Aernoud Fiolet

338 Chapter 14 ABSTRACT Importance: Evaluations of healthcare interventions in clinical trials are notoriously expensive. If routinely collected electronic healthcare data are accurate enough to serve as trial endpoint data, leveraging these data may reduce costs of clinical research and facilitate trials on a larger scale Objective: Investigate the accuracy of using routinely collected healthcare data from electronic healthcare records (EHRs) to identify major cardiovascular events in a multicenter randomized clinical trial. Design: A validation study. Setting: Participants of four sites of a multicenter cardiovascular outcome trial identified in EHR. Participants: Patients with chronic coronary disease. Exposures: Collection of cardiovascular endpoints using semi-automated text- mining software and investigator-reported endpoints as reference standard. Main Outcomes and measures: The primary endpoint was the composite of cardiovascular death, myocardial infarction, or ischemic stroke (MACE). Secondary endpoints were the composite of MACE plus ischemia-driven coronary revascularization (MACE+) and the individual components of the composite endpoints. Sensitivity and specificity were used as measurements of diagnostics accuracy and Kaplan-Meier curves of cumulative event-free survival probabilities were used to compare endpoints collection methods. Results: For the 945 trial participants, 154 endpoints were reported by trial investigators. EHR data retrieval identified 130 (84.4%) of these endpoints: 47 of the 51 investigator-reported MACE (sensitivity 92.2%, specificity 86.5%) and 94 of the 114 investigator-reported MACE+ (sensitivity 82.5%, specificity 81.8%). Cardiovascular death was identified in all six cases (100%). EHR data retrieval identified 36 of 38 investigator-reported myocardial infarctions (sensitivity 94.7%, specificity 85.8%), five of seven investigator-reported ischemic strokes (sensitivity 71.4%, specificity 99.3%) and 47 of the investigator-reported ischemia- driven coronary revascularizations (sensitivity 74.6%, specificity 96.7%). Positive