Detecting Bicuspid Aortic Valve From Echocardiographic Reports Using Natural Language Processing: A Veteran Affairs Study

Comprehensive Summary

Bowles et al. developed and validated a rule-based natural language processing (NLP) system to identify bicuspid aortic valve (BAV) from free-text echocardiographic reports within the U.S. Veterans Affairs health system. The model was trained on 555 manually annotated reports and evaluated on 170 held-out reports, achieving high accuracy for BAV detection (precision: 0.925, sensitivity: 0.939, F1-score: 0.932). When applied to 14,453,591 echocardiographic documents from 3,478,658 patients, the system identified 83,461 individuals (2.40%) with affirmed BAV, creating the largest BAV cohort reported to date. Comparison with the newly introduced BAV-specific ICD-10 code (Q23.81) showed 86.1% concordance, with manual review indicating that discordant cases largely reflected documentation or coding limitations rather than NLP errors.

Outcomes and Implications

This study demonstrates that unstructured echocardiographic text can be leveraged to accurately identify congenital valve morphology at scale, addressing longstanding limitations of structured diagnostic coding for BAV. The ability to assemble large, longitudinal BAV cohorts retrospectively opens new opportunities to study disease prevalence, progression, and outcomes, as well as patterns of surveillance and intervention. More broadly, this work highlights the value of NLP as a complementary approach to traditional EHR-based analyses and supports its application to other forms of structural heart disease where reliable diagnosis codes are absent or inconsistently used.