Simulation-Guided Trial Planning of a Cluster Randomized Trial With Probabilistic Covariate-Constrained Randomization for Vaccine Coverage

Author(s)

Rebecca Metcalfe, BA, MA, PhD1, Richard Yan, MSc2, Shomoita Alam, PhD3, Jay JH Park, PhD1.
1Core Clinical Sciences, Vancouver, BC, Canada, 2Department of Statistical and Actuarial Science, Simon Fraser University, Vancouver, BC, Canada, 3Department of Epidemiology, Biostatistics, and Occupational Health, McGill University, Montreal, QC, Canada.
OBJECTIVES: Cluster randomized trials are often at risk of imbalanceed prognostic factors when a small number of clusters are available. Covariate-constrained randomization restricts cluster randomization schemes to those that meet the pre-specified balancing covariate metrics can be used to avoid imbalances between trial arms. Given their complexity, using conventional sample size (closed-form) formulas can be challenging, especially under a multi-nested data structure where the units of randomization (e.g., health units) and outcome collection (e.g., villages) differ. In this study, we used simulations to plan a two-arm health area-level cluster trial of mass distribution of nutritional supplements embedded with an existing immunization program aimed at improving measles vaccination coverage in Chad.
METHODS: Our trial consisted of 12 health areas where random subsets of villages would be selected for the post-intervention survey. Causal reasoning via directed acyclic graphs was applied to select covariates to be balanced using 0.20 village-level standardized mean differences (SMDs) as balancing metrics. Using baseline census survey data collected, 1,000,000 assignments of health areas with different possible selections of villages were generated. In each assignment, village-level SMDs of village population size, distance to the nearest health center, and baseline measles vaccination rate were calculated to restrict the randomization. Operating characteristics of power, type I error rate, and sample size were compared for quasi-binomial regression and beta-regression methods.
RESULTS: The control of type I error rate was less optimal for quasi-binomial regression and naïve analysis than for beta-regression. For our base case, with an assumed 0.70 event rate and 0.24 intracluster correlation coefficient, beta regression showed control of type I error rate below 0.05, and 0.80 power to detect an absolute increase of 0.13 in vaccination coverage by including 80 villages per arm in the post-intervention survey.
CONCLUSIONS: Simulation-guided planning is valuable for planning complex cluster trials with limited clusters and a nested structure.

Conference/Value in Health Info

2025-11, ISPOR Europe 2025, Glasgow, Scotland

Value in Health, Volume 28, Issue S2

Code

MSR187

Topic

Clinical Outcomes, Methodological & Statistical Research, Study Approaches

Topic Subcategory

Confounding, Selection Bias Correction, Causal Inference

Disease

Nutrition, Pediatrics, Vaccines

Your browser is out-of-date

ISPOR recommends that you update your browser for more security, speed and the best experience on ispor.org. Update my browser now

×