Abstract: Objective Causal inference using observational data has long been the focus of epidemiologic researchers. In pharmacoepidemiologic studies, exploring effect of a drug or adverse drug reaction caused by a drug also fits into causal inference. Targeted maximum likelihood estimation (TMLE) proposed by van deer laan is proved of desirable properties, we introduce its principle and application in the study. Method The common methods for causal inference include inverse probability of treatment weighting (IPTW), G-formulation and some double robust estimators. We provide an overview of the theory, model, estimation, statistical inference and properties. Meanwhile, we also compare the performances of several causal effects estimators. Result Compared with other methods, TMLE has certain advantages. However, its implementation is difficult. Conclusion TMLE is a method with double robustness and could also produce efficient unbiased estimator of targeted parameter. There is potential value of TMLE enhancing drug risk management in active drug safety surveillance.

Key words: targeted maximum likelihood estimation, causal inference, pharmacoepidemiolog, active surve illance of adverse drug reactions