Abstract 15748: Resting Myocardial Computed Tomography Perfusion Provides Incremental Value Beyond Coronary Stenosis and Atherosclerotic Plaque Characteristics for Predicting Lesion Ischemia: A Machine Learning Approach
Introduction: Atherosclerotic plaque characteristics (APCs) including stenosis severity, positive arterial remodeling, low attenuation plaque, and spotty calcification by coronary CT angiography (CCTA) are associated with lesion-specific ischemia. Typically-acquired CCTA enables evaluation of first-pass contrast-enhanced myocardial attenuation patterns. Whether resting CCTA perfusion (CTP) augments identification of ischemia-causing coronary lesions is unknown.
Hypothesis: To evaluate the incremental value of resting CTP beyond APCs for determining lesion ischemia by fractional flow reserve (FFR).
Methods: 252 patients (mean age: 62.9±8.7 years, 70.6% men) with suspected CAD were examined. Coronary lesion ischemia was defined as ≤0.80 by invasive FFR, and CT stenosis ≥50% defined obstructive CAD. The number of APCs was classified as 0, 1, and ≥2 present. Resting CTP was ascertained using a gradient boosting classifier for supervised machine learning, and expressed as the percent likelihood (%) of ischemia. An optimal binary cut-point for percent likelihood (53.4%) was determined. Multivariable odds ratios (OR), area under the receiver operating characteristic curves (AUC), and category-free net reclassification improvement (cNRI) were estimated for predicting ischemia.
Results: Coronary lesion ischemia was identified among 129 (51.2%) patients. An increased likelihood of ischemia was observed for presence of obstructive CT stenosis (OR: 2.4, 95% CI: 1.2-4.5), increasing numbers of APCs (e.g., OR for ≥2 present: 9.1, 95% CI: 4.4-18.7), and abnormal resting CTP (OR: 2.9, 95% CI: 1.6-5.2). Resting CTP displayed improved discrimination (AUC: 0.82 vs. 0.79, P=0.034) and reclassification (cNRI: 0.49, P <0.001) for predicting ischemia beyond CT stenosis and number of APCs present.
Conclusions: The addition of resting CTP improves incremental utility for predicting coronary lesions that cause ischemia over and above CT stenosis and APCs.
Author Disclosures: J. Lee: None. G. Xiong: None. D. Han: None. A. Rizvi: None. H. Gransar: None. K. Elmore: None. B. Ó Hartaigh: None. F. Lin: None. J. Min: Research Grant; Modest; GE Healthcare, NIH/NHLBI R01 HL118019, NIH/NHLBI R01 HL115150, NIH/NHLBI R01 HL111141, NIH/NHLBI U01 HL105907, QNRF NPR-370-3-089. Ownership Interest; Modest; Autoplaq, MDDX. Consultant/Advisory Board; Modest; HeartFlow. Other; Modest; Arineta.
- © 2016 by American Heart Association, Inc.