Coronary Model Shows Limits in Cath-Referred Patients

An interpretable machine-learning model classified angiographic coronary artery disease in patients referred for coronary angiography, but high disease prevalence and unclear inflammatory signals limited clinical interpretation.

By
Andrea Surnit
June 22, 2026
5 min

Conexiant

Objective:

To evaluate the performance of an explainable machine-learning model in predicting angiographic coronary artery disease (CAD) among patients referred for coronary angiography.

Approach:

Key Findings:

The final model achieved a balanced accuracy of 77% and an area under the curve of 0.815.
Hypertension, hyperlipidemia, sex, diabetes, and triglyceride levels were the most influential predictors.
Inflammatory biomarkers had unclear roles, with some showing inverse relationships with CAD.

Interpretation:

The model was selected for interpretability and balanced performance.

Limitations:

The model has not been validated in an independent external cohort.
It was developed in patients already undergoing coronary angiography, limiting its applicability to lower-prevalence populations.
The study's retrospective design and limited inflammatory biomarker panel may affect results.