Machine-learning models presented at the ESC Digital & AI Summit predict mortality after stent treatment for severe heart attacks, beating the current guideline score. Researchers say the patient benefit still needs proving.

Machine-learning models can predict how likely a patient is to die after treatment for the most serious kind of heart attack, and they beat the current guideline score on both measures across all three databases tested. The results were presented on 13 November at the European Society of Cardiology's Digital & AI Summit

2

.

Source: News-Medical

Source: News-Medical

Why a new score was needed

These heart attacks cut blood flow through a main artery to the heart. The standard treatment is a procedure that reopens the blocked artery, usually with a stent. More than 2 million people worldwide have this kind of heart attack each year

2

. Many survive the procedure but some still die of it, in hospital or in the months that follow

2

.

Source: Medical Xpress

Source: Medical Xpress

The PERISCOPE project, short for Patient Risk StratifiCation for Post-STEMI Complications, set out to identify which patients need extra treatment and closer monitoring after the procedure

2

. The gap its researchers described is real. Current guideline scores estimate the risk of dying when the patient arrives, before the procedure, rather than afterwards

2

. Dr Andrea Rafanelli, the Amsterdam University Medical Centre researcher who presented the PERISCOPE project, said the models that do look at risk after the procedure lean on variables doctors do not routinely collect. That keeps them out of daily practice, she said

2

.

What the models did

The team built logistic regression models, a conventional statistical method rather than deep learning

2

. They trained them on about 38,500 American patient records, then tested them on roughly 119,000 more from a separate US database

2

. A final validation used about 6,800 records from Amsterdam's own electronic files

2

.

The models beat the score doctors currently rely on, the guideline-recommended standard, on both measures across all three databases tested

2

. In each comparison the gap was wide enough that the confidence intervals, the statistical margins around the results, did not overlap

2

. In-hospital death rates in the datasets ran from 6.6% to 9.9%

2

. Deaths within six months ran from 11.4% to 23.7%

2

.

But per the researchers themselves, the models predict risk well, and nobody has yet shown that acting on the prediction helps the patient. "The PERISCOPE models may offer an easy-to-implement system to help triage patients at high risk of future events, but first, their impact on outcomes must be tested in more diverse populations and prospective studies," she said

2

. She also suggested extending the models beyond this kind of heart attack to other acute coronary syndromes

2

.

© 2026 TheOutpost.AI All rights reserved