Comparison of Machine Learning Methods for Predicting 3-Year Survival in Elderly Esophageal squamous cancer Patients Based on Oxidative Stress

Comparison of Machine Learning Methods for Predicting 3-Year Survival in Elderly Esophageal squamous cancer Patients Based on Oxidative Stress | 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 Article Comparison of Machine Learning Methods for Predicting 3-Year Survival in Elderly Esophageal squamous cancer Patients Based on Oxidative Stress Jin-Biao Xie, Shi-Jie Huang, Lei Gao, Tian-Bao Yang, Wu Wang, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4281425/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 There is currently a lack of machine learning model studies exploring the relationship between oxidative stress score (OSS) and the prognosis of elderly Esophageal squamous cancer(ESCC) patients. Methods This study included elderly ESCC patients who underwent curative resection surgery from January 2013 to December 2020. Machine learning strategies including decision tree (DT), random forest (RF), and support vector machine (SVM) were employed to construct a predictive model for 3-year overall survival (OS) for elder ESCC base on OSS. Results Patients were divided into derivation cohort and validation cohort, and consisted of 340 and 145 patients, respectively. 8 important features which were the most important factors influencing 3-year OS (pathological N stage, pathological T stage, tumor histological type, vascular invasion, CEA, OSS, CA 19-9, and the amount of bleeding) were included in training the RF, DT and SVM. In the derivation cohort, the RF model exhibited the highest predictive performance with an AUC of 0.975(0.962-0.987), while the DT model is 0.784(0.739-0.830) and the SVM is 0.879(0.843-0.916). In the external validation cohort showed the similar trend . Conclusion The random forest clinical prediction model constructed based on OSS can effectively predict the prognosis of elderly ESCC patients after curative surgery. Health sciences/Gastroenterology/Oesophagogastroscopy Health sciences/Diseases/Cancer Elderly Esophageal squamous cancer Oxidative stress Machine learning Overall survival Prediction model Figures Figure 1 Figure 2 Figure 3 Introduction Esophageal squamous cancer (ESSC) is one of the malignant tumors with high incidence and mortality rates in the world 1 . Compared to younger patients, elderly cancer patients exhibit greater heterogeneity in terms of physical, functional, psychological, and social capabilities or weaknesses 2 , leading to the current TNM staging system, being unable to accurately reflect the prognostic characteristics of elderly ESSC patients. In recent years, studies have found that biological markers such as albumin (ALB), total bilirubin (TBIL), direct bilirubin (DBIL), blood urea nitrogen (BUN), creatinine (Crs), which reflect systemic oxidative stress, play important roles in the occurrence, development, and prognosis of elderly malignant tumor patients 3-6 . It is believed that the antioxidant enzyme activity in elderly individuals decreases, making them more vulnerable to damage from oxygen free radicals, which may lead to DNA, protein, and lipid damage, further accelerating cellular aging and disease progression 7-9 . Therefore, understanding the oxidative stress characteristics of elderly cancer patients is expected to help clinicians predict the complex prognosis of ESCC, optimize patient treatment, and improve quality of life. Currently, there is still a lack of predictive models for survival in elderly ESCC patients based on oxidative stress indicators. As a branch of artificial intelligence, supervised machine learning methods, due to their greater flexibility in capturing complex nonlinear relationships, particularly when dealing with large and sparse data, have been widely used in predicting the prognosis of biological properties 10-12 . However, previous models relied on known variables such as TNM staging, histopathological characteristics, surgery, and chemotherapy, making them unsuitable for complex elderly individuals. Given the crucial role of oxidative stress in elderly ESSC patients, this study aims to explore the relationship between oxidative stress and the prognosis of elderly ESSC patients and to construct a machine learning model for predicting 3-year survival post-surgery to assist in clinical decision-making. Methods Patient Selection and Study Methods This study included elderly ESCC patients who underwent curative ESCC resection surgery from January 2013 to December 2020 and were registered in the Thoracic Surgery Database of Putian University Affiliated Hospital (AHPTU) and Fujian Medical University Union Hospital (FMUUH). Inclusion criteria were as follows: (1) postoperative pathological diagnosis of Esophageal squamous cell carcinoma; (2) age ≥65 years at diagnosis; (3) underwent curative surgery with no evidence of distant metastasis; and (4) complete clinical and pathological data were available. Exclusion criteria were: (1) postoperative pathology confirmed a non-primary tumor originating from the esophagus; (2) presence of distant metastases; (3) incomplete clinical data. Finally, after exclusions, 340 patients from AHPTU were included in the study group as the derivation cohort, and 145 patients from FMUUH were included in the external validation cohort. The derivation cohort was used to develop a novel oxidative stress score (OSS) and construct a machine learning predictive model (Figure 1). The predictive performance of the models was evaluated in the derivation and validation cohorts using the area under the receiver operating characteristic curve (AUC) and calibration curve. Event occurrence time or review was determined based on the date of surgery to the last contact (death or last follow-up). This study was a retrospective analysis of anonymized data in the database, and the Institutional Review Board waived the requirement for obtaining informed consent. This study was approved by the Ethics Committee of Putian University Affiliated Hospital (Approval Number: [2022044]). The study protocol was reviewed and granted ethical approval by the ethics committee to ensure the protection of human subjects and compliance with ethical standards. Candidate predictive variables Complete data on preoperative tests, intraoperative conditions, postoperative recovery, and pathological results are essential. Routine blood and biochemical tests were conducted for each patient from the first day of admission. The TNM staging was reclassified according to the American Joint Committee on Cancer/Union for International Cancer Control 8th edition AJCC Cancer Staging Manual. Based on the optimal cutoff values identified by R 4.3.3 software, each biochemical parameter was converted into a categorical variable. The oxidative stress indicators in our study included ALB, TBIL, DBIL, BUN, Crs, lactate dehydrogenase (LDH), and uric acid (UA). Based on the optimal cutoff values, the values of the biochemical parameters were defined as low (values below the cutoff) or high (values above the cutoff). The derivation cohort was used to develop a new OSS based on the beta coefficients from multivariate stepwise Cox regression analysis. Patients were stratified into risk groups based on the calculated optimal cutoff value of OSS and validated in the validation cohort. Other clinically relevant features used in training the machine learning predictive model were established by researchers based on clinical reasoning, literature review, and consensus on routine availability to ensure wide applicability in various clinical settings. Specifically, the predictive model included preoperative hematological tests (white blood cell count (WBC), hemoglobin (HB), neutrophils (NE), lymphocytes (LYM), monocytes (NOM), prothrombin time (PT)), biochemical tests( (ALB), BUN, TBIL, DBIL, UA, Crs, LDH), tumor markers (alpha-fetoprotein (AFP), carcinoembryonic antigen (CEA), carbohydrate antigen 19-9 (CA199)), clinical variables (gender, age, OSS, body mass index, history of major abdominal surgery, history of previous malignancy, Charlson Comorbidity Index, American Society of Anesthesiologists (ASA) score, neoadjuvant therapy), intraoperative variables (extent of resection, intraoperative blood loss), postoperative variables (Clavien-Dindo complication grading, adjuvant chemotherapy), and pathological variables (differentiation grade, pathological T(pT) stage, pathological N(pN) stage, pathological TNM(pTNM) stage, vascular invasion, perineural invasion). Due to potential collinearity among variables, other potential predictive factors were excluded from the candidate predictive variables (using pT stage, pN stage without using pTNM stage). Variables were standardized to ensure comparability of scales. Establishment of machine learning model We used Random Forest (RF), Decision Tree (DT), and Support Vector Machine (SVM) models to predict the survival status at 3 years post-surgery. The selection of these three machine learning methods was based on their wide application in handling similar diseases, outcomes, and cohorts. In the RF package, the RF function was used to generate the RF model. This function defaults to generating 500 trees, with 4 variables sampled at each node, and a minimum node size of 1. For the DT model, we used the r part function, using the information gain criterion by setting the parms parameter. This process involved cross-validation to select the optimal model by analyzing the complexity parameter (CP) and cross-validation error in the cross-validation table. We determined the optimal CP value and pruned the model based on it to reduce overfitting and optimize model performance. SVM is a well-known supervised classification model that primarily learns the optimal linear separating hyperplane in a high-dimensional feature space. SVM also uses kernel functions to handle non-linear separable categories. We fitted the SVM model with various hyperparameter combinations using the RBF kernel function and conducted comparative analysis to find the best model configuration, systematically optimizing model performance. Follow-up Follow-up visits were scheduled every 3 months within the first 2 years postoperatively, and every 6 months from 2 to 5 years. The final assessment was conducted in December 2023. Most routine follow-up visits included physical examinations, laboratory tests, chest X-rays, abdominal ultrasound or computed tomography scans, and annual endoscopic examinations. The primary outcome was defined as overall survival (OS) post-discharge. OS was defined as the time from surgery to death from any cause or to the last follow-up date for censored observations. Statistical analysis We conducted data analysis using R version 4.3.3. For continuous data that did not follow a normal distribution, we used the Mann-Whitney test, for continuous data that followed a normal distribution, we used the independent t-test. Differences in the distribution of categorical variables between groups were analyzed using the Pearson's chi-squared test and Fisher's exact test. Overall survival (OS) curves were plotted using the Kaplan-Meier method, and differences between survival curves were assessed using the log-rank test. Validation was performed using bootstrap resampling. Model parameters were trained using derivation cohorts, and the performance of the trained model was evaluated using independent validation datasets. The performance of the trained classifier was measured using sensitivity, specificity, accuracy, AUC values, and Brier scores. All statistical analyses were conducted using R software version 4.3.3 (https://www.r-project.org/), and a two-sided p-value <0.05 was considered statistically significant. Results Research cohort A total of 485 elderly ESCC patients were included in this study. In the derivation cohort, there were 340 patients, of whom 261 (76.8%) were male and 79 (23.2%) were female, with a median age of 69 (67 - 73) years. In the validation cohort, there were 145 patients, with 119 (82.1%) males and 26 (17.9%) females, with a median age of 69 (67 - 73) years old. There were statistical differences in the clinical and pathological data of patients in the derivation and validation cohorts in terms of pathological N staging, lymphovascular invasion, and complication grading (p=0.007, 0.047, 0.002, respectively), while other variables showed no statistical differences (Table 1, p>0.05). In terms of survival rates, the 3-year overall survival rate in the