Building Distant Metastasis Models for HNSCC Using Machine Learning Techniques

Building Distant Metastasis Models for HNSCC Using Machine Learning Techniques | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Building Distant Metastasis Models for HNSCC Using Machine Learning Techniques Changping Ma, Yaoqi Chen, Jingjing Mao, Jiawen Xue, Xinyi Xu, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5667236/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background : Accurate assessment of the risk of distant metastasis of head and neck squamous cell carcinoma (HNSCC) is important for the development of personalized treatment and prognosis. The purpose of this study was to explore the risk factors for distant metastasis of HNSCC and to establish a predictive model using machine learning methods. Materials and methods: We designed a retrospective cross-sectional study with a cohort from the SEER database (affiliated with the National Cancer Institute). A total of 31,060 cases of head and neck cancer were included by our inclusion criteria. We constructed four machine learning models—Logistic Regression, Decision Tree, XGBoost, and Neural Network—to predict the risk of distant metastasis in HNSCC patients and compared the performance of the four models. Accuracy, precision, recall, and F1-score were used to evaluate the performance of the model. The evaluation ability and clinical practicability of the model were verified by comparing the area under the curve and the receiver operating characteristic curve. Results : The receiver operating characteristic of the four models ranged from 0.681 to 0.847. The average accuracy of all algorithms was 77 %, and XGBoost had the highest accuracy of 85.119 %. Among the four models, XGBoost and Logistic Regression had the highest precision, both with precision greater than 79. Neural Network had the highest recall and F1-score. Decision Tree had the lowest accuracy and recall. Among the four models, the area under the curve of Decision Tree was the lowest, at 0.690, whereas that of XGBoost was the highest, at 0.846. Overall, XGBoost had the best predictive effect. Conclusion : XGBoost had the highest classification accuracy, so this machine learning method could be used to predict distant metastasis of HNSCC. The application of machine learning algorithms can stratify patients with HNSCC in clinic, which is conducive to the development of personalized treatment plans. Clinical Relevance :The findings of this study have significant implications clinical management of HMSCC.The superior predictive performance of XGBoost,as demonstrated by its high precision and area under the Decision Curve,suggests that this machine learning algorithm could be effectively integrated into clinical practice to predict distant metastasis in HMSCC patients.This has the potential to enhance the accuracy of prognostic assessments, thereby facilitating more informed treatment planning and personalized care. head and neck squamous cell carcinoma machine learning predictive model Figures Figure 1 Figure 2 Figure 3 1. Introduction Head and neck cancer is a kind of malignant tumor originating from the upper respiratory tract, salivary gland, and thyroid gland [ 1 ]. It is the eighth most prevalent malignant tumor in the world [ 2 ]. Although its histopathological types are diverse, head and neck squamous cell carcinoma (HNSCC) is by far the most common, accounting for more than 90% of head and neck cancer cases [ 3 ]. The reported incidence of distant metastasis varies greatly, ranging from 3.1 to 30.7% [ 4 , 5 – 11 ]. Modern autopsy results show that the distant metastasis rate of head and neck cancer is as high as 37–47% [ 12 – 14 ]. HNSCC cells often spread to various distant organs, commonly involving the lung (70–85%), liver (10–30%), bone (15–39%), brain (2–8%), skin, mediastinum, and bone marrow [ 15 – 17 ]. Many studies have explored the risk factors related to the distant metastasis of HNSCC, such as tumor site, grade, tumor recurrence, advanced disease, and regional lymph node involvement [ 5 , 8 , 11 , 18 – 20 ], but the findings are very inconsistent. The multidisciplinary management approach has effectively extended the life expectancy of head and neck cancer patients. However, the survival of patients with distant metastasis of head and neck cancer is not promising, as the median survival is only 3.3 months [ 5 ]. More than 65% of patients with HNSCC eventually develop recurrent or metastatic disease [ 21 ]. Therefore, early detection of patients at high risk for distant metastasis is of great significance for the diagnosis and treatment of patients with head and neck squamous cell carcinoma. Machine learning is a type of artificial intelligence. Unlike statistics, which draws population inferences from samples, machine learning processes data to get prediction results [ 22 ]. Machine learning methods connect the concepts of learning and reasoning from data samples to generate models for performing classification, prediction, estimation, or other similar tasks [ 23 ]. Currently, numerous machine learning models are used in the medical field to assist with tasks such as diagnosis, prognosis, treatment, and medical record management to help clinicians improve work efficiency [ 24 ]. In recent years, research on survival prediction of cancer pa tients using machine learning algorithms has become increasingly common. In this study, we used clinical and histopathological parameters from the Surveillance, Epidemiology, and End Results (SEER) public database to develop a machine learning approach for predicting distant metastases in patients with HNSCC. The developed model can economically, conveniently, and efficiently identify patients with a high risk of distant metastasis of HNSCC and provide personalized treatment and prognosis prediction. 2. Materials and Methods 2.1 Data source Data were obtained from the SEER database (affiliated with the National Cancer Institute), which is an authoritative source for cancer statistics in the U.S. We selected the data from the Incidence-seer 18 registration custom data (with additional treatment fields) released in April 2021, based on a document submitted in November 2020. It currently collects and publishes cancer incidence and survival data from 18 population-based cancer registries covering approximately 27.8% of the U.S. population. The database is openly available, so ethics approval was not required for this study. 2.2 Study population From the SEER database, we enrolled 84,633 patients who were diagnosed with head and neck cancer between January 2010 and December 2018. Patients with a diagnosis of head and neck cancer were identified using the International Classification of Disease for Oncology (ICD-O-3) codes: C00-C14, excluding C07 and C08, which include oral cavity, larynx, and pharynx but not salivary glands. Cases were selected based on squamous cell histology (ICD-O-3 histology codes 8050 through 8076, 8078, 8083, 8084, and 8094). Patients with carcinoma in situ or missing TNM staging, regional lymph node examination, surgical method data were excluded, and tumor diagnosis from autopsy or death certificate was excluded. For uniform analysis,Clinicopathological data of all patients were obtained through the database, including age, race, sex, tumor site, histology, grade, surgical method, regional lymph node status, primary lesion, radiotherapy record, chemotherapy record, overall survival, and TNM staging data (according to AJCC classification). The case selection flowchart is shown in Fig. 1 . 2.3 Establishing models The patients with HNSCC data from the SEER cohort were initially analyzed by logistic regression to explore independent risk factors that might influence distant metastasis. We utilized the statistically significant risk factors, including race, sex, tumor site, grade, T stage, N stage, surgical method, and primary lesion, to construct four machine learning prediction models—logistic regression (LR), decision tree (DT), XGBoost (XGB), and neural network (NN)—to predict distant metastasis of HNSCC. An under-sampling method was used to improve the performance of a small classifier in unbalanced data. The processed dataset was divided, with 70% of the data used to train the machine learning algorithm and the remaining 30% used as the test set. 2.4 Model evaluation After establishing the machine learning prediction models, we validated the performance of each model in the test set. The performance of the models compared by means of confusion matrix, and four prediction indexes were obtained: TP, correct metastatic prediction; FP, incorrect metastatic prediction; TN, correct non-metastatic prediction; and FN, incorrect non-metastatic prediction. TP, FP, TN, and FN were used to calculate the recall, accuracy, precision, and F1-score of each model in order to quantify the performance of each model. We also used the area under the curve (AUC) and the receiver operating characteristic (ROC) to evaluate the models. 2.5 Statistical analysis All data were obtained from the SEER database using SEER*Stat software, version 8.3.9. All statistical analyses were performed using Python 3.8.6 and SPSS Statistics 22 software. Descriptive statistics were used to analyze demographic and clinical variables, chi-square tests were used to compare clinicopathological features between patients with distant metastasis and non-metastasis, and nonparametric tests were used for continuous variables. In addition, binary logistic regression analysis was used to explore the influential factors of distant metastasis in HNSCC patients. A P -value of less than or equal to 0.05 was considered to indicate statistical significance. 