ABSTRACT
Post-marketing data from clinical trials and observational studies are important for regulatory agencies “to monitor the safety of drugs after they reach the marketplace and to take corrective action if drugs risks are judged unacceptable in light of their benefits” (Institute of Medicine, 2012). Section 6.1 describes data from phase IV clinical trials and observational data from spontaneous reporting of adverse events by users of approved drugs. There are many challenges, which have in turn spurred the development of new methodologies, to analyze and infer from postmarketing data. The preceding chapter has considered challenges due to multiplicity and methods to address them for Tier 2 clinical data. In this chapter we consider challenges in, and approaches to, causal inference from post-marketing data. Section 6.2 begins with an introduction to causal inference from experimental studies and associated statistical models and methods. Section 6.3 focuses on non-experimental (observational) studies and causal inference methods in these studies. Section 6.4 addresses unmeasured confounding in observational studies by using instrumental variables (IVs) and study designs that provide natural substitutes for instruments. Section 6.5 introduces recent developments in structural causal models and symbolic causal calculus. The supplements and problems in Section 6.6 provide additional background for causal inference from experimental and non-experimental studies.
