Abstract 18578: Replication of a Risk Prediction Model for Ambulatory Incident Atrial Fibrillation Using Electronic Medical Record

Abstract

Introduction: Atrial fibrillation (AF) contributes to substantial increases in morbidity, mortality and increasing healthcare expenditures. Using data from community-based cohorts, CHARGE-AF investigators developed an AF risk prediction model. We aimed to replicate the CHARGE AF risk model using the Vanderbilt electronic medical records (EMR) repository. Methods: We identified a large ambulatory cohort of Caucasians and African Americans of age ≥ 40 years without a prior history of AF, followed by assessment of incident AF over a 5-year period (December 2005 to December 2010) using prospective chart review and validated AF algorithm. Baseline clinical variables (age, ethnicity, sex, height, weight, systolic and diastolic BP, PR interval, current use of antihypertensive medications, DM, MI, CHF, current smoking, and LVH on ECG) were extracted on or within 6 months prior to December 2005. We used multiple imputation for missing values and multivariable Cox proportional hazards analysis to construct our risk prediction model. Results: Our study cohort consisted of 63,772 subjects (age 58±11 years, 47% male, 90% Caucasians) with no prior history of AF. Over a mean follow up time of 3.9±0.95 years, 3,383 (5.3%) subjects developed AF in 2.6±1.3 years. In the multivariable model, older age, Caucasian ethnicity, male sex, taller height, greater weight, history of heart failure, LVH on ECG and prolonged PR interval were all significant risk predictors of AF (similar to CHARGE-AF risk model), whereas current use of antihypertensive medication was protective of developing AF (Table). The C-index from our model, 0.768, indicated relatively good discrimination. Conclusions: From clinical factors readily accessible in a large de-identified EMR database, we replicated the CHARGE AF risk prediction model to identify individuals at risk for developing AF in an ambulatory setting. This risk model may be used to target high risk-individuals for preventive measures.