derivation cohort was 37.22% (29.78%, 46.52%), while in the validation cohort, it was 49.87% (39.50%, 62.97%) (Figure s1). Developing a novel oxidative stress score In the derivation cohort, the optimal cutoff values for oxidative stress indicators were determined using the surv_cutpoint method as follows: ALB 39.4g/dL, BUN 4.65mg/dL, TBIL 5.7μmol/L, DBIL 1.5μmol/L, UA 215μmol/L, LDH 162U/L, and Crs 49.3μmol/L. These indicators were included in a Cox proportional hazards model to perform stepwise regression analysis based on the Akaike Information Criterion. It was found that ALB, BUN, UA, LDH, and Crs were the core factors affecting the OS of elderly ESCC patients in the final model (eTable 1). Based on the regression coefficients of these variables, a prognostic model called Esophageal Squamous Cancer Oxidative Stress Score (OSS) was further constructed as follows: OSS = ALB * -0.3197 + BUN * 0.2397 + UA * -0.4927 + LDH * 0.3392 + Crs * 0.6625. Patients were stratified into high, moderate, and low-risk groups using the optimal cutoff value of OSS (Figure s2). Kaplan-Meier survival curve analysis indicated that the survival rate of patients in the low OSS group was significantly lower than that of the moderate OSS group and high OSS group (p < 0.05). Similar results were obtained in the validation cohort (Figure s3). Variable selection In our study, we used the Boruta algorithm to select important variables related to the disease status. Boruta is a variable importance assessment method based on RF, which determines which variables are truly important by comparing the importance of the original variables with randomly generated "shadow" variables. After running 50 iterations, a set of variables such as pN, pT, Tumor histology, Lymphovascular invasion, CEA, OSS, CA 19-9, and Bleeding were identified as important variables. The importance of these variables was visualized in graphical charts, where the importance of each variable is shown by comparing its importance relative to the maximum importance of the shadow variables (Figure s4). Model Performance: Validation We included the selected variables in machine learning to construct three models (RF, DT, SVM), and Tables 2 show the performance metrics of these models in predicting 3-year OS in the derivation and validation cohorts. The AUC for RF was 0.975(0.962-0.987) and 0.791(0.717-0.864), DT had AUC values of 0.784(0.739-0.830) and 0.717(0.640-0.794) in the derivation and validation cohorts, respectively, and SVM had AUC values of 0.879(0.843-0.916) and 0.779(0.702-0.856). Compared to DT and SVM, RF had higher AUC values on validation dataset, indicating that the RF model has excellent predictive performance and good generalization ability (Figure s5A/B/C). In the derivation cohort, results from 1000 resamplings showed that the Brier scores of RF, SVM, and DT were 0.075, 0.143, and 0.168, respectively. The Brier score is used to measure the accuracy of predicted probabilities, with lower scores indicating smaller deviations between predictions and actual outcomes, i.e., higher prediction accuracy. These data suggest that in the derivation cohort, the RF model had the highest prediction accuracy, followed by SVM, while the DT model had relatively lower accuracy. In the resampling performance metrics on the test set, the Brier scores for RF, SVM, and DT were 0.191, 0.186, and 0.218, respectively. This highlights the advantage of RF and SVM over DT in terms of generalization ability (Tables 2). Tables 2 show the performance metrics of each model based on resampling. Model Performance: Calibration and Decision Curve The calibration curve plots show that the RF model performed well on all datasets, with predicted probabilities roughly matching the observed event frequencies. Particularly in the validation cohort, the predicted probabilities were closer to the actual outcomes, indicating better generalization ability for the RF model (Figure 2). In contrast, the DT model's predictions deviated significantly from the diagonal line on both datasets, indicating substantial differences between predicted and observed values in certain probability ranges, suggesting poor calibration within these specific prediction probability intervals. The SVM model's calibration was close to the 45-degree line in both the derivation and validation cohorts, with slight deviations in some probability intervals (Figure 2), indicating good calibration of the SVM model in predicting probabilities and consistency across different datasets.Additionally, we used decision curve analysis to compare the clinical usefulness of these models. The results showed that the RF model exhibited the highest net benefit in most threshold probability ranges in the derivation cohort, while the DT and SVM models had similar net benefits, but both were lower than the RF model (Figure 3A). In the validation cohort, the RF model maintained high net benefits in most threshold probability ranges, while the DT and SVM showed similar performance, both lower than the RF model (Figure 3B).。 Discussion The 3-year survival rate after ESCC resection can serve as a valuable audit indicator for evaluating the long-term quality of tumor surgical care. As a special population, elderly individuals may face greater risks from surgery due to their unique bio-psychosocial characteristics, such as increased risk of complications, frailty, decreased stress endurance, declining physical function, cognitive decline, and other factors that may complicate postoperative survival status 23 ， 24 . In recent years, machine learning methods have been widely applied in predicting survival outcomes for patients with liver cancer, gastric cancer, colorectal cancer, breast cancer, and prostate cancer 18-22 , demonstrating good predictive performance. Therefore, in this study, we successfully predicted the 3-year survival rate after surgery for elderly ESCC patients using machine learning models (DT, RF, SVM), with the RF model performing the best. Previous clinical studies have preliminarily confirmed the predictive value of some clinical pathological biomarkers in predicting recurrence, metastasis, and overall survival rates after ESCC surgery, such as tumor size, venous invasion, differentiation status, and tumor lymph node metastasis (TNM) 25-27 . However, these predictive biomarkers have high detection costs and may not be applicable to the complex prognosis of elderly ESCC patients. On the other hand, oxidative stress catalyzes glycolysis, activates tumor cell migration, and promotes tumor proliferation. At the same time, varying levels of oxidative stress can also alter phosphorylation levels, thereby regulating the malignancy and prognosis of malignant tumors 28 . Some researchers have proposed that oxidative stress may be associated with the expression of ferritin metabolism genes, thereby interfering with prognosis 29 . Animal model experiments have shown that in response to external stimuli, mice exhibit increased oxidative stress factors, leading to significant elevations in biochemical markers such as TBIL, lactate dehydrogenase (LDH), creatinine (CRE), and blood urea nitrogen (BUN), which can promote tumor initiation and progression 30 ， 31 . Prospective studies have shown that under oxidative stress, patients' levels of ALB、BUN、UA、LDH and Crs undergo changes. Following antioxidant therapy, patients' scores, mortality rates, and rates of sepsis are all superior to those of the control group 32 . Although oxidative stress is associated with cancer, there is limited research on using it to predict the prognosis of ESCC. Oxidative stress may be a potential indicator to improve the accuracy of predicting the prognosis of elderly ESCC patients. Based on this, our study proposed an oxidative stress index for ESCC, termed OSS, which is composed of ALB, DBIL, and BUN, based on preoperative hematological indicators closely associated with oxidative stress. Our research found that patients with lower OSS had a worse prognosis compared to those with higher OSS. It is worth noting that OSS was generated by training on a cohort of patients from one of our institutions, which allows for the use of highly detailed clinical data and long-term follow-up. Therefore, we reasonably speculate that a predictive model incorporating OSS status may better predict the prognosis of elderly ESCC patients. Several models have been reported for predicting postoperative survival in ESCC. Li 33 built nomograms for predicting progression-free survival (PFS) and OS based on the Cox model to determine independent prognostic factors for PFS and OS. Following internal cross-validation, the corrected concordance indices were 0.739 and 0.696. However, inherent selection bias in retrospective studies and the inclusion of a limited number of cases further magnified this limitation. On the other hand, Xie 34 utilized a prospective study to construct an OS prediction model for ESCC patients using LASSO regression. In the training and validation cohorts, the area under the 3-year ROC curve was 0.811 (0.67 - 0.952,95% CI) and 0.805 (0.638 - 0.973,95% CI) respectively, showing high predictive performance. However, this model did not specifically differentiate elderly ESCC patients, and further research is needed to confirm whether the established model is applicable to elderly patients. In contrast, Liu 35 conducted a stratified analysis of survival characteristics in elderly ESCC patients and developed a nomogram prognostic prediction model with a C-index of 0.706. However, this study used the SEER database as the validation cohort, resulting in the inclusion of fewer modeling indicators that might not reflect more information about the patients. To overcome this limitation, Xie 36 studied the predictive value of basic indicators such as age, sex, education level, as well as pathological indicators including pathological tumor staging, tumor histology, and margin status, along with surgical indicators like neoadjuvant therapy, reoperation, and Charlson comorbidity index for ESCC. However, this comprehensive model, which includes a wide range of information, was established only through multivariable regression analysis to create a predictive model, lacking the generalizability and automation that machine learning offers may lead to the omission of key prognostic indicators. In this scenario, we developed and validated different machine learning methods (RF, DT, SVM) to enhance the discriminative ability of 3-year OS in elderly ESCC patients. Compared to other models, the RF (RF) model demonstrated excellent performance and good calibration in predicting the 3-year survival status. Additionally, our model utilized routine and easily obtainable perioperative clinical data. At the same time, our model utilizes routine and easily obtainable perioperative clinical data. The most important variables influencing the 3-year postoperative survival in elderly patients are pN, pT, tumor histological type, vascular invasion, CEA, OSS, CA 19-9, and the amount of bleeding.. This provides a new opportunity to understand the importance of preoperative oxidative stress and body status, surgical performance, postoperative recovery, and tumor staging in predicting the 3-year survival rate after surgery in elderly ESCC patients. This may allow clinicians to better understand the accuracy and effectiveness of managing these patients. Decision curve analysis provides us with a tool to assess and compare the performance of different models at multiple thresholds, further aiding in model selection and application. Based on decision curve analysis, we observed that the RF model exhibited relatively superior performance in the training, testing, and validation datasets, implying that it may possess good generalizability. Our novel RF model demonstrated higher AUC values compared to previous models, possibly due to the inclusion of more comprehensive clinical assessment indicators specific to elderly patients, such as oxidative stress indicators, comorbidity indices, and complication status. Although pN and pT were identified as the most decisive variables for model prediction, their importance significantly exceeded that of other variables. Tumor histological type, vascular invasion, CEA, OSS, CA 19-9, and the amount of bleeding also showed relatively high importance. These results provide insights into the key variables that play crucial roles in the model, aiding in further optimization and interpretation of the model. Therefore, when establishing a predictive model for