Abstract 13469: A Risk Score Model to Identify Patients With High Likelihood of Left Main or 3-vessel Coronary Artery Disease

Abstract

Introduction: Revascularization improves outcomes only for stable coronary artery disease (CAD) patients with high-risk coronary anatomy (HRCA). While exercise stress test (EST) is the screening modality for CAD, identifying HRCA noninvasively is challenging. We aimed to derive a prediction model, incorporating clinical and EST characteristics, to screen for HRCA. Methods: We included adult patients who underwent EST followed by elective invasive coronary angiography (ICA) at our center between 2017 and 2019. HRCA was defined as left main (LM) or 3-vessel disease on ICA. Significant CAD was defined as ≥50% stenosis in the LM and/or ≥70% stenosis in the 3 main vessels. We excluded patients with acute coronary syndrome or ejection fraction <50%. Variables with p<0.1 at univariable analysis were selected as candidates to enter a multivariable stepwise backward regression model to derive a HRCA risk model. Results: 304 consecutive patients met inclusion criteria (64[13]years, 81.6% M). 15.5% were found to have HRCA on ICA. Median time from EST to ICA was 11[31]days. The following hit the threshold for inclusion into the multivariable model: hypertension, maximal predicted heart rate (MPHR), diabetes, rate pressure product, metabolic equivalents, exercise time, chronotropic response index ≥0.8, heart rate recovery, Duke Treadmill Score (DTS). After backward selection, hypertension (OR 17.08, p=0.008), MPHR (0.98, p=0.048), and very abnormal DTS (8.26, p=0.001) were the only significant predictors of HRCA ( Fig.1A ). The fully adjusted model showed good discrimination for HRCA (AUC 0.74, p<0.001, Fig.1B ). Conclusions: In an era of declining emphasis on EST in favor of imaging stress test, we identified a risk model based on clinical and EST-derived parameters, that can accurately predict likelihood of HRCA, enabling personalized risk-stratification. Further studies are needed to externally validate our model.