External Validation and Comparison of Two Clinical Prediction Models (PTP2013 and PTP2019) for Chest Pain in a Colombian Cohort

Abstract Aims The European Society of Cardiology (ESC) has proposed four pre-test probability (PTP) models for obstructive coronary artery disease (CAD). However, no studies have evaluated the diagnostic performance of any predictive model in the Latin American population. The aim of this study is to compare the PTP2013 and PTP2019 predictive models in order to determine which demonstrates a superior diagnostic performance for CAD in a cohort of Colombian patients. Methods A total of 408 patients who presented with chest pain and underwent coronary angiography (CA) and/or coronary computed tomography angiography (CCTA) at Fundación Santa Fe de Bogotá, between January 2019 and December 2023 were enrolled. Medical records were retrieved from the Hemodynamics and Radiology units. Pre-test probabilities were calculated for each patient using both the PTP2013 and PTP2019 models. CAD was defined as >50% stenosis on CA or CCTA. Each predictive model was assessed against CA and/or CCTA findings. The comparative performance of both models was evaluated. Results Prevalence of obstructive CAD of 24.9%. The PR2019 model underestimated the probability of CAD by 59%, whereas the PTP2013 model overestimated it by 35.6%. PTP2019 model yielded a C-statistic of 0.610 [95% CI: 0.544 - 0.676], while the PTP2013 model reported a C-statistic of 0.633 [95% CI: 0.570 - 0.696] (comparative p-value: 0.060). The net reclassification improvement was 14.7%). At a 15% threshold, the PTP2013 model demonstrated a sensitivity of 90% (82.38 - 95.10%), compared to 48% (37.9 - 58.22%) for the PTP2019 model. Conclusion The PTP2013 model is favored, as it showed higher sensitivity and a tendency to overestimate risk, in contrast to the PTP2019 model, which exhibited a concerning underdiagnosis of CAD. Consequently, the methodological challenge of identifying the predictive model with the highest diagnostic performance remains, highlighting the need to develop a tailored prediction model for the local population.