long-term survival after ESCC surgery, these prognostic factors should be considered comprehensively. Some limitations should be acknowledged, as the study was retrospective and selection bias cannot be completely avoided. The OSS was constructed by combining serum indicators such as ALB, BUN, UA, LDH, and Crs, and the score may not be sufficient to accurately reveal the oxidative stress status of patients. A more accurate assessment of oxidative stress usually requires the detection of reactive oxygen species and their associated markers, such as superoxide dismutase, malondialdehyde. Second, the performance of the predictive model we developed was evaluated in terms of discriminant ability and risk calibration, and the self-resampling bootstrap program offset overfitting to some extent. However, due to the limited sample size, there is still a risk of insufficient generalization. Finally, well-known factors that influence ESCC, such as high-risk genetic mutations, immunotherapy drug use, and socioeconomic status, were not available from our database, despite the potential for these factors to improve model performance. In the future, we recommend prospective studies to further validate the results of this study. Conclusion The clinical predictive model for OSS constructed using machine learning methods can effectively predict the prognosis of elderly patients with ESCC after curative surgery. Declarations Compliance with ethical standards Author Disclosures: Jin-Biao Xie, Shi-Jie Huang, Tian-Bao Yang, Lei Gao, Wu Wang, Bo-Yang Chen, have no conflicts of interest or financial ties to disclose. Informed Consent: Informed consent was obtained from each participant. Funding: Science and Technology Foundation of Putian, Grant/Award (Number: 2023S3F005),the Health Science and Technology Foundation of Fujian Province，Grant/Award(Number: 2022QNA100), Joint Funds for the innovation of Science and Technology，Fujian province (NO.2023Y9168). Human rights statement and informed consent: All procedures followed were in accordance with the ethical standards of the responsible committee on human experimentation (institutional and national) and with the Helsinki Declaration of 1964 and later versions. Informed consent or substitute for it was obtained from all patients for being included in the study. Data availability statement ： All data generated or analysed during this study are included in the supplementary information files. References Ferlay J, Colombet M, Soerjomataram I, et al. Estimating the global cancer incidence and mortality in 2018: GLOBOCAN sources and methods. Int J Cancer. 2019;144(8):1941-1953. Bron D, Soubeyran P, Fulop T. Innovative approach to older patients with malignant hemopathies. Haematologica. 2016;101(8):893-895. Hayes J, Dinkova-Kostova A, Tew K. Oxidative Stress in Cancer. Cancer cell. 2020;38(2):167-197. Arfin S, Jha N, Jha S, et al. Oxidative Stress in Cancer Cell Metabolism. Antioxidants Basel, Switzerland.2021;10(5). Abdelhamid R, Nagano S. Crosstalk between Oxidative Stress and Aging in Neurodegeneration Disorders. Cells. 2023;12(5). Wang J, Sun Y, Zhang X, et al. Oxidative stress activates NORAD expression by H3K27ac and promotes oxaliplatin resistance in gastric cancer by enhancing autophagy flux via targeting the miR-433-3p. Cell death & disease. 2021;12(1):90. Toh D, Lee W, Zhou H, et al. Lycium barbarumWolfberry () Consumption with a Healthy Dietary Pattern Lowers Oxidative Stress in Middle-Aged and Older Adults: A Randomized Controlled Trial. Antioxidants (Basel, Switzerland). 2021;10(4). Cortés-Jofré M, Rueda J, Asenjo-Lobos C, Madrid E, Bonfill Cosp X. Drugs for preventing lung cancer in healthy people. The Cochrane database of systematic reviews. 2020;3(3):CD002141. Sanidad K, Sukamtoh E, Xiao H, McClements D, Zhang G. Curcumin: Recent Advances in the Development of Strategies to Improve Oral Bioavailability. Annual review of food science and technology. 2019;10:597-617. Deo R. Machine Learning in Medicine. Circulation. 2015;132(20):1920-1930. Van Calster B, Wynants L. Machine Learning in Medicine. The New England journal of medicine. 2019;380(26):2588. Jiang Y, Zhang Z, Yuan Q, et al. Predicting peritoneal recurrence and disease-free survival from CT images in gastric cancer with multitask deep learning: a retrospective study. The Lancet Digital health. 2022;4(5):e340-e350. Li X, Zhai Z, Ding W, et al. An artificial intelligence model to predict survival and chemotherapy benefits for gastric cancer patients after gastrectomy development and validation in international multicenter cohorts. International journal of surgery. 2022;105:106889. Rahman S, Maynard N, Trudgill N, et al. Prediction of long-term survival after gastrectomy using random survival forests. The British journal of surgery. 2021;108(11):1341-1350. Collins G, Reitsma J, Altman D, Moons K. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): the TRIPOD statement. Annals of internal medicine. 2015;162(1):55-63. Gao Q, Zhou H, Wang G, et al. A Model for Predicting Clinical Prognosis in Patients with WHO Grade 2 Glioma. Journal of oncology. 2022;2022:2795939. Fazel S, Långström N, Hjern A, Grann M, Lichtenstein P. Schizophrenia, substance abuse, and violent crime. JAMA. 2009;301(19):2016-2023. Famularo S, Donadon M, Cipriani F, et al. Machine Learning Predictive Model to Guide Treatment Allocation for Recurrent Hepatocellular Carcinoma After Surgery. JAMA surgery. 2023;158(2):192-202. Lee K, Jang J, Yu Y, et al. Usefulness of artificial intelligence for predicting recurrence following surgery for pancreatic cancer: Retrospective cohort study. International journal of surgery (London, England). 2021;93:106050. van den Bosch T, Warps A, de Nerée Tot Babberich M, et al. Predictors of 30-Day Mortality Among Dutch Patients Undergoing Colorectal Cancer Surgery, 2011-2016. JAMA network open. 2021;4(4):e217737. Clift A, Dodwell D, Lord S, et al. Development and internal-external validation of statistical and machine learning models for breast cancer prognostication: cohort study. BMJ (Clinical research ed). 2023;381:e073800. Lee C, Light A, Alaa A, Thurtle D, van der Schaar M, Gnanapragasam V. Application of a novel machine learning framework for predicting non-metastatic prostate cancer-specific mortality in men using the Surveillance, Epidemiology, and End Results (SEER) database. The Lancet Digital health. 2021;3(3):e158-e165. Bollschweiler E, Plum P, Mönig SP, Hölscher AH. Current and future treatment options for ESCC in the elderly. Expert Opin Pharmacother. 2017;18(10):1001-1010.. Lu HW, Chen CC, Chen HH, Yeh HL. The clinical outcomes of elderly ESCC patients who received definitive chemoradiotherapy. J Chin Med Assoc. 2020;83(10):906-910. Powell AGMT, Eley C, Chin C, et al. Prognostic significance of serum inflammatory markers in ESCC [published correction appears in Esophagus. 2021 Jul;18(3):710. Xi Y, Lin Y, Guo W, et al. Multi-omic characterization of genome-wide abnormal DNA methylation reveals diagnostic and prognostic markers for esophageal squamous-cell carcinoma. Signal Transduct Target Ther. 2022;7(1):53.. Hagens E, Tukanova K, Jamel S, et al. Prognostic relevance of lymph node regression on survival in ESCC: a systematic review and meta-analysis. Dis Esophagus. 2022;35(1):doab021. Liang J, Cao R, Wang X, et al. Mitochondrial PKM2 regulates oxidative stress-induced apoptosis by stabilizing Bcl2. Cell Res. 2017;27(3):329-351 Wang X, Xu Y, Dai L, et al. A novel oxidative stress- and ferroptosis-related gene prognostic signature for distinguishing cold and hot tumors in colorectal cancer. Front Immunol. 2022;13:1043738. Liu M, Rao H, Liu J, et al. The histone methyltransferase SETD2 modulates oxidative stress to attenuate experimental colitis. Redox biology. 2021;43:102004. Sandesc M, Rogobete A, Bedreag O, et al. Analysis of oxidative stress-related markers in critically ill polytrauma patients: An observational prospective single-center study. Bosnian journal of basic medical sciences. 2018;18(2):191-197. Lima W, Martins-Santos M, Chaves V. Uric acid as a modulator of glucose and lipid metabolism. Biochimie. 2015;116:17-23. Li B, Wang R, Zhang T, et al. Development and validation of a nomogram prognostic model for ESCC patients with oligometastases. Sci Rep. 2020;10(1):11259. Xie C, Yang P, Zhang X, et al. Sub-region based radiomics analysis for survival prediction in oesophageal tumours treated by definitive concurrent chemoradiotherapy. EBioMedicine. 2019;44:289-297. Liu X, Guo W, Shi X, et al. Construction and verification of prognostic nomogram for early-onset ESCC. Bosn J Basic Med Sci. 2021;21(6):760-772. Xie SH, Santoni G, Mälberg K, Lagergren P, Lagergren J. Prediction Model of Long-term Survival After ESCC Surgery. Ann Surg. 2021;273(5):933-939. Tables Table 1. Comparison of clinicopathologic characteristics between patients included in the derivation and validation set. level Overall Derivation Validation p n 485 340 145 Sex (%) Female 105 (21.6) 79 (23.2) 26 (17.9) 0.239 Male 380 (78.4) 261 (76.8) 119 (82.1) Age[median(QR)] 69(67 - 73) 69(67 - 73) 69(67 - 73) 0.904 Smoking history (%) No 477 (98.4) 333 (97.9) 144 (99.3) 0.487 Yes 8 (1.6) 7 (2.1) 1 (0.7) Alcohol consumption (%) No 484 (99.8) 339 (99.7) 145 (100.0) 1.000 Yes 1 (0.2) 1 (0.3) 0 (0.0) Charlson Comorbidity Index (%) 0 405 (83.5) 285 (83.8) 120 (82.8) 0.728 1 62 (12.8) 43 (12.6) 19 (13.1) 2 2 (0.4) 2 (0.6) 0 (0.0) 3 15 (3.1) 9 (2.6) 6 (4.1) 5 1 (0.2) 1 (0.3) 0 (0.0) History of malignancy (%) No 472 (97.3) 332 (97.6) 140 (96.6) 0.706 Yes 13 (2.7) 8 (2.4) 5 (3.4) Surgical history (%) No 431 (88.9) 302 (88.8) 129 (89.0) 1.000 Yes 54 (11.1) 38 (11.2) 16 (11.0) BMI (mean (SD)) 21.70 (3.01) 21.63 (3.03) 21.87 (2.97) 0.436 Differentation (%) G1 47 (9.7) 35 (10.3) 12 (8.3) 0.895 G2 183 (37.7) 129 (37.9) 54 (37.2) G3 246 (50.7) 170 (50.0) 76 (52.4) G4 9 (1.9) 6 (1.8) 3 (2.1) pT (%) 1 88 (18.1) 67 (19.7) 21 (14.5) 0.594 2 50 (10.3) 34 (10.0) 16 (11.0) 3 20 (4.1) 14 (4.1) 6 (4.1) 4 327 (67.4) 225 (66.2) 102 (70.3) pN (%) 0 195 (40.2) 149 (43.8) 46 (31.7) 0.007 1 73 (15.1) 40 (11.8) 33 (22.8) 2 87 (17.9) 62 (18.2) 25 (17.2) 3 130 (26.8) 89 (26.2) 41 (28.3) Lymphovascular invasion (%) No 365 (75.3) 265 (77.9) 100 (69.0) 0.047 Yes 120 (24.7) 75 (22.1) 45 (31.0) Perineural invasion (%) No 350 (72.2) 253 (74.4) 97 (66.9) 0.114 Yes 135 (27.8) 87 (25.6) 48 (33.1) Neoadjuvant therapy (%) No 417 (86.0) 289 (85.0) 128 (88.3) 0.419 Yes 68 (14.0) 51 (15.0) 17 (11.7) Resection margin status (%) R0 477 (98.4) 336 (98.8) 141 (97.2) 0.388 R1-2 8 (1.6) 4 (1.2) 4 (2.8) Tumor histology (%) Adenocarcinoma 197 (40.6) 144 (42.4) 53 (36.6) 0.276 Squamous cell carcinoma 288 (59.4) 196 (57.6) 92 (63.4) Adjuvant chemotherapy (%) 0 208 (42.9) 153 (45.0) 55 (37.9) 0.18 1 277 (57.1) 187 (55.0) 90 (62.1) Clavien Dindo Classification (%) 0 175 (36.1) 133 (39.1) 42 (29.0) 0.002 ＜3 212 (43.7) 152 (44.7) 60 (41.4) ≥3 98 (20.2) 55 (16.2) 43 (29.7) Table 2. Classification performance of the individual model. RF DT SVM Derivation AUC (95% CI) 0.975(0.962-0.987) 0.784(0.739-0.830) 0.879(0.843-0.916) Brier (95% CI) 0.075(0.065-0.086) 0.168(0.145-0.197) 0.143(0.124-0.161) Accuracy (95% CI) 0.903(0.872-0.934) 0.786(0.741-0.818) 0.803(0.756-0.843) Precision (95% CI) 0.885(0.828-0.926) 0.763(0.694-0.808) 0.816(0.74-0.87) Recall (95% CI) 0.943(0.904-0.982) 0.878(0.825-0.917) 0.822(0.772-0.869) F1 Score (95% CI) 0.913(0.888-0.943) 0.816(0.774-0.843) 0.819(0.77-0.855) Validation AUC (95% CI) 0.791(0.717-0.864) 0.717(0.640-0.794) 0.779(0.702-0.856) Brier (95% CI) 0.191(0.149-0.229) 0.218(0.185-0.266) 0.186(0.155-0.231) Accuracy (95% CI) 0.721(0.655-0.783) 0.684(0.614-0.766) 0.741(0.676-0.8) Precision (95% CI) 0.738(0.639-0.816) 0.675(0.587-0.765) 0.78(0.687-0.867) Recall (95% CI) 0.742(0.652-0.82) 0.79(0.712-0.868) 0.717(0.626-0.794) F1 Score (95% CI) 0.739(0.668-0.797) 0.727(0.66-0.801) 0.746(0.676-0.814) Additional Declarations No competing interests reported. Supplementary Files tables.docx SupplementalFigure1.tif SupplementalFigure2.tif SupplementalFigure3.tif SupplementalFigure4.tif SupplementalFigure5.tif Supplementalfigurelegends.pdf data.xlsx 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-4281425","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":294240107,"identity":"79e176a4-289c-400b-af71-ee8c575f36d9","order_by":0,"name":"Jin-Biao Xie","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA5ElEQVRIiWNgGAWjYBACxmYGNoYEIIONgbHx4YcKCTl5ErQwNxtLnLEwNmwgbBEblGZvk+Btq0hkOEBAPXM777EHD2psEvvYGxskJOdJJDA2MD98dAOvw/jSDRKOpRmz8RxsMCjcJpHHzsBmbJyDVwuPmUQC22E5NonEhgTJbRLFjA08bNKEtfw7zMMm/7DhAO8coMYDxGhJbAPZwtjYwNtAnBZzg8Q+kF8Sm5kljkkYGzYT8Ith/xmzhz++2STObz/+/OeHmjo5efbmh4/xamnAEGLGoxwECCaPUTAKRsEoGAUMAK6QRX+36JEpAAAAAElFTkSuQmCC","orcid":"","institution":"The