Abstract: Objective To identify the factors influencing distant metastasis(DM)in major salivary gland cancer(MaSGC)and to develop a predictive model for the occurrence of DM in MaSGC. Methods The study included 1729 MaSGC patients from the Surveillance, Epidemiology, and End Results(SEER)database,as well as 218 patients from the Affiliated Yantai Yuhuangding Hospital of Qingdao University. Univariate and multivariate logistic regression analyses were used to screen for risk factors for DM in MaSGC patients. Predictive models were built utilizing six machine learning models including support vector machine(SVM), logistic regression(LR), adaptive boosting(AdaBoost), decision tree(DT), random forest(RF), eXtreme Gradient Boosting(XGB)and externally validated. The performance of the model was evaluated using a range of metrics, including accuracy, recall, area under the receiver operating characteristic(ROC)curve(AUC), precision, and the F1 score. The SHapley Additive exPlanation(SHAP)was utilized for the purpose of generating explanations of the model's decisions. Results Gender(OR=0.20, P<0.001), primary tumor site(OR=2.83, P=0.014), T stage(OR=1.88, P=0.004), histological type(P=0.008), tumor size(OR=1.84, P<0.001), and lymph node ratio(LNR)(OR=7.67, P<0.001)were found to be significant influencing factorsfor DM in MaSGC patients. The XGB prediction model built based on the above six factors significantly outperforms the other prediction models(AUCs of 0.97, 0.94, and 0.75 for the internal training cohort, internal validation cohort, and external validation cohort, respectively). Conclusion Gender, primary tumor site, T stage, histological type, tumor size, and LNR were significant risk factorsfor predicting DM in MaSGC patients. The development of the XGB model was predicated on the aforementioned factors, with the objective being the provision of individualised risk predictions for DM in MaSGC patients.

Key words: Major salivary gland cancer, Distant metastasis, Machine learning, Prediction model