Development and Internal Validation of an Explainable Machine-Learning Model to Predict 3-Year overall survival rate After Radical Cystectomy

Development and Internal Validation of an Explainable Machine-Learning Model to Predict 3-Year overall survival rate After Radical Cystectomy | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Development and Internal Validation of an Explainable Machine-Learning Model to Predict 3-Year overall survival rate After Radical Cystectomy Yunze Wang, Aikeshanjiang Ailiyaer, Shiming Chen, Wenguang Wang This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8754027/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 10 You are reading this latest preprint version Abstract Background: This study aimed to develop and internally validate an explainable machine-learning model using routinely available clinicopathologic and laboratory variables for predicting 3-year overall survival (OS) after radical cystectomy. Methods: We retrospectively included 300 patients who underwent radical cystectomy between January 2018 and December 2022. Predictors were selected in the training set using LASSO logistic regression followed by random-forest recursive feature elimination. Ten variables were retained. Seven algorithms (logistic regression, KNN, SVM-RBF, random forest, XGBoost, LightGBM, and CatBoost) were trained on a 70% training set and evaluated on a 30% internal validation set. Discrimination, calibration, and clinical utility were assessed, and the final model was interpreted using Shapley additive explanations (SHAP). Results: In internal validation, AUCs ranged from 0.834 to 0.950. CatBoost achieved the best overall classification performance (AUC = 0.931, accuracy = 0.862, sensitivity = 0.647, specificity = 0.951, PPV = 0.846, and NPV = 0.867). SHAP analyses identified tumor stage (T, N, and M stage) as the dominant drivers of predicted risk, with additional contributions from age, BMI, albumin, globulin, lymphocyte count, platelet count, and preoperative creatinine. Conclusions: We developed an internally validated, SHAP-interpretable CatBoost model for predicting 3-year overall survival (OS) after radical cystectomy. External validation and recalibration in independent cohorts are required before clinical use. Bladder cancer Radical cystectomy Machine Learning SHAP Predictive Modeling Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 1 Introduction Bladder cancer is one of the most common malignancies of the urinary tract and its incidence has been increasing worldwide. In 2020, approximately 570,000 new cases were diagnosed globally, accounting for about 3% of all cancers, and by 2022 this number had risen to roughly 610,000 [ 1 ]. China bears a particularly heavy burden, with about 93,000 new cases and more than 270,000 patients living within 5 years of diagnosis in 2022 [ 2 ].Clinically, bladder cancer is classified into non–muscle-invasive bladder cancer (NMIBC) and muscle-invasive bladder cancer (MIBC), which differ markedly in prognosis and management [ 3 ]. For patients with MIBC or high-risk NMIBC, radical cystectomy (RC) with pelvic lymph node dissection remains the standard treatment [ 4 ]. Despite advances in surgical techniques and perioperative care, long-term outcomes after RC remain unsatisfactory. The 5-year recurrence-free survival rate of patients with invasive bladder cancer undergoing RC is approximately 60–70%, and overall only about half of patients achieve durable survival [ 5 ]. Consistent with these long-term outcomes, the RAZOR trial reported that the 36-month (3-year) overall survival after radical cystectomy was 73.9% in the robot-assisted group and 68.5% in the open-surgery group [ 6 ].Early identification of patients at high risk of death in the first few years after RC is therefore crucial to guide follow-up intensity and adjuvant therapy. Postoperative prognostic assessment currently relies mainly on traditional clinicopathological variables, including pathological T and N stage, tumor grade, tumor size and multiplicity, surgical margins, and lymph node status [ 7 , 8 ]. Nomograms derived from these factors can outperform