Predicting the 14-Day Hospital Readmission of Patients with Pneumonia Using Artificial Neural Networks (ANN)

Area of Science:

• Healthcare Informatics
• Clinical Prediction Modeling
• Artificial Intelligence in Medicine

Background:

• Unplanned patient readmission (UPRA) presents a significant financial burden and quality concern in healthcare.
• Current clinical practices lack effective tools for identifying patients at high risk of UPRA during hospitalization.

Purpose of the Study:

• To develop and validate a predictive model for early detection of 14-day UPRA in pneumonia patients.
• To compare the performance of artificial neural network (ANN) and convolutional neural network (CNN) models for UPRA prediction.
• To create a user-friendly application for real-time UPRA risk assessment.

Main Methods:

• Utilized a dataset of 21,892 pneumonia patients from three Taiwanese hospitals (2016-2018).
• Developed and compared ANN and CNN models using training (70%) and testing (30%) datasets.
• Extracted 17 feature variables for model development and evaluated performance using Area Under the Curve (AUC).

Main Results:

• The ANN model demonstrated a high Area Under the Receiver Operating Characteristic Curve (AUC) of 0.75 with 17 extracted features.
• ANN model performance (AUC=0.73) surpassed CNN model performance (AUC=0.50).
• A functional application was successfully developed for predicting UPRA.

Conclusions:

• The developed ANN model effectively predicts 14-day UPRA in pneumonia patients.
• The prediction tool can assist clinicians in identifying high-risk individuals for proactive intervention.
• Early detection via the app facilitates timely preparedness plans, potentially reducing readmission rates.