AHA Scientific Sessions 2026 Abstract
AI model outperformed heart disease risk tools in US records
The model outperformed conventional methods, with one-year scores of 0.946 in TriNetX and 0.945 in eMERGE for distinguishing future cases from noncases.The model outperformed conventional methods, with one-year scores of 0.946 in TriNetX and 0.945 in eMERGE for distinguishing future cases from noncases.
Computational studyBiomarkers
Who was studiedPeople with electronic health records, against LASSO, random forest, and PCE
What changedThe model predicted heart disease better than comparison models, with a one-year score of 0.946 in TriNetX
The catchPrediction scores measured model performance, not improvements in care or health
