Restricting randomization reduces imbalance in trials with clusters. Both clustering and restricted randomization induce correlations among outcomes. Special techniques are needed to control error rates for correlated data. Failure to account for correlations in power calculations can yield sample sizes that are either too large or too small. Failure to account for correlations in analysis can lead to large differences between the actual and nominal Type I error probabilities. No study has quantified the exact effects of ignoring correlations due to clustering and pair matching on power and the actual probability of a Type I error. Two reviewers with expertise in trials with clusters identified and reviewed a set of NCI-funded trials mentioning clusters published between 1/1/2018 and 6/1/2024 . Exact derivations gave actual Type I error probabilities and power values when clustering, pair-matching, or both were ignored. Of 232 manuscripts, 96 met inclusion criteria. Only 21% had both correct power and correct data analyses; 63% had neither. Incorrect power and data analyses of cluster trials remain common, even among NCI-funded investigators. New closed-form exact derivations show that the consequences may be substantial: actual Type I error probabilities can be far from nominal, and sample sizes substantially over- and underestimated.

Dr. Deborah H. Glueck is the senior biostatistician at the Lifecourse Epidemiology of Adiposity & Diabetes (LEAD) Center at the Colorado School of Public Health and Professor of Pediatrics at the University of Colorado Denver School of Medicine. Dr. Glueck has an A.B. cum laude in mathematics from Harvard College and M.S. and Ph.D. degrees in biostatistics from the University of North Carolina at Chapel Hill. She completed a postdoctoral fellowship in health services research at Robert Wood Johnson Medical School. Dr. Glueck has a career-long focus on power and sample size for multilevel and longitudinal studies and the analysis of longitudinal epidemiological cohort studies and randomized controlled clinical trials.

For more information about Deborah H. Glueck, Ph.D., please visit their online bio: https://www.researchgate.net/profile/Deborah-Glueck