Machine learning based quantitative pain assessment for the perioperative period.

Area of Science:

• Biomedical Engineering
• Anesthesiology
• Medical Informatics

Background:

• Accurate pain assessment during surgery is crucial for patient well-being and optimal anesthetic management.
• Photoplethysmography (PPG) offers a non-invasive method to monitor physiological changes potentially related to pain.
• Existing methods for intraoperative pain assessment may have limitations in accuracy and real-time applicability.

Purpose of the Study:

• To develop and validate a machine learning model using PPG data for perioperative pain assessment.
• To compare the performance of the developed model against a commercial surgical pain index.
• To identify key PPG waveform features indicative of intraoperative and postoperative pain.

Main Methods:

• Development of XGBoost-based models using extracted photoplethysmogram waveform features.
• Collection of PPG data and pain scores (numerical rating scale or clinical criteria) from 242 surgical patients at 2-minute intervals.
• Evaluation of model performance using area under the receiver operating characteristics curve (AUC).

Main Results:

• The developed intraoperative model achieved an AUC of 0.819, comparable to the commercial index (0.829).
• The developed postoperative model achieved a significantly higher AUC of 0.927 compared to the commercial index (0.577).
• Key features identified include waveform skewness and diastolic phase rate decrease for intraoperative pain, and systolic phase area or baseline fluctuation for postoperative pain.

Conclusions:

• The developed PPG-based model demonstrates high effectiveness for perioperative pain assessment.
• The model shows particular strength in postoperative pain evaluation, surpassing commercial alternatives.
• Specific PPG waveform characteristics are valuable indicators for real-time pain monitoring during and after surgery.