Demonstration of information and confounding bias with simulated outbreak data.

<p>(A) Muizenberg Mathematical Fever 2011 outbreak data to illustrate how using a case definition with imperfect sensitivity (symptomatic disease) can cause nondifferential misclassification bias (the category of information bias where exposed and unexposed individuals are equally likely to be misclassified). Nondifferential misclassification biases the association between a risk factor and a disease outcome towards the null hypothesis of no association (odds ratio = 1). While attendance at the prior year's clinic was actually protective (black square and 95% CI), this bias was sufficient to cause the confidence interval for the odds ratio of this very protective variable to overlap (gray). (B) illustrates how a risk factor (arrival a day or more early to the clinic) that has no real association to a disease outcome can appear associated through confounding. Individuals who had attended the clinic in prior years were less likely to come to the clinic early and were also protected (i.e., A). Consequently, early attendance appeared associated with a higher risk of disease in a univariate analysis (gray) though the CI contains the null hypothesis of no association in a multivariate analysis that adjusts for prior attendance (black). The script for production of and further detail on this figure are given in <a href="http://www.plosbiology.org/article/info:doi/10.1371/journal.pbio.1001295#pbio.1001295.s018" target="_blank">Text S3</a>.