3. Results 3.1 Baseline characteristics A total of 84,633 head and neck squamous cell carcinoma patients were collected from 2010 to 2018, among which 28,144 patients were excluded from the study due to tumor grade, and 25,429 patients were excluded due to insufficient T or N staging, lymph node information, treatment information, age or diagnostic mode. Ultimately, 31,060 patients diagnosed with HNSCC were included in the study. We divided the patients into metastatic ( n = 1,105, 3.6 %) and non-metastatic ( n = 29,995, 96.4 %) groups. In the metastatic group, the proportion of males was significantly higher than the proportion of females ( P < 0.001). There was a significant difference in racial distribution between the two groups ( P < 0.001). The proportion of oropharynx, hypopharynx, and nasopharynx in the metastatic group was significantly higher than that in the non-metastatic group ( P < 0.001). The incidence of poorly differentiated carcinoma and N2 stage was significantly higher in the metastatic group. In the metastatic group,T1, T2, T3, and T4 patients accounted for 11.86 %, 25.16 %, 22.53 %, and 40.45 %, respectively, indicating that most HNSCC patients were in advanced disease when metastasis was found. Table 1 shows the baseline characteristics in the study population. Tab 1 Baseline characteristics Nonmetastasis Metastasis（n=1105） p- value Age (year) 62.53±10.87 62.61±10.65 0.772 Sex <0.001 Male 22084(73.72%) 879(79.55%) Female 7871(26.28%) 226(20.45%) Race <0.001 White 25516(85.18%) 813(73.57%) Black 2488(8.31%) 187(16.92%) Other 1951(6.51%) 105(9.50%) Site <0.001 Oral cavity 17943(59.90%) 455(41.18%) Oropharynx 1338(4.47%) 104(9.41%) Larynx 7383(24.65%) 231(20.90%) Hypopharynx 1971(6.58%) 175(15.84%) Nasopharynx 1320(4.41%) 140(12.67%) Grade <0.001 I 4918(16.42%) 61(5.52%) II 13895(46.39%) 472(42.71%) III 10651(35.56%) 541(48.96%) IV 491(1.64%) 31(2.81%) Surgery <0.001 No 14652(48.91%) 949(85.88%) Yes 15303(51.09%) 156(14.12%) Radiation record <0.001 Beam radiation 18695(62.41%) 603(54.57%) None/Unknown 11260(37.59%) 502(45.43%) Chemotherapy record <0.001 No 15121(50.48%) 370(33.48%) Yes 14834(49.52%) 735(66.52%) Primary lesion <0.001 No 6877(22.96%) 200(18.10%) Yes 23078(77.04%) 905(81.90%) T <0.001 T1 11181(37.33%) 131(11.86%) T2 9161(30.58%) 278(25.16%) T3 4495(15.01%) 249(22.53%) T4 5118(17.09%) 447(40.45%) N <0.001 N0 14162(47.28%) 166(15.02%) N1 4374(14.60%) 201(18.19%) N2 10612(35.43%) 619(56.02%) N3 807(2.69%) 119(10.77%) Variables that predicted distant metastasis status were evaluated. Univariate analysis showed that age , race, tumor site, grade, radiotherapy, chemotherapy, surgical treatment, T stage, regional lymph node involvement, and other factors were correlated with distant metastasis ( P < 0.05). The results of univariable and multivariable analysis are shown in Table 2. According to multivariate analysis, the main risk factors of distant metastasis were being older, being black, and having a tumor with the oropharynx, hypopharynx, or nasopharynx as the primary site. Compared with well-differentiated and undifferentiated tumors, medium and poorly differentiated tumors were found to be more likely to develop distant metastasis. Chemotherapy was shown to have a certain effect on distant metastasis. Advanced T staging was also found to be a predictor of distant metastasis. Finally, positive lymph nodes were shown to have a strong predictive effect on distant metastasis. Table 2 Univariable and Multivariable analysis of potential factors associated with distant metastasis Variable Univariable analysis Multivariable analysis HR (95% CI) p -value HR (95% CI) p -value Age (year) ≦55 56-70 1.040 （0.891-1.214） 0.622 1.042 （0.893-1.216） 0.602 >70 Sex Female 1.336 （1.107-1.612） 0.002 1.327 （1.104-1.596） 0.003 Male Race White 1.112 （0.950-1.302） 0.187 Black 1.426 （1.195-1.701） 0.000 1.423 （1.192-1.697） 0.000 Other Site 1.241 （0.981-1.570） 0.072 1.237 （0.978-1.566） 0.076 Oral cavity Oropharynx 1.582 （1.252-2.000） 0.000 1.577 （1.248-1.993） 0.000 Larynx 0.966 （0.814-1.147） 0.696 0.970 （0.818-1.152） 0.732 Hypopharynx 1.901 （1.567-2.305） 0.000 1.908 （1.573-2.313） 0.000 Nasopharynx Grade 2.434 （1.923-3.082） 0.000 2.419 （1.911-3.061） 0.000 I II 1.579 （1.191-2.094） 0.002 1.590 （1.199-2.108） 0.001 III 2.017 （1.515-2.684） 0.000 2.039 （1.533-2.712） 0.000 IV 1.572 （0.964-2.563） 0.070 1.581 （0.970-2.578） 0.066 Surgery No Yes 0.330 （0.269-0.404） 0.000 0.327 （0.267-0.401） 0.000 Radiation d No Yes 0.330 （0.269-0.404） 0.000 0.327 （0.267-0.401） 0.000 Chemotherapy d No Yes 1.585 （1.332-1.885） 0.000 1.587 （1.334-1.888） 0.000 Primary lesion No Yes T 1.005 （0.847-1.192） 0.955 T1 T2 1.605 （1.284-2.006） 0.000 1.604 （1.284-2.005） 0.000 T3 2.042 （1.617-2.578） 0.000 2.044 （1.619-2.581） 0.000 T4 2.901 （2.334-3.606） 0.000 2.898 （2.332-3.603） 0.000 Lymph node Negative Positive 3.351 （2.753-4.079） 0.000 3.367 （2.772-4.091） 0.000 3.2 Performance of developed models Figure 2 shows the ROC curves of the four models. The ROC values of the four models ranged from 0.681 to 0.847. The average accuracy of all algorithms was 77 %, and XGB had the highest accuracy of 85.119 % (Table 3). T A B L E 3 Comparison of the performances of the developed models Model AUC Precision Accuracy Recall F1-score LR 0.838 79.240 74.908 0.807 0.799 DT 0.690 76.758 70.295 0.747 0.757 XBG 0.846 79.666 85.119 0.823 0.773 NN 0.837 74.134 76.568 0.955 0.834 Abbreviations: AUC, area under curve . LR, Logisitc Regression; DT, Decision; Tree; XGB, XGBoost; NN, Neural Network Among the four models, XGB and LR had the highest precision, both with precision greater than 79. NN had the highest recall and F1-score. DT had the lowest accuracy and recall. Among the four models, the AUC of DT was the lowest, at 0.690, and the AUC of XGB was the highest, at 0.846. Overall, XGB had the best predictive effect. 3.3 Permutation feature of importance The importance of permutation features was used in the test set to quantify variables affecting metastasis of HNSCC (Figure 3). With XGB, the most important factor influencing distant metastasis was the surgical method, followed by radiotherapy. Discussion The purpose of this study was to establish a machine learning model to predict distant metastasis of HNSCC to help clinicians evaluate the disease and prognosis of patients and make personalized treatment plans. First, we systematically analyzed the distant metastasis of HNSCC patients using data from the SEER database. According to our results, distant metastases were detected in 3.3 % of patients. The main risk factors of distant metastasis in HNSCC patients were found to be being older, being black, having tumors with the primary sites of oropharynx, hypopharynx or nasopharynx, tumor grade II to III stage, undergoing chemotherapy, having advanced T classification, and having regional lymph node involvement. We then constructed the machine learning models based on clinical and pathological data to predict 31,060 cases of distant metastasis of HNSCC. Among the four machine learning models, XGB had the highest AUC, and XGB and NN had the highest accuracy. The prognosis of patients with head and neck cancer varies with lymph node involvement. Similar findings have been reported for head and neck cancer. Hosni et al. found that an increase of 1 % in the positive rate of lymph nodes in oral cancer patients would lead to a 6 % increase in the risk of regional metastasis and a 3 % increase in the risk of distant metastasis and death [25]. Another study showed that the possibility of distant metastasis in HPV-positive oropharyngeal cancer patients with more than five lymph nodes involved was over 50 % [26]. Leemans [27] indicated that the incidence of distant metastases with extranodal spread was three times higher compared with patients without this feature. Remco found that extranodal spread and matted nodes were high risk factors for distant metastasis of HNSCC through CT and/or MRI images before treatment [19]. Studies have also shown that lymph node density has an important impact on patients with distant metastasis of head and neck cancer or local regional treatment failure [28]. In our study, we found a threefold increased risk of distant metastasis when regional lymph nodes were involved. In recent years, owing to the growing maturity of artificial intelligence, the application of metastatic prediction models for head and neck cancer has been increasing. Leitheiser [29] used machine learning models based on DNA methylation to identify the primary tumor of HNSCC presenting with metastases with an unknown primary site. The results showed that NN, support vector machine, and LOGREG models were significantly better than p16-based prediction in overall accuracy (92 % vs. 77 %) and sensitivity (88 % vs. 52 %). Other scholars [30] used the random forest model to predict lymph node metastasis by dual-energy CT texture analysis for HNSCC, and the results showed that the accuracy, sensitivity, and specificity of this model in predicting cervical lymph node metastasis were 88, 100, and 67, respectively. Diamant et al. [31] used preoperative CT imaging data to construct a convolutional neural network model to predict the therapeutic effect of patients with HNSCC. The model predicted the distant metastasis with an AUC of 0.88 and an AUC of local failure of 0.65. According to our analysis, XGB and NN showed good predictive ability. XGB is an optimized distributed gradient boosting library designed to be highly efficient, flexible, and portable. Gradient boosting is an approach where new models are created to predict the residuals or