Affiliated Hospital of Putian University","correspondingAuthor":true,"prefix":"","firstName":"Jin-Biao","middleName":"","lastName":"Xie","suffix":""},{"id":294240108,"identity":"977228e8-3633-42bd-8886-bf7ef7f8f00c","order_by":1,"name":"Shi-Jie Huang","email":"","orcid":"","institution":"The Affiliated Hospital of Putian University","correspondingAuthor":false,"prefix":"","firstName":"Shi-Jie","middleName":"","lastName":"Huang","suffix":""},{"id":294240109,"identity":"3af353ec-bfce-4a51-842c-0f5a060e305d","order_by":2,"name":"Lei Gao","email":"","orcid":"","institution":"Fujian Medical University Union Hospital","correspondingAuthor":false,"prefix":"","firstName":"Lei","middleName":"","lastName":"Gao","suffix":""},{"id":294240111,"identity":"6666ff83-ca51-44f9-8a3b-12b371f8d44b","order_by":3,"name":"Tian-Bao Yang","email":"","orcid":"","institution":"The Affiliated Hospital of Putian University","correspondingAuthor":false,"prefix":"","firstName":"Tian-Bao","middleName":"","lastName":"Yang","suffix":""},{"id":294240113,"identity":"44b76087-70ff-4b62-8633-638d3127c807","order_by":4,"name":"Wu Wang","email":"","orcid":"","institution":"The Affiliated Hospital of Putian University","correspondingAuthor":false,"prefix":"","firstName":"Wu","middleName":"","lastName":"Wang","suffix":""},{"id":294240115,"identity":"4bd16bbb-5ead-4566-b604-1a8a5869919e","order_by":5,"name":"Bo-Yang Chen","email":"","orcid":"","institution":"The Affiliated Hospital of Putian University","correspondingAuthor":false,"prefix":"","firstName":"Bo-Yang","middleName":"","lastName":"Chen","suffix":""}],"badges":[],"createdAt":"2024-04-17 10:59:33","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4281425/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4281425/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":55527997,"identity":"ad97d4ec-4e77-406a-b800-de822c7644ff","added_by":"auto","created_at":"2024-04-29 15:07:12","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":125018,"visible":true,"origin":"","legend":"\u003cp\u003eFlow diagram of the study population.\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-4281425/v1/1b8199012694167df7538b47.png"},{"id":55527998,"identity":"bde9039f-037f-4d76-96ee-04b3b81aefe9","added_by":"auto","created_at":"2024-04-29 15:07:12","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":2327254,"visible":true,"origin":"","legend":"\u003cp\u003eCalibration curves of models.\u003c/p\u003e\n\u003cp\u003e(A) Calibration curves of RSF model in derivation dataset; (B) Calibration curves of RSF model in validation dataset; (C) Calibration curves of DT model in derivation dataset; (D) Calibration curves of DT model in validation dataset; (E) Calibration curves of SVM model in derivation dataset; (F) Calibration curves of SVM model in validation dataset.\u003c/p\u003e","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-4281425/v1/c644e809f8637e78b938106b.png"},{"id":55527996,"identity":"0e949b39-042d-4d01-b00e-af66162c68be","added_by":"auto","created_at":"2024-04-29 15:07:12","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1380009,"visible":true,"origin":"","legend":"\u003cp\u003eDecision curve analysis curve of prediction models.\u003c/p\u003e\n\u003cp\u003e(A) Decision curve analysis curve in derivation dataset; (B) Decision curve analysis curve in validation dataset.\u003c/p\u003e","description":"","filename":"Figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-4281425/v1/dd1fbca1260ccb3479a83628.png"},{"id":59079970,"identity":"cbd811f5-78c5-44cf-b54c-f97a05d5dcc9","added_by":"auto","created_at":"2024-06-26 06:55:36","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3961143,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4281425/v1/b9449718-fe72-4883-bd91-a4b84b2e4798.pdf"},{"id":55528000,"identity":"b9440938-cb2b-426a-8d7c-f52a190112cc","added_by":"auto","created_at":"2024-04-29 15:07:12","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":25412,"visible":true,"origin":"","legend":"","description":"","filename":"tables.docx","url":"https://assets-eu.researchsquare.com/files/rs-4281425/v1/fe1c2a2ecd3038f3de8b3836.docx"},{"id":55528006,"identity":"2322eede-59be-478a-be49-0303da09a3d6","added_by":"auto","created_at":"2024-04-29 15:07:13","extension":"tif","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":3045108,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementalFigure1.tif","url":"https://assets-eu.researchsquare.com/files/rs-4281425/v1/5583fabb41b8ab2120bfc433.tif"},{"id":55528566,"identity":"a771ee17-b17a-4c2e-97b6-f89984b494fb","added_by":"auto","created_at":"2024-04-29 15:15:12","extension":"tif","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":2560218,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementalFigure2.tif","url":"https://assets-eu.researchsquare.com/files/rs-4281425/v1/8306fdccbc61caac5c999c84.tif"},{"id":55528004,"identity":"b19021b1-1826-483d-a84a-4839576ca2ec","added_by":"auto","created_at":"2024-04-29 15:07:12","extension":"tif","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":5143400,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementalFigure3.tif","url":"https://assets-eu.researchsquare.com/files/rs-4281425/v1/36f0a2fc04040a906346fe38.tif"},{"id":55528002,"identity":"4bf31d01-19f1-438a-bc85-e02245809218","added_by":"auto","created_at":"2024-04-29 15:07:12","extension":"tif","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":5973354,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementalFigure4.tif","url":"https://assets-eu.researchsquare.com/files/rs-4281425/v1/0b66cac61b28ec18b66ed2aa.tif"},{"id":55528001,"identity":"8ef9327d-30c9-4e9a-93a2-164d0a8b6ea9","added_by":"auto","created_at":"2024-04-29 15:07:12","extension":"tif","order_by":6,"title":"","display":"","copyAsset":false,"role":"supplement","size":1665928,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementalFigure5.tif","url":"https://assets-eu.researchsquare.com/files/rs-4281425/v1/9eb530fb1042041c04954bac.tif"},{"id":55528007,"identity":"6df5692e-0a4a-486d-b57e-048d4073f3dc","added_by":"auto","created_at":"2024-04-29 15:07:13","extension":"pdf","order_by":7,"title":"","display":"","copyAsset":false,"role":"supplement","size":76917,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementalfigurelegends.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4281425/v1/029b60d55af3c81e708f996d.pdf"},{"id":55528005,"identity":"cc77eb22-64b3-42ba-a278-45cf21268a02","added_by":"auto","created_at":"2024-04-29 15:07:13","extension":"xlsx","order_by":8,"title":"","display":"","copyAsset":false,"role":"supplement","size":89139,"visible":true,"origin":"","legend":"","description":"","filename":"data.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-4281425/v1/e72a2133dd7dc2a097ff600d.xlsx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Comparison of Machine Learning Methods for Predicting 3-Year Survival in Elderly Esophageal squamous cancer Patients Based on Oxidative Stress","fulltext":[{"header":"Introduction","content":"\u003cp\u003eEsophageal squamous cancer (ESSC) is one of the malignant tumors with high incidence and mortality rates in the world\u003csup\u003e1\u003c/sup\u003e. Compared to younger patients, elderly cancer patients exhibit greater heterogeneity in terms of physical, functional, psychological, and social capabilities or weaknesses\u003csup\u003e2\u003c/sup\u003e, leading to the current TNM staging system, being unable to accurately reflect the prognostic characteristics of elderly ESSC patients. In recent years, studies have found that biological markers such as albumin (ALB), total bilirubin (TBIL), direct bilirubin (DBIL), blood urea nitrogen (BUN), creatinine (Crs), which reflect systemic oxidative stress, play important roles in the occurrence, development, and prognosis of elderly malignant tumor patients\u003csup\u003e3-6\u003c/sup\u003e. It is believed that the antioxidant enzyme activity in elderly individuals decreases, making them more vulnerable to damage from oxygen free radicals, which may lead to DNA, protein, and lipid damage, further accelerating cellular aging and disease progression\u003csup\u003e7-9\u003c/sup\u003e. Therefore, understanding the oxidative stress characteristics of elderly cancer patients is expected to help clinicians predict the complex prognosis of ESCC, optimize patient treatment, and improve quality of life. Currently, there is still a lack of predictive models for survival in elderly ESCC patients based on oxidative stress indicators.\u003c/p\u003e\n\u003cp\u003eAs a branch of artificial intelligence, supervised machine learning methods, due to their greater flexibility in capturing complex nonlinear relationships, particularly when dealing with large and sparse data, have been widely used in predicting the prognosis of biological properties\u003csup\u003e10-12\u003c/sup\u003e. However, previous models relied on known variables such as TNM staging, histopathological characteristics, surgery, and chemotherapy, making them unsuitable for complex elderly individuals. Given the crucial role of oxidative stress in elderly ESSC patients, this study aims to explore the relationship between oxidative stress and the prognosis of elderly ESSC patients and to construct a machine learning model for predicting 3-year survival post-surgery to assist in clinical decision-making.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cstrong\u003ePatient Selection and Study Methods\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study included elderly ESCC patients who underwent curative ESCC resection surgery from January 2013 to December 2020 and were registered in the Thoracic Surgery Database of Putian University Affiliated Hospital (AHPTU) and Fujian Medical University Union Hospital (FMUUH). Inclusion criteria were as follows: (1) postoperative pathological diagnosis of Esophageal squamous cell carcinoma; (2) age \u0026ge;65 years at diagnosis; (3) underwent curative surgery with no evidence of distant metastasis; and (4) complete clinical and pathological data were available. Exclusion criteria were: (1) postoperative pathology confirmed a non-primary tumor originating from the esophagus; (2) presence of distant metastases; (3) incomplete clinical data. Finally, after exclusions, 340 patients from AHPTU were included in the study group as the derivation cohort, and 145 patients from FMUUH were included in the external validation cohort. The derivation cohort was used to develop a novel oxidative stress score (OSS) and construct a machine learning predictive model (Figure 1). The predictive performance of the models was evaluated in the derivation and validation cohorts using the area under the receiver operating characteristic curve (AUC) and calibration curve. Event occurrence time or review was determined based on the date of surgery to the last contact (death or last follow-up). This study was a retrospective analysis of anonymized data in the database, and the Institutional Review Board waived the requirement for obtaining informed consent. This study was approved by the Ethics Committee of Putian University Affiliated Hospital (Approval Number: [2022044]). The study protocol was reviewed and granted ethical approval by the ethics committee to ensure the protection of human subjects and compliance with ethical standards.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCandidate predictive variables\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eComplete data on preoperative tests, intraoperative conditions, postoperative recovery, and pathological results are essential. Routine blood and biochemical tests were conducted for each patient from the first day of admission. The TNM staging was reclassified according to the American Joint Committee on Cancer/Union for International Cancer Control 8th edition AJCC Cancer Staging Manual. Based on the optimal cutoff values identified by R 4.3.3 software, each biochemical parameter was converted into a categorical variable. The oxidative stress indicators in our study included ALB, TBIL, DBIL, BUN, Crs, lactate dehydrogenase (LDH), and uric acid (UA). Based on the optimal cutoff values, the values of the biochemical parameters were defined as low (values below the cutoff) or high (values above the cutoff). The derivation cohort was used to develop a new OSS based on the beta coefficients from multivariate stepwise Cox regression analysis. Patients were stratified into risk groups based on the calculated optimal cutoff value of OSS and validated in the validation cohort.