Prediction of hospital-acquired influenza using machine learning algorithms: a comparative study

Area of Science:

• Healthcare epidemiology
• Medical informatics
• Infectious disease modeling

Background:

• Hospital-acquired influenza (HAI) presents significant morbidity and is often under-recognized.
• Early detection of HAI is critical for controlling transmission within healthcare facilities.

Purpose of the Study:

• To identify factors associated with HAI.
• To develop and compare machine learning models for predicting HAI occurrence.
• To determine the optimal algorithm for HAI prediction.

Main Methods:

• Retrospective observational study of 111 HAI and 73,748 non-HAI patients across two influenza seasons.
• Analysis of patient characteristics, comorbidities, vital signs, lab results, and room information.
• Development of predictive models using Logistic Regression, Random Forest, Extreme Gradient Boosting, and Artificial Neural Network.

Main Results:

• Significant differences observed in patient demographics, comorbidities, vital signs, and lab results between HAI and non-HAI groups.
• The Random Forest model achieved the highest performance with an AUC of 83.3% and only four false negatives.
• Staying in double rooms, vital signs, and laboratory results were the most significant predictors of HAI.

Conclusions:

• Patient characteristics and room ventilation are key factors in HAI incidence.
• Machine learning models, especially Random Forest, can facilitate early intervention for HAI.
• Findings support the development of targeted infection prevention strategies in hospitals to mitigate influenza spread.