Readmission prediction using deep learning on electronic health records

Area of Science:

• Biomedical Informatics
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Unscheduled 30-day readmissions are a significant challenge for Congestive Heart Failure (CHF) patients, increasing health risks and healthcare expenditures.
• Effective identification of high-risk patients at hospital discharge is crucial for implementing targeted interventions to reduce readmissions.

Purpose of the Study:

• To develop and validate a deep learning framework for predicting 30-day unscheduled readmissions in Congestive Heart Failure patients.
• To integrate expert-derived and machine-derived features, sequential patterns, and address class imbalance within a unified Electronic Health Record (EHR)-driven prediction model.

Main Methods:

• Utilized a real-world dataset of over 7500 CHF patients hospitalized in Sweden (2012-2016).
• Developed a cost-sensitive Long Short-Term Memory (LSTM) neural network incorporating expert features and contextual embedding of clinical concepts.
• Evaluated the model's performance using ROC-AUC and F1-measure, assessing the contribution of each component and potential cost savings.

Main Results:

• The comprehensive deep learning model achieved a high discrimination ability with an AUC of 0.77 and an F1-measure of 0.51.
• The model demonstrated a cost-saving potential of 22% of the maximum possible savings, outperforming reduced models in key metrics.
• Individual component analysis confirmed the value of integrating expert features, sequential patterns, and addressing class imbalance.

Conclusions:

• The proposed deep learning framework effectively predicts 30-day readmissions for CHF patients, offering a valuable tool for clinical decision support.
• Integrating multiple data elements and advanced deep learning techniques significantly enhances prediction accuracy and cost-effectiveness.
• Targeted interventions based on this model's predictions hold substantial potential for reducing healthcare costs and improving CHF patient management.