Designing case-control studies: decisions about the controls.

The authors quantified, first, the effect of misclassified controls (i.e., individuals who are affected with the disease under study but who are classified as controls) on the ability of a case-control study to detect an association between a disease and a genetic marker, and second, the effect of leaving misclassified controls in the study, as opposed to removing them (thus decreasing sample size). The authors developed an informativeness measure of a study's ability to identify real differences between cases and controls. They then examined this measure's behavior when there are no misclassified controls, when there are misclassified controls, and when there were misclassified controls but they have been removed from the study. The results show that if, for example, 10% of controls are misclassified, the study's informativeness is reduced to approximately 81% of what it would have been in a sample with no misclassified controls, whereas if these misclassified controls are removed from the study, the informativeness is only reduced to about 90%, despite the reduced sample size. If 25% are misclassified, those figures become approximately 56% and 75%, respectively. Thus, leaving the misclassified controls in the control sample is worse than removing them altogether. Finally, the authors illustrate how insufficient power is not necessarily circumvented by having an unlimited number of controls. The formulas provided by the authors enable investigators to make rational decisions about removing misclassified controls or leaving them in.