Machine Learning Model Predicts Heart Attack Mortality

New PERISCOPE models outperformed the GRACE 2.0 score in forecasting deaths after STEMI heart attacks across three patient databases.

Updated on Nov. 13, 2026 in Heart Disease

Isometric editorial illustration of a heart form in oxblood beside three glass vials on a cream tray, with a stepped row of teal blocks below, evoking measured mortality risk prediction.
Researchers reported that machine-learning models called PERISCOPE outperformed the GRACE 2.0 score in predicting in-hospital and six-month deaths after STEMI heart attacks across three patient databases. AI Illustration. Upload story photo >

Live Poll

Would you trust a computer-based tool to help decide your heart care?

Researchers developed machine learning models called PERISCOPE to predict in-hospital and six-month mortality after ST-elevation myocardial infarction (STEMI) heart attacks. The models outperformed the GRACE 2.0 score across all three databases tested.

Why it matters

Existing GRACE 2.0 scores estimate mortality at presentation, before percutaneous coronary intervention (PCI), while many post-PCI risk models rely on numerous or hard-to-collect variables. A simpler post-procedure tool could help clinicians identify high-risk patients more readily in everyday practice.

The PERISCOPE models were trained on 38,489 patients, independently tested on 119,035 patients, and externally validated on 6,776 patients, with in-hospital mortality ranging from 6.6% to 9.9% and six-month mortality from 11.4% to 23.7%. Their real-world effect on outcomes has not yet been tested in more diverse populations or prospective studies.

The players

Amsterdam University Medical Centre

A large academic medical center in the Netherlands whose electronic medical record data was used to externally validate the PERISCOPE models.

The details

The team trained a set of logistic regression models on the US PREMIER PINC AI Healthcare Database, then independently tested them on US TriNetX data and externally validated them using electronic medical record data from Amsterdam UMC. Across all three databases, the PERISCOPE models predicted both in-hospital and six-month mortality more accurately than the GRACE 2.0 score.

Timeline

  1. - Results were presented on November 13, 2026, at the ESC Digital & AI Summit 2026.

Health Landscape

The GRACE 2.0 score has long served as the standard presentation-stage tool for estimating mortality after acute coronary syndromes. The PERISCOPE models mark a shift toward post-procedure risk estimation using a small set of routinely collected variables.

If validated in wider populations, a simpler post-procedure risk tool could help clinicians decide how closely to monitor heart attack patients after treatment. The models are still research tools and are not yet part of routine clinical care.

The takeaway

Risk prediction after a heart attack treatment may soon become more precise, but patients should continue to rely on their own cardiologists for decisions about care and follow-up.

Further reading

Explore more research on heart disease and how risk prediction is changing cardiac care.

Source note: This article includes information reported by News-Medical.

Live Poll

Would you trust a computer-based tool to help decide your heart care?