Application of XGBoost in the prediction of acute postoperative pain after major noncardiac surgery in older patients.

Abstract

Background: Acute postoperative pain (APP) are key factors in the recovery of surgical patients after surgery. This study used the machine learning eXtreme Gradient Boosting (XGBoost) algorithm for the prediction of acute postoperative pain after major noncardiac surgery in older patients.

Methods: This was a secondary analysis of data from a randomized controlled trial containing 1720 older patients undergoing general anesthesia. The training and test sets were divided according to the timeline. The Boruta function was made to screen for relevant characteristic variables. The XGBoost model was built on the training set using 10-fold cross-validation and hyperparameter optimization, and the tuned optimal model plotted the importance ranking diagram of feature variables, partial dependence profile (PDP) and Break down profile (BDP). The optimal model was used to calculate the confusion matrices and their parameters for the training and validation sets, and to plot the receiver operating characteristic curve (ROC), precision recall curve (PRC), calibration curve and Clinical decision curve (CDC) on the validation set.

Results: The Boruta function was used to screen the relevant characteristic variables, and the screened postoperative acute pain characteristic variables were CHARLSON score, Mini-Mental State Examination (MMSE), duration of surgery, preoperative depression score, smoking or not, duration of anesthesia, intraoperative mean heart rate, lidocaine dosage, age, intraoperative morphine dosage, grouping, preoperative anxiety score, loperamide dosage, intraoperative colloid amount, APACHE -II score, postoperative ICU or not, surgical site and postoperative tracheal intubation or not. Test set and validation set accuracy (ACC) for acute postoperative pain: 0.921 and 0.871; AUC-ROC: 0.964 and 0.920; AUC-PRC: 0.983 and 0.959; Brier: 0.067 and 0.098; Matthews Correlation Coefficient (MCC): 0.847 and 0.746.

Conclusions: A high-performance algorithm was developed and validated to predict the degree of change in postoperative pain; controlling important characterizing variables may be helpful for postoperative analgesia.