External validation of prediction models for obstructive coronary artery disease in patients with suspected stable coronary artery disease

Abstract

Abstract Background Pre-test probability (PTP) is an important tool in the diagnostic work-up for obstructive coronary artery disease (CAD) but must be calibrated to the declining disease prevalence in patients referred to diagnostic testing. Purpose To externally validate the published basic and clinical PTP models in a contemporary angina cohort with low prevalence of CAD and to compare with the reference European Society of Cardiology 2019 PTP (ESC 2019 PTP). Methods The validation cohort consisted of 42.328 patients (54% women, age ≥30 years, no previous CAD) with symptoms of CAD referred for cardiac computed tomography angiography in the western region of Denmark from 2008–2017 (3.3 million inhabitants). Obstructive CAD was defined from either invasive angiography as stenosis >50%, or when performed, from FFR <0.8 in coronary arteries with diameters >2 mm. The basic prediction model included type of angina, sex, and age, and the clinical model added diabetes, family history of CAD, and dyslipidemia. The ESC 2019 PTP was calculated from age, sex, and angina symptoms. Discrimination, calibration, and negative predictive value (NPV) were measured for all three models. Results Obstructive CAD was present in 3718 (8.8%). In the ESC 2019 PTP model, the basic model, and the clinical model 19.5%, 48.5%, and 55.7% were classified as very low risk and only 1.6%, 3.7%, and 3.5% of these had obstructive CAD, respectively (figure 1). Discrimination was similar for the three models with AUC of 0.76 (95% CI 0.75–0.77), 0.74 [0.73–0.75], and 0.76 [0.75–0.76], for the ESC 2019 PTP, basic, and clinical model, respectively. At the clinically relevant very low predicted probability (≤5%) of CAD, the clinical and basic model were very well calibrated, whereas the ESC 2019 PTP model overestimated the CAD prevalence. NPV at cut-off ≤5% were 98.4% [98.1–98.7] for the ESC 2019 PTP model, 96.3% [96.1–96.6] for the basic model, and 96.5% [96.3–96.7] for the clinical model. At cut-off <15%, NPVs were 96.1% [95.8–96.3] for the ESC 2019 PTP model, 94.5% [94.2–94.7] for the basic model, and 94.2% [94.0–94.5] for the clinical model. Conclusion In a population with a prevalence of 8.8% obstructive CAD, a clinical prediction model including diabetes, family history of CAD, and dyslipidemia in addition to the variables of the ESC 2019 PTP model ruled out 36.2% more patients than the ESC 2019 PTP model (23.592 vs. 8245 patients) while only overlooking 1.9% more cases of obstructive CAD when choosing a cut-off ≤5%. Use of this model is therefore potentially cost saving. Funding Acknowledgement Type of funding sources: None. Figure 1Figure 2