Inter-organ correlation based multi-task deep learning model for dynamically predicting functional deterioration in...

Area of Science:

• Critical Care Medicine
• Artificial Intelligence in Healthcare
• Computational Biology

Background:

• Functional deterioration (FD) is a leading cause of death in ICU patients.
• Existing multi-task (MT) models struggle to predict FD across multiple organ systems simultaneously.

Purpose of the Study:

• To propose and evaluate a novel MT deep learning model, the inter-organ correlation based multi-task model (IOC-MT).
• To dynamically predict FD in six organ systems by capturing inter-organ correlations.

Main Methods:

• Developed the IOC-MT model incorporating Graph Attention Networks (GAT) for inter-organ correlation and an adaptive adjustment mechanism (AAM).
• Trained and validated the model on three public ICU databases, comparing it against five single-task (ST) baseline models.
• Assessed model performance using AUROC, AUPRC, and calibration curves, with ablation studies to validate component contributions.

Main Results:

• IOC-MT demonstrated comparable discrimination and calibration to advanced ST models (LSTM-ST, GRU-ST, Transformer-ST) and significantly outperformed simpler ST models (GRU-ST, RF-ST).
• Ablation studies confirmed that GAT, AAM, and missing indicators enhance overall model performance.
• The model's inter-organ correlation analysis and prediction adjustments were found to be intuitive, comprehensible, and biologically plausible.

Conclusions:

• The IOC-MT model shows significant promise as an effective tool for the dynamic prediction of functional deterioration in multiple organ systems within ICU settings.
• The model's ability to leverage inter-organ correlations offers a novel approach to improving predictive accuracy and understanding complex physiological interactions.