David M. Murray, PhD,
Director, Office of Disease Prevention, National Institutes of Health, Rockville, MD
Sheri L. Pals, PhD, Mathematical Statistician, Center for Global Health, Centers for Disease Prevention and Control, Atlanta, GA
Stephanie George, PhD, Senior Epidemiologist, Office of Disease Prevention, National Institutes of Health, Rockville, MD
Jocelyn A. Lee, PhD, Health Scientist Administrator, Office of Disease Prevention, National Institutes of Health, Rockville, MD
Ranell L. Myles, PhD, Health Science Policy Analyst, Office of Disease Prevention, National Institutes of Health, Bethesda, MD
Gabriel Lai, PhD, Program Director, National Cancer Institute, National Institutes of Health, Rockville, MD
Shakira Nelson, PhD, Postdoctoral Fellow, National Cancer Institute, National Institutes of Health, Rockville, MD
Andrey Kuzmichev, PhD, Managing Editor, Public Health Reports, Office of the Surgeon General, Washington, DC
Introduction: Group-randomized trial designs are widely used in public health and medicine to evaluate the effects of interventions that operate at a group level, manipulate the social or physical environment, or cannot be delivered to individuals without substantial risk of contamination. Periodic reviews of the methods used in these trials since 1990 have reported disappointing results. The most recent review was published in 2011 and found no improvement in the reporting on important methodological issues after the publication in 2004 of the Consolidated Standards of Reporting Trials (CONSORT) extension to group-randomized trials. We published a review of group-randomized trials that focused on cancer-related outcomes in 2008, covering studies published between 2002 and 2006. We found only 18% of the studies analyzed documented appropriate methods for sample size calculation, only 45% limited their analyses to methods that were appropriate given the design, and only 60% reported matching or stratification in the design to limit potential confounding. In addition, only 51% randomized 9 or more groups to each condition, a number that is generally required to have 80% power to detect an intervention effect of 0.20 standard deviation units given the level of intraclass correlation often seen in these studies and 50 or more members per group.
Methods: In order to determine whether this situation has improved, we identified 147 group-randomized trials focused on cancer-related outcomes published in print or online between 2011 and 2015, inclusive, and coded them for issues related to design, analysis, and sample size. Three reviewers read and coded each paper independently, then met as a group to discuss their coding and reach consensus on each paper. Coding will be complete by the end of the calendar year and results will be compiled and presented at the annual meeting in May.
Results: We will report on the characteristics of the reviewed studies, including the number of study conditions; whether the design was a cohort or cross-sectional design, or some combination; whether matching, stratification, or constrained randomization was employed; the number of groups assigned to each condition and the size of those groups; the number of time points included in the design and in the analysis; and the type of primary outcome variable. In addition, we will report on the documentation provided for sample size calculations and on the distribution of articles across analytic methods, grouped by whether they were appropriate or inappropriate given the design. We will evaluate the new results to determine whether improvements have occurred since our previous review in 2008.
Conclusions: We will make recommendations for the field based on these findings.
![[ Visit Client Website ]](images/banner.gif)