Abstract 17956: External Validation of a Prediction Model for the Development of Atrial Fibrillation in a Repository of Electronic Medical Records

Background: Atrial fibrillation (AF) contributes to substantial morbidity, mortality, and healthcare costs. Accurate prediction of incident AF might enhance patient care and improve outcomes. We aimed to externally validate the AF risk model developed by the CHARGE-AF investigators utilizing a large repository of electronic medical records (EMR). Methods: Using a database of de-identified EMRs, we conducted a retrospective cohort study of subjects serially followed in internal medicine clinics at our institution (minimum 3 visits in a 24 month window). Subjects were followed for incident AF from 2005 until 2010. We applied the published CHARGE-AF Cox proportional hazards model beta coefficients to our cohort. Predictors included age, race, height, weight, systolic and diastolic blood pressure, treatment for hypertension, smoking status, diabetes, heart failure, history of myocardial infarction, left ventricular hypertrophy, and PR interval. Calibration and discrimination were assessed by generating calibration plots and calculating C-statistics. Results: The study included 33,494 subjects with median age 57 years (25th to 75th percentile: 49 - 67), 57% women, 86% whites, and 14% African Americans. During the mean follow-up period of 4.8 ± 0.85 years, 2455 (7.3%) subjects developed AF. After correcting for baseline hazard, the CHARGE-AF model over-predicted AF at the highest risk deciles but was otherwise well-calibrated (Figure) and showed good discrimination, with a C-statistic of 0.746 (95% confidence interval: 0.738 to 0.754). Conclusion: From clinical factors readily accessible in a large de-identified EMR repository, we externally validated the CHARGE-AF risk prediction model to identify individuals at risk for developing AF in an ambulatory setting. These data not only provide strong validation for the CHARGE-AF risk prediction tool, but also indicate that the tool, and thus primary prevention strategies, can be implemented in an EMR context.