errors of prior models and then combined to make the final prediction. There is evidence that gradient boosting is the algorithm of choice for winners of Kaggle's competition. NN is a kind of imitation biological neural network, and the learning function of NN is similar to the human brain. Like other machine learning methods, neural networks have been used to solve various problems, such as image recognition and natural language processing. The good performance of the NN and XGB models in this study indicates that machine learning has potential in clinical application. To our knowledge, there has been no research on the use of clinical and pathological features to construct a prediction model for distant metastasis of HNSCC in a large cohort database. Therefore, we divided the population into a training set and a test set with a 7:3 ratio. The risk factors of distant metastasis in HNSCC were analyzed, and four machine learning models were established and validated to predict distant metastasis in patients with HNSCC. The prediction of distant metastasis includes seven factors—race, tumor site, grade, chemotherapy, therapy, T stage, and N stage—all of which are clinically and pathologically available before or after treatment. This is the first study using pretreated clinicopathological features to predict distant metastasis of HNSCC in a large cohort data set, demonstrating the feasibility and potential of using machine learning algorithms to assess head and neck cancer. In this database-based study, we enrolled 31,060 patients who met our inclusion criteria, and the median follow-up time was 51 months. We analyzed the data using appropriate statistical methods and drew compelling conclusions. However, there are some limitations to our study. First, data integrity affects the accuracy of the model to a large extent in machine learning research, and the absence of some data in the SEER database may lead to potential systematic bias. Second, to be more consistent with clinical practice, it is not enough to include only clinical data of patients. The further development of this study should seek to combine patient imaging data with clinical data. Third, to confirm the availability of the predicting models in this study, additional validation with external data should be performed. The prediction models should be continuously optimized to meet the needs of clinical application. Conclusion In summary, we used a large population database and machine learning methods to establish four prediction models for distant metastasis in patients with HNSCC. Among the four prediction models, XGB showed good properties in predictive performance. Hence, XGB holds promise for helping physicians make clinical decisions for patients with HNSCC, particularly decisions regarding diagnostic investigation, individualized treatment, and follow-up treatment strategies. Declarations Data Availability The data that support the findings of this study are available from the corresponding author upon reasonable request. Conflict of Interest The authors declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the research reported. Funding Statement This work was supported by the Natural Science Foundation of Ningxia (2021AAC05019). Acknowledgments We thank LetPub (www.letpub.com) for its linguistic assistance during the preparation of this manuscript. References Cramer, J.D., et al., The changing therapeutic landscape of head and neck cancer. Nat Rev Clin Oncol, 2019. 16(11): p. 669-683. Sung, H., et al., Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin, 2021. 71(3): p. 209-249. Du, M., et al., Incidence Trends of Lip, Oral Cavity, and Pharyngeal Cancers: Global Burden of Disease 1990-2017. J Dent Res, 2020. 99(2): p. 143-151. Papac, R.J., Distant metastases from head and neck cancer. Cancer, 1984. 53(2): p. 342-345. Duprez, F., et al., Distant metastases in head and neck cancer. Head Neck, 2017. 39(9): p. 1733-1743. Ellis, E.R., et al., Does node location affect the incidence of distant metastases in head and neck squamous cell carcinoma? International Journal of Radiation Oncology*Biology*Physics, 1989. 17(2): p. 293-297. de Bree, R., et al., Screening for distant metastases in patients with head and neck cancer. Laryngoscope, 2000. 110(3 Pt 1): p. 397-401. Spector, J.G., et al., Delayed regional metastases, distant metastases, and second primary malignancies in squamous cell carcinomas of the larynx and hypopharynx. Laryngoscope, 2001. 111(6): p. 1079-87. Probert, J.C., R.W. Thompson, and M.A. Bagshaw, Patterns of spread of distant metastases in head and neck cancer. Cancer, 1974. 33(1): p. 127-133. Len, X., et al., Distant metastases in head and neck cancer patients who achieved loco-regional control. Head & Neck, 2000. 22(7): p. 680-686. Liu, J.C., et al., Patterns of distant metastasis in head and neck cancer at presentation: Implications for initial evaluation. Oral Oncol, 2019. 88: p. 131-136. Nishijima, W., et al., Analyses of distant metastases in squamous cell carcinoma of the head and neck and lesions above the clavicle at autopsy. Arch Otolaryngol Head Neck Surg, 1993. 119(1): p. 65-8. Zbaren, P. and W. Lehmann, Frequency and sites of distant metastases in head and neck squamous cell carcinoma. An analysis of 101 cases at autopsy. Arch Otolaryngol Head Neck Surg, 1987. 113(7): p. 762-4. Kotwall, C., et al., Metastatic patterns in squamous cell cancer of the head and neck. The American Journal of Surgery, 1987. 154(4): p. 439-442. Ferlito, A., et al., Incidence and sites of distant metastases from head and neck cancer. ORL J Otorhinolaryngol Relat Spec, 2001. 63(4): p. 202-7. Takes, R.P., et al., Distant metastases from head and neck squamous cell carcinoma. Part I. Basic aspects. Oral Oncol, 2012. 48(9): p. 775-9. Bree, R.D., et al., Intracranial Metastases in Patients with Squamous Cell Carcinoma of the Head and Neck. Otolaryngology–Head and Neck Surgery, 2016. 124(2): p. 217-221. Coca-Pelaz, A., J.P. Rodrigo, and C. Suarez, Clinicopathologic analysis and predictive factors for distant metastases in patients with head and neck squamous cell carcinomas. Head Neck, 2012. 34(6): p. 771-5. de Bree, R., et al., Radiologic extranodal spread and matted nodes: Important predictive factors for development of distant metastases in patients with high-risk head and neck cancer. Head Neck, 2016. 38 Suppl 1: p. E1452-8. Peters, T.T., et al., Pretreatment screening on distant metastases and head and neck cancer patients: Validation of risk factors and influence on survival. Oral Oncol, 2015. 51(3): p. 267-71. Argiris, A., et al., Head and neck cancer. The Lancet, 2008. 371(9625): p. 1695-1709. Bzdok, D., N. Altman, and M. Krzywinski, Statistics versus machine learning. Nat Methods, 2018. 15(4): p. 233-234. Kourou, K., et al., Machine learning applications in cancer prognosis and prediction. Comput Struct Biotechnol J, 2015. 13: p. 8-17. Rajkomar, A., J. Dean, and I. Kohane, Machine Learning in Medicine. N Engl J Med, 2019. 380(14): p. 1347-1358. Hosni, A., et al., Lymph node ratio relationship to regional failure and distant metastases in oral cavity cancer. Radiother Oncol, 2017. 124(2): p. 225-231. Lee, N.C.J., et al., Patterns of failure in high-metastatic node number human papillomavirus-positive oropharyngeal carcinoma. Oral Oncol, 2018. 85: p. 35-39. Leemans, C.R., et al., Regional lymph node involvement and its significance in the development of distant metastases in head and neck carcinoma. Cancer, 1993. 71(2): p. 452-456. Rudra, S., et al., Lymph node density--prognostic value in head and neck cancer. Head Neck, 2014. 36(2): p. 266-72. Leitheiser, M., et al., Machine Learning Models Predict the Primary Sites of Head and Neck Squamous Cell Carcinoma Metastases Based on DNA Methylation. J Pathol, 2021. Forghani, R., et al., Head and neck squamous cell carcinoma: prediction of cervical lymph node metastasis by dual-energy CT texture analysis with machine learning. Eur Radiol, 2019. 29(11): p. 6172-6181. Diamant, A., et al., Deep learning in head & neck cancer outcome prediction. Sci Rep, 2019. 9(1): p. 2764. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5667236","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":394406585,"identity":"d0e34707-ea9e-4464-8956-0c22c1baa6f8","order_by":0,"name":"Changping Ma","email":"","orcid":"","institution":"Ningxia Medical University","correspondingAuthor":false,"prefix":"","firstName":"Changping","middleName":"","lastName":"Ma","suffix":""},{"id":394406586,"identity":"f93e2094-32c1-40f4-a8dd-44dd3f889127","order_by":1,"name":"Yaoqi Chen","email":"","orcid":"","institution":"Ningxia Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yaoqi","middleName":"","lastName":"Chen","suffix":""},{"id":394406587,"identity":"54b3a779-18ef-4993-920e-ce00c2867dd2","order_by":2,"name":"Jingjing Mao","email":"","orcid":"","institution":"Ningxia Key Laboratory of Oral Diseases Research,Ningxia Medical University","correspondingAuthor":false,"prefix":"","firstName":"Jingjing","middleName":"","lastName":"Mao","suffix":""},{"id":394406588,"identity":"fd8d6cee-9191-474d-a139-9f6ff511ce60","order_by":3,"name":"Jiawen Xue","email":"","orcid":"","institution":"People’s Hospital of Ningxiang","correspondingAuthor":false,"prefix":"","firstName":"Jiawen","middleName":"","lastName":"Xue","suffix":""},{"id":394406589,"identity":"3075e3d4-e9d3-4cb9-9400-2296e43dc78f","order_by":4,"name":"Xinyi Xu","email":"","orcid":"","institution":"Southern Central Hospital of Yunnan Province","correspondingAuthor":false,"prefix":"","firstName":"Xinyi","middleName":"","lastName":"Xu","suffix":""},{"id":394406593,"identity":"21f4ea6b-c6b7-47a1-b5da-69206acb56de","order_by":5,"name":"Zhongwei