Intraoperative Hypotension Prediction Model Based on Systematic Feature Engineering and Machine Learning.

Area of Science:

• Anesthesiology and Critical Care Medicine
• Biomedical Engineering
• Data Science in Healthcare

Background:

• Postoperative arterial hypotension increases the risk of serious complications like myocardial infarction and acute kidney injury.
• Current research primarily focuses on identifying high-risk patients rather than real-time prediction of hypotension events.
• Forecasting the precise timing of hypotension is significantly more challenging than general risk assessment.

Purpose of the Study:

• To develop a systematic feature engineering method applicable across various vital signs for hypotension prediction.
• To create a machine learning model capable of real-time hypotension prediction 5 minutes prior to occurrence.
• To address the challenge of predicting the exact timing of hypotensive events.

Main Methods:

• Engineered features using statistical, peak, change, and frequency analyses from invasive blood pressure (IBP) data.
• Developed a random forest machine learning model utilizing the engineered features.
• Trained the model to distinguish between hypotensive events and normal physiological states.

Main Results:

• The random forest model achieved a high overall accuracy of 0.974.
• The model demonstrated a precision of 0.904 in identifying hypotensive events.
• A recall of 0.511 was obtained for predicting hypotensive events 5 minutes in advance.

Conclusions:

• The proposed feature engineering and machine learning approach enables accurate real-time prediction of arterial hypotension.
• The model shows promise for improving patient safety by providing early warnings of impending hypotensive events.
• Further research can explore broader vital sign integration and model optimization for enhanced predictive performance.