ABSTRACT
A fundamental design issue of cluster randomized trials (CRTs) is that a simple randomized CRT is often as effective in reducing imbalance of baseline prognostic factors, threatening the internal validity of the trial. Stratified cluster randomized design can also be described as a randomized block design or a multi-site cluster randomized design with treatment at the cluster level, where the terminology of block or site is used instead of stratum. Stratification factors adopted often include cluster size, cluster-level socio-economic status, geographic location, and categorized levels of prognostic factors. Similar to simple randomized CRTs, intracluster correlation coefficient needs to be properly accounted for in the design of stratified CRTs. Moreover, since stratification is implemented in the design phase and stratification factors are often associated with outcome variables, a sample size estimation based on such stratification is necessary and more appropriate than an estimation ignoring the stratification.
