A pooled multi-national validation study of a machine learning, high-sensitivity troponin-based multi-proteomic model to predict the presence of obstructive coronary artery disease

Abstract

Abstract Background Undetected obstructive coronary artery disease (oCAD) is a global health problem associated with significant morbidity and mortality. A need exists for an accurate and easily accessible diagnostic test for oCAD. Using machine learning, a multi-biomarker blood diagnostic test for oCAD based on high-sensitivity cardiac troponin-I (hs-cTnI) has been developed. Purpose To validate the performance of a previously developed, algorithmically weighted, multiple protein diagnostic panel to diagnose oCAD in a pooled multi-national cohort and to compare the diagnostic panel's performance to predict oCAD to hs-cTnI alone. Methods Three clinical factors (sex, age, and previous coronary percutaneous intervention) and three biomarkers (hs-cTnI, Adiponectin, and Kidney Injury Molecule-1) were combined. hs-cTnI blood samples were assayed on the Siemens Atellica and Abbott Diagnostics ARCHITECT immunoassay platforms. Adiponectin and Kidney Injury Molecule-1 were measured with a multiplex assay on blood samples via the Luminex 100/200 xMAP platform. Individual data from a total of 924 patients with a mixture of acute and lesser acute presentations from three centers were pooled (Table 1). oCAD was defined as >50% coronary obstruction in at least one coronary artery (for the University Hospital Hamburg-Eppendorf cohort) or >70% coronary obstruction in at least one coronary artery (for the other two cohorts). The multiple biomarker diagnostic panel's performance to predict oCAD was also compared to hs-cTnI alone. Results The multiple protein panel had an area under the receiver-operating characteristic curve of 0.80 (95% CI, 0.77, 0.83, p<0.001) for the presence of oCAD (Figure 1). At optimal cutoff, the score had 74% sensitivity, 72% specificity, and a positive predictive value of 81% for oCAD. The multiple biomarker panel had a diagnostic odds ratio of 7.48 (95% CI 5.55, 10.09, p<0.001). In comparison, in patients without an acute MI, hs-cTnI alone had an area under the receiver-operating characteristic curve of 0.63 (95% CI, 0.60, 0.67, p<0.001)) for oCAD (Figure 1). Conclusions In this multinational pooled cohort, a previously described novel machine learning, multiple biomarker panel provided high accuracy to diagnose patients for oCAD. Funding Acknowledgement Type of funding sources: Private company. Main funding source(s): Prevencio, Inc. Table 1. Pooled Variable DataFigure 1. ROC for HART CADhs and hs-cTnI