. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Selecting the right cohorts and endpoints for the validation of pre-test probability models for obstructive coronary artery disease Simon Winther 1 *, Samuel Emil Schmidt 2 , Juhani Knuuti 3 and Morten Bøttcher 1 1 Department of Cardiology, Regional Hospital Unit West, Gl. Landevej 61, Herning 7400, Denmark; 2 Department of Health Science and Technology, Aalborg University, Fredrik Bajers Vej 7D2. 9220 Aalborg Øst, Denmark; and 3 Department of Clinical Physiology, Nuclear Medicine and PET and Turku PET Centre, Turku University Hospital, Kiinamyllynkatu 4-8, FI20520 Turku, Finland Online publish-ahead-of-print 20 June 2021 This commentary refers to ‘Validation of the European Society of Cardiology pre-test probability model for ob- structive coronary artery disease’, by S. Winther et al., https://doi.org/10.1093/eurheartj/ehaa755. and the discus- sion piece ‘Coronary artery stenosis prediction does not mean coronary artery stenosis obstruction’, by F. Derimay et al., https://doi.org/10.1093/eurheartj/ehab332. We thank Dr Rioufol for the interest in our paper. External validation of existing pre-test probability models for coronary artery disease (CAD) is important for the clinical implementation of these models. Naturally, any cohort used for such a validation task has some limi- tations and is often affected by several types of selection biases. In addition, the definition of CAD is also a crucial component in these validation studies. Not surprisingly, clinical site reading of coronary computed tomography angiography (CTA), core-lab reading of CTA, myocardial perfusion imaging, invasive angiography diameter stenosis or fractional flow reserve will provide different estimations of ob- structive CAD in the same population. Therefore, we agree with many aspects raised by Dr Rioufol. However, we disagree with some of the comments. In our study, we used a large real-world cohort to validate the ESC-2019 PTP model. 1 Only the patients without previously diag- nosed CAD were included. We chose an endpoint that is similar to the endpoint used for developing the ESC model. In our primary ana- lysis, we used a combined endpoint of CTA and invasive measure- ments which allowed both accurate exclusion of anatomical obstructive CAD (CTA or invasive stenosis diameter <50%) and confirmation of truly obstructive disease (CTA and invasive stenosis diameter >_50% with FFR included on a clinical indication). Even the use of FFR is included in the endpoint, this was a real-life clinical registry and it may happen that not all suspected lesions were inter- rogated by FFR. However, in our opinion, this is the best that can be achieved in clinical routine in a scale that can be used in defining PTP. In our other recent paper, we proposed a clinical likelihood model for obstructive CAD in which we incorporated also clinical risk fac- tors and coronary artery calcium score. 2 The Dan-NICAD study co- hort was also analyzed in that study and all patients with suspected stenosis at CTA underwent ICA with mandatory FFR measurements. That analysis also showed good calibration with the ESC-2019 model. 3 We do not see that defining the patients with <50% stenosis at CTA as having non-obstructive disease is a major source of bias. It is well known that CTA is overestimating the degree of stenosis and that the negative predictive value of CTA compared to invasive FFR is very high. 4 Therefore, we can reliably assume that these patients do not have obstructive CAD. We disagree also that it is methodologically questionable to calcu- late sensitivity and specificity for CAD if invasive coronary angiog- raphy (ICA) is not systematically used. In contrast, demanding that all patients should undergo ICA with further FFR would lead to severe population selection bias. Such population would not represent the real life patients which we face in outpatient clinics and in whom the PTP estimations are applied. The suggestion also ignores the fact that FFR is not without limitations. Even though FFR is regarded today as a reference standard for estimating hemodynamic impact of specific coronary lesion, its association with myocardial ischemia and the role in guidance of therapy warrants further studies. 5 Conflict of interest: S.W., S.E.S., and M.B. received support from Acarix in the form of an institutional research grant. M.B. discloses ad- visory board participation for NOVO Nordisk, Astra-Zeneca, Bayer, * Corresponding author. Tel: þ45 78430000, Email: sw@dadlnet.dk Published on behalf of the European Society of Cardiology. All rights reserved. V C The Author(s) 2021. For permissions, please email: journals.permissions@oup.com. European Heart Journal (2021) 42, 4402–4403 DISCUSSION FORUM doi:10.1093/eurheartj/ehab336 Downloaded from https://academic.oup.com/eurheartj/article/42/42/4402/6306827 by guest on 28 October 2022