Improved estimates of discordance ratios in twin studies

In the study of monozygotic twins relative to disease and risk factors, particular interest focuses on the subset who are discordant for some suspected risk factor (for example, smoking), since such twins constitute a natural case‐control pair. In such studies, questionnaires designed to identify the status of all twin pairs are sometimes error prone and can yield misleading estimates of the concordance‐discordance ratios. Greater efforts to verify the characteristics of apparently discordant pairs than to verify those of apparently concordant pairs can result in the ‘unequal ascertainment’ fallacy. Using the results of a questionnaire with known error rates and the ‘apparent’ frequencies yielded, we present unbiased, maximum likelihood estimates of the ‘true’ proportions of concordant and discordant pairs. We also present approximate covariances among these estimates.