Construction and validation of a machine learning-based prediction model for intraoperative hypothermia in general anesthesia surgery patients
10.3760/cma.j.cn115682-20240901-04867
- VernacularTitle:基于机器学习的全身麻醉手术患者术中低体温预测模型的构建与验证
- Author:
Min FENG
1
;
Muhataile JIAYINAER
;
Wenjuan MA
;
Li LI
Author Information
1. 新疆医科大学护理学院，乌鲁木齐 830000
- Publication Type:Journal Article
- Keywords:
Machine learning;
Intraoperative hypothermia;
Prediction model;
Nomograms;
Risk factors
- From:
Chinese Journal of Modern Nursing
2025;31(21):2837-2844
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To construct and validate a prediction model for intraoperative hypothermia in general anesthesia surgery patients based on machine learning algorithm.Methods:Convenience sampling was used to retrospectively collect data from 1 075 general anesthesia surgery patients in the First Teaching Hospital of Xinjiang Medical University from March to August 2023, which were randomly divided into modeling set and validation set in the ratio of 7∶3. Combining LASSO regression with the random forest algorithm, intraoperative hypothermia risk factors were screened. Models were constructed based on six machine learning algorithms, Logistic regression, decision tree, support vector machine (SVM) , extreme gradient boosting (XGBoost) , multilayer perceptron (MLP) and K-Nearest Neighbors (KNN) , and evaluate the performance of all models. Web-based dynamic nomogram was developed and interpretable analysis of the optimal model was performed using SHAP graph.Results:According to LASSO regression and random forest algorithm, length of anesthesia, intraoperative blood loss, baseline body temperature, age, intraoperative urine volume, and type of surgery were risk factors for intraoperative hypothermia in patients under general anesthesia, and the difference was statistically significant ( P<0.05) . The areas under the receiver operating characteristic curve for Logistic regression, decision tree, XGBoost, KNN, MLP, and SVM models were 0.777, 0.746, 0.793, 0.743, 0.768, 0.793, and the F1 scores were 0.667, 0.719, 0.861, 0.756, 0.820, and 0.842, respectively. Decision curve showed that the net benefit of the XGBoost model for predicting intraoperative hypothermia in patients was high when the threshold probability was between 0 and 1. A web-based dynamic nomogram was developed with good clinical applicability and generalizability. Conclusions:A dynamic nomogram of intraoperative hypothermia in general anesthesia surgery patients constructed and validated based on the machine learning algorithm can assist medical and nursing staff in identifying patients at high risk of intraoperative hypothermia and implementing personalized interventions.
