ABSTRACT
Measured . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 468 14.6.2 Unmeasured Confounding . . . . . . . . . . . . . . . . . . . . . . . . . 469
14.7 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 471 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 472
ABSTRACT Electronic health records (EHRs) provide a huge opportunity for comparative effectiveness research (CER). As electronic renderings of health and health-related information, EHRs give researchers unprecedented access to data that can help address questions that would otherwise be very difficult to answer. Crucially, as data collected from EHRs often reflect large and diverse populations of individuals seeking care, EHR-based CER studies can be used to help guide patient decision-making in real-world settings. Notwithstanding their potential benefits, however, the use of EHR data for research purposes requires care. Specifically, since EHRs are typically implemented to support clinical and/or billing systems, researchers
interested in using EHR data for CER face challenges regarding (i) accessing the data, (ii) ensuring the data are research quality, and (iii) the control of confounding bias. Adding to these challenges is that, since there is no single universally accepted format, EHRs vary substantially in their structure and in the information they contain. This chapter reviews the current methods for addressing these challenges, focusing on the context of observational CER. While many of these methods borrow from the existing literature, novel methods that consider issues specific to EHRs, including their complex, highdimensional nature, have been developed in recent years. Progress has not been uniform, however, and the gaps that remain represent opportunities for the development of new innovative methods.
