Abstract 14965: A Multivariate Proteomic Classifier for the Presence of Obstructive Coronary Artery Disease in Symptomatic, Non Diabetic Patients Referred for Invasive Coronary Angiography
Background: Diagnosis of patients presenting with new symptoms consistent with obstructive coronary artery disease (CAD) remains challenging. Despite the existence of a number of non-invasive modalities for the assessment of CAD, the yield of CAD in patients who undergo invasive coronary angiography remains low. To address this issue we have constructed a classifier comprising protein biomarkers and clinical risk factors for the diagnosis of CAD in non-diabetic patients.
Methods: 14 proteins chosen from multiple biological pathways and previously shown to be associated with CAD were assessed in plasma from 496 non-diabetic patients (38% obstructive CAD) from the PREDICT study (NCT00500617). Cases were defined as patients with ≥ 50% stenosis in ≥ 1 major coronary artery, as determined by quantitative coronary angiography (QCA, 73% of patients) or as 70% stenosis by clinical read in the remainder. Logistic regression and cross-validation were used to examine models of varying complexity which included the 14 proteins and clinical risk factors; model complexity allowed for marker correlation, interactions between clinical factors and the proteins as well as non-linear effects. Model performance was assessed by corrected AIC (AICc) and AUC.
Results: A linear model consisting of the 14 protein markers including the mean values of three pairs of correlated proteins (S100A12-TNFAIP6, S100A8-MPO, and Adiponectin-APOA1), age, and sex displayed the best performance by AICc. The model had a cross-validated AUC = 0.79 (95% CI = 0.75-0.83) and showed good sensitivity and negative predictive value (89% and 91% respectively) using a threshold of 20% CAD probability. Comparison to a model containing only clinical covariates was highly significant (p < 0.001), indicating that the protein markers were independently associated with the presence of CAD.
Conclusion: We have constructed a multivariate protein model which is predictive of obstructive CAD in non-diabetic subjects referred for invasive angiography. The protein markers provide additional predictive ability, independent of clinical risk factors. This type of classifier may provide a low-cost approach for eliminating CAD as a source of symptoms in stable, symptomatic patients.
Author Disclosures: K. Fitch: Employment; Significant; CardioDx, Inc. A. Johnson: Employment; Significant; CardioDx, Inc. M. Doctolero: Employment; Significant; CardioDx, Inc. S. Rosenberg: Consultant/Advisory Board; Significant; CardioDx, Inc. J.A. Wingrove: Employment; Significant; CardioDx.
- © 2016 by American Heart Association, Inc.