Misclassification in more than one factor in a case-control study: a combination of Mantel-Haenszel and maximum likelihood approaches.

Misclassification can arise in the case-control setting in both the risk factor of interest, and in a potential confounder. If true status is not ascertainable, but an independent replicated sample is available, adjusted maximum likelihood estimates of the individual binomial proportions can be obtained, and hence a relative risk estimate of the Mantel-Haenszel type constructed. An interval estimate on the latter can be calculated using the asymptotic properties of the likelihood function. The method is illustrated by application to a case-control study of breast cancer risk and such lifestyle attributes as diet, smoking and alcohol consumption, in which misclassification was anticipated and consequently repeatability data were collected.