Development and validation of a machine learning-based predictive model for postoperative nausea and vomiting in...

The purpose of the study is to construct a postoperative nausea and vomiting (PONV) risk prediction model for day-case laparoscopic cholecystectomy (LC) using a machine learning combination algorithm and evaluate its performance.

A retrospective cohort study.

The Hospital Information System (HIS) and the Surgical Anaesthesia Information Management System (SAIMS).

Patient data are collected from the day surgery ward of Sichuan Provincial People's Hospital from February 2023 to April 2024. The research subjects are adult patients (18-75) who underwent day-case LC, excluding patients with unexpected termination of the day surgery plan, such as the patient who was transferred to hepatobiliary surgery due to intraoperative conversion to laparotomy.

The study employed two data filling methods, two data sampling methods, two variable screening methods and six machine learning algorithms to construct 48 predictive models. Area under curve (AUC), accuracy, precision, recall rate and F1 value were used to evaluate the predictive performance of the model. The AUC of the test set is mainly used to evaluate the prediction performance, and the Shapley weighted explanatory value is used to determine the weight of the variable's prediction contribution. We will collect patient data from this unit in July 2025 to evaluate the model's performance.

A total of 2709 patients were selected for model construction in the study. 20 input variables were retained for developing the predictive model. The combined model of KNN, BSMOTE, RFEL and GBM shows the best AUC performance (0.9600). The five most important variables in the prediction model were postoperative pain, LESS method, citraturia dosage, gender and sufentanil dosage. An additional 211 patients were collected to validate the model performance with an AUC of 0.79.

The study finds that postoperative pain, LESS method and cisatracurium dosage are closely related to the occurrence of PONV in day-case LC. However, these three variables have rarely been reported in the previous literature and worth further research. The prediction model obtained in this study provides a meaningful reference for the perioperative prevention and treatment of PONV in day surgery.