Detecting Bicuspid Aortic Valve From Echocardiographic Reports Using Natural Language Processing: A Veterans Affairs...

Area of Science:

• Cardiovascular Medicine
• Medical Informatics
• Computational Biology

Background:

• Bicuspid aortic valve (BAV) is the most prevalent congenital heart defect, often diagnosed late due to varied symptoms.
• Lack of a specific BAV diagnosis code before October 2024 hindered retrospective patient identification.

Purpose of the Study:

• To develop and validate a natural language processing (NLP) system for automated extraction of heart valve morphology from echocardiographic reports.
• To specifically focus on the accurate detection of bicuspid aortic valve (BAV).

Main Methods:

• A rule-based NLP system utilizing MedSpaCy was developed to analyze echocardiographic reports.
• The system was trained on 555 annotated reports and validated on 170 reports, focusing on valve leaflet structure identification.

Main Results:

• The NLP system demonstrated high performance for BAV detection with precision (0.925), sensitivity (0.939), and F1-score (0.932).
• Applied to over 14 million echocardiographic documents, the system identified 83,461 patients with BAV (2.40%).
• High concordance (86.1%) was observed with ICD-10 code Q23.81, with manual review confirming accuracy in discordant cases.

Conclusions:

• This NLP approach facilitates large-scale retrospective identification of BAV patients from clinical text.
• It establishes the largest BAV cohort to date, supporting future cardiovascular research.
• The system aids in clinical decision-making and timely diagnosis of BAV.