Tri-Service General Hospital (National Defense Medical Center)
AI-enabled ECG alert for hospitalized patients at high risk of death
A randomized trial of an AI ECG alert at two Taiwanese hospitals found 90-day mortality of 3.6% against 4.3% with usual care
Problem
Doctors on busy wards can miss patients whose condition is about to worsen. An ECG is taken for many inpatients, but the signs of high mortality risk in it are hard to spot by eye.
Approach
Researchers at Tri-Service General Hospital in Taipei built a neural network trained on more than 450,000 ECGs, with all-cause mortality as the label. When it flagged a patient as high risk, a warning went to the attending physician's smartphone with a link to the ECG and the prediction. From December 15, 2021 to April 30, 2022, 15,965 patients were randomized: 8,001 to the alert group and 7,964 to usual care. Follow-up ran to July 31, 2022. The system did not change standard treatment rules.
Results
Takeaway
In the randomized trial, patients the model flagged as high risk saw a larger reduction in mortality. A later cost analysis of the same trial found the 90-day cost was higher in the alert group, and its confidence intervals crossed zero for cost per death averted.
Categories
- Size
- Large enterprise
- Industry
- Healthcare and life sciences
- Business function
- Production and quality
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and ML
- Adoption
- Built in-house
- Goal
- Quality and risk
- Evidence
- Independent study
Sources
Source summaries
Researchers analysed costs in the Tri-Service General Hospital trial of 15,965 randomized patients. They reported 90-day mortality of 3.6% with the AI-ECG alert against 4.3% with usual care. They also reported an added cost of $402 per patient over 90 days and stated that the uncertainty range on cost per death averted includes negative values.
Physician and researcher Eric Topol described the trial published in Nature Medicine that month, involving nearly 16,000 patients at two Taiwanese hospitals. He reported a 17% reduction in 90-day all-cause mortality and a 31% reduction in the pre-specified high-risk ECG subgroup.
Full details are in the original sources.