Misclassification Models

Abstract

AbstractClassification errors in categorical data may distort the results of statistical analyses. Misclassification models have been developed in various areas of biostatistical application to account for this distortion. This article reviews selected methods that have been proposed for (i) modeling the natural history of a disease or growth process, (ii) evaluating diagnostic tests, and (iii) identifying epidemiological risk factors. We consider methods that incorporate known estimates of misclassification rates, methods that estimate misclassification rates from a subsample in which the presumably “true” classification is known, and methods in which both the true process and the misclassification rates are simultaneously modeled from misclassified data. The various methodological approaches are discussed along with their limitations, and we note an important issue for additional research.