Enhanced Detection of Heart Valve Disease Using Integrated Artificial Intelligence at Scale

The Annals of Thoracic Surgery Manuscript
Authors Daniel P. O’Hair, Hemal Gada, Miguel R. Sotelo, Loren Wagner, Cara M. Feind, Logan Brigman, Chris Rogers, and Navjot S. Kohli


Undertreatment of heart valve disease creates unnecessary patient risk. Poorly integrated healthcare data systems are unequipped to solve this problem. A software program using a rules-based algorithm to search the electronic health record for heart valve disease among patients treated by healthcare systems in the United States may provide a solution.


A software interface allowed concurrent access to picture archiving communication systems, the electronic health record, and other sources. The software platform was created to programmatically run a rules engine to search structured and unstructured data for identification of moderate or severe heart valve disease using guideline-reported values. Incidence and progression of disease as well as compliance with a care pathway were assessed.


In 2 health institutions in the United States 60,145 patients had 77,215 echocardiograms. Moderate or severe aortic stenosis (AS) was identified at a rate of 9.1% of patients (5474 and 6910 echocardiograms) in this population. The precision and accuracy of the algorithm for the detection of moderate or severe AS was 92.9% and 98.6%, respectively. Thirty-five percent of patients (441/1265) with moderate stenosis and a subsequent echocardiogram progressed to severe stenosis (mean interval, 358 days). In 1 sample 70.3% of moderate AS patients lacked a 6-month echocardiogram or appointment. The platform enabled 100% accountability for all patients with severe AS.


A rules-based software program enhances detection of heart valve disease and can be used to measures disease progression and care pathway compliance.