Abstract: Objective: To construct and validate the risk prediction model of Intraoperative Hypothermia in lung cancer patients, and to provide a reference for clinical medical staff to identify the high-risk groups of Intraoperative Hypothermia in lung cancer patients. Methods: Patients with lung cancer from June to November 2022 in a tertiary grade A cancer hospital in Sichuan province were randomly divided into training set (770 cases) and validation set (330 cases). Four machine learning algorithms (Logistic regression, XGBoost, random forest, support vector machine) in Recovery Component were used to construct prediction models, and their performance was compared to obtain the best prediction model algorithm for Intraoperative Hypothermia in lung cancer patients. The model was verified in the validation set. Results: The incidence of Intraoperative Hypothermia was 53.2%. Intraoperative blood loss, intraoperative infusion volume, operation time, anesthesia time, operating room temperature, core body temperature after anesthesia and surgical resection site were the influencing factors of Intraoperative Hypothermia in patients with lung cancer. The AUC of the random forest model in the training set was 0.968, and the AUC of the validation set was 0.968, which was better than the other three prediction models. Conclusion: The model based on random forest algorithm is the optimal Intraoperative Hypothermia prediction model for lung cancer patients, which is conducive to clinical screening of high-risk groups, can provide reference for medical staff to take targeted preventive measures early.

Key words: lung cancer; hypothermia; nomogram; machine learning; XGBoost; random forest; support vector machine