Clinical Report: Utilizing Calibrated Machine Learning to Forecast In-Hospital Cardiac Arrest

Overview

This study identifies clinical predictors of in-hospital cardiac arrest (IHCA) and develops a calibrated machine-learning model for early warning in general wards. Key predictors include comorbidities and laboratory results.

Background

In-hospital cardiac arrest (IHCA) poses significant risks to patient survival and neurological outcomes, with a high incidence in general wards. This study utilizes machine learning to enhance prediction capabilities for IHCA.

Data Highlights

Key Findings

• Atrial fibrillation, heart failure, and end-stage renal disease are significant predictors of IHCA.
• MEWS is associated with IHCA risk up to 16 hours prior to the event.
• Lower SpO₂ levels are predominant predictors in the final 8 hours before IHCA.
• The calibrated XGBoost model achieved an AUROC of 0.89 and an average precision of 0.88.
• At a sensitivity of 0.95, the model's specificity was 0.47 in the matched test set.

Clinical Implications

Further prospective multicenter validation is necessary to confirm the model's applicability.

Conclusion

This study demonstrates a calibrated machine-learning model for early warning for IHCA in general wards.

Related Resources & Content

1. Author(s)/Org, Source, Year -- Title
2. npj Digital Medicine, 2025 -- Development and evaluation of a machine learning model predicting out-of-hospital cardiac arrest using environmental factors
3. Frontiers in Cardiovascular Medicine, 2026 -- Artificial intelligence applied to post-resuscitation ECGs for early prognostication after out-of-hospital cardiac arrest
4. JMIR Medical Informatics, 2026 -- Prediction of Early Hospital Admission (≤24 Hours) After Stroke Using Machine Learning and Deep Learning: Multicenter Study From China
5. DIGITAL HEALTH — Incremental domain adaptation-based ICU patient mortality prediction
6. Medical Emergency Systems/ Rapid Response Teams for adult in-hospital patients: EIT 6309 TF SR
7. Early Warning Scores With and Without Artificial Intelligence
8. Prospective, multicenter validation of the deep learning-based cardiac arrest risk management system for predicting in-hospital cardiac arrest or unplanned intensive care unit transfer in patients admitted to general wards - PubMed