\u003c/p\u003e\n\u003cp\u003eOther clinically relevant features used in training the machine learning predictive model were established by researchers based on clinical reasoning, literature review, and consensus on routine availability to ensure wide applicability in various clinical settings. Specifically, the predictive model included preoperative hematological tests (white blood cell count (WBC), hemoglobin (HB), neutrophils (NE), lymphocytes (LYM), monocytes (NOM), prothrombin time (PT)), biochemical tests( (ALB), BUN, TBIL, DBIL, UA, Crs, LDH), tumor markers (alpha-fetoprotein (AFP), carcinoembryonic antigen (CEA), carbohydrate antigen 19-9 (CA199)), clinical variables (gender, age, OSS, body mass index, history of major abdominal surgery, history of previous malignancy, Charlson Comorbidity Index, American Society of Anesthesiologists (ASA) score, neoadjuvant therapy), intraoperative variables (extent of resection, intraoperative blood loss), postoperative variables (Clavien-Dindo complication grading, adjuvant chemotherapy), and pathological variables (differentiation grade, pathological T(pT) stage, pathological N(pN) stage, pathological TNM(pTNM) stage, vascular invasion, perineural invasion). Due to potential collinearity among variables, other potential predictive factors were excluded from the candidate predictive variables (using pT stage, pN stage without using pTNM stage). Variables were standardized to ensure comparability of scales.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEstablishment of machine learning model\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe used Random Forest (RF), Decision Tree (DT), and Support Vector Machine (SVM) models to predict the survival status at 3 years post-surgery. The selection of these three machine learning methods was based on their wide application in handling similar diseases, outcomes, and cohorts. In the RF package, the RF function was used to generate the RF model. This function defaults to generating 500 trees, with 4 variables sampled at each node, and a minimum node size of 1. For the DT model, we used the r part function, using the information gain criterion by setting the parms parameter. This process involved cross-validation to select the optimal model by analyzing the complexity parameter (CP) and cross-validation error in the cross-validation table. We determined the optimal CP value and pruned the model based on it to reduce overfitting and optimize model performance. SVM is a well-known supervised classification model that primarily learns the optimal linear separating hyperplane in a high-dimensional feature space. SVM also uses kernel functions to handle non-linear separable categories. We fitted the SVM model with various hyperparameter combinations using the RBF kernel function and conducted comparative analysis to find the best model configuration, systematically optimizing model performance.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFollow-up\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFollow-up visits were scheduled every 3 months within the first 2 years postoperatively, and every 6 months from 2 to 5 years. The final assessment was conducted in December 2023. Most routine follow-up visits included physical examinations, laboratory tests, chest X-rays, abdominal ultrasound or computed tomography scans, and annual endoscopic examinations. The primary outcome was defined as overall survival (OS) post-discharge. OS was defined as the time from surgery to death from any cause or to the last follow-up date for censored observations.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStatistical analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe conducted data analysis using R version 4.3.3. For continuous data that did not follow a normal distribution, we used the Mann-Whitney test, for continuous data that followed a normal distribution, we used the independent t-test. Differences in the distribution of categorical variables between groups were analyzed using the Pearson\u0026apos;s chi-squared test and Fisher\u0026apos;s exact test. Overall survival (OS) curves were plotted using the Kaplan-Meier method, and differences between survival curves were assessed using the log-rank test. Validation was performed using bootstrap resampling. Model parameters were trained using derivation cohorts, and the performance of the trained model was evaluated using independent validation datasets. The performance of the trained classifier was measured using sensitivity, specificity, accuracy, AUC values, and Brier scores. All statistical analyses were conducted using R software version 4.3.3 (https://www.r-project.org/), and a two-sided p-value \u0026lt;0.05 was considered statistically significant.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eResearch cohort\u003c/p\u003e\n\u003cp\u003eA total of 485 elderly ESCC patients were included in this study. In the derivation cohort, there were 340 patients, of whom 261 (76.8%) were male and 79 (23.2%) were female, with a median age of 69 (67 - 73) years. In the validation cohort, there were 145 patients, with 119 (82.1%) males and 26 (17.9%) females, with a median age of 69 (67 - 73) years old. There were statistical differences in the clinical and pathological data of patients in the derivation and validation cohorts in terms of pathological N staging, lymphovascular invasion, and complication grading (p=0.007, 0.047, 0.002, respectively), while other variables showed no statistical differences (Table 1, p\u0026gt;0.05).\u003c/p\u003e\n\u003cp\u003eIn terms of survival rates, the 3-year overall survival rate in the derivation cohort was 37.22% (29.78%, 46.52%), while in the validation cohort, it was 49.87% (39.50%, 62.97%) (Figure s1).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDeveloping a novel oxidative stress score\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn the derivation cohort, the optimal cutoff values for oxidative stress indicators were determined using the surv_cutpoint method as follows: ALB 39.4g/dL, BUN 4.65mg/dL, TBIL 5.7\u0026mu;mol/L, DBIL 1.5\u0026mu;mol/L, UA 215\u0026mu;mol/L, LDH 162U/L, and Crs 49.3\u0026mu;mol/L. These indicators were included in a Cox proportional hazards model to perform stepwise regression analysis based on the Akaike Information Criterion. It was found that ALB, BUN, UA, LDH, and Crs were the core factors affecting the OS of elderly ESCC patients in the final model (eTable 1). Based on the regression coefficients of these variables, a prognostic model called Esophageal\u0026nbsp;Squamous Cancer Oxidative Stress Score (OSS) was further constructed as follows: OSS = ALB * -0.3197 + BUN * 0.2397 + UA * -0.4927 + LDH * 0.3392 + Crs * 0.6625. Patients were stratified into high, moderate, and low-risk groups using the optimal cutoff value of OSS (Figure s2). Kaplan-Meier survival curve analysis indicated that the survival rate of patients in the low OSS group was significantly lower than that of the moderate OSS group and high OSS group (p \u0026lt; 0.05). Similar results were obtained in the validation cohort (Figure s3).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eVariable selection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn our study, we used the Boruta algorithm to select important variables related to the disease status. Boruta is a variable importance assessment method based on RF, which determines which variables are truly important by comparing the importance of the original variables with randomly generated \u0026quot;shadow\u0026quot; variables. After running 50 iterations, a set of variables such as pN, pT, Tumor histology, Lymphovascular invasion, CEA, OSS, CA 19-9, and Bleeding were identified as important variables. The importance of these variables was visualized in graphical charts, where the importance of each variable is shown by comparing its importance relative to the maximum importance of the shadow variables (Figure s4).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModel Performance: Validation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe included the selected variables in machine learning to construct three models (RF, DT, SVM), and Tables 2 show the performance metrics of these models in predicting 3-year OS in the derivation and validation cohorts. The AUC for RF was 0.975(0.962-0.987) and 0.791(0.717-0.864), DT had AUC values of 0.784(0.739-0.830) and 0.717(0.640-0.794) in the derivation and validation cohorts, respectively, and SVM had AUC values of 0.879(0.843-0.916) and 0.779(0.702-0.856). Compared to DT and SVM, RF had higher AUC values on validation dataset, indicating that the RF model has excellent predictive performance and good generalization ability (Figure s5A/B/C).\u003c/p\u003e\n\u003cp\u003eIn the derivation cohort, results from 1000 resamplings showed that the Brier scores of RF, SVM, and DT were 0.075, 0.143, and 0.168, respectively. The Brier score is used to measure the accuracy of predicted probabilities, with lower scores indicating smaller deviations between predictions and actual outcomes, i.e., higher prediction accuracy. These data suggest that in the derivation cohort, the RF model had the highest prediction accuracy, followed by SVM, while the DT model had relatively lower accuracy. In the resampling performance metrics on the test set, the Brier scores for RF, SVM, and DT were 0.191, 0.186, and 0.218, respectively. This highlights the advantage of RF and SVM over DT in terms of generalization ability (Tables 2). Tables 2 show the performance metrics of each model based on resampling.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModel Performance: Calibration and Decision Curve\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe calibration curve plots show that the RF model performed well on all datasets, with predicted probabilities roughly matching the observed event frequencies. Particularly in the validation cohort, the predicted probabilities were closer to the actual outcomes, indicating better generalization ability for the RF model (Figure 2). In contrast, the DT model\u0026apos;s predictions deviated significantly from the diagonal line on both datasets, indicating substantial differences between predicted and observed values in certain probability ranges, suggesting poor calibration within these specific prediction probability intervals. The SVM model\u0026apos;s calibration was close to the 45-degree line in both the derivation and validation cohorts, with slight deviations in some probability intervals (Figure 2), indicating good calibration of the SVM model in predicting probabilities and consistency across different datasets.Additionally, we used decision curve analysis to compare the clinical usefulness of these models. The results showed that the RF model exhibited the highest net benefit in most threshold probability ranges in the derivation cohort, while the DT and SVM models had similar net benefits, but both were lower than the RF model (Figure 3A). In the validation cohort, the RF model maintained high net benefits in most threshold probability ranges, while the DT and SVM showed similar performance, both lower than the RF model (Figure 3B).。\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe 3-year survival rate after ESCC resection can serve as a valuable audit indicator for evaluating the long-term quality of tumor surgical care. As a special population, elderly individuals may face greater risks from surgery due to their unique bio-psychosocial characteristics, such as increased risk of complications, frailty, decreased stress endurance, declining physical function, cognitive decline, and other factors that may complicate postoperative survival status\u003csup\u003e23\u003c/sup\u003e\u003csup\u003e，\u003c/sup\u003e\u003csup\u003e24\u003c/sup\u003e. In recent years, machine learning methods have been widely applied in predicting survival outcomes for patients with liver cancer, gastric cancer, colorectal cancer, breast cancer, and prostate cancer\u003csup\u003e18-22\u003c/sup\u003e, demonstrating good predictive performance. Therefore, in this study, we successfully predicted the 3-year survival rate after surgery for elderly ESCC patients using machine learning models (DT, RF, SVM), with the RF model performing the best.