Diagnostic performance of combined biomarkers and phonocardiography vs. the 2024 ESC risk factor-weighted clinical likelihood model for detecting coronary artery disease

TL;DR

This study compares a new device for detecting heart disease with a clinical risk model, finding the device is good at ruling out disease but not better than existing methods.

Contribution

The study evaluates a novel non-invasive CAD detection system and its combination with biomarkers against a new clinical likelihood model.

Findings

The CADScor©System showed high sensitivity (93.8%) and negative predictive value (90.0%) for ruling out CAD.

Combining CADScore with hs-TnI did not significantly improve diagnostic accuracy over the CADScore alone.

The 2024 ESC RF-CL model remains a strong primary method for risk stratification.

Abstract

Non-invasive rule-out of coronary artery disease (CAD) using phonocardiography (CADScor©System) has been introduced as an alternative to the calculation of clinical risk-scores in populations with low-to intermediate CAD-likelihood. This study aims to (i) evaluate the diagnostic performance of the CADScor©System in a population with an intermediate CAD-likelihood and (ii) investigate whether a combined strategy of selected biomarkers and use of the device improves diagnostic accuracy compared to the 2024 ESC risk factor-weighted clinical likelihood (RF-CL) model.

A total of 167 patients with symptoms suggesting CAD and scheduled for coronary computed tomography angiography (CCTA) or elective invasive coronary angiography (ICA) were prospectively enrolled. Blood samples and heart sound recordings, obtained with the CADScor©System, were performed within 24 h before scheduled examinations. A resulting score (CADScore) ranging from 0 to 99 determined the risk of CAD. Diagnostic performance was calculated as sensitivity, specificity, negative predictive value, and positive predictive value. Diagnostic accuracy incorporating cardiovascular biomarkers was assessed using AUC-ROC analysis. Bootstrap validation (n = 2000 resamples) was performed to assess estimate stability. Net Reclassification Index (NRI) and Decision Curve Analysis (DCA) were conducted to evaluate reclassification performance and clinical utility. Obstructive CAD was detected in 56 (33.6%) patients. The diagnostic performance of the CADScor©System at a cut-off value of ≤20 was characterized by a negative predictive value of 90.0% (95% confidence interval [CI], 74.2–96.6), a positive predictive value of 37.8% (95% CI, 34.6–41.1), a sensitivity of 93.8% (95% CI, 82.8–98.7), and a specificity of 26.7% (95% CI, 18.4–36.5), respectively. Diagnostic accuracy of the CADScor©System was defined by an AUC of 0.743 (95% CI, 0.656–0.830), while the RF-CL model demonstrated an AUC of 0.719 (95% CI, 0.629–0.807, P = 0.559). Bootstrap validation confirmed estimate stability, with no statistically significant differences between methods. Net Reclassification Index analysis showed modest improvement with CADScore (Total NRI = 0.039), and decision curve analysis demonstrated clinical utility for both models. High-sensitivity troponin I (hs-TnI) was a significant predictor of obstructive CAD. Combining hs-TnI values with CADScore using logistic regression yielded an AUC of 0.768 (95% BCa CI, 0.670–0.842), showing no superior diagnostic performance compared to the RF-CL model.

The CADScor©System is a reliable non-invasive test method for ruling out obstructive CAD in a population with intermediate disease prevalence, characterized by high sensitivity (93.8%) and negative predictive value (90.0%). Bootstrap validation confirmed the stability of diagnostic performance estimates across all methods. While incorporation of hs-TnI as a predictor for obstructive CAD did not significantly improve diagnostic accuracy beyond the CADScore alone, Net Reclassification Index analysis revealed modest improvements in patient classification. Decision curve analysis demonstrated complementary clinical utility patterns between both models. The 2024 ESC RF-CL model remains a strong primary approach for risk stratification, while CADScore offers excellent rule-out capability.

EK 21-012-0221

Graphical AbstractFor image description, please refer to the figure legend and surrounding text.

Figures