302 Biomarker Profiling for Obstructive Coronary Artery Disease: A PROMISE Substudy

Abstract

Coronary artery disease (CAD) is a prevalent condition, frequently diagnosed by imaging studies. Laboratory-based adjuncts would be useful but currently play a limited role in diagnosis. We hypothesized that a multimarker blood-based panel could exclude obstructive CAD such that further noninvasive imaging would not be required. We used banked samples from the randomized PROMISE trial. Patients were symptomatic outpatients without diagnosed CAD referred for noninvasive cardiovascular testing for suspected CAD who were randomized to either coronary computed tomography (CT) or functional testing (exercise ECG or stress echocardiogram or SPECT imaging). Traditional clinical CAD risk factor information was collected including age, sex, race, and a history of hypertension, diabetes, dyslipidemia smoking, or family history of CAD. Baseline blood samples were collected, banked, and tested for a panel of 13 cardiovascular risk biomarkers (high sensitivity troponin I (hsTn), Cystatin C, Galectin-3, NT-pro-BNP, Creatinine, Homocysteine, Beta2, Lipoprotein A, Glucose, Insulin, ALT, Free T3, and Uric Acid) using least absolute shrinkage and selection operator (LASSO) regression analysis with 10-fold nested cross validation to estimate prediction performance. Our endpoint was obstructive CAD as defined by >70% stenosis on coronary CT. CT scans were read at a core lab for study purposes. We compared models using only clinical variables to those using all biomarker variables, hsTn alone, or either of these plus clinical variables. The PROMISE biomarker repository included 1,716 patients who were randomized to and received an evaluable coronary CT, with mean age 60.2 years, 52.5 % female and 87.3% white. On average, patients had 2.4 clinical risk factors for CAD and a ASCVD score of 14. A total of 102 patients had obstructive CAD; these patients were mostly male (71.6%) and smokers (65.7%). A predictive model using only clinical variables had similar negative predictive value (NPV), AUC, and Youden index (sensitivity + specificity -1) compared to a biomarker + clinical model. Similar results were found for hsTn in combination with clinical variables (Table 1). A biomarker-only model and hsTn alone had lower AUC and Youden index, with the former superior to the latter. In this population of symptomatic outpatients being evaluated for suspected CAD, clinical variables had only modest overall accuracy but good negative predictive value. A biomarker panel of 13 different analytes also had modest overall accuracy but was inferior to clinical factors alone, or hsTN alone. Combining the biomarkers with clinical data did not provide significant incremental accuracy to exclude CAD.Table 1Model Performance Summary for Obstructive CAD.ModelAUC (95% CI)NPV %, (95% CI)PPV %, (95% CI)YoudenSensitivity %, (95% CI)Specificity %, (95% CI)Clinical0.681 (0.630, 0.732)88.2% (85.6, 89.3%)11.1% (10.1, 16.3%)0.33866.0% (55.8, 75.2%)67.8% (65.5, 70.0%)Biomarkers0.604 (0.547, 0.662)85.7% (82.6, 87.0%)7.2% (6.6, 11.2%)0.17570.8% (60.7, 79.7%)46.7% (44.2, 49.1%)Clinical + Biomarkers0.691 (0.64, 0.743)89.2% (86.6, 90.2%)10.2% (9.3, 15.8%)0.34771.7% (61.4, 80.6%)63.0% (60.6, 65.4%)Hs Troponin I alone0.651 (0.598, 0.703)85.0% (81.9, 86.5%)8.8% (8.0, 12.9%)0.23762.7% (52.6, 72.1%)61.0% (58.7, 63.4%)Clinical + Hs Troponin I0.695 (0.642, 0.747)(85.2, 89.0%)88.0%10.7% (9.8, 16.2%)0.35871.4% (61.4, 80.1%)64.4% (62.0, 66.7%) Open table in a new tab