8000012 Information bias: misclassification and mismeasurement of exposure and disease

<h3>Objective</h3> To present methods for evaluating bias related to misclassification and mismeasurement of exposure. <h3>Methods</h3> Potential bias due to misclassification and mismeasurements of exposures and disease are a serious concern in nearly all epidemiologic studies. This presentation will describe methods that may be used to evaluate the direction and magnitude of the bias. <h3>Results</h3> Errors in exposure may be differential or non-differential with respect to disease. It is commonly assumed that non-differential errors in exposure bias results towards the null. However, the direction of the bias is determined by the type of exposure metric and the error model. Non-differential errors of binary exposure variables are expected to bias results towards the null, but this is only an expectation, and results may be biased in either direction. Non-differential exposure misclassification using several categories of exposure may result in inflated estimates in intermediate categories, and underestimates of risk in the highest category. Errors in continuous exposure measures are expected to bias the results towards the null when they are ‘Classical’ and are not expected to bias study findings when they are ‘Berksonian’. Methods are available for evaluating the potential direction and magnitude of bias from misclassification and mismeasurement errors using validation data. Alternatively, one can conduct sensitivity analyses with assumptions about sensitivity and specificity. These methods can be extended to analyses that are multl-dimensional, and probabilistic. <h3>Conclusion</h3> Misclassification and mismeasurement of exposure is common in epidemiology research. Methods are available to evaluate the direction and magnitude of potential biases resulting from these errors.