Study Of The Influencing Factors And Prediction Model Of The Risk Of Death In Oral Cavity Cancer Based On SEER Database

Objective:The study aims to explore the influencing factors of the risk of all-cause death and cancer-specific death in oral cavity cancer patients by collecting and analysing information of the demographics and clinic-pathological characteristics and therapeutic of oral cavity cancer patients enrolled from the SEER database,and to construct prediction models of the risk of death for quantitative assessment of the risk of death in oral cancer patients,providing a basis for early judgment and screening of patients with poor prognosis of oral cavity cancer.Methods:16413 patients diagnosed with oral cancer from 2010 to2014 were enrolled from the SEER database follow-up data according to inclusion and exclusion criteria.These data were used to analyse the influencing factors and construct the prediction model of death risk.One-wayχ~2 test and multiple logistic regression were used to analyse influencing factors of death risk.Logistic regression,support vector machine and random forest were used to construct prediction models of death risk.The statistical analysis in this study were performed via R4.1.3and SPSS 26.0.The statistical level was two-sidedα=0.05.Results:1.A total of 16,413 oral cancer patients were included in this study,6,677 of whom occurred all-cause death and 2,369 occurred cancer-specific death.2.The results of multiple logistic regression of all-cause death in oral cavity cancer patients showed that:Compared with the age<45,the age45～59(OR=1.240,95%CI:1.045-1.472),the age 60～74(OR=1.801,95%CI:1.522-2.131),the age≥75(OR=4.777,95%CI:3.998-5.707)were risk factors.Compared with the white,the black(OR=1.545,95%CI:1.357-1.760)was a risk factor.Compared with lip primary site,tongue(OR=1.415,95%CI:1.209-1.657),gum(OR=1.795,95%CI:1.481-2.175),floor of mouth(OR=2.470,95%CI:2.058-2.965),palate(OR=1.959,95%CI:1.602-2.395)and buccal mucosa(OR=2.672,95%CI:2.227-3.207)were risk factors.Compared with high differentiation,moderate differentiation(OR=1.428,95%CI:1.298-1.571)and poor differentiation(OR=1.143,95%CI:1.023-1.278)were the risk factors.Compared with non-squamous cell carcinoma,squamous cell carcinoma(OR=3.023,95%CI:2.504-3.651)was a risk factor.Compared with stageⅠ,stageⅡ(OR=1.997,95%CI:1.777-2.244),stageⅢ(OR=2.821,95%CI:2.446-3.254)and stageⅣ(OR=3.848,95%CI:3.361-4.405)were the risk factors.Compared with non-lymphatic involvement,involvement in level 1-3(OR=1.365,95%CI:1.201-1.552),level 1-3+4-5±retropharyngeal(OR=2.338,95%CI:1.972-2.773),level 4-5±retropharyngeal(OR=1.801,95%CI:1.261-2.573)were the risk factors.Neck dissection(OR=1.378,95%CI:1.204-1.577)was a risk factor.Female(OR=0.900,95%CI:0.833-0.973),married(OR=0.603,95%CI:0.562-0.646),primary tumor resection(OR=0.652,95%CI:0.580-0.734),and radiotherapy(OR=0.634,95%CI:0.577-0.697)were the protective factors.3.The results of multiple logistic regression of cancer-specific death in oral cavity cancer patients showed that:Compared with the age<45,the age≥75(OR=2.114,95%CI:1.676-2.666)was a risk factor.Compared with the white,the black(OR=1.266,95%CI:1.076-1.480)and other races(OR=1.423,95%CI:1.198-1.691)were risk factors.Compared with lip primary site,tongue(OR=5.296,95%CI:3.609-7.773),gum(OR=3.064,95%CI:2.021-4.647),floor of mouth(OR=6.014,95%CI:4.026-8.983),palate(OR=3.382,95%CI:2.205-5.186)and buccal mucosa(OR=4.858,95%CI:3.250-7.264)were risk factors.Compared with high differentiation,moderate differentiation(OR=1.177,95%CI:1.029-1.346)was a risk factor.Compared with