TNM staging but still have limitations in modelling nonlinear relationships and complex interactions, and often omit potentially relevant markers such as systemic inflammation, nutritional status, and perioperative complications [ 9 ]. More flexible methods are needed to integrate heterogeneous features and refine individual risk stratification. Machine learning (ML) models, such as random forests, support vector machines, and gradient boosting, have shown superior predictive performance over conventional approaches in various oncologic settings, including bladder cancer after RC [ 10 , 11 ]. However, many ML models function as “black boxes” and provide little insight into how specific predictors influence the outcome. Shapley additive explanations (SHAP) offer a practical solution by decomposing model predictions into feature-level contributions, thereby quantifying the impact of each variable on the predicted risk at both population and individual levels [ 12 ]. To date, few studies have combined multiple machine learning algorithms with SHAP to predict medium-term survival, such as 3-year overall survival, after radical cystectomy. Most existing studies focus on a single algorithm, such as logistic regression or random forest, without addressing model interpretability. Additionally, many studies have not validated their models externally, limiting the generalizability of their findings. Therefore, in this study we retrospectively analysed patients who underwent RC for bladder cancer at our institution and developed several 3-year survival prediction models based on both conventional and ML algorithms. We compared their discrimination, calibration, and clinical usefulness, and used SHAP to interpret the contribution of key predictors. Our aim was to establish an accurate and interpretable tool to support individualized postoperative risk stratification in patients with bladder cancer undergoing RC. 2 Materials and methods 2.1 Data Source and Study population This retrospective cohort study was conducted at the First Affiliated Hospital of Xinjiang Medical University. We consecutively identified patients who underwent radical cystectomy for bladder cancer between January 2018 and December 2022 from the hospital electronic medical record system (Jiahe 6.0). Demographic characteristics, preoperative laboratory parameters, perioperative clinical variables, radiologic findings, and postoperative pathological data were retrieved and curated using unique inpatient hospitalization identifiers. Follow-up data were obtained from outpatient records and standardized telephone interviews, and were used to determine survival status and postoperative outcomes. Patients were included if they met all of the following criteria: (1) underwent radical cystectomy at our institution and had a complete preoperative work-up, including adequate cross-sectional imaging that allowed reliable assessment of tumor location and extent; (2) had postoperative pathological confirmation of urothelial carcinoma of the bladder; (3) had at least 3 years of postoperative follow-up with sufficient documentation to ascertain 3-year survival status. Patients were excluded if: (1) bladder involvement represented metastasis from another primary malignancy; (2) concomitant upper tract urothelial carcinoma or other synchronous malignancies were present, except for incidentally detected prostate adenocarcinoma in the cystoprostatectomy specimen; (3) the final histology was non-urothelial (e.g., squamous cell carcinoma, adenocarcinoma, or neuroendocrine carcinoma). 2.2 Ethics statement The study was conducted in accordance with the Declaration of Helsinki. The study protocol was reviewed and approved by the Institutional Review Board of the First Affiliated Hospital of Xinjiang Medical University, and the requirement for written informed consent was waived because of the retrospective nature of the study and the use of de-identified data. All analyses were performed on anonymized datasets. Hospitalization identifiers were used exclusively for data linkage and quality control and did not contain direct personal identifiers. 2.3 Data preprocessing and feature selection Our study extracted all clinicopathological and perioperative variables from the electronic medical record system. We applied predefined inclusion and exclusion criteria and manually verified data quality. We examined continuous variables for distributional properties. We standardized continuous features to z-scores for scale-sensitive algorithms (logistic regression, support vector machine, and k-nearest neighbours). We left variables on their original scale for tree-based models (random forest, XGBoost, LightGBM, and CatBoost). Categorical variables were one-hot encoded (including T/N/M stages). Binary variables were encoded as 0/1. The primary outcome was 3-year overall survival (OS), defined as survival status at 3 years after radical cystectomy; coded as survivor vs non-survivor. We randomly split the cohort into training and internal validation sets at a 7:3 ratio and stratified the split by 3-year survival status. The study conducted feature selection in two stages using only the training set. First, we entered all candidate predictors into a LASSO logistic regression model. We used ten-fold cross-validation to choose the penalty parameter λ that minimized the mean cross-validated error (Fig. 1 A–B). At this λ, we retained variables with non-zero coefficients, leaving 16 preliminary features. Second, we applied recursive feature elimination with cross-validation (RFECV) using a random forest classifier to capture potential non-linear effects and interactions. In each iteration, we removed the least important feature based on the mean decrease in Gini impurity and evaluated performance with ten-fold cross-validated AUC. We selected the feature number that maximized the mean AUC (n = 10) as the final subset (Fig. 1 C). We used these 10 variables as inputs for all machine-learning models. 2.4 Model establishment and performance Using the final set of 10 predictors derived from the feature-selection procedure, we developed several prognostic models for 3-year overall survival after radical cystectomy. Seven supervised machine-learning algorithms were employed: logistic regression (LR), k-nearest neighbours (KNN), support vector machine with radial basis function kernel (SVM-RBF), random forest (RF), extreme gradient boosting (XGBoost), Light Gradient Boosting Machine (LightGBM) and CatBoost. All models were implemented in Python using the scikit-learn, xgboost, lightgbm and catboost libraries. The endpoint was 3-year overall survival, coded as a binary outcome (0 = alive at 3 years, 1 = death within 3 years), and each model was trained to output the predicted probability of death within 3 years after radical cystectomy . Hyperparameters for each machine-learning algorithm were tuned in the training cohort by grid search combined with 10-fold cross-validation, with the mean area under the receiver operating characteristic curve (AUC) used as the primary optimisation criterion. For LR, the type and strength of regularisation were optimised; for KNN, the number of neighbours and distance metric; for SVM-RBF, the penalty parameter \(\:\text{C}\) and kernel coefficient \(\:\text{γ}\) ; for RF, the number of trees, maximum depth and minimum number of samples required for node split; and for XGBoost, LightGBM and CatBoost, key boosting parameters (learning rate, maximum tree depth, number of estimators, subsample ratio and column subsample ratio) were tuned. After selection of the optimal hyperparameters, each model was refitted on the entire training set and then applied to the independent validation set to obtain predicted probabilities. Model performance was evaluated from multiple perspectives. Discrimination was assessed by ROC curves and corresponding AUC values in both the training and validation cohorts (Figs. 2 A and 3 A). Calibration was examined by plotting observed versus predicted 3-year overall survival (OS) rate across deciles of predicted risk (Figs. 2 B and 3 B). Clinical utility was evaluated using decision-curve analysis (DCA), in which the net benefit of each model was calculated across a range of threshold probabilities and compared with the default strategies of “treat none” and “treat all” (Figs. 2 C and 3 C). In addition, precision–recall (PR) curves and the associated average precision (AP) were generated to characterise performance under class imbalance (Figs. 2 D and 3 D). Accuracy, sensitivity, specificity, positive predictive value, negative predictive value and F1-score in the validation cohort were also reported. The model that achieved the most favourable balance of discrimination, calibration and net clinical benefit was regarded as the optimal model and was further interpreted using SHAP analysis, as described below. The final model was prespecified to be the algorithm showing the most favorable overall profile across discrimination (AUC), calibration, and decision-curve net benefit in the validation set, rather than the highest AUC alone. 2.5 Interpretability analysis To improve model transparency, we used Shapley Additive Explanations (SHAP) to quantify the contribution of each predictor to the model output (predicted probability of death within 3 years) for each patient. SHAP analyses were primarily performed for the final selected CatBoost model. For tree-based models (random forest and CatBoost), SHAP values were computed using TreeExplainer; for SVM-RBF, SHAP values were estimated using KernelExplainer based on predicted probabilities. Global interpretability was evaluated using a SHAP summary (beeswarm) plot and by comparing normalized mean absolute SHAP values across CatBoost, random forest, and SVM-RBF models. Local interpretability was assessed using SHAP waterfall and force plots for representative low- and high-risk individuals. 3 Results 3.1 Baseline characteristics Between January 2018 and December 2022, 300 patients with pathologically confirmed bladder cancer who underwent radical cystectomy at the First Affiliated Hospital of Xinjiang Medical University were included in this study. According to 3-year overall survival status, 236 patients (78.7%) were classified as survivors and 64 (21.3%) as non-survivors. Baseline clinicopathological and perioperative characteristics are summarised in Table 1 . Table 1 Baseline characteristics of patients stratified by 3-year survival status Variable Category Survival (n = 236) Death (n = 64) P value Age group (years) < 40 2 (0.8%) 1 (1.6%) < 0.001 40–44 1 (0.4%) 1 (1.6%) 45–49 7 (3.0%) 1 (1.6%) 50–54 22 (9.3%) 2 (3.1%) 55–59 36 (15.3%) 6 (9.4%) 60–64 62 (26.3%) 7 (10.9%) 65–69 50 (21.2%) 15 (23.4%) 70–74 36 (15.3%) 10 (15.6%) 75–79 10 (4.2%) 7 (10.9%) 80–84 10 (4.2%) 12 (18.8%) 85+ 0 (0.0%) 2 (3.1%) Sex Female 44 (18.6%) 11 (17.2%) 0.932 Male 192 (81.4%) 53 (82.8%) Preoperative weight loss No 170 (72.0%) 32 (50.0%) 0.001 Yes 66 (28.0%) 32 (50.0%) Previous abdominal surgery No 191 (80.9%) 48 (75.0%) 0.411 Yes 45 (19.1%) 16 (25.0%) Neoadjuvant immunotherapy No 219 (92.8%) 60 (93.8%) 1.000 Yes 17 (7.2%) 4 (6.2%) Neoadjuvant radiotherapy No 228 (97.5%) 62 (96.9%) 0.682 Yes 6 (2.5%) 2 (3.1%) Neoadjuvant chemotherapy No 176 (74.6%) 50 (78.1%) 0.674 Yes 60 (25.4%) 14 (21.9%) Smoking history No 174 (73.7%) 51 (79.7%) 0.310 Yes 62 (26.3%) 12 (18.8%) Heart disease No 215 (91.1%) 58 (90.6%) 1.000 Yes 21 (8.9%) 5 (7.8%) Diabetes mellitus No 214 (90.7%) 59 (92.2%) 0.617 Yes 22 (9.3%) 4 (6.2%) Hypertension No 190 (80.5%) 46 (71.9%) 0.262 Yes 46 (19.5%) 17 (26.6%) Adjuvant radiotherapy No 221 (93.6%) 55 (85.9%) 0.083 Yes 15 (6.4%) 9 (14.1%) Adjuvant immunotherapy No 215 (91.1%) 52 (81.2%) 0.025 Yes 19 (8.1%) 12 (18.8%) Adjuvant chemotherapy No 183 (77.5%) 45 (70.3%) 0.249 Yes 51 (21.6%) 19 (29.7%) Carcinoma in situ No 213 (90.3%) 61 (95.3%) 0.435 Yes 21 (8.9%) 3 (4.7%) Tumor grade Moderately differentiated 1 (0.4%) 0 (0.0%) 0.047 Low grade 27 (11.4%) 1 (1.6%) High grade 208 (88.1%) 63 (98.4%) Number of tumors Single 140 (59.3%) 42 (65.6%) 0.441 Multiple 96 (40.7%) 22 (34.4%) Tumor size (cm) < 3 135 (57.2%) 30 (46.9%) 0.053 ≥3 101 (42.8%) 34 (53.1%) Pathological T stage T1 83 (35.2%) 1 (1.6%) < 0.001 T2 107 (45.3%) 12 (18.8%) T3 39 (16.5%) 28 (43.8%) T4 7 (3.0%) 23 (35.9%) Pathological N stage N0 199 (84.3%) 28 (43.8%) < 0.001 N1 31 (13.1%) 15 (23.4%) N2 6 (2.5%) 21 (32.8%) Pathological M stage M0 230 (97.5%) 40 (62.5%) < 0.001 M1 6 (2.5%) 23 (35.9%) Surgical approach Open 151 (64.0%) 54 (84.4%) 0.002 Robotic 10 (4.2%) 1 (1.6%) Laparoscopic 75 (31.8%) 7 (10.9%) Urinary diversion Orthotopic neobladder 88 (37.3%) 11 (17.2%) < 0.001 Ileal neobladder 109 (46.2%) 29 (45.3%) Cutaneous ureterostomy 39 (16.5%) 24 (37.5%) Perioperative blood transfusion No 166 (70.3%) 35 (54.7%) 0.027 Yes 70 (29.7%) 29 (45.3%) Postoperative complications No 146 (61.9%) 23 (35.9%) < 0.001 Yes 90 (38.1%) 41 (64.1%) Preoperative creatinine (µmol/L) 72.8 (62.2–87.0) 75.6 (62.2–98.2) 0.140 Albumin (g/L) 37.7 (34.5–40.5) 37.0 (34.8–40.5) 0.648 Globulin (g/L) 29.8 (27.1–32.2) 30.8 (27.9–34.9) 0.039 Albumin-to-globulin ratio (A/G) 1.3 (1.1–1.4) 1.2 (1.1–1.4) 0.074 Platelet count (×10⁹/L) 228.5 (180.8–268.0) 237.5 (201.0-296.8) 0.116 Neutrophils (×10⁹/L) 3.7 (2.8–4.8) 4.0 (3.2-5.0) 0.102 Lymphocytes (×10⁹/L) 1.5 (1.2–1.9) 1.5 (1.2–1.9) 0.868 Neutrophil-to-lymphocyte ratio (NLR) 2.3 (1.6–3.7) 2.6 (1.9–3.7) 0.263 BMI (kg/m²) 24.0 (22.0–26.0) 21.0 (20.0–25.0) 0.002 Length of stay (days) 24.0 (19.8–29.2) 24.5 (20.0–28.0) 0.814 Estimated blood loss (mL) 300.0 (200.0-500.0) 300.0 (200.0-525.0) 0.448 Operative time, (min) 430.0 (370.0-505.0) 397.5 (338.8–450.0) 0.011 Time to ambulation (days) 3.0 (3.0–4.0) 3.0 (3.0–4.0) 0.384 Time to flatus, (days) 3.0 (3.0–4.0) 3.0 (3.0–4.0) 0.574 Time to oral intake (days) 4.0 (3.0–4.0) 4.0 (3.0–5.0) 0.386 Notes: Survival indicates alive at 3 years after surgery; Death indicates death within 3 years. Categorical variables are presented as n (%), and continuous variables as median (IQR). P values < 0.001 are shown as < 0.001. BMI, body mass index; NLR, neutrophil-to-lymphocyte ratio; A/G, albumin-to-globulin ratio. Overall, the cohort was predominantly male (approximately 80%), and most patients were middle-aged or elderly. Non-survivors tended to be older than survivors, with a markedly higher proportion of patients aged ≥ 75 years (32.8% vs 8.4%; P < 0.001). Preoperative weight loss was also more frequent in the non-survivor group (50.0% vs 28.0%; P = 0.001). By contrast, the prevalence of previous abdominal surgery, smoking, cardiovascular disease, diabetes and hypertension did not differ significantly between the two groups (all P > 0.05). Regarding laboratory and nutritional indices, non-survivors had a lower body mass index than survivors (21.0 [20.0–25.0] vs 24.0 [22.0–26.0] kg/m²; P = 0.002) and slightly higher serum globulin levels (30.8 [27.9–34.9] vs 29.8 [27.1–32.2] g/L; P = 0.039), whereas preoperative creatinine, albumin, platelet count, neutrophil and lymphocyte counts, and NLR were broadly comparable between groups (all P > 0.05). Tumour-related and perioperative variables showed more pronounced differences. Non-survivors were more likely to have high-grade disease, advanced pathological T stage (T3–4: 79.7% vs 19.5%), nodal involvement (N1–2: 56.3% vs 15.7%) and distant metastasis (M1: 35.9% vs 2.5%; all P < 0.001). Open radical cystectomy was performed more frequently in non-survivors, whereas minimally invasive approaches were more common among survivors ( P = 0.002). With respect to urinary diversion, orthotopic neobladder reconstruction was less frequent and cutaneous ureterostomy more frequent in the non-survivor group ( P < 0.001). Non-survivors were also more likely to receive perioperative blood transfusion (45.3% vs 29.7%; P = 0.027), experience postoperative complications (64.1% vs 38.1%; P < 0.001), and undergo adjuvant immunotherapy (18.8% vs 8.1%; P = 0.025). 3.2 Model performance analysis Table 2 and Figs. 2 – 3 summarize the discrimination and classification performance of the candidate models. Although Random Forest achieved the highest AUC (0.950) in the validation set, CatBoost showed a better overall balance of discrimination, calibration, and clinical utility (accuracy 0.862, specificity 0.951, PPV 0.846) and was therefore selected as the final model for SHAP interpretation. Given the class imbalance in the data (21.3% mortality rate), we also reported precision–recall curves and average precision (AP) as additional metrics for assessing model performance, as they are more informative under class-imbalanced conditions. Table 2 Performance of machine learning models Model Training set Validation set AUC Accuracy Sensitivity Specificity PPV NPV AUC Accuracy Sensitivity Specificity PPV NPV Random Forest 0.913 0.842 0.590 0.947 0.821 0.848 0.950 0.828 0.647 0.902 0.733 0.860 SVM (RBF) 0.910 0.887 0.718 0.957 0.875 0.891 0.943 0.828 0.529 0.951 0.818 0.830 CatBoost 0.891 0.827 0.513 0.957 0.833 0.826 0.931 0.862 0.647 0.951 0.846 0.867 XGBoost 0.887 0.812 0.564 0.915 0.733 0.835 0.882 0.845 0.647 0.927 0.786 0.864 Logistic Regression 0.885 0.872 0.692 0.947 0.844 0.881 0.872 0.879 0.706 0.951 0.857 0.886 LightGBM 0.841 0.827 0.590 0.926 0.767 0.845 0.859 0.776 0.412 0.927 0.700 0.792 KNN 0.894 0.835 0.564 0.947 0.815 0.840 0.834 0.862 0.588 0.976 0.909 0.851 Notes: AUC, area under the receiver operating characteristic curve; PPV, positive predictive value; NPV, negative predictive value; RBF, radial basis function; KNN, k-nearest neighbors. Classification metrics were calculated using a probability threshold of 0.5 (or the optimal Youden index threshold) in the validation set. 3.3 Explanation of risk factors Global interpretability analyses are presented in Figs. 4 – 5 . The radar plot (Fig. 4 ) compares normalized mean absolute SHAP values across CatBoost, random forest, and SVM-RBF models, showing consistent prioritization of tumor stage variables (T stage, N stage, and M stage). The SHAP summary (beeswarm) plot for the final CatBoost model (Fig. 5 ) further illustrates both the magnitude and direction of each feature’s effect on the predicted probability of death within 3 years overall survival rate, highlighting the dominant contribution of tumor stage alongside host-related markers (age, BMI, albumin, globulin, lymphocyte count, platelet count, and preoperative creatinine). Local SHAP explanations for representative patients are shown in the waterfall plots (Figs. 6 – 7 ) and force plots (Figs. 8 – 9 ). In a low-risk patient (f(x) = 0.298), node-negative and non-metastatic status reduced the predicted probability, whereas higher T stage and adverse host-related features increased risk (Figs. 6 and 8 ). Conversely, in a high-risk patient (f(x) = 0.728), metastatic disease, advanced T stage, and markedly elevated preoperative creatinine were the main contributors shifting the prediction toward mortality (Figs. 7 and 9 ). 4 Discussion In this single-center retrospective cohort of patients undergoing radical cystectomy for urothelial carcinoma, we developed and internally validated a machine-learning framework to predict 3-year overall survival using routinely available perioperative variables together with TNM stage. [ 13 , 14 ].A two-stage feature-selection strategy (LASSO followed by RFECV) yielded a parsimonious 10-variable signature capturing both tumor burden (pathological T/N/M stage) and host-related status (albumin, lymphocyte count, BMI, creatinine, platelet count, and globulin). Among the evaluated algorithms, CatBoost achieved the most favorable balance of discrimination, calibration, and clinical utility in internal validation and was selected as the final model. To enhance transparency, we used SHAP-based explanations to attribute predicted risk at both cohort and individual levels, supporting clinical interpretability and informing future research on explainable AI-assisted risk stratification in healthcare [ 15 ].Feature selection for prognostic modeling after radical cystectomy used cross-validated LASSO to minimize dimensionality and RFECV to refine predictors, balancing model complexity with clinical applicability while reducing overfitting and information leakage [ 16 ]. Using SHAP, We enhanced the interpretability of our machine learning model using SHAP, which provided clinically coherent and biologically plausible insights into 3-year overall survival following radical cystectomy.