Why the study?
Patients with HFrEF face high risks of adverse outcomes and worsening HF after hospitalization, making it crucial to compare different machine learning prediction models and feature construction methods for predicting readmissions and worsening HF events.
Do machine learning algorithms accurately predict hospital readmissions and worsening heart failure events in patients with HFrEF?
Population
30,687 adult patients with HFrEF and HF-related hospitalization
Comparison
Different ML prediction models and feature construction methods
Design
Retrospective registry-linked modeling study
Follow-up
1 year
Key result
Machine learning models, such as XGBoost, demonstrated mediocre performance in predicting 365-day hospital readmission (AUC=0.649) and worsening heart failure events (AUC=0.640) in HFrEF patients.
Authors
Loading...
ML models lack accuracy for HFrEF readmission prediction; leaves open whether refined features or algorithms improve performance.
Observational (n=30,687)
Yes
Do machine learning algorithms accurately predict hospital readmissions and worsening heart failure events in patients with HFrEF?
Effect estimate: AUC 0.649
Machine learning models using administrative and clinical codes showed mediocre discriminative ability (AUC 0.595-0.649) for predicting readmissions and worsening heart failure in HFrEF patients.
Ru et al. (2022) conducted an observational in Heart failure with reduced ejection fraction (HFrEF) (n=30,687). Machine learning prediction models was evaluated on 30-, 90-, and 365-day hospital readmissions and worsening HF events (WHFEs) (AUC 0.649). Machine learning models, such as XGBoost, demonstrated mediocre performance in predicting 365-day hospital readmission (AUC=0.649) and worsening heart failure events (AUC=0.640) in HFrEF patients.