Abstract 14347: Rapid Calculation of Cardiology Risk Scores by Natural Language Processing of Unstructured Medical Record

Abstract

Introduction: Risk score calculators such as ASCVD (for coronary artery disease), CHA2DS2-VASc (for atrial fibrillation), and H2FPEF (for diastolic heart failure) optimize outpatient cardiology care, but may not be routinely performed because calculation is time consuming. Natural Language Processing (NLP) methods can rapidly capture clinical data from unstructured medical records. Our aim was to create an algorithm that can rapidly calculate these scores using NLP methods. Methods: In the algorithm we created (named MYTH), demographic, history and physical, ICD10 diagnostic codes, laboratory, EKG, echocardiography data were captured by analysis of textual syntax, domain-specific knowledge graph, and regular expressions. The data captured are then fed into various risk score calculators. We compared the speed of the algorithm with manual calculation using a commonly used online calculator. Twenty outpatient cardiology unstructured medical records were used for each of the three risk scores. Results: NLP is faster than manual calculation. See attached table. Conclusions: NLP-based algorithm increases the speed of risk score calculation. The improvement is especially dramatic in complex risk scores, such as diastolic heart failure risk scores. NLP method can optimize cardiology care in time-constrained settings.