Medical-AI trial: AI predicts one-year mortality risk from the ECG — a Software as a Medical Device validated on 462,000 records (NCT07659262)

An electrocardiogram (ECG) is a simple, widely available test that records the electrical activity of the heart for about ten seconds. Traditional assessment of mortality or cardiovascular risk has also used values such as blood tests, but these are not always present in the electronic health record and cannot always be computed.

The Software as a Medical Device this study validates avoids that constraint, seeking to estimate the probability of cardiac-related death within one year from the standard ECG waveform alone using AI.

Per the registry summary, the software, the Chang Gung ECG Mortality Risk Prediction Software, is an AI-based Software as a Medical Device (SaMD) that analyzes a standard 10-second, 12-lead resting ECG signal to predict the probability of cardiac-related mortality within one year.

The study validates it in a multicenter retrospective cohort, with the primary outcome being the area under the ROC curve (AUC) for predicting one-year cardiac-related mortality. The scale used for validation is large, at 461,982 records.

If risk can be estimated from an ECG that can be taken anywhere, without relying on blood tests, it could help flag high-risk people even where testing facilities are limited and connect them to further work-up or preventive intervention. Retrospective validation on about 462,000 records is aimed at confirming predictive performance under conditions close to real practice.

That said, this study evaluates predictive performance by AUC and the like in a retrospective design; it does not assert the prognosis of an individual patient or establish clinical adoption or intervention effect. Risk prediction is a probability, and medical decisions should rest on professionals and official information.