LIMITATIONS OF RELATIVE SENSITIVITY IN DETECTING DIFFERENTIAL MISCLASSIFICATION IN CASE-CONTROL STUDIES

Apparent relative sensitivity, based on an investigator's external sta ndard, is the ratio of observed case to control exposure sensitivity. An apparent relative sensitivity different from 1.0 is usually interpr eted as evidence for differential misclassification of exposure status . We undertook this in vestigation to determine the conditions under w hich an apparent: relative sensitivity exceeding 1.0 is actually due t o differential misclassification. We also consider whether apparent re lative sensitivity correctly quantifies the degree of differential mis classification. To achieve these goals, we derived an algebraic relati on involving apparent relative sensitivity, true sensitivities and spe cificities, true odds ratio, an index of how well the external standar d classifies true exposure, and the incidence of the disease among the nonexposed. We found that an apparent relative sensitivity greater th an 1.0 correctly indicates differential misclassification when either (1) the investigator's external standard classifies true exposure perf ectly, or (2) the investigator's external standard is imperfect, but t he true odds ratio equals 1.0, true relative sensitivity is greater th an 1.0, and true relative specificity is less than 1.0. We also found that apparent relative sensitivity greater than 1.0 falsely suggests d ifferential misclassification when true relative sensitivity equals 1. 0, the investigator's external standard is imperfect, and the true odd s ratio is greater than 1.0. Furthermore, even when apparent relative sensitivity correctly detects the presence of differential misclassifi cation, it may misrepresent the degree.