Misclassification bias

Misclassification bias occurs in epidemiologic investigations when the infection or exposure status of individuals is incorrectly measured due to poor diagnostic tests or incorrect assessment of exposure. This results in biased measures of association (e.g. relative risk or RR, odds ratio or OR, etc.).

In epidemiologic investigations, the null hypothesis states that there is no association between the risk factor (e.g. modified-live respiratory viral vaccine on arrival in feeder calves) and the outcome of interest (e.g. bovine respiratory disease). Epidemiologic studies often attempt to disprove, or reject, the null hypothesis. Failure to reject the null hypothesis does not inherently mean no association exists, only that the data could not detect an association. It is important to understand that misclassification leads to bias in measures of association towards (i.e. RR or OR closer to 1) or away (i.e. RR or OR >>1 or <<1) from the null hypothesis.

Misclassification bias can be either differential or nondifferential. Differential misclassification occurs when the rate of misclassification differs across exposure groups. For example, in a study of risk factors for bovine anaplasmosis (BA), cattle producers who experienced losses from BA may be more likely to recall seeing ticks on their cattle, than producers who haven’t experienced loss. This is recall bias, and occurs when subjects under duress (i.e. cattle producers with sick or dead cattle) notice or recall details that other subjects wouldn’t, or may unintentionally exaggerate details if they believe they are related to the problem. As a result, unexposed cases may be incorrectly considered exposed cases more often than unexposed non-cases are considered exposed non-cases. The result may be an apparent association between seeing ticks on cattle and having cases of BA, when no association exists. Differential misclassification can lead to belief in a non-existent association, or failure to find an association between risk factor and outcome.

Nondifferential misclassification occurs when the rate of misclassification is the same across the exposed and nonexposed groups, and is often due to problems with data collection. Problems with diagnosis (e.g. false positives or false negatives) may result in nondifferential misclassification. The result of nondifferential misclassification is dilution of measures of association towards the null (e.g. RR or OR closer to 1, indicating little to no association).