Pre-test probability prediction in patients with a low to intermediate probability of coronary artery disease: a prospective study with a fractional flow reserve endpoint.

Aims European and North American guidelines currently recommend pre-test probability (PTP) stratification based on simple probability models in patients with suspected coronary artery disease (CAD). However, no unequivocal recommendation has yet been established. We aimed to compare the ability of risk factors and different PTP stratification models to predict haemodynamically obstructive CAD with fractional flow reserve (FFR) as reference in low to intermediate probability patients. Methods and results We prospectively included 1675 patients with low to intermediate risk who had been referred to coronary computed tomography angiography (CTA). Patients with coronary stenosis were subsequently investigated by invasive coronary angiography (ICA) with FFR measurement if indicated. Discrimination and calibration were assessed for four models: the updated Diamond-Forrester (UDF), the CAD Consortium Basic, the Clinical, and the Clinical+Coronary artery calcium score (CACS). At coronary CTA, 24% of patients were diagnosed with a suspected stenosis and 10% had haemodynamically obstructive CAD at the ICA. Calibration for all CAD Consortium models increased compared with the UDF score. However, all models overestimated the probability of haemodynamically obstructive CAD. Discrimination increased by area under the receiver operating curve from 67% to 86% for UDF vs. CAD Consortium Clinical+CACS. The proportion of low-probability patients (pre-test score< 15%) was for the UDF, CAD Consortium Basic, Clinical, and Clinical+CACS: 14%, 58%, 51%, and 66%, respectively. The corresponding negative predictive values were 97%, 94%, 95%, and 98%, respectively. Conclusion CAD Consortium models improve PTP stratification compared with the UDF score, mainly due to superior calibration in low to intermediate probability patients. Adding the coronary calcium score to the models substantially increases discrimination.