Intraoperative Hypotension Prediction Based on Features Automatically Generated Within an Interpretable Deep Learning...

Area of Science:

• Medical Technology
• Artificial Intelligence in Medicine
• Anesthesiology

Background:

• Monitoring arterial blood pressure (ABP) is critical in anesthesia to prevent hypotension and adverse outcomes.
• Existing artificial intelligence (AI) hypotension prediction indices often lack clear interpretability.
• A need exists for predictive models that offer physiological insights into hypotension development.

Purpose of the Study:

• To develop an interpretable deep learning model for forecasting hypotension.
• To predict hypotension occurrence 10 minutes prior to an event based on ABP.
• To provide physiological interpretations of ABP trends associated with hypotension.

Main Methods:

• Development of a deep learning model utilizing arterial blood pressure (ABP) data.
• Forecasting hypotension 10 minutes ahead of a 90-second ABP record.
• Internal and external validation of the model's predictive performance.

Main Results:

• The model achieved high accuracy, with Area Under the Receiver Operating Characteristic Curves (AUROCs) of 0.9145 (internal) and 0.9035 (external).
• The prediction mechanism is physiologically interpretable through automatically generated ABP trend predictors.
• Demonstrated the clinical applicability of an accurate, interpretable deep learning model.

Conclusions:

• An interpretable deep learning model accurately predicts hypotension in anesthetized patients.
• The model provides valuable physiological insights into ABP dynamics preceding hypotensive events.
• This approach enhances clinical practice by offering a transparent and reliable tool for hypotension management.