Effects of errors in classification and diagnosis in various types of epidemiological-studies.

THE AUTHORS' interest in the effect of misclassification on inferences made from epidemiological studies was stimulated by the results of a study, reported in 1958, in which one of the authors participated.' This study, made on a consecutive series of 192 male patients admitted to the Roswell Park Memorial Institute, was concerned with the comparison of their statements concerning their circumcision status and physical examination findings. The results are summarized in Table 1. We note that in 34.4 per cent of the patients, their statements, and the physical examination findings were not in agreement. Of the 84 patients found on examination to be circumcised, only 37 or 44.0 per cent said they were and of the 108 found on examination to be uncircumcised, 19 or 17.6 per cent said they were circumcised. Conversely, of the 56 patients who said they were circumcised, 37 or 66.1 per cent were found to be circumcised on examination and of the 136 patients who said they were not circumcised, 47 or 34.6 per cent were found on examination to be circumcised. Similiar results have been obtained in two subsequent studies by other investigators.2'3 This study of the validity of statements concerning circumcision status was conducted because of an interest in epidemiological studies of cancer of the cervix in which the lack of circumcision was suggested as being of etiological importance. As a result of these observations it was suggested that this type of misclassification may produce an observed association between cancer of the cervix and lack of circumcision, when in truth no association exists. This possibility can be illustrated with data reported by Wynder in his study on the relationship of circumcision to cancer of the cervix.4 Some data taken from the results of this study are shown in Table 2. In order to facilitate discussion and consideration of the issues the data have been simplified in that only two classes of circumcision status are considered. The actual situation was more complex in that there were pre-, extra-, and postmarital partners, and in the case of multiple marriages some women had been exposed to both circumcised and uncircumcised husbands. Let us assume that the degree of misclassification of circumcision status observed in the Roswell Park Memorial Institute study was present in the population studied by Wynder. Applying the proportions found there and