In-Hospital Mortality Prediction Model of Heart Failure Patients Using Imbalanced Registry Data: A Machine Learning Approach Publisher

Summary: Can registry data predict deaths during heart-failure admission? In 3,968 records, an XGBoost model reached 76% accuracy, but needs validation. #HeartFailure #MachineLearning

Sabahi H ; Vali M ; Shafie D
Source: Scientia Iranica Published:2025

Abstract

Heart Failure (HF) is a cardiac dysfunction disease with a high mortality rate that is mostly calculated via registry data. The objective of this work was to predict in-hospital mortality in patients hospitalized with HF utilizing their pre-hospitalization registry data. The data include 3968 HF records extracted from the Persian Registry of cardio Vascular diseasE (PROVE)/HF registry. We proposed a method that contains an imbalanced ensemble probabilistic model which using registry data to predict HF patients who die during hospitalization from those who survive. The suggested ensemble model uses machine learning models, namely Decision Tree, Random Forest, Linear Discriminant Analysis (LDA), Logistic Regression, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Extreme Gradient Boosting (XGBoost), which were evaluated. We also used feature importance analysis to identify the important features and reduce the complexity. The results illustrated that the proposed method can predict in-hospital mortality of HF patients using XGBoost that outperformed all others. The feature importance ranking obtained by XGBoost demonstrated that the proposed method can achieve acceptable performance with the first 18 important features and XGBoost (accuracy: 76.4%±1.6%, sensitivity: 76.8%±6.9%, specificity: 76.4%±1.8%). Moreover, statistical analysis presented significant predictors of in-hospital mortality (P-value<0.01). In conclusion, the proposed method can effectively predict in-hospital mortality of HF patients using the imbalanced data. © 2025 Sharif University of Technology.