An integrated LSTM-HeteroRGNN model for interpretable opioid overdose risk prediction

Area of Science:

• Biomedical Informatics
• Artificial Intelligence in Healthcare
• Public Health Informatics

Background:

• Opioid overdose (OD) is a leading cause of accidental death in the US, with deaths escalating during the COVID-19 pandemic.
• Identifying individuals at high risk of OD is crucial for combating the opioid crisis.
• Deep learning models using electronic health records (EHR) show promise but face challenges with data complexity and clinical explainability.

Purpose of the Study:

• To develop an integrated deep learning model for predicting patient opioid overdose risk.
• To enhance the interpretability of deep learning models for clinical decision support.

Main Methods:

• Developed LIGHTED, a deep learning model combining Long Short-Term Memory (LSTM) and Graph Neural Networks (GNN).
• Incorporated temporal disease progression and feature interactions within EHR data.
• Evaluated the model on the Cerner Health Facts database (over 5 million patients).
• Proposed a novel interpretability method using GNN embeddings for patient and feature clustering.

Main Results:

• The LIGHTED model demonstrated superior performance compared to traditional machine learning and other deep learning methods.
• The interpretability method allowed for qualitative analysis of clinical features and patient clusters.
• The model effectively leverages longitudinal EHR data and the graph structure of patient data.

Conclusions:

• LIGHTED provides effective and interpretable opioid overdose risk predictions.
• The model's ability to utilize longitudinal EHR data and intrinsic data structures offers potential for improved clinical decision support.
• This approach can aid in targeting high-need populations and mitigating the opioid crisis.