Frederick K Korley1, Constantine Gatsonis2, Bradley S Snyder3, Richard T George4, Thura Abd5, Stefan L Zimmerman6, Harold I Litt7, Judd E Hollander8. 1. Department of Emergency Medicine, University of Michigan Medical School, Ann Arbor, MI, United States. Electronic address: korley@med.umich.edu. 2. Center for Statistical Sciences and Department of Biostatistics, Brown University School of Public Health, Providence, RI, United States. Electronic address: gatsonis@stat.brown.edu. 3. Center for Statistical Sciences, Brown University School of Public Health, Providence, RI, United States. Electronic address: bsnyder@stat.brown.edu. 4. Adjunct Faculty, Division of Cardiology, Department of Internal Medicine, Johns Hopkins University School of Medicine, Baltimore, MD, United States. Electronic address: rtgeorge3@gmail.com. 5. Division of Cardiology, Department of Internal Medicine, Johns Hopkins University School of Medicine, Baltimore, MD, United States. Electronic address: tabd1@jh.edu. 6. Department of Radiology, Johns Hopkins University School of Medicine, Baltimore, MD, United States. Electronic address: stefan.zimmerman@jhmi.edu. 7. Department of Radiology and Division of Cardiovascular Medicine, Department of Internal Medicine, Perelman School of Medicine of the University of Pennsylvania, Philadelphia, PA, United States. Electronic address: Harold.litt@uphs.upenn.edu. 8. Department of Emergency Medicine, Thomas Jefferson University, Philadelphia, PA, United States. Electronic address: judd.hollander@jefferson.edu.
Abstract
OBJECTIVE: We sought to derive and validate a model for identifying suspected ACS patients harboring undiagnosed significant coronary artery disease (CAD). METHODS: This was a secondary analysis of data from a randomized control trial (RCT). Patients randomized to the CTA arm of an RCT examining a CTA-based strategy for ruling-out acute coronary syndrome (ACS) constitute the derivation cohort, which was randomly divided into a training dataset (2/3, used for model derivation) and a test dataset (1/3, used for internal validation (IV)). ED patients from a different center receiving CTA to evaluate for suspected ACS constitute the external validation (EV) cohort. Primary outcome was CTA-assessed significant CAD (stenosis of ≥50% in a major coronary artery). RESULTS: In the derivation cohort, 11.2% (76/679) of subjects had CTA-assessed significant CAD, and in the EV cohort, 8.2% of subjects (87/1056) had CTA-assessed significant CAD. Age was the strongest predictor of significant CAD among the clinical risk factors examined. Predictor variables included in the derived logistic regression model were: age, sex, tobacco use, diabetes, and race. This model exhibited an area under the receiver operating characteristic curve (ROC AUC) of 0.72 (95% CI: 0.61-0.83) based on IV, and 0.76 (95% CI: 0.70, 0.82) based on EV. The derived random forest model based on clinical risk factors yielded improved but not sufficient discrimination of significant CAD (ROC AUC = 0.76 [95% CI: 0.67-0.85] based on IV). Coronary artery calcium score was a more accurate predictor of significant CAD than any combination of clinical risk factors (ROC AUC = 0.85 [95% CI: 0.76-0.94] based on IV; ROC AUC = 0.92 [95% CI: 0.88-0.95] based on EV). CONCLUSIONS: Clinical risk factors, either individually or in combination, are insufficient for accurately identifying suspected ACS patients harboring undiagnosed significant coronary artery disease.
RCT Entities:
OBJECTIVE: We sought to derive and validate a model for identifying suspected ACS patients harboring undiagnosed significant coronary artery disease (CAD). METHODS: This was a secondary analysis of data from a randomized control trial (RCT). Patients randomized to the CTA arm of an RCT examining a CTA-based strategy for ruling-out acute coronary syndrome (ACS) constitute the derivation cohort, which was randomly divided into a training dataset (2/3, used for model derivation) and a test dataset (1/3, used for internal validation (IV)). ED patients from a different center receiving CTA to evaluate for suspected ACS constitute the external validation (EV) cohort. Primary outcome was CTA-assessed significant CAD (stenosis of ≥50% in a major coronary artery). RESULTS: In the derivation cohort, 11.2% (76/679) of subjects had CTA-assessed significant CAD, and in the EV cohort, 8.2% of subjects (87/1056) had CTA-assessed significant CAD. Age was the strongest predictor of significant CAD among the clinical risk factors examined. Predictor variables included in the derived logistic regression model were: age, sex, tobacco use, diabetes, and race. This model exhibited an area under the receiver operating characteristic curve (ROC AUC) of 0.72 (95% CI: 0.61-0.83) based on IV, and 0.76 (95% CI: 0.70, 0.82) based on EV. The derived random forest model based on clinical risk factors yielded improved but not sufficient discrimination of significant CAD (ROC AUC = 0.76 [95% CI: 0.67-0.85] based on IV). Coronary artery calcium score was a more accurate predictor of significant CAD than any combination of clinical risk factors (ROC AUC = 0.85 [95% CI: 0.76-0.94] based on IV; ROC AUC = 0.92 [95% CI: 0.88-0.95] based on EV). CONCLUSIONS: Clinical risk factors, either individually or in combination, are insufficient for accurately identifying suspected ACS patients harboring undiagnosed significant coronary artery disease.