Illustrating problems faced by stroke researchers: a review of cluster-randomized controlled trials

Abstract

The cluster-randomized controlled trial is a design increasingly used in prevention and health care evaluation studies and is highly relevant to stroke research. However, there are methodological issues that make it complex to implement. These are not always fully appreciated, with reviews continuing to reveal deficiencies. We searched PUBMED and CENTRAL databases to March 31, 2011 for cluster-randomized controlled trials in stroke. Two investigators independently reviewed citations for eligibility and extracted data on key aspects of each trial. Fifteen trials met the eligibility criteria. No trial fully met CONSORT cluster-randomized controlled trial guidelines, although good design and reporting practice were usually present. Twelve trials included the term ‘cluster-randomized’ (or ‘group-randomized’) in the title, and 12 trials stated the intraclass correlation coefficient used to plan the number of clusters and cluster size. However, few provided a clear, evidence-based justification for the choice of intraclass correlation coefficient, and only two-thirds reported the intraclass correlation coefficient for primary outcomes. Several trials appeared underpowered because of problems in determining an appropriate sample size, defining appropriate clusters, and recruiting and retaining clusters and patients. Cluster-randomized controlled trials are difficult to design and perform due to the combination of methodological and practical difficulties. It is important that further improvements are made to reporting cluster-randomized trials and intraclass correlation coefficients should be estimated using a standardized approach and reported consistently; this would be beneficial for stroke researchers when designing future cluster-randomized trials.