Methodologic considerations in the design and conduct of randomized trials: the U.S. Physicians' Health Study

For treatments whose hypothesized risk reductions are small to moderate in size, on the order of 10%-30%, observational epidemiologic studies may not be reliable, since the amount of uncontrolled confounding could easily be as great as the likely reduction in risk. In such cases, randomized trials may be the only epidemiologic design strategy able to provide clear answers to research questions. However, the choice of design features and proper conduct of a randomized trial are crucial to the ability of the study to provide reliable results. In planning a study, five principal issues should be considered: design efficiency, completeness of follow-up, level of compliance, size of the probable risk reduction, and accumulation of sufficient numbers of outcome measures. Careful attention to these design considerations will enhance a trial's ability to provide definitive research results. The importance of these issues can be illustrated by examining the design and conduct of the ongoing Physicians' Health Study, a randomized, double-blind, placebo-controlled trial of aspirin in the reduction of cardiovascular disease and beta-carotene in the prevention of cancer.