CLEAR-Shock: Contrastive LEARning for Shock.

Area of Science:

• Critical Care Medicine
• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Shock is a life-threatening circulatory failure requiring prompt intervention.
• Early prediction and continuous monitoring of physiological signs are crucial for timely treatment.
• Existing machine learning models often overlook post-onset shock monitoring.

Purpose of the Study:

• To develop a novel prognostic monitoring system for predicting shock onset and continuation.
• To categorize patient physiological states into pre-shock and post-shock onset scenarios.
• To improve early detection and real-time risk assessment of shock.

Main Methods:

• Implemented a supervised contrastive learning framework for predictive modeling.
• Utilized a feature tokenizer transformer to extract effective contrastive embeddings.
• Categorized patient states into pre-shock (within 8 hours) and post-shock (within 1 hour) predictions.

Main Results:

• The novel framework demonstrated improved predictive performance over baseline models.
• The system achieved early shock detection an average of 6 hours before onset.
• Validated predictive accuracy using both internal and external datasets.

Conclusions:

• The developed system offers hourly predictions of shock occurrence and continuation risk.
• This prognostic monitoring system facilitates early prevention and treatment evaluation.
• The framework enhances early detection and real-time patient state assessment for critical care.