Long-Term Postoperative Pain Prediction Using Higher-Order Singular Value Decomposition of Intraoperative...

Area of Science:

• Anesthesiology and Pain Management
• Biomedical Engineering
• Data Science

Background:

• Long-term postoperative pain (POP) remains poorly understood, with limited insight into patient responses to pain relief.
• Current nociception indices under general anesthesia lack correlation with long-term POP outcomes.

Purpose of the Study:

• To analyze the dynamics of physiological parameters during surgery and their association with long-term POP.
• To develop a predictive model for long-term POP using intraoperative physiological data.

Main Methods:

• Employed complex higher-order singular value decomposition (HOSVD) for multivariate-temporal analysis of patient physiological responses.
• Extracted interconnected factors (multivariate, temporal, patient-specific) from intraoperative vital signs of 175 patients.

Main Results:

• HOSVD clarified dynamic correlations in physiological responses, enabling discrimination between mild and severe POP at 30 and 90 days post-surgery.
• Achieved high predictive accuracy (AUC up to 1.0) for long-term POP across various surgical types (thoracic, orthopedic, colorectal, transplant, pancreatic).

Conclusions:

• Surgery-specific principal descriptors derived from physiological dynamics are linked to long-term POP development.
• The study establishes a correlation between intraoperative physiological response dynamics and long-term POP, offering predictive value.