Zhuo","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA1klEQVRIiWNgGAWjYJACAwYGCzkDCNOCaC0SxgYMzCCmBNEWSSRuAGthIEKLfP8Zg4IPFRLp29n7j274USDBwN/enYBXC2PDGQPDGWckcnf2HGa72QN0mMSZsxvwamFm7N1gzNsmkbvhRjLbDR6gFgMgG68WNmZeoJZ/EukGQC03/xCjhYcNpKVBIgGk5TZRtkjw8H8wnHFMwnDDmcNmt2UMJHgI+kW+/1iawYcaG3mD443Pbr75YyPH396LXwvIOwYoLiWkHASYHxCjahSMglEwCkYwAACQX0C+Iz7gTgAAAABJRU5ErkJggg==","orcid":"","institution":"The General Hospital of Ningxia Medical University","correspondingAuthor":true,"prefix":"","firstName":"Zhongwei","middleName":"","lastName":"Zhuo","suffix":""}],"badges":[],"createdAt":"2024-12-18 07:53:25","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5667236/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5667236/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":72609637,"identity":"757509e9-641f-4dc5-82b9-e18eb419641c","added_by":"auto","created_at":"2024-12-30 10:10:44","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":25744,"visible":true,"origin":"","legend":"\u003cp\u003eResearch workflow. LR, Logisitc Regression; DT, Decision Tree; XGB, XGBoost; NN, Neural Network\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-5667236/v1/dabda5513c7271d5b91b9baf.png"},{"id":72609638,"identity":"ea4328cf-f3d3-4654-adbe-cb8a52ddeb55","added_by":"auto","created_at":"2024-12-30 10:10:44","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":83110,"visible":true,"origin":"","legend":"\u003cp\u003eEvaluation of the prediction model for distant metastasis in head and neck squamous carcinoma, the average receiver operating characteristic curves from the four machine learning-based models. LR, Logisitc Regression; DT, Decision Tree; XGB, XGBoost; NN, Neural Network\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-5667236/v1/e8f175b1062bfcf4e1d8d084.png"},{"id":72609639,"identity":"d754d54d-aa28-485e-9ee0-e1e4bd162996","added_by":"auto","created_at":"2024-12-30 10:10:44","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":15539,"visible":true,"origin":"","legend":"\u003cp\u003eThe factor importanc e of developed XGB model\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-5667236/v1/8e9bf2d8ea5d440ef037db7d.png"},{"id":72612680,"identity":"186bc732-792c-40fd-a3d1-ab898903b0e4","added_by":"auto","created_at":"2024-12-30 10:34:46","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":630556,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5667236/v1/a1aa24fd-4271-4e84-9448-5e84b3ef1094.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Building Distant Metastasis Models for HNSCC Using Machine Learning Techniques","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eHead and neck cancer is a kind of malignant tumor originating from the upper respiratory tract, salivary gland, and thyroid gland [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. It is the eighth most prevalent malignant tumor in the world [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Although its histopathological types are diverse, head and neck squamous cell carcinoma (HNSCC) is by far the most common, accounting for more than 90% of head and neck cancer cases [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. The reported incidence of distant metastasis varies greatly, ranging from 3.1 to 30.7% [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan additionalcitationids=\"CR6 CR7 CR8 CR9 CR10\" citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Modern autopsy results show that the distant metastasis rate of head and neck cancer is as high as 37\u0026ndash;47% [\u003cspan additionalcitationids=\"CR13\" citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. HNSCC cells often spread to various distant organs, commonly involving the lung (70\u0026ndash;85%), liver (10\u0026ndash;30%), bone (15\u0026ndash;39%), brain (2\u0026ndash;8%), skin, mediastinum, and bone marrow [\u003cspan additionalcitationids=\"CR16\" citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Many studies have explored the risk factors related to the distant metastasis of HNSCC, such as tumor site, grade, tumor recurrence, advanced disease, and regional lymph node involvement [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan additionalcitationids=\"CR19\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e], but the findings are very inconsistent. The multidisciplinary management approach has effectively extended the life expectancy of head and neck cancer patients. However, the survival of patients with distant metastasis of head and neck cancer is not promising, as the median survival is only 3.3 months [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. More than 65% of patients with HNSCC eventually develop recurrent or metastatic disease [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. Therefore, early detection of patients at high risk for distant metastasis is of great significance for the diagnosis and treatment of patients with head and neck squamous cell carcinoma.\u003c/p\u003e \u003cp\u003eMachine learning is a type of artificial intelligence. Unlike statistics, which draws population inferences from samples, machine learning processes data to get prediction results [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Machine learning methods connect the concepts of learning and reasoning from data samples to generate models for performing classification, prediction, estimation, or other similar tasks [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Currently, numerous machine learning models are used in the medical field to assist with tasks such as diagnosis, prognosis, treatment, and medical record management to help clinicians improve work efficiency [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn recent years, research on survival prediction of cancer pa tients using machine learning algorithms has become increasingly common. In this study, we used clinical and histopathological parameters from the Surveillance, Epidemiology, and End Results (SEER) public database to develop a machine learning approach for predicting distant metastases in patients with HNSCC. The developed model can economically, conveniently, and efficiently identify patients with a high risk of distant metastasis of HNSCC and provide personalized treatment and prognosis prediction.\u003c/p\u003e"},{"header":"2. Materials and Methods","content":"\u003cp\u003e2.1 Data source\u003c/p\u003e \u003cp\u003eData were obtained from the SEER database (affiliated with the National Cancer Institute), which is an authoritative source for cancer statistics in the U.S. We selected the data from the Incidence-seer 18 registration custom data (with additional treatment fields) released in April 2021, based on a document submitted in November 2020. It currently collects and publishes cancer incidence and survival data from 18 population-based cancer registries covering approximately 27.8% of the U.S. population. The database is openly available, so ethics approval was not required for this study.\u003c/p\u003e \u003cp\u003e2.2 Study population\u003c/p\u003e \u003cp\u003eFrom the SEER database, we enrolled 84,633 patients who were diagnosed with head and neck cancer between January 2010 and December 2018. Patients with a diagnosis of head and neck cancer were identified using the International Classification of Disease for Oncology (ICD-O-3) codes: C00-C14, excluding C07 and C08, which include oral cavity, larynx, and pharynx but not salivary glands. Cases were selected based on squamous cell histology (ICD-O-3 histology codes 8050 through 8076, 8078, 8083, 8084, and 8094). Patients with carcinoma in situ or missing TNM staging, regional lymph node examination, surgical method data were excluded, and tumor diagnosis from autopsy or death certificate was excluded. For uniform analysis,Clinicopathological data of all patients were obtained through the database, including age, race, sex, tumor site, histology, grade, surgical method, regional lymph node status, primary lesion, radiotherapy record, chemotherapy record, overall survival, and TNM staging data (according to AJCC classification). The case selection flowchart is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e2.3 Establishing models\u003c/p\u003e \u003cp\u003eThe patients with HNSCC data from the SEER cohort were initially analyzed by logistic regression to explore independent risk factors that might influence distant metastasis. We utilized the statistically significant risk factors, including race, sex, tumor site, grade, T stage, N stage, surgical method, and primary lesion, to construct four machine learning prediction models\u0026mdash;logistic regression (LR), decision tree (DT), XGBoost (XGB), and neural network (NN)\u0026mdash;to predict distant metastasis of HNSCC. An under-sampling method was used to improve the performance of a small classifier in unbalanced data. The processed dataset was divided, with 70% of the data used to train the machine learning algorithm and the remaining 30% used as the test set.\u003c/p\u003e \u003cp\u003e2.4 Model evaluation\u003c/p\u003e \u003cp\u003eAfter establishing the machine learning prediction models, we validated the performance of each model in the test set. The performance of the models compared by means of confusion matrix, and four prediction indexes were obtained: TP, correct metastatic prediction; FP, incorrect metastatic prediction; TN, correct non-metastatic prediction; and FN, incorrect non-metastatic prediction. TP, FP, TN, and FN were used to calculate the recall, accuracy, precision, and F1-score of each model in order to quantify the performance of each model. We also used the area under the curve (AUC) and the receiver operating characteristic (ROC) to evaluate the models.