Abstract 12: Accuracy and Validation of an Automated Electronic Medical Record Algorithm to Identify Patients with Atrial Fibrillation at Risk for Stroke

Background: Anticoagulation rates remain suboptimal for patients with atrial fibrillation (AF). Although these rates could potentially be improved with performance measures, there is no universal algorithm for identifying AF patients at risk for stroke using the electronic medical record. Methods: Patients with AF seen between 6/1/2011-5/31/2012 were identified. CHADS¬2 and CHA2DS2-Vasc scores were calculated with a broader algorithm that uses codes dating back 10 years, and a restrictive algorithm requiring comorbidity diagnoses within the past year. The accuracies of a claims-based AF definition and of each of four stroke risk classification algorithms (broad and restrictive each using CHADS¬2 and CHA2DS2-Vasc scores ≥2) were evaluated using data obtained from manual chart reviews for 300 patients. Electronic medication data were used to evaluate rates of anticoagulation among AF patients identified as at risk for stroke. Results: 6397 patients with prevalent AF were identified. Chart reviews confirmed a diagnosis of AF or atrial flutter in 95.7%. Patients with AF at risk for stroke were most accurately assessed using a restrictive algorithm based on a CHA2DS2-Vasc score ≥2 (PPV 97.5%, NPV 65.1%). The PPV of the broader algorithm based on CHADS2 score was 88.0%; 12% of those identified as high-risk electronically had CHADS2 scores <2 by chart reviews. Rates of anticoagulation evaluated using CHADS2 scores were identical using the broad (n=4498 with score ≥2, 58.3% anticoagulated) and restrictive algorithms (n=4008 with score ≥2, 58.3% anticoagulated). Anticoagulation rates calculated for patients with electronic CHA2DS2-Vasc scores ≥2 were similar based on broad comorbidity definitions (n=5711 with score ≥2, 56.0% anticoagulated) and restrictive comorbidity definitions (n=5533 with score ≥2, 56.0% anticoagulated). Discussion: Automated methods can be used to identify patients with prevalent AF indicated for anticoagulation, but may suffer from misclassification of up to 12%. Misclassification is minimized by requiring a diagnosis of AF within the prior year and using a CHA2DS2-Vasc based algorithm. Despite differences in accuracy between definitions, system-wide anticoagulation rates assessed using these definitions were similar. The diagnosis codes validated in this study can be applied for internal quality improvement and observational studies, and might be adapted for use in nationwide quality reporting programs.