AI Models Improve Mortality Risk Prediction After PCI for STEMI

By HospiMedica International staff writers
Posted on 18 Nov 2026

ST-elevation myocardial infarction (STEMI) is a heart attack caused by complete blockage of one or more coronary arteries. More than 2 million people worldwide develop STEMI each year. Percutaneous coronary intervention (PCI) is effective for many patients, but mortality risk can remain high after treatment. Researchers have now developed machine-learning models to identify patients at increased risk of death after PCI.

The PERISCOPE (Patient Risk StratifiCation for Post-STEMI Complications) project, involving Amsterdam University Medical Centre, was developed to predict in-hospital and six-month mortality after percutaneous coronary intervention (PCI). The models are designed to support risk stratification after the procedure, a clinically relevant time point because existing guideline-recommended scores, including GRACE 2.0, estimate risk at presentation rather than after PCI.


Image Credit: Adobe Stock

The research team trained logistic regression models using the U.S. PREMIER PINC AI Healthcare Database, which included 38,489 patients. The models were then independently tested using U.S. TriNetX data from 119,035 patients and externally validated using electronic medical record data from 6,776 patients at Amsterdam UMC.

Across the three databases, in-hospital mortality rates ranged from 6.6% to 9.9%, while six-month mortality rates ranged from 11.4% to 23.7%. The PERISCOPE models outperformed the GRACE 2.0 score in predicting both in-hospital and six-month mortality across all three databases. In each comparison, the lower bound of the 95% confidence interval for the area under the receiver operating characteristic curve for PERISCOPE exceeded the upper bound observed with GRACE 2.0.

The findings were presented at the ESC Digital & AI Summit 2026. The results suggest that post-PCI machine-learning models may help identify patients who could benefit from additional treatment or closer monitoring. Further evaluation in more diverse populations and prospective studies is still needed before the models can be integrated into clinical care pathways.

“The machine-learning models provided good discrimination of both in-hospital and six-month mortality. One of the most encouraging aspects was that the models showed consistent performances across the three datasets, which were geographically and clinically diverse, suggesting good generalizability,” said Dr. Andrea Rafanelli, presenter from Amsterdam University Medical Centre.

Related Links
European Society of Cardiology
Amsterdam University Medical Centre


Latest Critical Care News