\u003c/p\u003e \u003cp\u003e2.5 Statistical analysis\u003c/p\u003e \u003cp\u003eAll data were obtained from the SEER database using SEER*Stat software, version 8.3.9. All statistical analyses were performed using Python 3.8.6 and SPSS Statistics 22 software. Descriptive statistics were used to analyze demographic and clinical variables, chi-square tests were used to compare clinicopathological features between patients with distant metastasis and non-metastasis, and nonparametric tests were used for continuous variables. In addition, binary logistic regression analysis was used to explore the influential factors of distant metastasis in HNSCC patients. A \u003cem\u003eP\u003c/em\u003e-value of less than or equal to 0.05 was considered to indicate statistical significance.\u003c/p\u003e"},{"header":"3. Results","content":"\u003cp\u003e3.1 Baseline characteristics\u003c/p\u003e\n\u003cp\u003eA total of 84,633 head and neck squamous cell carcinoma patients were collected from 2010 to 2018, among which 28,144 patients were excluded from the study due to tumor grade, and 25,429 patients were excluded due to insufficient T or N staging, lymph node information, treatment information, age\u0026nbsp;or diagnostic mode. Ultimately, 31,060 patients diagnosed with HNSCC were included in the study. We divided the patients into metastatic (\u003cem\u003en\u003c/em\u003e = 1,105, 3.6 %) and non-metastatic (\u003cem\u003en\u003c/em\u003e = 29,995, 96.4 %) groups. In the metastatic group, the proportion of males was significantly higher than the proportion of females (\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.001). There was a significant difference in racial distribution between the two groups (\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.001). The proportion of oropharynx, hypopharynx, and nasopharynx in the metastatic group was significantly higher than that in the non-metastatic group (\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.001). The incidence of poorly differentiated carcinoma and N2 stage was significantly higher in the metastatic group. In the metastatic group,T1, T2, T3, and T4 patients accounted for 11.86 %, 25.16 %, 22.53 %, and 40.45 %, respectively, indicating that most HNSCC patients were in advanced disease when metastasis was found. Table 1 shows the baseline characteristics in the study population.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTab 1 Baseline characteristics\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" align=\"\" width=\"437\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003eNonmetastasis\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eMetastasis（n=1105）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 81px;\"\u003e\n \u003cp\u003e\u003cem\u003ep-\u003c/em\u003evalue\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003eAge (year)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 104px;\"\u003e\n \u003cp\u003e62.53\u0026plusmn;10.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 126px;\"\u003e\n \u003cp\u003e62.61\u0026plusmn;10.65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 81px;\"\u003e\n \u003cp\u003e0.772\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 126px;\"\u003e\n \u003cp\u003eSex\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eMale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e22084(73.72%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e879(79.55%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e7871(26.28%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e226(20.45%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eRace\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eWhite\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e25516(85.18%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e813(73.57%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eBlack\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e2488(8.31%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e187(16.92%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eOther\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e1951(6.51%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e105(9.50%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eSite\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eOral cavity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e17943(59.90%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e455(41.18%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eOropharynx\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e1338(4.47%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e104(9.41%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eLarynx\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e7383(24.65%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e231(20.90%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eHypopharynx\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e1971(6.58%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e175(15.84%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eNasopharynx\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e1320(4.41%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e140(12.67%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eGrade\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e4918(16.42%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e61(5.52%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eII\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e13895(46.39%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e472(42.71%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eIII\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e10651(35.56%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e541(48.96%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eIV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e491(1.64%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e31(2.81%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eSurgery\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e14652(48.91%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e949(85.88%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e15303(51.09%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e156(14.12%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eRadiation record\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eBeam radiation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e18695(62.41%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e603(54.57%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eNone/Unknown\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e11260(37.59%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e502(45.43%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eChemotherapy record\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e15121(50.48%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e370(33.48%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e14834(49.52%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e735(66.52%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003ePrimary lesion\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e6877(22.96%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e200(18.10%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e23078(77.04%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e905(81.90%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eT1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e11181(37.33%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e131(11.86%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eT2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e9161(30.58%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e278(25.16%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eT3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e4495(15.01%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e249(22.53%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eT4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e5118(17.09%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e447(40.45%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eN0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e14162(47.28%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e166(15.02%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eN1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e4374(14.60%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e201(18.19%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eN2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e10612(35.43%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e619(56.02%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003eN3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e807(2.69%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 126px;\"\u003e\n \u003cp\u003e119(10.77%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eVariables that predicted distant metastasis status were evaluated. Univariate analysis showed that\u003cu\u003eage\u003c/u\u003e, race, tumor site, grade, radiotherapy, chemotherapy, surgical treatment, T stage, regional lymph node involvement, and other factors were correlated with distant metastasis (\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05). The results of univariable and multivariable analysis are shown in Table 2. According to multivariate analysis, the main risk factors of distant metastasis were being older, being black, and having a tumor with the oropharynx, hypopharynx, or nasopharynx as the primary site. Compared with well-differentiated and undifferentiated tumors, medium and poorly differentiated tumors were found to be more likely to develop distant metastasis. Chemotherapy was shown to have a certain effect on distant metastasis. Advanced T staging was also found to be a predictor of distant metastasis. Finally, positive lymph nodes were shown to have a strong predictive effect on distant metastasis.