Monte Carlo simulation of uncertainties in epidemiological studies: an example of false-positive findings due to misclassification

The 95% confidence intervals for the risk ratios (RR) reported in epidemiological studies reflect only sampling errors and do not include uncertainty caused by misclassification and confounding. Analysis of uncertainties in epidemiological studies can be improved using Monte Carlo simulations. For case-control studies, we show how differential misclassification of exposure status increases the probability of getting a statistically significant positive result. The misclassification error is relatively more important when several studies are pooled together. Simulations enable the uncertainties in epidemiologic results to be reported similarly to natural science where systematic and statistical uncertainties are carefully combined. We illustrate this by showing how false positives can result from misclassification