Aernoud Fiolet

340 Chapter 14 INTRODUCTION D rug research and development comes at high costs, among others due to the scale of randomized controlled trials (RCTs) with clinical outcomes. Especially in populations with low to moderate absolute risk for study endpoints, over 10,000 participants may be required to detect a clinically relevant treatment effect. Costs of such trials can rise up to 45,000 US dollars per participant. 1,2 The number of new drugs approved relative to the spending on research and development consequently halves every ten years, compromising innovation in drug research. 3,4 Lowering costs in drug research could reduce barriers for all trial phases, facilitate head-to-head comparisons of treatments, and improve patient care. 5 Data collection is a major cost component in trial conduct, among others driven by personnel costs and site monitoring costs. 6–9 Current practice in trials is to collect data during planned patient-clinician interactions, after which data are registered manually in (electronic) clinical report forms. Most of this information is already collected in electronic healthcare records (EHRs) during routine practice. 9 Extracting information from EHRs may complement or even substitute some parts of data collection in trials. 9,10 However, the use of EHR data in RCTs is limited to date. 11 Aprerequisite to use EHR extracted information in trials is the accurate recognition and mining of this data. 9 Commonly, accuracy of endpoint data collection in clinical trials is safeguarded by strict monitoring, and reporting of potential events by patients and investigators. Best practice prescribes blinded adjudication by an adjudication committee to homogenize endpoint assesment. 12 While studies have investigated the accuracy of insurance and registry data for cardiovascular outcome trials, investigations of the accuracy of data extracted from EHRs are limited. 13,14 This study aimed to investigate the accuracy of using routinely collected healthcare data from EHRs to identify major cardiovascular events for a multicenter randomized clinical trial. METHODS This study compared conventional and semi-automated EHR text-mining methods for the collection of endpoint data as a post hoc analysis of the Low-Dose Colchicine 2 (LoDoCo2) trial (Figure 1).