Department of Biostatistics Seminar/Workshop Series
Performance of Summary Measures in Pharmacoepidemiology
Patrick Arbogast, PhD
Associate Professor, Department of Biostatistics, Vanderbilt University School of Medicine
Wednesday, October 20, 1:30-2:30pm, MRBIII Conference Room 1220
Propensity scores are widely used in cohort studies to improve performance of regression models when considering very large numbers of covariates. Recently, another type of summary score, the disease risk score, a variant of Miettinen’s multivariate confounder score, has been suggested. However, little is known how it compares to propensity scores. Monte Carlo simulations were conducted comparing regression models using the disease risk score, the multivariate confounder score, and propensity scores to models that directly adjust for all of the individual potential confounders. For moderate correlation between exposure and covariates, all three summary scores had performance comparable to that of the traditional regression models. For strong correlation between exposure and covariates, the multivariate confounder score and propensity scores had comparable performance with traditional regression models. When traditional methods may be subject to model misspecification, propensity scores and the multivariate confounder score had superior performance. All four models were affected by the number of events per covariate, but propensity scores and traditional multiple regression had better performance. These data suggest that for cohort studies, traditional multiple regression and propensity scores perform well, and when covariates are not highly correlated with exposure, disease risk and multivariate confounder scores are useful alternatives.