Construction of prediction models for postoperative nausea and vomiting in patients with thyroid surgery based on machine learning
Objective To construct machine learning models for predicting postoperative nausea and vomiting(PONV)in patients undergoing thyroid surgery.Methods Clinical data of 196 patients who underwent elective thyroid surgery at Affiliated Hospital of Jiaxing University from September 2022 to March 2023 were collected,including 73 patients with postoperative nausea and vomiting(PONV group)and 123 patients without PONV(non-PONV group).A comprehensive model and predictive evaluation system were constructed using the Python programming language.Five classic machine learning methods were used,including support vector machine,decision tree,random forest,logistic regression,and adaptive boost(Adaboost).To ensure the robustness of the model,90％of the data were randomly selected as the training set,and 10％of the data as the validation set.To further evaluate the performance of the model,ten-fold cross-validation was used to comprehensively examine the accuracy of the model.Results Univariate analysis showed that there were significant differences in gender,smoking history,drinking history,motion sickness history,fluid intake,sufentanil and remifentanil dosage between the PONV amd non-PONV groups(all P＜0.05).Among prediction models constructed by five algorithms of machine learning,the model based on Adaboost algorithm was the most accurate one to predict PONV in thyroid surgery with a mean AUC of 0.74.Conclusion The Adaboost model is the most accurate model for predicting postoperative nausea and vomiting in thyroid surgery,and can be developed into software for clinical practice.Based on the prediction results,targeted prevention and treatment measures can be taken to effectively prevent postoperative nausea and vomiting.
Machine learningThyroid surgeryPostoperative nausea and vomiting
万方数据
维普
