Journal of School of Public Health and

Misclassification of disease status and risk factors is one of the main sources of error in studies. Wrong assignment of individuals into exposed and non-exposed groups may seriously distort the results in case-control studies. This study investigates the effect of misclassification error on odds ratio estimates and attempts to introduce a correction method. Data on 3332 men aged 30-69 years from Tehran Lipid and Glucose Study (TLGS) were considered for investigating the relationship between angina pectoris and smoking. The “Rose” questionnaire was used to evaluate angina status. Two different definitions of smoking were included: I) active smokers versus non-smokers, II) active and ex-smokers versus never smoked. The relation between variables was assessed both with and without considering misclassification in smoking. Definition I (commonly used in many epidemiologic studies), resulted in a non-significant OR=1.09 (95% CI: 0.84, 1.41). Using the correction method based on the principle of non-differential misclassification, the OR increased to 1.13 (95% CI: 0.89, 1.43), which was still non-significant. However, a correction procedure that included the probability of differential misclassification produced a significant OR=1.46 (95% CI: 1.15, 1.85), p<0.05. It is evident that misclassification in risk factors can lead to inaccurate results. This study showed that the relation between variables may have not been discovered if the probability of misclassification was ignored. Moreover, in case of probable differential misclassification, applying correction methods for non-differential misclassification would be inefficient.