Development of methods for correcting for bias due to exposure misclassification in propensity scores:

This method is particularly useful in settings where there are a large number of potential confounders, such as research, electronic medical records and administrative databases. When the prevalence of the exposure of interest is higher than that of the outcome, using the propensity score to control for confounding may be more stable in finite samples than standard multivariate adjustment. We have derived estimators for the exposure-outcome association in a Cox proportional hazard model which have corrected the bias due to exposure misclassification when confounding is controlled for through a propensity score. We have derived the variance of the proposed estimators. An extensive simulation study showed satisfactory performance proposed point and interval estimators in finite samples. As an illustrative example, Dr. Wang is now working with Billing Hong, a Master level statistician in Dr. Spiegelman’s group to apply the method to estimate the association of PM2.5 and the risk of lung cancer based on a the Nurses’ Health Study (NHS). The bias in the association estimate due to measurement error in PM2.5 measurement will be corrected using the new method. A manuscript related to this work will be submitted this year.