\u003c/p\u003e\n\u003cp\u003ePrevious clinical studies have preliminarily confirmed the predictive value of some clinical pathological biomarkers in predicting recurrence, metastasis, and overall survival rates after ESCC surgery, such as tumor size, venous invasion, differentiation status, and tumor lymph node metastasis (TNM)\u003csup\u003e\u0026nbsp;25-27\u003c/sup\u003e. However, these predictive biomarkers have high detection costs and may not be applicable to the complex prognosis of elderly ESCC patients. On the other hand, oxidative stress catalyzes glycolysis, activates tumor cell migration, and promotes tumor proliferation. At the same time, varying levels of oxidative stress can also alter phosphorylation levels, thereby regulating the malignancy and prognosis of malignant tumors\u003csup\u003e28\u003c/sup\u003e. Some researchers have proposed that oxidative stress may be associated with the expression of ferritin metabolism genes, thereby interfering with prognosis\u003csup\u003e29\u003c/sup\u003e. Animal model experiments have shown that in response to external stimuli, mice exhibit increased oxidative stress factors, leading to significant elevations in biochemical markers such as TBIL, lactate dehydrogenase (LDH), creatinine (CRE), and blood urea nitrogen (BUN), which can promote tumor initiation and progression\u003csup\u003e30\u003c/sup\u003e\u003csup\u003e，\u003c/sup\u003e\u003csup\u003e31\u003c/sup\u003e. Prospective studies have shown that under oxidative stress, patients\u0026apos; levels of ALB、BUN、UA、LDH and Crs undergo changes. Following antioxidant therapy, patients\u0026apos; scores, mortality rates, and rates of sepsis are all superior to those of the control group\u003csup\u003e32\u003c/sup\u003e. Although oxidative stress is associated with cancer, there is limited research on using it to predict the prognosis of ESCC. Oxidative stress may be a potential indicator to improve the accuracy of predicting the prognosis of elderly ESCC patients. Based on this, our study proposed an oxidative stress index for ESCC, termed OSS, which is composed of ALB, DBIL, and BUN, based on preoperative hematological indicators closely associated with oxidative stress. Our research found that patients with lower OSS had a worse prognosis compared to those with higher OSS. It is worth noting that OSS was generated by training on a cohort of patients from one of our institutions, which allows for the use of highly detailed clinical data and long-term follow-up. Therefore, we reasonably speculate that a predictive model incorporating OSS status may better predict the prognosis of elderly ESCC patients.\u003c/p\u003e\n\u003cp\u003eSeveral models have been reported for predicting postoperative survival in ESCC. Li\u003csup\u003e33\u003c/sup\u003e built nomograms for predicting progression-free survival (PFS) and OS based on the Cox model to determine independent prognostic factors for PFS and OS. Following internal cross-validation, the corrected concordance indices were 0.739 and 0.696. However, inherent selection bias in retrospective studies and the inclusion of a limited number of cases further magnified this limitation. On the other hand, Xie\u003csup\u003e34\u003c/sup\u003e utilized a prospective study to construct an OS prediction model for ESCC patients using LASSO regression. In the training and validation cohorts, the area under the 3-year ROC curve was 0.811 (0.67 - 0.952,95% CI) and 0.805 (0.638 - 0.973,95% CI) respectively, showing high predictive performance. However, this model did not specifically differentiate elderly ESCC patients, and further research is needed to confirm whether the established model is applicable to elderly patients. In contrast, Liu\u003csup\u003e35\u003c/sup\u003e conducted a stratified analysis of survival characteristics in elderly ESCC patients and developed a nomogram prognostic prediction model with a C-index of 0.706. However, this study used the SEER database as the validation cohort, resulting in the inclusion of fewer modeling indicators that might not reflect more information about the patients. To overcome this limitation, Xie\u003csup\u003e36\u003c/sup\u003e studied the predictive value of basic indicators such as age, sex, education level, as well as pathological indicators including pathological tumor staging, tumor histology, and margin status, along with surgical indicators like neoadjuvant therapy, reoperation, and Charlson comorbidity index for ESCC. However, this comprehensive model, which includes a wide range of information, was established only through multivariable regression analysis to create a predictive model,\u0026nbsp;lacking the generalizability and automation that machine learning offers may lead to the omission of key prognostic indicators. In this scenario, we developed and validated different machine learning methods (RF, DT, SVM) to enhance the discriminative ability of 3-year OS in elderly ESCC patients. Compared to other models, the RF (RF) model demonstrated excellent performance and good calibration in predicting the 3-year survival status. Additionally, our model utilized routine and easily obtainable perioperative clinical data. At the same time, our model utilizes routine and easily obtainable perioperative clinical data. The most important variables influencing the 3-year postoperative survival in elderly patients are pN, pT, tumor histological type, vascular invasion, CEA, OSS, CA 19-9, and the amount of bleeding.. This provides a new opportunity to understand the importance of preoperative oxidative stress and body status, surgical performance, postoperative recovery, and tumor staging in predicting the 3-year survival rate after surgery in elderly ESCC patients. This may allow clinicians to better understand the accuracy and effectiveness of managing these patients. Decision curve analysis provides us with a tool to assess and compare the performance of different models at multiple thresholds, further aiding in model selection and application. Based on decision curve analysis, we observed that the RF model exhibited relatively superior performance in the training, testing, and validation datasets, implying that it may possess good generalizability.\u003c/p\u003e\n\u003cp\u003eOur novel RF model demonstrated higher AUC values compared to previous models, possibly due to the inclusion of more comprehensive clinical assessment indicators specific to elderly patients, such as oxidative stress indicators, comorbidity indices, and complication status. Although pN and pT were identified as the most decisive variables for model prediction, their importance significantly exceeded that of other variables. Tumor histological type, vascular invasion, CEA, OSS, CA 19-9, and the amount of bleeding also showed relatively high importance. These results provide insights into the key variables that play crucial roles in the model, aiding in further optimization and interpretation of the model. Therefore, when establishing a predictive model for long-term survival after ESCC surgery, these prognostic factors should be considered comprehensively.\u003c/p\u003e\n\u003cp\u003eSome limitations should be acknowledged, as the study was retrospective and selection bias cannot be completely avoided. The OSS was constructed by combining serum indicators such as ALB, BUN, UA, LDH, and Crs, and the score may not be sufficient to accurately reveal the oxidative stress status of patients. A more accurate assessment of oxidative stress usually requires the detection of reactive oxygen species and their associated markers, such as superoxide dismutase, malondialdehyde. Second, the performance of the predictive model we developed was evaluated in terms of discriminant ability and risk calibration, and the self-resampling bootstrap program offset overfitting to some extent. However, due to the limited sample size, there is still a risk of insufficient generalization. Finally, well-known factors that influence ESCC, such as high-risk genetic mutations, immunotherapy drug use, and socioeconomic status, were not available from our database, despite the potential for these factors to improve model performance. In the future, we recommend prospective studies to further validate the results of this study.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThe clinical predictive model for OSS constructed using machine learning methods can effectively predict the prognosis of elderly patients with ESCC after curative surgery.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eCompliance with ethical standards\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Disclosures:\u0026nbsp;\u003c/strong\u003eJin-Biao Xie, Shi-Jie Huang, Tian-Bao Yang, Lei Gao, Wu Wang, Bo-Yang Chen, have no conflicts of interest or financial ties to disclose.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInformed Consent:\u003c/strong\u003e Informed consent was obtained from each participant.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding:\u003c/strong\u003e Science and Technology Foundation of Putian, Grant/Award (Number: 2023S3F005),the Health Science and Technology Foundation of Fujian Province，Grant/Award(Number: 2022QNA100), Joint Funds for the innovation of Science and Technology，Fujian province (NO.2023Y9168).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eHuman rights statement and informed consent:\u0026nbsp;\u003c/strong\u003eAll procedures followed were in accordance with the ethical standards of the responsible committee on human experimentation (institutional and national) and with the Helsinki Declaration of 1964 and later versions. Informed consent or substitute for it was obtained from all patients for being included in the study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e \u003cstrong\u003estatement\u003c/strong\u003e\u003cstrong\u003e：\u003c/strong\u003eAll data generated or analysed during this study are included in the supplementary information files.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eFerlay J, Colombet M, Soerjomataram I, et al. Estimating the global cancer incidence and mortality in 2018: GLOBOCAN sources and methods. Int J Cancer. 2019;144(8):1941-1953.\u003c/li\u003e\n\u003cli\u003eBron D, Soubeyran P, Fulop T. Innovative approach to older patients with malignant hemopathies. Haematologica. 2016;101(8):893-895.\u003c/li\u003e\n\u003cli\u003eHayes J, Dinkova-Kostova A, Tew K. Oxidative Stress in Cancer. Cancer cell. 2020;38(2):167-197.\u003c/li\u003e\n\u003cli\u003eArfin S, Jha N, Jha S, et al. Oxidative Stress in Cancer Cell Metabolism. Antioxidants Basel, Switzerland.2021;10(5).\u003c/li\u003e\n\u003cli\u003eAbdelhamid R, Nagano S. Crosstalk between Oxidative Stress and Aging in Neurodegeneration Disorders. Cells. 2023;12(5).\u003c/li\u003e\n\u003cli\u003eWang J, Sun Y, Zhang X, et al. Oxidative stress activates NORAD expression by H3K27ac and promotes oxaliplatin resistance in gastric cancer by enhancing autophagy flux via targeting the miR-433-3p. Cell death \u0026amp; disease. 2021;12(1):90.\u003c/li\u003e\n\u003cli\u003eToh D, Lee W, Zhou H, et al. Lycium barbarumWolfberry () Consumption with a Healthy Dietary Pattern Lowers Oxidative Stress in Middle-Aged and Older Adults: A Randomized Controlled Trial. Antioxidants (Basel, Switzerland). 2021;10(4).\u003c/li\u003e\n\u003cli\u003eCort\u0026eacute;s-Jofr\u0026eacute; M, Rueda J, Asenjo-Lobos C, Madrid E, Bonfill Cosp X. Drugs for preventing lung cancer in healthy people. The Cochrane database of systematic reviews. 2020;3(3):CD002141.\u003c/li\u003e\n\u003cli\u003eSanidad K, Sukamtoh E, Xiao H, McClements D, Zhang G. Curcumin: Recent Advances in the Development of Strategies to Improve Oral Bioavailability. Annual review of food science and technology. 2019;10:597-617.\u003c/li\u003e\n\u003cli\u003eDeo R. Machine Learning in Medicine. Circulation. 2015;132(20):1920-1930.\u003c/li\u003e\n\u003cli\u003eVan Calster B, Wynants L. Machine Learning in Medicine. The New England journal of medicine. 2019;380(26):2588.\u003c/li\u003e\n\u003cli\u003eJiang Y, Zhang Z, Yuan Q, et al. Predicting peritoneal recurrence and disease-free survival from CT images in gastric cancer with multitask deep learning: a retrospective study. The Lancet Digital health. 2022;4(5):e340-e350.\u003c/li\u003e\n\u003cli\u003eLi X, Zhai Z, Ding W, et al. An artificial intelligence model to predict survival and chemotherapy benefits for gastric cancer patients after gastrectomy development and validation in international multicenter cohorts. International journal of surgery. 2022;105:106889.