Key Issues in the Statistical Interpretation of Randomized Clinical Trials

Abstract

Traditional statistical analyses of randomized clinical trials typically use P values within a null hypothesis significance testing paradigm. Clinical adoption of randomized trials results, by guideline writers and individual clinicians, are often unequivocally determined by a deification of P < 0.05. Although the many limitations and large potential for misinterpretations with this approach have long been appreciated in the statistical literature, these issues are less frequently considered in the clinical literature. Discussions of these limitations might be viewed as esoteric, so that many cardiovascular specialists believe they are of little clinical importance. Using contemporary examples from the cardiovascular literature the potential, interpretative pitfalls in assessing “negative,” “positive,” and “noninferiority” trials are discussed accompanied by caveats on the statistical issues of the strength of the evidence and researcher degrees of freedom. Considerations such as effect size overestimation, robustness of results, statistical power issues, pitfalls in noninferiority trial interpretation, researcher degrees of freedom, and strength of evidence are discussed, using examples from well known trials in the areas of surgical and interventional management of coronary artery disease, heart failure management, and pharmacotherapy. The examples provided illustrate how even landmark trials in prestigious journals can present statistical problems that substantially undermine their conclusions or even make them unreliable. The goal of this article is to help clinicians more accurately reflect on and interpret research findings, by emphasizing the importance of the choice and interpretation of some of the current statistical techniques used in the analyses of randomized clinical trials.