Heart failure & cardiomyopathies
Development of Interpretable Machine Learning Models to Predict in-Hospital Prognosis of Acute Heart Failure Patients
Authors' conclusion
The XGBoost models with SHAP analysis provide high prediction performance, interpretability, and reproducible risk stratification for in-hospital mortality and WHF for patients with AHF.
Study context
- Population
- Patients with acute HF in the Kyoto Congestive Heart Failure registrytraining 3232, test 808
- Participants
- 4,040 patients
- Comparison
- XGBoost vs classification and regression trees (CART) vs random forest (RF) models
- Outcome
- Prediction of in-hospital mortality and worsening HF (WHF)
- Follow-up
- In-hospital
Methodological detailsregistry-based, 4:1 training/test split, Kyoto Congestive Heart Failure registry; dataset split 4:1 into training (n = 3232) and test (n = 808) datasets, 16 patients who died on the first or second day of admission excluded
- Registry-based
- 4:1 training/test split
- Kyoto Congestive Heart Failure registry; dataset split 4:1 into training (n = 3232) and test (n = 808) datasets
- 16 patients who died on the first or second day of admission excluded
Performance
As reported in the abstract
AUC in test dataset
| Test or model | AUC, in-hospital mortality | AUC, worsening heart failure (WHF) |
|---|---|---|
| XGBoost | 0.816 (95% CI 0.815-0.818) | 0.766 (95% CI 0.765-0.768) |
| CART | 0.683 (95% CI 0.680-0.685) | 0.688 (95% CI 0.686-0.689) |
| RF | 0.755 (95% CI 0.753-0.757) | 0.713 (95% CI 0.711-0.714) |
AUC, in-hospital mortality
XGBoost0.816 (95% CI 0.815-0.818)
CART0.683 (95% CI 0.680-0.685)
RF0.755 (95% CI 0.753-0.757)
AUC, worsening heart failure (WHF)
XGBoost0.766 (95% CI 0.765-0.768)
CART0.688 (95% CI 0.686-0.689)
RF0.713 (95% CI 0.711-0.714)
Further findings
Event rates, total study population
- In-hospital mortality
- 6.3% (255/4040)
- WHF
- 22.3% (900/4040)
SHAP clusters (XGBoost)
- Interpretable clusters formed; in-hospital mortality and WHF rates similar among each cluster in training and test datasets
Original publication
Development of Interpretable Machine Learning Models to Predict in-Hospital Prognosis of Acute Heart Failure Patients. ESC Heart Fail. Published 15 May 2024. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →