Measures of association in clinical trials: definition and interpretation

Abstract

The choice of randomized controlled trials as the best evidence on therapeutic (or preventive) measures is based on the fact that this design is the only one that allows the establishment of causality, i.e. a relationship between an intervention and an outcome. For example, it is essential for the cardiologist treating patients with congestive heart failure (CHF) from systolic dysfunction of the left ventricle to know if the use of selective beta-blockers will reduce mortality. In order to find this out, he must have access to the published evidence on beta-blockers use in CHF, either through primary sources--RCTs--or secondary sources--systematic reviews. The results from clinical trials must be presented in a standardized way: in RCTs, authors usually present proportions (or rates, or percentages) of events (acute myocardial infarction, stroke, death) in each study group: experimental and control. These events are usually dichotomous, i.e. they are either present or not. In this paper, we will discuss the meaning as well as the interpretation of a number of measures of association in RCTs: absolute risk and absolute risk reduction, relative risk and relative risk reduction, odds ratios, number needed to treat and number needed to harm.