\u003c/li\u003e\n\u003cli\u003eRahman S, Maynard N, Trudgill N, et al. Prediction of long-term survival after gastrectomy using random survival forests. The British journal of surgery. 2021;108(11):1341-1350.\u003c/li\u003e\n\u003cli\u003eCollins G, Reitsma J, Altman D, Moons K. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): the TRIPOD statement. Annals of internal medicine. 2015;162(1):55-63.\u003c/li\u003e\n\u003cli\u003eGao Q, Zhou H, Wang G, et al. A Model for Predicting Clinical Prognosis in Patients with WHO Grade 2 Glioma. Journal of oncology. 2022;2022:2795939.\u003c/li\u003e\n\u003cli\u003eFazel S, L\u0026aring;ngstr\u0026ouml;m N, Hjern A, Grann M, Lichtenstein P. Schizophrenia, substance abuse, and violent crime. JAMA. 2009;301(19):2016-2023.\u003c/li\u003e\n\u003cli\u003eFamularo S, Donadon M, Cipriani F, et al. Machine Learning Predictive Model to Guide Treatment Allocation for Recurrent Hepatocellular Carcinoma After Surgery. JAMA surgery. 2023;158(2):192-202.\u003c/li\u003e\n\u003cli\u003eLee K, Jang J, Yu Y, et al. Usefulness of artificial intelligence for predicting recurrence following surgery for pancreatic cancer: Retrospective cohort study. International journal of surgery (London, England). 2021;93:106050.\u003c/li\u003e\n\u003cli\u003evan den Bosch T, Warps A, de Ner\u0026eacute;e Tot Babberich M, et al. Predictors of 30-Day Mortality Among Dutch Patients Undergoing Colorectal Cancer Surgery, 2011-2016. JAMA network open. 2021;4(4):e217737.\u003c/li\u003e\n\u003cli\u003eClift A, Dodwell D, Lord S, et al. Development and internal-external validation of statistical and machine learning models for breast cancer prognostication: cohort study. BMJ (Clinical research ed). 2023;381:e073800.\u003c/li\u003e\n\u003cli\u003eLee C, Light A, Alaa A, Thurtle D, van der Schaar M, Gnanapragasam V. Application of a novel machine learning framework for predicting non-metastatic prostate cancer-specific mortality in men using the Surveillance, Epidemiology, and End Results (SEER) database. The Lancet Digital health. 2021;3(3):e158-e165.\u003c/li\u003e\n\u003cli\u003eBollschweiler E, Plum P, M\u0026ouml;nig SP, H\u0026ouml;lscher AH. Current and future treatment options for ESCC in the elderly. Expert Opin Pharmacother. 2017;18(10):1001-1010..\u003c/li\u003e\n\u003cli\u003eLu HW, Chen CC, Chen HH, Yeh HL. The clinical outcomes of elderly ESCC patients who received definitive chemoradiotherapy. J Chin Med Assoc. 2020;83(10):906-910.\u003c/li\u003e\n\u003cli\u003ePowell AGMT, Eley C, Chin C, et al. Prognostic significance of serum inflammatory markers in ESCC [published correction appears in Esophagus. 2021 Jul;18(3):710.\u003c/li\u003e\n\u003cli\u003eXi Y, Lin Y, Guo W, et al. Multi-omic characterization of genome-wide abnormal DNA methylation reveals diagnostic and prognostic markers for esophageal squamous-cell carcinoma. Signal Transduct Target Ther. 2022;7(1):53..\u003c/li\u003e\n\u003cli\u003eHagens E, Tukanova K, Jamel S, et al. Prognostic relevance of lymph node regression on survival in ESCC: a systematic review and meta-analysis. Dis Esophagus. 2022;35(1):doab021.\u003c/li\u003e\n\u003cli\u003eLiang J, Cao R, Wang X, et al. Mitochondrial PKM2 regulates oxidative stress-induced apoptosis by stabilizing Bcl2. Cell Res. 2017;27(3):329-351\u003c/li\u003e\n\u003cli\u003eWang X, Xu Y, Dai L, et al. A novel oxidative stress- and ferroptosis-related gene prognostic signature for distinguishing cold and hot tumors in colorectal cancer. Front Immunol. 2022;13:1043738.\u003c/li\u003e\n\u003cli\u003eLiu M, Rao H, Liu J, et al. The histone methyltransferase SETD2 modulates oxidative stress to attenuate experimental colitis. Redox biology. 2021;43:102004.\u003c/li\u003e\n\u003cli\u003eSandesc M, Rogobete A, Bedreag O, et al. Analysis of oxidative stress-related markers in critically ill polytrauma patients: An observational prospective single-center study. Bosnian journal of basic medical sciences. 2018;18(2):191-197.\u003c/li\u003e\n\u003cli\u003eLima W, Martins-Santos M, Chaves V. Uric acid as a modulator of glucose and lipid metabolism. Biochimie. 2015;116:17-23.\u003c/li\u003e\n\u003cli\u003eLi B, Wang R, Zhang T, et al. Development and validation of a nomogram prognostic model for ESCC patients with oligometastases. Sci Rep. 2020;10(1):11259.\u003c/li\u003e\n\u003cli\u003eXie C, Yang P, Zhang X, et al. Sub-region based radiomics analysis for survival prediction in oesophageal tumours treated by definitive concurrent chemoradiotherapy. EBioMedicine. 2019;44:289-297.\u003c/li\u003e\n\u003cli\u003eLiu X, Guo W, Shi X, et al. Construction and verification of prognostic nomogram for early-onset ESCC. Bosn J Basic Med Sci. 2021;21(6):760-772.\u003c/li\u003e\n\u003cli\u003eXie SH, Santoni G, M\u0026auml;lberg K, Lagergren P, Lagergren J. Prediction Model of Long-term Survival After ESCC Surgery. Ann Surg. 2021;273(5):933-939.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTable 1. Comparison of clinicopathologic characteristics between patients included in the derivation and validation set.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"562\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003elevel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003eOverall\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003eDerivation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003eValidation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003ep\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003en\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e485\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e340\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e145\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003eSex (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e105 (21.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e79 (23.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e26 (17.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e0.239\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eMale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e380 (78.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e261 (76.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e119 (82.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003eAge[median(QR)]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e69(67 - 73)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e69(67 - 73)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e69(67 - 73)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e0.904\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003eSmoking history (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e477 (98.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e333 (97.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e144 (99.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e0.487\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e8 (1.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e7 (2.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e1 (0.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003eAlcohol consumption (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e484 (99.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e339 (99.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e145 (100.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e1 (0.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e1 (0.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e0 (0.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003eCharlson Comorbidity Index (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e405 (83.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e285 (83.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e120 (82.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e0.728\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e62 (12.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e43 (12.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e19 (13.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e2 (0.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e2 (0.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e0 (0.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e15 (3.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e9 (2.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e6 (4.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e1 (0.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e1 (0.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e0 (0.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003eHistory of malignancy (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e472 (97.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e332 (97.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e140 (96.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e0.706\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e13 (2.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e8 (2.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e5 (3.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003eSurgical history (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e431 (88.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e302 (88.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e129 (89.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e54 (11.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e38 (11.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e16 (11.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003eBMI (mean (SD))\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e21.70 (3.01)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e21.63 (3.03)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e21.87 (2.97)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e0.436\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003eDifferentation (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eG1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e47 (9.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e35 (10.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e12 (8.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e0.895\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eG2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e183 (37.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e129 (37.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e54 (37.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eG3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e246 (50.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e170 (50.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e76 (52.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eG4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e9 (1.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e6 (1.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e3 (2.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003epT (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e88 (18.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e67 (19.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e21 (14.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e0.594\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e50 (10.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e34 (10.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e16 (11.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e20 (4.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e14 (4.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e6 (4.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e327 (67.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e225 (66.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e102 (70.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003epN (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e195 (40.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e149 (43.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e46 (31.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e0.007\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e73 (15.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e40 (11.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e33 (22.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e87 (17.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e62 (18.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e25 (17.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e130 (26.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e89 (26.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e41 (28.