To the Editor.
—In the April Archives (1983;143:753-755), Sacks et al quantitated the performance of randomized control trials (RCTs) and the historical control trials (HCTs) in a format used for the evaluation of tests, and reported that the sensitivity and specificity values were 0.90 and 0.11 for HCTs and 0.12 and 0.88 for RCTs, respectively. The authors considered that the sensitivity and specificity estimates "fell in narrow, nonoverlapping ranges and were quite similar no matter which therapies were assumed effective." Apparently, based on their findings, the authors thought that well-designed RCTs are better for determining the efficacy of new therapies.A nonbiased analysis of the data reported is made possible by using the receiver operating characteristics curves defined by the sensitivity and specificity pairs mentioned. The receiver operating characteristic analysis allows for the transformation of each pair into one value, the discrimination index (dI).1,2 The greater dI