In a retrospective study of 4,656 patients undergoing coronary angiography for suspected CAD, an explainable machine learning model improved CAD detection with an AUC of 0.815, with SHAP analysis highlighting hypertension, hyperlipidaemia, sex, triglycerides and interleukin-6 as the top predictors. The study suggests that pairing inflammatory biomarkers with traditional risk factors may sharpen early CAD diagnosis while keeping model decisions interpretable for clinicians.

Why It Matters To Your Practice

• Inflammatory markers may add diagnostic value beyond standard risk factors in patients being worked up for CAD.

• IL-6 and triglycerides emerged as high-impact features, suggesting a potential role for routinely available metabolic data plus selective biomarker use.

• An interpretable model may be easier to trust and discuss than a black-box tool when supporting referral or testing decisions.

Clinical Implications

• The final generalised linear model delivered balanced performance rather than pursuing maximal complexity.

• HTN was among the most influential predictors, reinforcing that familiar clinical variables still carry major signal in AI-assisted CAD assessment.

• If validated externally, similar tools could help prioritize patients with suspected CAD before or alongside invasive evaluation.

Insights

• Investigators started with demographic, clinical, biochemical and 12 inflammatory cytokine variables, then narrowed to 21 candidate features using LASSO regression.

• Ten machine learning algorithms were trained with cross-validation before selecting the final model.

• SHAP explainability was used to show which variables most influenced predictions, helping translate model output into clinically understandable drivers.

The Bottom Line

• For clinicians interested in AI, this study offers a practical example of explainable ML that combines standard risk factors with inflammation signals to detect CAD.

• The immediate takeaway is not to order IL-6 broadly today, but to watch for validated tools that use interpretable biomarker-plus-clinical models to improve diagnostic triage.