\u003c/p\u003e\n\u003cp\u003eTable 2 Univariable and Multivariable analysis of potential factors associated with distant metastasis\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" align=\"\" width=\"581\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eVariable\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 232px;\"\u003e\n \u003cp\u003eUnivariable analysis\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 261px;\"\u003e\n \u003cp\u003eMultivariable analysis\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 141px;\"\u003e\n \u003cp\u003eHR\u0026nbsp;(95%\u0026nbsp;CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u003cem\u003ep\u003c/em\u003e-value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 158px;\"\u003e\n \u003cp\u003eHR\u0026nbsp;(95%\u0026nbsp;CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e\u003cem\u003ep\u003c/em\u003e-value\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eAge\u0026nbsp;(year)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e≦55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e56-70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e1.040\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e（0.891-1.214）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.622\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e1.042\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e（0.893-1.216）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e0.602\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e\u0026gt;70\u003c/p\u003e\n \u003cp\u003eSex\u003c/p\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e1.336\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e（1.107-1.612）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e1.327\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e（1.104-1.596）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e0.003\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eMale \u0026nbsp;Race \u0026nbsp; \u0026nbsp; White\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e1.112\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e（0.950-1.302）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.187\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eBlack\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e1.426\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e（1.195-1.701）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e1.423\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e（1.192-1.697）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eOther Site\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e1.241\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e（0.981-1.570）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.072\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e1.237\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e（0.978-1.566）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e0.076\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eOral\u0026nbsp;cavity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eOropharynx\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e1.582\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e（1.252-2.000）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e1.577\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e（1.248-1.993）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eLarynx\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e0.966\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e（0.814-1.147）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.696\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e0.970\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e（0.818-1.152）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e0.732\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eHypopharynx\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e1.901\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e（1.567-2.305）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e1.908\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e（1.573-2.313）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eNasopharynx\u0026nbsp;Grade\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e2.434\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e（1.923-3.082）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e2.419\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e（1.911-3.061）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eII\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e1.579\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e（1.191-2.094）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e1.590\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e（1.199-2.108）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eIII\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e2.017\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e（1.515-2.684）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e2.039\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e（1.533-2.712）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eIV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e1.572\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e（0.964-2.563）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.070\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e1.581\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e（0.970-2.578）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e0.066\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eSurgery\u0026nbsp;No\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e0.330\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e（0.269-0.404）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e0.327\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e（0.267-0.401）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eRadiation\u003c/p\u003e\n \u003cp\u003ed No\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e0.330\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e（0.269-0.404）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e0.327\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e（0.267-0.401）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eChemotherapy\u003c/p\u003e\n \u003cp\u003ed No\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e1.585\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e（1.332-1.885）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e1.587\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e（1.334-1.888）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003ePrimary\u0026nbsp;lesion No\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eYes\u0026nbsp;T\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e1.005\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e（0.847-1.192）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.955\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eT1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eT2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e1.605\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e（1.284-2.006）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e1.604\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e（1.284-2.005）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eT3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e2.042\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e（1.617-2.578）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e2.044\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e（1.619-2.581）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eT4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e2.901\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e（2.334-3.606）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e2.898\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e（2.332-3.603）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eLymph\u0026nbsp;node Negative\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003ePositive\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e3.351\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003e（2.753-4.079）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003e3.367\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e（2.772-4.091）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e3.2 Performance of developed models\u003c/p\u003e\n\u003cp\u003eFigure 2 shows the ROC curves of the four models. The ROC values of the four models ranged from 0.681 to 0.847. The average accuracy of all algorithms was 77 %, and XGB had the highest accuracy of 85.119 % (Table 3).