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003eLymphovascular invasion (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e365 (75.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e265 (77.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e100 (69.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e0.047\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e120 (24.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e75 (22.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e45 (31.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003ePerineural invasion (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e350 (72.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e253 (74.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e97 (66.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e0.114\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e135 (27.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e87 (25.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e48 (33.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003eNeoadjuvant therapy (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e417 (86.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e289 (85.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e128 (88.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e0.419\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e68 (14.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e51 (15.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e17 (11.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003eResection margin status (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eR0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e477 (98.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e336 (98.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e141 (97.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e0.388\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eR1-2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e8 (1.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e4 (1.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e4 (2.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003eTumor histology (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eAdenocarcinoma\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e197 (40.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e144 (42.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e53 (36.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e0.276\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003eSquamous cell carcinoma\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e288 (59.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e196 (57.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e92 (63.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003eAdjuvant chemotherapy (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e208 (42.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e153 (45.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e55 (37.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e0.18\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e277 (57.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e187 (55.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e90 (62.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003eClavien Dindo Classification (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e175 (36.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e133 (39.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e42 (29.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e＜3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e212 (43.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e152 (44.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e60 (41.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"15.836298932384341%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.946619217081851%\"\u003e\n \u003cp\u003e\u0026ge;3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e98 (20.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e55 (16.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.86120996441281%\"\u003e\n \u003cp\u003e43 (29.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.633451957295375%\"\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\u003eTable 2. Classification performance of the individual model.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"555\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"12.23021582733813%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"20.323741007194243%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.48201438848921%\" valign=\"top\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.48201438848921%\" valign=\"top\"\u003e\n \u003cp\u003eDT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.48201438848921%\" valign=\"top\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"12.23021582733813%\" rowspan=\"6\" valign=\"top\"\u003e\n \u003cp\u003eDerivation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"20.323741007194243%\" valign=\"top\"\u003e\n \u003cp\u003eAUC (95% CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.48201438848921%\" valign=\"top\"\u003e\n \u003cp\u003e0.975(0.962-0.987)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.48201438848921%\" valign=\"top\"\u003e\n \u003cp\u003e0.784(0.739-0.830)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.48201438848921%\" valign=\"top\"\u003e\n \u003cp\u003e0.879(0.843-0.916)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"23.15573770491803%\" valign=\"top\"\u003e\n \u003cp\u003eBrier (95% CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"top\"\u003e\n \u003cp\u003e0.075(0.065-0.086)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"top\"\u003e\n \u003cp\u003e0.168(0.145-0.197)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"top\"\u003e\n \u003cp\u003e0.143(0.124-0.161)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"23.15573770491803%\" valign=\"bottom\"\u003e\n \u003cp\u003eAccuracy (95% CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.903(0.872-0.934)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.786(0.741-0.818)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.803(0.756-0.843)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"23.15573770491803%\" valign=\"bottom\"\u003e\n \u003cp\u003ePrecision (95% CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.885(0.828-0.926)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.763(0.694-0.808)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.816(0.74-0.87)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"23.15573770491803%\" valign=\"bottom\"\u003e\n \u003cp\u003eRecall (95% CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.943(0.904-0.982)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.878(0.825-0.917)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.822(0.772-0.869)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"23.15573770491803%\" valign=\"bottom\"\u003e\n \u003cp\u003eF1 Score (95% CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.913(0.888-0.943)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.816(0.774-0.843)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.819(0.77-0.855)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"12.23021582733813%\" rowspan=\"6\" valign=\"top\"\u003e\n \u003cp\u003eValidation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"20.323741007194243%\" valign=\"top\"\u003e\n \u003cp\u003eAUC (95% CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.48201438848921%\" valign=\"top\"\u003e\n \u003cp\u003e0.791(0.717-0.864)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.48201438848921%\" valign=\"top\"\u003e\n \u003cp\u003e0.717(0.640-0.794)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.48201438848921%\" valign=\"top\"\u003e\n \u003cp\u003e0.779(0.702-0.856)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"23.15573770491803%\" valign=\"top\"\u003e\n \u003cp\u003eBrier (95% CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"top\"\u003e\n \u003cp\u003e0.191(0.149-0.229)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"top\"\u003e\n \u003cp\u003e0.218(0.185-0.266)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"top\"\u003e\n \u003cp\u003e0.186(0.155-0.231)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"23.15573770491803%\" valign=\"bottom\"\u003e\n \u003cp\u003eAccuracy (95% CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.721(0.655-0.783)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.684(0.614-0.766)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.741(0.676-0.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"23.15573770491803%\" valign=\"bottom\"\u003e\n \u003cp\u003ePrecision (95% CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.738(0.639-0.816)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.675(0.587-0.765)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.78(0.687-0.867)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"23.15573770491803%\" valign=\"bottom\"\u003e\n \u003cp\u003eRecall (95% CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.742(0.652-0.82)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.79(0.712-0.868)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.717(0.626-0.794)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"23.15573770491803%\" valign=\"bottom\"\u003e\n \u003cp\u003eF1 Score (95% CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.739(0.668-0.797)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.727(0.66-0.801)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.614754098360656%\" valign=\"bottom\"\u003e\n \u003cp\u003e0.746(0.676-0.814)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\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":"Elderly Esophageal squamous cancer, Oxidative stress, Machine learning, Overall survival, Prediction model","lastPublishedDoi":"10.21203/rs.3.rs-4281425/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4281425/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground \u003c/strong\u003eThere is currently a lack of machine learning model studies exploring the relationship between oxidative stress score (OSS) and the prognosis of elderly Esophageal squamous cancer(ESCC) patients.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods\u003c/strong\u003eThis study included elderly ESCC patients who underwent curative resection surgery from January 2013 to December 2020. Machine learning strategies including decision tree (DT), random forest (RF), and support vector machine (SVM) were employed to construct a predictive model for 3-year overall survival (OS) for elder ESCC base on OSS.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults\u003c/strong\u003ePatients were divided into derivation cohort and validation cohort, and consisted of 340 and 145 patients, respectively. 8 important features which were the most important factors influencing 3-year OS (pathological N stage, pathological T stage, tumor histological type, vascular invasion, CEA, OSS, CA 19-9, and the amount of bleeding) were included in training the RF, DT and SVM. In the derivation cohort, the RF model exhibited the highest predictive performance with an AUC of 0.975(0.962-0.987), while the DT model is 0.784(0.739-0.830) and the SVM is 0.879(0.843-0.916). In the external validation cohort showed the similar trend .\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion \u003c/strong\u003eThe random forest clinical prediction model constructed based on OSS can effectively predict the prognosis of elderly ESCC patients after curative surgery.\u003c/p\u003e","manuscriptTitle":"Comparison of Machine Learning Methods for Predicting 3-Year Survival in Elderly Esophageal squamous cancer Patients Based on Oxidative Stress","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-04-29 15:07:07","doi":"10.21203/rs.3.rs-4281425/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":"e61701a1-acfb-4035-abe5-60d463487d96","owner":[],"postedDate":"April 29th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":31016933,"name":"Health sciences/Gastroenterology/Oesophagogastroscopy"},{"id":31016934,"name":"Health sciences/Diseases/Cancer"}],"tags":[],"updatedAt":"2024-06-26T06:47:24+00:00","versionOfRecord":[],"versionCreatedAt":"2024-04-29 15:07:07","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4281425","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4281425","identity":"rs-4281425","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}