"Predicting Mortality Across Hospital Departments: A Machine Learning Approach for Various Healthcare-Associated Infections".

Abstract

Background: Healthcare-associated infections (HAIs) pose a serious challenge to healthcare systems. Early identification of high-risk patients is crucial for optimizing resource allocation and preventive screening. This study develops and evaluates machine learning (ML) models to predict mortality in HAI patients across different hospital wards.

Methods: This cross-sectional study analyzed a dataset of 4,346 HAI-diagnosed patients from a 700-bed hospital in Tehran, Iran, spanning March 2018 to January 2023. The dataset included demographics, clinical factors, and laboratory results. We applied four ML algorithms: multilayer perceptron (MLP), extreme gradient boosting (XGBoost), gradient boosting machines (GBM), and decision trees. Model performance was assessed using accuracy, precision, recall, F1 score, and area under the receiver operating characteristic curve (AUC-ROC).

Results: Among all models, MLP achieved the highest accuracy (91%) and AUC-ROC (0.95), outperforming XGBoost, GBM, and decision trees. Learning curves and cross-validation confirmed its robustness and generalizability.

Conclusion: ML techniques, particularly MLP, effectively predict mortality in HAI patients across hospital departments. By enabling targeted interventions and optimized resource allocation, MLP models can significantly improve HAI management and patient outcomes. Integrating these models into clinical decision support systems may enhance patient care and reduce the burden of HAIs.