Machine learning approach to predict postoperative pain after spinal morphine administration during caesarean...

Area of Science:

• Anesthesiology
• Pain Management
• Health Informatics

Background:

• Optimal pain management is hindered by difficulties in timely prediction and assessment of high-risk patients.
• Fragmented clinical information exacerbates delays in pain reviews, increasing the risk of untreated pain.
• This study focuses on predicting significant postoperative pain following caesarean delivery with spinal morphine.

Purpose of the Study:

• To evaluate and compare the predictive performance of six modeling techniques for significant postoperative pain.
• To identify key clinical variables for accurate pain prediction in women undergoing caesarean delivery.
• To improve timely risk stratification and pain management strategies.

Main Methods:

• Retrieved medical records of 6561 women who received postoperative spinal morphine after caesarean delivery (Aug 2019-Aug 2022).
• Extracted and selected 23 clinical variables from 120 initially retrieved variables to enhance algorithm accuracy.
• Utilized an 80% training and 20% validation data split for model development and assessment.

Main Results:

• Ridge regression showed the best predictive performance (AUC: 0.719) with selected features for significant postoperative pain (7.9% incidence).
• Reduced feature sets with Ridge, LASSO, Elastic net, and XGBoost demonstrated comparable performance metrics (AUC: 0.704-0.719).
• Key predictive variables included the last recorded pain score on movement and pain score variability within the first 12 hours post-surgery.

Conclusions:

• Ridge regression and other machine learning models show promise in predicting significant postoperative pain after caesarean delivery.
• The models utilize readily available clinical data, including previous pain scores, for risk stratification.
• Future research should focus on refining these models for clinical implementation to enhance real-time pain management.