Predictive Effect of Postoperative Recovery in General Anesthesia Patients Using Interpretable Models Based on Swarm Intelligence Machine Learning

Chenqiao HuaYeyuan ChuMinshu ZhouXin Xu^*

• Sir Run Run Shaw Hospital,affiliated with Zhejiang University School of Medicine, Hangzhou, China

To analyze the clinical value of predicting postoperative recovery in patients undergoing general anesthesia using an interpretable model based on swarm intelligence machine learning. Methods: This study retrospectively collected data from 1128 patients who underwent general anesthesia at Sir Run Run Shaw Hospital, affiliated with Zhejiang University School of Medicine, from January 2021 to January 2024. Based on predefined inclusion and exclusion criteria, a total of 1,128 patients were included in the study, comprising Dataset A. Additionally, patients meeting the same criteria from Sir Run Run Shaw Hospital Affiliated to Zhejiang University School of Medicine Alar Hospital during the period January 2021 -January 2024 were selected, constituting Dataset B. Dataset a was used for model training and testing, and dataset b was used for external validation of the model. Patients who experienced at least one of the following conditions-hypothermia upon admission to the Post-Anesthesia Care Unit (PACU), delayed discharge from PACU, or delayed awakening-were classified into the poor postoperative recovery group. The remaining patients were classified into the good postoperative recovery group. Clinical data were analyzed using a swarm intelligence machine learning algorithm to develop a predictive model for postoperative recovery in patients undergoing general anesthesia. The value of the identified features was analyzed, and a visualization system was constructed. Results: LASSO regression identified seven variables: surgery duration, anesthesia duration, neutrophil-to-lymphocyte ratio (NLR), C-reactive protein (CRP), serum creatinine, body mass index (BMI), and age. The swarm intelligence machine learning model, with XGBoost as the base learner, demonstrated the best performance. It achieved an F1 score of 0.8447 and an area under the curve (AUC) of 0.9265 on the training set, and an F1 score of 0.7735 and an AUC of 0.8808 on the test set. The validation results demonstrated that the model achieved: ROC-AUC: 0.8383, PR-AUC: 0.8241 This model can be used to predict postoperative recovery in patients undergoing general anesthesia.The application of an interpretable swarm intelligence machine learning model can assist in predicting postoperative recovery in patients undergoing general anesthesia, thereby aiding clinicians in formulating subsequent intervention plans.