Multi-Scale Spatiotemporal Dynamic Graph Neural Network for Early Prediction of Mortality Risks in Heart Failure...

Area of Science:

• Biomedical Informatics
• Artificial Intelligence in Healthcare
• Cardiovascular Medicine

Background:

• Heart Failure (HF) presents a significant global health challenge, straining healthcare resources.
• Current prognostic models for HF patients have limitations in capturing dynamic physiological interdependencies.
• Accurate early mortality risk prediction is crucial for effective HF patient management.

Purpose of the Study:

• To introduce a novel Multi-scale Spatiotemporal Dynamic Graph Neural Network (MSTD-GNN) for enhanced early mortality prediction in Heart Failure patients.
• To dynamically extract and analyze spatio-temporal information from physiological parameters in Electronic Health Records (EHRs).
• To reveal dynamic relationships between physiological variables across multiple time scales.

Main Methods:

• Development of MSTD-GNN, a novel deep learning model utilizing dynamic graphs.
• Modeling of multivariate time series data from ICU patient EHRs to capture inter-parameter dependencies.
• Utilizing MIMIC-III and MIMIC-IV datasets for model training and validation.

Main Results:

• MSTD-GNN demonstrated superior performance in predicting early mortality risk for HF patients compared to existing methods.
• Achieved Area Under the Curve (AUC) scores of 83.93% on MIMIC-III and 81.74% on MIMIC-IV.
• The model successfully unveiled dynamic relationships between physiological variables at different time scales.

Conclusions:

• MSTD-GNN offers a significant advancement in predicting early mortality for Heart Failure patients.
• The model's ability to capture dynamic spatio-temporal information enhances prognostic accuracy.
• This approach holds promise for improving clinical decision-making and patient outcomes in HF care.