Predicting the likelihood of coronary revascularisation in a suspected acute coronary syndrome, in the era of machine learning.

Abstract

Background Accurate discrimination of patients who require early coronary revascularisation from a suspected acute coronary syndrome (ACS) population is challenging and relies heavily on coronary angiography. Alternatives, such as CT coronary angiography (CTCA) are safe, cost-effective and more readily available and may be suitable for low-moderate risk patients. Angiographic procedures that do not lead to early revascularisation are invasive and resource intensive. Thus, we aimed to develop and evaluate machine learning (ML) and logistic regression models to predict which ACS patients need early revascularisation in emergency departments (EDs) and to help guide urgent angiography referrals. Methods Prospectively collected data from 14,131 suspected ACS patients across 12 South Australian hospitals was analysed. Two logistic regression models were developed in Stata 18.1 to predict early revascularisation, with the second model incorporating a ML indicator for Type 1 myocardial injury (T1MI). ML models, including Random Forest, Neural Network and Extreme Gradient Boosting were developed in TensorFlow 2.17.0. Results The analysis included 10,470 participants, with 7329 in the training set and 3141 in the testing set. The logistic regression models achieved AUCs of 0.853 +/- 0.3% and 0.902 +/- 0.1%, while a Neural Network model reached an AUC of 0.939 +/- 0.2% for predicting revascularisation. Using the model, cost analysis showed that implementing CTCA for patients with moderate T1MI suspicion could increase the angiography-to-revascularisation rate by 25% and conservatively save approximately $ 29,747 across sites by avoiding 189 unnecessary angiograms. Using a ML model to more accurately select patients for early angiography could reduce rates of recurrent myocardial infarction and cardiovascular mortality within 6 months from 8.0% to 3.5% (p=0.02). Conclusion Logistic regression and ML models can help predict revascularisation needs in EDs, supporting safer, cost-effective decisions. While their use may reduce recurrent MI and cardiovascular mortality, external prospective validation is needed to assess clinical impact.Copyright © 2025 American College of Cardiology Foundation