Prediction of in-hospital cardiac arrest on general wards using calibrated machine learning
-
By
-
Wen-Ying Yu
-
Mei-Li Pan
-
Chung-Yu Chen
-
July 8, 2026
-
Clinical Scorecard: Utilizing Calibrated Machine Learning to Forecast In-Hospital Cardiac Arrest in General Ward Settings
At a Glance
| Category | Detail |
|---|
| Condition | In-Hospital Cardiac Arrest (IHCA) |
| Key Mechanisms | Combination of comorbidity burden, laboratory indices, and time-windowed physiology. |
| Target Population | Patients in general wards without do-not-resuscitate (DNR) orders. |
| Care Setting | Tertiary hospital general wards. |
Key Highlights
- Atrial fibrillation, heart failure, and end-stage renal disease are independent predictors of IHCA.
- MEWS is informative up to 16 hours before IHCA, with lower SpO₂ being significant in the final 8 hours.
- Calibrated XGBoost model achieved AUROC of 0.89 and average precision of 0.88.
Guideline-Based Recommendations
Diagnosis
- Utilize MEWS and calibrated machine learning models for early identification of IHCA risk.
Management
- Integrate calibrated, time-windowed risk estimates into ward surveillance and rapid-response workflows.
Monitoring & Follow-up
- Monitor vital signs and laboratory indices regularly to identify deterioration.
Risks
- Consider comorbidities and laboratory abnormalities as significant risk factors for IHCA.
Patient & Prescribing Data
General ward patients at a tertiary hospital in Taiwan.
Focus on early warning systems that incorporate physiological and laboratory data.
Clinical Best Practices
- Implement machine learning models for improved prediction of IHCA.
- Recalibrate risk estimates to reflect real-world prevalence before clinical deployment.
Related Resources & Content
