Abstract

AIMS: Accurate stratification of mortality risk is essential for management of chronic coronary syndromes (CCS), but existing models focus primarily on short-term outcomes and acute settings. We aimed to develop and validate a machine learning model (ISCHEMIA-PREDICT) for cardiovascular and all-cause death risk in CCS patients. METHODS AND RESULTS: Machine learning analysis of the ISCHEMIA randomized controlled trial (median follow-up 3.2 years) with trial and registry validation. The development cohort comprised 5179 CCS patients; external validation used a 23 303 patient multicentre registry (median follow-up 7.4 years). ISCHEMIA-PREDICT combined the glomerular filtration rate, exercise-induced ST-segment depression, exercise duration, systolic blood pressure, multivessel CAD, and invasive- vs. conservative-treatment strategy. Model discrimination was high for both cardiovascular [area under the curve (AUC) 0.94; 95% CI 0.92-0.96] and all-cause (AUC 0.94; 0.91-0.96) mortality and remained robust in registry-based external validation for all-cause mortality (AUC 0.82; 0.81-0.84). Cardiovascular mortality differed by 6.0% between highest and lowest quartiles (7.10 vs. 1.10%; hazard ratio 6.65; 3.15-14.03). CONCLUSION: The ISCHEMIA-PREDICT score provides a practical tool to stratify cardiovascular and all-cause mortality among CCS patients, enabling risk-guided clinical decision-making. REGISTRATION: ISCHEMIA ClinicalTrials.gov number, NCT01471522. LAY SUMMARY: We created and validated ISCHEMIA-PREDICT, a machine learning-based score using six routine clinical measures to estimate the risk of cardiovascular and all-cause death in people with chronic coronary syndrome. KEY FINDINGS: What we did and found: Using the randomized ISCHEMIA trial (5179 patients), we built a risk score from routinely available data-glomerular filtration rate, ST-segment depression and exercise duration on a stress test, systolic blood pressure, multivessel coronary disease, and initial treatment strategy (invasive vs. conservative). The score clearly separated lower- from higher-risk patients and showed strong discrimination. Its performance for all-cause death was confirmed in an independent, contemporary multicentre registry of over 23 000 patients with long-term follow-up. Why it matters for patients and clinicians: Because it relies on familiar information collected in everyday practice, ISCHEMIA-PREDICT can help clinicians quickly gauge mortality risk in chronic coronary syndrome patients and tailor care accordingly-supporting intensified medical therapy and consideration of invasive management in those at highest risk. A web calculator enables point-of-care use and risk-guided decision-making.