JMIR Publications

ABSTRACT

Background:

Valvular heart disease (VHD) is a leading cause of cardiovascular morbidity and mortality. Administrative diagnostic codes are unable to capture the completeness of VHD.

Objective:

To develop a natural language processing (NLP) algorithm to identify patients with valve stenosis and regurgitation from echocardiography reports within a large integrated healthcare system.

Methods:

We utilized the reports from echocardiograms performed at Kaiser Permanente Southern California between 1/1/2011-12/31/2022. Related terms/phrases of heart valve stenosis and regurgitation and their severities were compiled from literature and enriched with input from clinicians. An NLP algorithm was iteratively developed and fine-trained via multiple rounds of chart review followed by adjudication. The developed algorithm was applied to 200 annotated echocardiography reports to assess its performance, then the study echocardiography reports.

Results:

At report level, valve lesions identified included 111,300 (9.1%) aortic stenosis, 20,246 (1.7%) mitral stenosis, 397 (0.03%) tricuspid stenosis, 2,585 (0.2%) pulmonic stenosis, 345,115 (28.2%) aortic regurgitation, 802,103 (65.5%) mitral regurgitation, 903,965 (73.8%) tricuspid regurgitation, and 286,903 (23.4%) pulmonic regurgitation. Males had a higher frequency of aortic stenosis and all four valvular regurgitations while females had more mitral stenosis, tricuspid stenosis, and pulmonic stenosis. Non-Hispanic whites had the highest frequency among all four valvular stenosis and regurgitations. Frequencies of aortic stenosis, mitral stenosis, and regurgitation of all heart valves were increased with age. Validation of the NLP algorithm against the 200 annotated echocardiography reports showed excellent precision, recall, and F1-scores.

Conclusions:

The developed and validated NLP computerized algorithm could support clinical research and monitoring of VHD.