Considerations for the Analysis of Longitudinal Electronic Health Records Linked to Claims Data to Study the Effectiveness and Safety of Drugs KJ Lin 1,2,3 and S Schneeweiss 1,3 Health insurance claims and electronic health records (EHR) databases have been considered the preferred data sources with which to study drug safety and effectiveness in routine care. Linking claims data to EHR allows researchers to leverage the complementary advantages of each data source to enhance study validity. We propose a framework to evaluate the need for supplementing claims data with EHR and vice versa to optimize outcome ascertainment, exposure assessment, and confounding adjustment. BACKGROUND Data from premarketing trials are most valuable in establishing the efficacy of drugs. However, they are often considered of lim- ited value for decision making in routine care, because: 1) data derived from a selected population in clinical trials often fail to translate into effectiveness and safety profiles for patients encoun- tered in routine practice, which include a more vulnerable popu- lation with a wide range of comorbidities and adherence patterns; and 2) the study size and duration of follow-up consid- ered sufficient to demonstrate drug efficacy often do not confer adequate statistical power for some rare but serious safety out- comes. 1,2 Large drug-safety studies often need to be completed rapidly, as emerging adverse drug events could cause considerable public health damage. Computerized databases of health infor- mation have been considered preferred and cost-effective data sources for conducting pharmacoepidemiology research in rou- tine healthcare settings, 3 with claims databases and EHRs being the two most commonly used categories. 4 As by-products of rou- tine medical care for purposes other than drug safety investiga- tion, both claims data and EHR have inherent limitations. 5,6 In the US, most EHR systems capture only the healthcare activity provided by the institutions participating in the EHR network, but not that provided outside of the network. Therefore, depend- ing on a single EHR system for data may result in substantial misclassification of exposure, outcome, and confounding varia- bles. On the other hand, lack of detailed clinical data and health behavior information (e.g., body mass index (BMI), alcohol con- sumption, smoking status, etc.) in the claims data often compro- mises researchers’ ability to achieve adequate confounding adjustment. Linking EHR to claims could supplement the limita- tions of the individual databases and help combat various sources of bias. 7 In this review we compare the characteristics of EHR vs. claims data relevant to pharmacoepidemiology, summarize the existing literature on linking EHR to claims data to study drug safety and effectiveness, and propose a framework to evaluate when such linkage is most helpful in improving study validity, using published studies as examples. METHODOLOGY We searched PubMed, EMBASE, and Web of Science for studies conducted until January 6, 2016 with the following key words: [claims, administrative database, insurance claim review, or insur- ance claim reporting] AND [electronic health records, personal health records, personal medical record, computerized (6patient) record, automated health/medical record, EHR, or EMR] AND [drug therapy, pharmaceutical preparations, pharmacologic actions, drug-related side effects, and adverse reactions, drug interactions, comparative effectiveness research, drug, medication, safety, pharmacoepidemiology, pharmacology, or pharmaceuti- cal]. All of the above terms were entered as MeSH terms (Medi- cal Subject Headings; when a MeSH term was used to search for relevant literature, all the terms under that subject heading were searched as well. We also used exploded terms in EMBASE for similar purposes) and text words when appropriate. All entries retrieved by this strategy were examined to identify studies satisfy- ing the following predefined inclusion criteria: 1) The study must 1 Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women’s Hospital, Harvard Medical School, Boston, Massachusetts, USA; 2 Department of Medicine, Massachusetts General Hospital, Harvard Medical School, Boston, Massachusetts, USA; 3 Department of Epidemiology, Harvard School of Public Health, Boston, Massachusetts, USA. Correspondence: KJ Lin (jklin@mgh.harvard.edu) Received 23 November 2015; accepted 18 February 2016; advance online publication 25 February 2016. doi:10.1002/cpt.359 CLINICAL PHARMACOLOGY & THERAPEUTICS | VOLUME 100 NUMBER 2 | AUGUST 2016 147 REVIEWS