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eT A B L E 3 Comparison of the performances of the developed models\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"555\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 92px;\"\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eAUC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003ePrecision\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eRecall\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003eF1-score\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 92px;\"\u003e\n \u003cp\u003eLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e0.838\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003e79.240\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e74.908\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e0.807\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e0.799\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 92px;\"\u003e\n \u003cp\u003eDT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e0.690\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003e76.758\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e70.295\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e0.747\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e0.757\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 92px;\"\u003e\n \u003cp\u003eXBG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e0.846\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003e79.666\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e85.119\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e0.823\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e0.773\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 92px;\"\u003e\n \u003cp\u003eNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e0.837\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003e74.134\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 103px;\"\u003e\n \u003cp\u003e76.568\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e0.955\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 88px;\"\u003e\n \u003cp\u003e0.834\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eAbbreviations: AUC, area under curve . LR, Logisitc Regression; DT, Decision; Tree; XGB, XGBoost; NN, Neural Network\u003c/p\u003e\n\u003cp\u003eAmong the four models, XGB and LR had the highest precision, both with precision greater than 79. NN had the highest recall and F1-score. DT had the lowest accuracy and recall. Among the four models, the AUC of DT was the lowest, at 0.690, and the AUC of XGB was the highest, at 0.846. Overall, XGB had the best predictive effect.\u003c/p\u003e\n\u003cp\u003e3.3 Permutation feature of importance\u003c/p\u003e\n\u003cp\u003eThe importance of permutation features was used in the test set to quantify variables affecting metastasis of HNSCC (Figure 3). With XGB, the most important factor influencing distant metastasis was the surgical method, followed by radiotherapy.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe purpose of this study was to establish a machine learning model to predict distant\u0026nbsp;metastasis of HNSCC to help clinicians evaluate the disease and prognosis of patients and make personalized treatment plans. First, we systematically analyzed the distant metastasis of HNSCC patients using data from the SEER database. According to our results, distant metastases were detected in 3.3 % of patients. The main risk factors of distant metastasis in HNSCC patients were found to be being older, being black, having tumors with the primary sites of oropharynx, hypopharynx or nasopharynx, tumor grade II to III stage, undergoing chemotherapy, having advanced T classification, and having regional lymph node involvement. We then constructed the machine learning models based on clinical and pathological data to predict 31,060 cases of distant metastasis of HNSCC. Among the four machine learning models, XGB had the highest AUC, and XGB and NN had the highest accuracy.\u003c/p\u003e\n\u003cp\u003eThe prognosis of patients with head and neck cancer varies with lymph node involvement. Similar findings have been reported for head and neck cancer. Hosni et al. found that an increase of 1 % in the positive rate of lymph nodes in oral cancer patients would lead to a 6 % increase in the risk of regional metastasis and a 3 % increase in the risk of distant metastasis and death [25]. Another study showed that the possibility of distant metastasis in HPV-positive oropharyngeal cancer patients with more than five lymph nodes involved was over 50 % [26]. Leemans [27] indicated that the incidence of distant metastases with extranodal spread was three times higher compared with patients without this feature. Remco found that extranodal spread and matted nodes were high risk factors for distant metastasis of HNSCC through CT and/or MRI images before treatment [19]. Studies have also shown that lymph node density has an important impact on patients with distant metastasis of head and neck cancer or local regional treatment failure [28]. In our study, we found a threefold increased risk of distant metastasis when regional lymph nodes were involved.\u003c/p\u003e\n\u003cp\u003eIn recent years, owing to the growing maturity of artificial intelligence, the application of metastatic prediction models for head and neck cancer has been increasing. Leitheiser [29] used machine learning models based on DNA methylation to identify the primary tumor of HNSCC presenting with metastases with an unknown primary site. The results showed that NN, support vector machine, and LOGREG models were significantly better than p16-based prediction in overall accuracy (92 % vs. 77 %) and sensitivity (88 % vs. 52 %). Other scholars [30] used the random forest model to predict lymph node metastasis by dual-energy CT texture analysis for HNSCC, and the results showed that the accuracy, sensitivity, and specificity of this model in predicting cervical lymph node metastasis were 88, 100, and 67, respectively. Diamant et al. [31] used preoperative CT imaging data to construct a convolutional neural network model to predict the therapeutic effect of patients with HNSCC. The model predicted the distant metastasis with an AUC of 0.88 and an AUC of local failure of 0.65.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAccording to our analysis, XGB and NN showed good predictive ability. XGB is an optimized distributed gradient boosting library designed to be highly efficient, flexible, and portable. Gradient boosting is an approach where new models are created to predict the residuals or errors of prior models and then combined to make the final prediction. There is evidence that gradient boosting is the algorithm of choice for winners of Kaggle\u0026apos;s competition.\u0026nbsp;NN is a kind of imitation biological neural network, and the learning function of NN is similar to the human brain. Like other machine learning methods, neural networks have been used to solve various problems, such as image recognition and natural language processing. The good performance of the NN and XGB models in this study indicates that machine learning has potential in clinical application.\u003c/p\u003e\n\u003cp\u003eTo our knowledge, there has been no research on the use of clinical and pathological features to construct a prediction model for distant metastasis of HNSCC in a large cohort database. Therefore, we divided the population into a training set and a test set with a 7:3 ratio. The risk factors of distant metastasis in HNSCC were analyzed, and four machine learning models were established and validated to predict distant metastasis in patients with HNSCC. The prediction of distant metastasis includes seven factors\u0026mdash;race, tumor site, grade, chemotherapy, therapy, T stage, and N stage\u0026mdash;all of which are clinically and pathologically available before or after treatment. This is the first study using pretreated clinicopathological features to predict distant metastasis of HNSCC in a large cohort data set, demonstrating the feasibility and potential of using machine learning algorithms to assess head and neck cancer.\u003c/p\u003e\n\u003cp\u003eIn this database-based study, we enrolled 31,060 patients who met our inclusion criteria, and the median follow-up time was 51 months. We analyzed the data using appropriate statistical methods and drew compelling conclusions. However, there are some limitations to our study. First, data integrity affects the accuracy of the model to a large extent in machine learning research, and the absence of some data in the SEER database may lead to potential systematic bias. Second, to be more consistent with clinical practice, it is not enough to include only clinical data of patients. The further development of this study should seek to combine patient imaging data with clinical data. Third, to confirm the availability of the predicting models in this study, additional validation with external data should be performed. The prediction models should be continuously optimized to meet the needs of clinical application.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn summary, we used a large population database and machine learning methods to establish four prediction models for distant metastasis in patients with HNSCC. Among the four prediction models, XGB showed good properties in predictive performance. Hence, XGB holds promise for helping physicians make clinical decisions for patients with HNSCC, particularly decisions regarding diagnostic investigation, individualized treatment, and follow-up treatment strategies.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eData Availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data that support the findings of\u0026nbsp;this\u0026nbsp;study are available from the corresponding author upon reasonable\u0026nbsp;request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of Interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe\u0026nbsp;authors\u0026nbsp;declare\u0026nbsp;that\u0026nbsp;there\u0026nbsp;is\u0026nbsp;no\u0026nbsp;conflict\u0026nbsp;of\u0026nbsp;interest\u0026nbsp;that\u0026nbsp;could\u0026nbsp;be\u0026nbsp;perceived\u0026nbsp;as prejudicing\u0026nbsp;the\u0026nbsp;impartiality\u0026nbsp;of\u0026nbsp;the\u0026nbsp;research\u0026nbsp;reported.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by the Natural Science Foundation of Ningxia (2021AAC05019).