Abstract 14929: Strokes Prevented: Biosurveillance of NVAF Patient Cohorts CHA2DS2-VASc and HAS-BLED Scores Using Natural Language Processing and SNOMED CT

Introduction: Nonvalvular Atrial Fibrillation (NVAF), is estimated to affect 5.8 million people in the US. NVAF results in a five times greater stroke risk. This study compared the accuracy of structured ICD9 vs. electronic health record (EHR) data including clinical note text using Natural Language Processing (NLP), to identify NVAF cases and the CHA2DS2- VASc and HAS-BLED Scores. Methods: The retrospective EHR cohort study included patients of age 18 to 90 with a diagnosis of NVAF. Following application of the inclusion / exclusion criteria, an electronic model for structured data using ICD-9 criteria and for unstructured data using a NLP to SNOMED CT algorithm, a high throughput phenotyping system that rapidly assigns ontology terms to text in patient records, was applied to identify the NVAF population and their CHA2DS2-VASc and HAS-BLED Scores. A random sample of 300 patients was reviewed independently by two or three clinicians to create the gold standard NVAF cohort with CHA2DS2-VASc and HAS-BLED S...