Heart failure & cardiomyopathies
Machine Learning Prediction of Heart Failure Readmissions: Insights From a Multicenter Emergency Department Trial
Authors' conclusion
An XGB-based ML model accurately predicted 12-month HF readmission following ED presentation and outperformed traditional regression methods, supporting its potential role in early risk stratification and targeted preventive care.
Study context
- Population
- Patients presenting to emergency department (EDs) with suspected cardiac conditions, with symptoms warranting high-sensitivity cardiac troponin T testing, enrolled April-December 2023ED patients (mean age 61 years, 49% female)
- Participants
- 14,131 patients
- Comparison
- Four machine learning (ML) algorithms were compared with logistic regression
Methodological details12 South Australia EDs, training (70%) and test (30%) sets, Four ML algorithms were compared with logistic regression
- 12 South Australia EDs
- Training (70%) and test (30%) sets
- Four ML algorithms were compared with logistic regression
Primary result
As reported in the abstract
HF readmission · 12 months
Readmitted for heart failure (HF)
808
5.71%
Median time to readmission
83 days
IQR: 27–178
Further findings
| Outcome | Values | P value |
|---|---|---|
| Prediction of HF readmission (AUC), test set · XGB outperformed logistic regression | eXtreme Gradient Boosting (XGB) AUC 0.861 (95% CI 0.847–0.875) · logistic regression AUC 0.846 (95% CI 0.828–0.861) · random forest AUC 0.845 · least absolute shrinkage and selection operator AUC 0.851 · neural networks AUC 0.842 | P = 0.015 |
Calibration of XGB
- Brier score for XGB
- 0.151
- Demonstrating good model calibration
Original publication
Machine Learning Prediction of Heart Failure Readmissions: Insights From a Multicenter Emergency Department Trial. JACC Adv. Published 22 June 2026. doi:10.
Summarised from the published abstract; every number is checked against it