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe thank LetPub (www.letpub.com) for its linguistic assistance during the preparation of this manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eCramer, J.D., et al., The changing therapeutic landscape of head and neck cancer. Nat Rev Clin Oncol, 2019. 16(11): p. 669-683.\u003c/li\u003e\n\u003cli\u003eSung, H., et al., Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin, 2021. 71(3): p. 209-249.\u003c/li\u003e\n\u003cli\u003eDu, M., et al., Incidence Trends of Lip, Oral Cavity, and Pharyngeal Cancers: Global Burden of Disease 1990-2017. J Dent Res, 2020. 99(2): p. 143-151.\u003c/li\u003e\n\u003cli\u003ePapac, R.J., Distant metastases from head and neck cancer. Cancer, 1984. 53(2): p. 342-345.\u003c/li\u003e\n\u003cli\u003eDuprez, F., et al., Distant metastases in head and neck cancer. Head Neck, 2017. 39(9): p. 1733-1743.\u003c/li\u003e\n\u003cli\u003eEllis, E.R., et al., Does node location affect the incidence of distant metastases in head and neck squamous cell carcinoma? International Journal of Radiation Oncology*Biology*Physics, 1989. 17(2): p. 293-297.\u003c/li\u003e\n\u003cli\u003ede Bree, R., et al., Screening for distant metastases in patients with head and neck cancer. Laryngoscope, 2000. 110(3 Pt 1): p. 397-401.\u003c/li\u003e\n\u003cli\u003eSpector, J.G., et al., Delayed regional metastases, distant metastases, and second primary malignancies in squamous cell carcinomas of the larynx and hypopharynx. Laryngoscope, 2001. 111(6): p. 1079-87.\u003c/li\u003e\n\u003cli\u003eProbert, J.C., R.W. Thompson, and M.A. Bagshaw, Patterns of spread of distant metastases in head and neck cancer. Cancer, 1974. 33(1): p. 127-133.\u003c/li\u003e\n\u003cli\u003eLen, X., et al., Distant metastases in head and neck cancer patients who achieved loco-regional control. Head \u0026amp; Neck, 2000. 22(7): p. 680-686.\u003c/li\u003e\n\u003cli\u003eLiu, J.C., et al., Patterns of distant metastasis in head and neck cancer at presentation: Implications for initial evaluation. Oral Oncol, 2019. 88: p. 131-136.\u003c/li\u003e\n\u003cli\u003eNishijima, W., et al., Analyses of distant metastases in squamous cell carcinoma of the head and neck and lesions above the clavicle at autopsy. Arch Otolaryngol Head Neck Surg, 1993. 119(1): p. 65-8.\u003c/li\u003e\n\u003cli\u003eZbaren, P. and W. Lehmann, Frequency and sites of distant metastases in head and neck squamous cell carcinoma. An analysis of 101 cases at autopsy. Arch Otolaryngol Head Neck Surg, 1987. 113(7): p. 762-4.\u003c/li\u003e\n\u003cli\u003eKotwall, C., et al., Metastatic patterns in squamous cell cancer of the head and neck. The American Journal of Surgery, 1987. 154(4): p. 439-442.\u003c/li\u003e\n\u003cli\u003eFerlito, A., et al., Incidence and sites of distant metastases from head and neck cancer. ORL J Otorhinolaryngol Relat Spec, 2001. 63(4): p. 202-7.\u003c/li\u003e\n\u003cli\u003eTakes, R.P., et al., Distant metastases from head and neck squamous cell carcinoma. Part I. Basic aspects. Oral Oncol, 2012. 48(9): p. 775-9.\u003c/li\u003e\n\u003cli\u003eBree, R.D., et al., Intracranial Metastases in Patients with Squamous Cell Carcinoma of the Head and Neck. Otolaryngology\u0026ndash;Head and Neck Surgery, 2016. 124(2): p. 217-221.\u003c/li\u003e\n\u003cli\u003eCoca-Pelaz, A., J.P. Rodrigo, and C. Suarez, Clinicopathologic analysis and predictive factors for distant metastases in patients with head and neck squamous cell carcinomas. Head Neck, 2012. 34(6): p. 771-5.\u003c/li\u003e\n\u003cli\u003ede Bree, R., et al., Radiologic extranodal spread and matted nodes: Important predictive factors for development of distant metastases in patients with high-risk head and neck cancer. Head Neck, 2016. 38 Suppl 1: p. E1452-8.\u003c/li\u003e\n\u003cli\u003ePeters, T.T., et al., Pretreatment screening on distant metastases and head and neck cancer patients: Validation of risk factors and influence on survival. Oral Oncol, 2015. 51(3): p. 267-71.\u003c/li\u003e\n\u003cli\u003eArgiris, A., et al., Head and neck cancer. The Lancet, 2008. 371(9625): p. 1695-1709.\u003c/li\u003e\n\u003cli\u003eBzdok, D., N. Altman, and M. Krzywinski, Statistics versus machine learning. Nat Methods, 2018. 15(4): p. 233-234.\u003c/li\u003e\n\u003cli\u003eKourou, K., et al., Machine learning applications in cancer prognosis and prediction. Comput Struct Biotechnol J, 2015. 13: p. 8-17.\u003c/li\u003e\n\u003cli\u003eRajkomar, A., J. Dean, and I. Kohane, Machine Learning in Medicine. N Engl J Med, 2019. 380(14): p. 1347-1358.\u003c/li\u003e\n\u003cli\u003eHosni, A., et al., Lymph node ratio relationship to regional failure and distant metastases in oral cavity cancer. Radiother Oncol, 2017. 124(2): p. 225-231.\u003c/li\u003e\n\u003cli\u003eLee, N.C.J., et al., Patterns of failure in high-metastatic node number human papillomavirus-positive oropharyngeal carcinoma. Oral Oncol, 2018. 85: p. 35-39.\u003c/li\u003e\n\u003cli\u003eLeemans, C.R., et al., Regional lymph node involvement and its significance in the development of distant metastases in head and neck carcinoma. Cancer, 1993. 71(2): p. 452-456.\u003c/li\u003e\n\u003cli\u003eRudra, S., et al., Lymph node density--prognostic value in head and neck cancer. Head Neck, 2014. 36(2): p. 266-72.\u003c/li\u003e\n\u003cli\u003eLeitheiser, M., et al., Machine Learning Models Predict the Primary Sites of Head and Neck Squamous Cell Carcinoma Metastases Based on DNA Methylation. J Pathol, 2021.\u003c/li\u003e\n\u003cli\u003eForghani, R., et al., Head and neck squamous cell carcinoma: prediction of cervical lymph node metastasis by dual-energy CT texture analysis with machine learning. Eur Radiol, 2019. 29(11): p. 6172-6181.\u003c/li\u003e\n\u003cli\u003eDiamant, A., et al., Deep learning in head \u0026amp; neck cancer outcome prediction. Sci Rep, 2019. 9(1): p. 2764.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"head and neck squamous cell carcinoma, machine learning, predictive model","lastPublishedDoi":"10.21203/rs.3.rs-5667236/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5667236/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground\u003c/strong\u003e: Accurate assessment of the risk of distant metastasis of head and neck squamous cell carcinoma (HNSCC) is important for the development of personalized treatment and prognosis. The purpose of this study was to explore the risk factors for distant metastasis of HNSCC and to establish a predictive model using machine learning methods.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMaterials and methods:\u003c/strong\u003e We designed a retrospective cross-sectional study with a cohort from the SEER database (affiliated with the National Cancer Institute). A total of 31,060 cases of head and neck cancer were included by our inclusion criteria. We constructed four machine learning models—Logistic Regression, Decision Tree, XGBoost, and Neural Network—to predict the risk of distant metastasis in HNSCC patients and compared the performance of the four models. Accuracy, precision, recall, and F1-score were used to evaluate the performance of the model. The evaluation ability and clinical practicability of the model were verified by comparing the area under the curve and the receiver operating characteristic curve.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults\u003c/strong\u003e: The receiver operating characteristic of the four models ranged from 0.681 to 0.847. The average accuracy of all algorithms was 77 %, and XGBoost had the highest accuracy of 85.119 %. Among the four models, XGBoost and Logistic Regression had the highest precision, both with precision greater than 79. Neural Network had the highest recall and F1-score. Decision Tree had the lowest accuracy and recall. Among the four models, the area under the curve of Decision Tree was the lowest, at 0.690, whereas that of XGBoost was the highest, at 0.846. Overall, XGBoost had the best predictive effect.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion\u003c/strong\u003e: XGBoost had the highest classification accuracy, so this machine learning method could be used to predict distant metastasis of HNSCC. The application of machine learning algorithms can stratify patients with HNSCC in clinic, which is conducive to the development of personalized treatment plans.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClinical Relevance\u003c/strong\u003e:The findings of this study have significant implications clinical management of HMSCC.The superior predictive performance of XGBoost,as demonstrated by its high precision and area under the Decision Curve,suggests that this machine learning algorithm could be effectively integrated into clinical practice to predict distant metastasis in HMSCC patients.This has the potential to enhance the accuracy of prognostic assessments, thereby facilitating more informed treatment planning and personalized care.\u003c/p\u003e","manuscriptTitle":"Building Distant Metastasis Models for HNSCC Using Machine Learning Techniques","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-12-30 10:10:39","doi":"10.21203/rs.3.rs-5667236/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"241fa4d6-f53a-4f98-8c28-5bb61ba62288","owner":[],"postedDate":"December 30th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-12-30T10:10:42+00:00","versionOfRecord":[],"versionCreatedAt":"2024-12-30 10:10:39","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5667236","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5667236","identity":"rs-5667236","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}