TITLE:
Development and Validation of a Machine Learning Model for Predicting Postoperative Nausea and Vomiting in Gynecological Day Surgery
AUTHORS:
Lantian Li, Yue Li, Caiyan Huang, Hui Zhang, Jinghao Huang, Dinggui Lu
KEYWORDS:
Machine Learning, Prediction Model, Postoperative Nausea and Vomiting, Day Surgery, Hysteroscopy, XGBoost
JOURNAL NAME:
Open Journal of Obstetrics and Gynecology,
Vol.15 No.6,
June
24,
2025
ABSTRACT: Objective: To develop and validate a machine learning-based risk prediction model for postoperative nausea and vomiting (PONV) following gynecological day hysteroscopy, providing decision-making support for individualized clinical interventions. Methods: Clinical data from 478 patients undergoing gynecological day hysteroscopy at the Affiliated Hospital of Youjiang Medical University for Nationalities between January 2022 and December 2024 were retrospectively collected, including: ① Demographic characteristics: age, body mass index (BMI), ASA classification, history of PONV, motion sickness, etc.; ② Anesthesia-related data: anesthesia type, opioid dosage, use of inhaled anesthetics, intraoperative fluid volume, etc.; ③ Surgery-related data: operating time, intraoperative blood loss, Surgical type, etc.; ④ Postoperative data: PONV occurrence (yes/no), postoperative pain score, early postoperative feeding time, etc.; and ⑤ ERAS (Enhanced Recovery After Surgery)-related indicators: preoperative anxiety score, early postoperative mobilization time. Predictive variables were screened using LASSO regression. Models were constructed using XGBoost, LightGBM, AdaBoost, logistic regression, KNN, and GNB, with a 7:3 split for training and validation sets. Model performance was evaluated via the area under the ROC curve (AUC) and calibration curves, while SHAP values were used to interpret feature importance. Results: Seven predictive variables were included. The XGBoost model demonstrated optical performance in the validation set (AUC = 0.911). Key predictors included intraoperative opioid dosage, delayed postoperative feeding, age, preoperative anxiety score, operating time, blood loss, and intraoperative fluid volume. SHAP analysis revealed a dose-dependent positive correlation between opioid dosage and PONV risk. The XGBoost model exhibited higher accuracy in predicting and distinguishing patients with PONV. Conclusion: The machine learning-based PONV prediction model demonstrates high predictive efficiency and clinical interpretation, enabling preoperative risk strategy and postoperative prediction. It serves as a precision tool for guiding individualized PONV prevention strategies.