Components of variance in a group-randomized trial analysed via a random-coefficients model: the Rapid Early Action for Coronary Treatment (REACT) trial

Rapid Early Action for Coronary Treatment (REACT) was a multisite trial testing a community intervention to reduce the delay between onset of symptoms of acute myocardial infarction (MI) and patients' arrival at a hospital emergency department. The study employed a group-randomized trial design, with ten communities randomized from within matched pairs to each of two conditions. REACT also employed continuous data collection, based on surveillance of heart attack patients in community emergency departments. They analysed their data by comparing the mean slope for delay time in the ten intervention communities to the mean slope estimated in the ten control communities. Because no estimates of slope variation were available a priori, REACT was sized using approximations based on more traditional designs. In this paper, we present the slope and residual error variances as estimated from the REACT data and examine their influence on the power of the trial post hoc. We also examine the power of the trial as it would have been given a more traditional pretest-post-test design with analysis via a comparison of the net difference in condition means pretest vs post-test.