Analyzing a single electrocardiogram (ECG) taken before surgery with artificial intelligence (AI) can effectively identify patients at high risk of death after surgery, as well as patients with a low need for detailed cardiac examinations, according to a research finding.
Seoul National University Bundang Hospital said on the 20th that a research team led by cardiology professors Choi Hong-mi and Cho Young-jin, resident Kim Ye-rin, and anesthesiology professor Song In-ae reached these findings after analyzing data from 46,000 non-cardiac surgeries performed in 2020-2021.
An ECG is a test that detects the electrical signals needed for heart contraction through electrodes attached to the skin, recording them as waveforms on a graph. While it is useful for non-invasively observing the electrical activity of the heart to diagnose conditions such as arrhythmia and coronary artery disease, it has faced the limitation that subtle differences and complex patterns are difficult to fully interpret with the naked eye. Professor Cho, noting that AI could quantify even risks that were difficult to detect through conventional readings, previously developed "ECG Buddy" together with Kim Joong-hee, professor of emergency medicine at Seoul National University Bundang Hospital. ECG Buddy is an AI-based ECG analysis solution that evaluates arrhythmia, emergency situations, and cardiac function abnormalities by analyzing ECG images with a smartphone. In this study, the various heart disease-related indicators provided by ECG Buddy were applied to pre-surgery patient evaluation. This is because all surgeries, not just cardiac surgery, can place strain on the cardiovascular system throughout the entire process from anesthesia to the operation itself, making it necessary to accurately assess risks in advance.
Based on the AI ECG, the research team calculated an "AI Critical score" (QCG-Critical score) representing the risk of severe disease occurring, and then evaluated the accuracy of predicting death risk within 30 days after surgery. According to the analysis, the mortality rate of the group with a risk score below 10 points was 0.1%, while the group exceeding 40 points had a rate of 11.7%, showing a clear correlation between the AI prediction and mortality. The AI's 30-day post-surgery death prediction power (AUROC) was measured at 0.909. This is not only higher than international standard assessment methods such as the European Society of Cardiology's evaluation method (0.728) and the cardiac risk index RCRI (0.725), but also slightly ahead of the "ASA classification" (0.886), which evaluates a patient's overall condition.
The research team additionally analyzed whether evaluating pre-surgery ECGs with AI could efficiently screen out patients who need detailed examinations such as echocardiography and coronary computed tomography (CT). Combining eight AI ECG biomarkers related to cardiac function and disease, patients were classified as "low-risk" when all indicators were within normal range, and "high-risk" when one or more showed abnormal findings. As a result, 92.3% of all surgeries were classified as low-risk on the AI ECG, and among these, the proportion who died or underwent emergency coronary procedures within 30 days after surgery was only 0.2%. In actual pre-surgery detailed examinations, 91% of these patients turned out normal, and the cost spent on this accounted for 62.8% of all pre-surgery cardiovascular detailed examinations. This demonstrated that examination burdens can be greatly reduced simply by analyzing a single ECG taken before surgery with AI, without the need to input complex data such as test values.
The research team said, "Considering the low cost and simplicity of the ECG test, its value for use in clinical settings is very high," while also assessing that "at present, AI ECG is not a complete replacement for existing pre-surgery detailed examinations, but has great value as a gateway for screening high-risk groups."
Professor Cho explained, "AI-based ECG analysis technology is being actively used as a screening tool before detailed examinations in emergency settings, leveraging its advantages of being fast, simple, and low-cost," adding, "It can also be used in general surgical procedures to lower examination burdens and screen high-risk groups."
The findings of this study were published in the international journals European Heart Journal-Digital Health and the Journal of Medical Internet Research.
