Background:Hospital-acquired influenza (HAI) is under-recognized despite high morbidity and poor health outcomes. It is important to detect influenza infections early to prevent its spread in hospitals.
Aim:This study was conducted to identify characteristics of HAI and develop HAI prediction models based on electronic medical records using machine learning.
Methods:This was a retrospective observational study including 111 HAI and 73,748 non-HAI patients. General characteristics, comorbidities, vital signs, laboratory results, chest X-ray results, and room information in EMR were analysed. Univariate analyses were performed to identify characteristics and logistic regression, random forest, extreme gradient boosting and artificial neural network were used to develop prediction models.
Results:HAI patients had significantly different general characteristics, comorbidities, vital signs, laboratory results, chest X-ray results and room status from non-HAI patients. The random forest model showed best performance in terms of AUC (83.4%) and the least number of false negatives. Staying in double rooms contributed most to prediction power followed by vital signs, laboratory results.
Conclusion: This study found HAI patients’ characteristics and the importance of ventilation to prevent influenza infection. They would help hospitals plan infection prevention strategies and prediction models could be used to early intervene spread of influenza in hospitals.