non-squamous cell carcinoma,squamous cell carcinoma(OR=2.939,95%CI:2.109-4.096)was a risk factor.Compared with stageⅠ,stageⅡ(OR=2.125,95%CI:1.786-2.528),stageⅢ(OR=2.719,95%CI:2.236-3.307)and stageⅣ(OR=3.491,95%CI:2.902-4.199)were the risk factors.Compared with non-lymphatic involvement,involvement in level1-3(OR=1.388,95%CI:1.175-1.639)and level 1-3+4-5±retropharyngeal(OR=1.793,95%CI:1.458-2.204)were risk factors.Female(OR=1.266,95%CI:1.146-1.399),Neck dissection(OR=1.213,95%CI:1.023-1.437)were risk factors.Married(OR=0.815,95%CI:0.734-0.894),primary tumor resection(OR=0.816,95%CI:0.702-0.947)and radiotherapy(OR=0.836,95%CI:0.742-0.942)were protective factors.4.Among the prediction models of all-cause-death risk of oral cavity cancer patients on the training set,the AUC of random forest model was0.777(95%CI:0.707-0.785),with an accuracy of 77.7%,a sensitivity of76.5%,and a specificity of 78.9%;the AUC of logistic model was 0.742(95%CI:0.733-0.751),with an accuracy of 68.0%,a sensitivity of 68.5%,and a specificity of 67.6%;the AUC of support vector machine model was0.692(95%CI:0.684-0.701),with an accuracy of 67.3%,a sensitivity of68.8%,and a specificity of 69.7%.On the validation set,the AUC of random forest model was 0.684(95%CI:0.671-0.698),and the accuracy,sensitivity and specificity of random forest model were 68.6%,67.6%and 69.3%,respectively.The AUC of logistic model was 0.727(95%CI:0.713-0.742),and the accuracy,sensitivity and specificity of logistic model were 66.7%,67.2%,and 66.4%,respectively.The AUC of support vector machine model was 0.680(95%CI:0.667-0.693),and the accuracy,sensitivity and specificity were67.9%,68.2%,and 67.7%,respectively.5.Among the prediction models of cancer-specific-death risk in oral cavity cancer patients on the training set,the AUC of random forest model was 0.796(95%CI:0.7888-0.803),with an accuracy of 79.6%,a sensitivity of 84.5%and a specificity of 74.7%.The AUC of logistic model was 0.709(95%CI:0.699-0.718),and the accuracy,sensitivity,and specificity were65.7%,71.8%,and 59.7%,respectively.The AUC of support vector machine model was 0.673(95%CI:0.665-0.682),and the accuracy,sensitivity,and specificity were 67.3%,73.4%,and 61.3%,respectively.On the validation set,the AUC of random forest model was 0.648(95%CI:0.629-0.677),and the accuracy,sensitivity,and specificity were66.6%,62.3%,and 67.3%.The AUC of logistic model was 0.687(95%CI:0.668-0.707),and the accuracy,sensitivity,and specificity were 59.4%,70.5%,and 57.4%,respectively.The AUC of support vector machine model was 0.637(95%CI:0.619-0.656),with an accuracy of 60.4%,a sensitivity of 68.4%and a specificity of 59.0%.Conclusions:1.Older age,black,non-lip primary site,high differentiation,squamous cell carcinoma,farther range of involvement,late clinical stage,neck dissection were the risk factors for all-cause death of oral cavity cancer patients,while female,married,primary tumor resection and radiotherapy were the protective factors.2.Older age,female,black race,non-lip primary site of tumor,high differentiation,squamous cell carcinoma,farther range of involvement,late clinical stage,neck dissection were the risk factors for cancer-specific death of oral cavity cancer patients,while married,primary tumor resection and radiotherapy were the protective factors.3.Whether in the training set or validation set,compared with logistic regression and support vector machine,the random forest model performed better and could provide a basis for predicting the death risk of oral cavity cancer patients,but further exploration of model optimization is still needed.