Misclassification

Abstract

The accurate measurement of exposure prevalence, disease occurrence, and relevant covariates is necessary to assess causal relations between exposures and outcomes. However, in all epidemiologic research, there exists the opportunity for measurement errors. When the variables being measured are categorical, these errors are referred to as misclassification. We will consider two ways that measurement error can arise. The first is when information is not correctly recorded in the study database; these errors could be due to faulty instruments (e.g., the scale used to weigh people is incorrectly calibrated), respondents not answering truthfully to sensitive questions (e.g., their use of illegal substances), or mistakes entering the data in medical records (e.g., a wrong ICD code was entered in a Medicare claim). The second source of misclassification is conceptual and occurs when there is discordance between the study’s definition of the variable and the true definition. For example, this discordance can occur when the definition of exposure or a confounder is incorrectly operationalized with regard to dose, duration, or induction period. Studies of smoking and lung cancer may use smoking status at diagnosis, but this definition will cause some subjects to be incorrectly classified with respect to their etiologically relevant smoking status. Former smokers might have had their lung cancer risk affected by their smoking history, and current smokers might have initiated smoking so near their diagnosis that it would have had no impact on their lung cancer risk. With both types of errors, when analyses divide study participants into exposed and unexposed or diseased and undiseased, some respondents will be classified in the wrong category, which can bias results.