Machine Learning for Preoperative Prediction of Intraoperative Hypothermia in Gynecological Laparoscopic Surgery: A Retrospective Cohort Study.

BACKGROUND

Intraoperative hypothermia (IOH, core temperature < 36.0°C) is common during gynecological laparoscopic surgery and is associated with adverse outcomes. However, predicting its occurrence using only preoperative indicators remains challenging.

METHODS

This retrospective cohort study included patients who underwent gynecological laparoscopic surgery at a single center. Candidate predictors were extracted from electronic health records (EHRs). Least absolute shrinkage and selection operator (LASSO) regression was applied for feature selection. Logistic regression (LR) and extreme gradient boosting (XGBoost) were developed and compared. The SHapley Additive exPlanations (SHAP) method was used to interpret the model and identify key predictors.

RESULTS

A total of 301 patients were included in this study, of which 118 cases (39.2%) developed IOH during gynecological laparoscopic surgery. Using LASSO regression, five predictors were retained: age, American Society of Anesthesiologists (ASA) physical status, basal temperature, estimated duration of surgery, and hypertension. The XGBoost model exhibited the best performance, achieving an area under the curve (AUC) of 0.980 in the training set and an AUC of 0.905 in the test set. SHAP analysis indicated that estimated duration of surgery was the most important predictive factor.

CONCLUSIONS

The XGBoost model best predicted IOH in patients undergoing gynecological laparoscopic surgery. SHAP analysis identified estimated duration of surgery as the most important predictor.