Abstract 14364: External Validation of the CAD Consortium Prediction Models for the Presence of Coronary Artery Disease in Patients With Stable Chest Pain
Introduction: A better assessment of the probability of coronary artery disease (CAD) may improve the identification of stable chest pain patients who benefit from non-invasive testing.
Methods: To externally validate existing prediction models for the presence of CAD, we used stable chest pain patients in the PROMISE trial with CT angiography (CTA), invasive coronary angiography (ICA), or both. Obstructive CAD was defined as stenosis of ≥50% in ≥1 vessel on ICA. We assumed that patients with CTA showing 0% stenosis + coronary artery calcium (CAC) score = 0 were free of obstructive CAD, and we multiply imputed the remaining missing ICA results based on clinical variables and the CTA results for 2106 patients (61%). Predicted probabilities of CAD were calculated using published coefficients for three models: a basic model (age, sex, type of chest pain), a clinical model (basic model + diabetes, hypertension, dyslipidemia, and smoking), and a clinical + CAC score model. We assessed model discrimination, constructed calibration plots, and compared published predictor effects to observed predictor effects.
Results: The validation cohort included 3468 patients (1805 women; mean age 60 years; 779 (23%) with obstructive CAD on CTA). The models demonstrated moderate to good discrimination, with c-statistics (95%CI) of 0.69 (0.67-0.72), 0.72 (0.69-0.74), and 0.86 (0.85-0.88), for the basic, clinical, and clinical + CAC score models respectively. Calibration was satisfactory for all models even though the observed predictive effects of typical chest pain and diabetes were weaker, and the observed predictive effect of the CAC score was stronger than suggested by the prediction models. Among the 31% of patients in the validation cohort for whom the clinical model predicted a low (≤10%) probability of CAD, actual prevalence of CAD was 7%; among the 48% for whom the clinical + CAC score model predicted a low probability, the observed prevalence of CAD was 2%.
Conclusions: Existing prediction models for CAD in stable chest pain patients based on clinical information provide well-calibrated predictions and can identify those with a low probability of CAD. Obtaining the CAC score may be useful to further refine the estimate of the probability.
- Coronary artery disease
- Risk factors
- Coronary artery calcification (CAC)
- CT angiography
- Heart catheterization
Author Disclosures: T.S. Genders: None. A. Coles: None. U. Hoffmann: Research Grant; Significant; American College of Radiology Imaging Network, HeartFlow, Siemens Healthcare. M.R. Patel: Research Grant; Significant; Astra Zeneca, Jansen, HeartFlow. Consultant/Advisory Board; Modest; AstraZeneca, Jansen, Bayer, Genzyme. D.B. Mark: Research Grant; Significant; Eli Lilly & Company, Gilead, Astra Zeneca, AGA Medical, Bristol Myers Squibb, Merck & Company, Oxygen Therapeutics. Consultant/Advisory Board; Modest; Medtronic, Inc., CardioDx, St Jude Medical. K.L. Lee: None. E.W. Steyerberg: None. M. Hunink: Research Grant; Modest; European Society of Radiology (ESR). Other Research Support; Modest; European Institute for Biomedical Imaging Research. Other; Modest; Royalties for textbook from Cambridge University Press. P.S. Douglas: Research Grant; Significant; GE, HeartFlow.
- © 2016 by American Heart Association, Inc.