 2 × 2 factorial trials: application to a diabetes...

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Health Services Research

Background:

• Cluster randomized trials (CRTs) involve randomization at the group level, posing challenges for covariate balance with small cluster numbers.
• Imbalance in cluster-level covariates can distort treatment effects, reduce power, and increase variability in CRTs.
• Covariate-constrained randomization (CR) is a strategy to mitigate covariate imbalance in CRTs, with existing methods not covering factorial designs.

Purpose of the Study:

• To develop and evaluate methods for covariate-constrained randomization (CR) in 2x2 factorial cluster randomized trials.
• To assess the performance of CR against simple randomization for estimating treatment effects in factorial CRTs.

Main Methods:

• Developed CR methods for 2x2 factorial CRTs with continuous outcomes and covariates.
• Applied methods to the BEGIN study (a weight loss CRT for pre-diabetes).
• Conducted simulations varying key parameters (number of clusters, covariate association, randomization space size, analysis strategies) to compare CR and simple randomization.

Main Results:

• Covariate-constrained randomization (CR) effectively balances cluster-level covariates in factorial CRTs, leading to more precise inferences compared to simple randomization.
• Including cluster-level covariates in the analysis model with CR increases statistical power to detect treatment effects.
• Power can be lower with adjusted analyses compared to unadjusted ones when the number of clusters is small.

Conclusions:

• Covariate-constrained randomization (CR) is recommended over simple randomization for factorial CRTs to prevent imbalanced designs and improve inference precision.
• Including cluster-level covariates in the analysis model is advised for factorial CRTs (unless the number of clusters is very small) to enhance power and maintain statistical properties.
• CR offers a robust approach for complex factorial cluster randomized trial designs.