[ 15 ].Across the SHAP summary plot and radar visualization, pathological T stage, N stage, and M stage were consistently the three most influential predictors, reinforcing the central oncologic principle that anatomical tumor burden underpins prognosis in bladder cancer [ 14 , 17 ].Beyond tumor stage, the model highlighted host status—markers of nutrition (lower BMI and albumin), immune competence (lower lymphocyte count), and renal function (higher creatinine)—as independent contributors to mortality risk [ 18 , 19 ].This pattern aligns with evidence linking systemic inflammation, malnutrition, and frailty-related physiology to adverse outcomes in urothelial carcinoma and major oncologic surgery [ 18 – 20 ].Consistently, our baseline comparisons showed a lower BMI in non-survivors, supporting the clinical relevance of nutritional vulnerability in this setting [ 19 , 20 ].Taken together, these findings suggest that perioperative optimization of modifiable host factors (e.g., nutritional support and renal optimization) may complement oncologic management and merits prospective evaluation [ 14 , 20 , 21 ].At the individual level, SHAP waterfall and decision plots (Figs. 6 – 9 ) decomposed risk into patient-specific contributions, improving interpretability for clinicians and patients. For example, a patient with T3 disease but M0 status and preserved host physiology could receive a lower predicted risk because protective host features partially offset anatomical risk. Conversely, a patient with advanced disease (e.g., T4M1) and severe renal dysfunction would show additive risk contributions from both tumor burden and comorbidity, making the rationale for a high-risk classification explicit. Such transparent explanations are particularly relevant for shared decision-making and for increasing clinician trust in model-assisted risk stratification [ 15 , 22 ]. All predictors used by our final CatBoost model are routinely collected in perioperative care, which may support future evaluation of implementation as a postoperative risk-stratification tool. Nevertheless, because this was a single-center retrospective study with internal validation only, external validation and potential recalibration are required across different case-mix, baseline risk, and treatment patterns. In addition, retrospective data collection may introduce selection bias and unmeasured confounding; follow-up ascertainment and missing data mechanisms should be carefully assessed in future prospective, multi-center studies. Our calibration plots showed jagged curves in some probability ranges, particularly in internal validation, which is consistent with limited event counts, binning effects, and instability in score-to-probability mapping [ 24 ].This issue is especially relevant for models whose native outputs are not probabilities, because additional calibration steps may be required to obtain reliable absolute risks [ 25 ].Therefore, while the model performed strongly overall, raw predicted probabilities should be interpreted cautiously prior to clinical implementation. Before deployment, formal recalibration should be performed and reported using calibration intercept/slope and the Brier score, ideally within a nested resampling framework. External validation is then required to assess transportability under different baseline risks and treatment patterns and to guide model updating if calibration drift is observed. This study is limited by its single-center retrospective design, which may introduce selection bias and restrict generalizability [ 23 , 24 , 26 ].We modeled a binary 3-year OS endpoint rather than fully leveraging time-to-event information, and future work should extend to survival modeling with time-dependent validation. The relatively limited number of death events likely contributed to mild calibration instability, emphasizing the need for larger cohorts and principled recalibration and model updating. In addition, we did not include granular treatment variables (e.g., chemotherapy regimen details or dose intensity), which could confound observed associations and should be incorporated in subsequent studies. Future research should prospectively validate this model, compare it head-to-head against established clinical nomograms, and test whether model-guided strategies improve patient-centered outcomes [ 27 ].Finally, integrating emerging biomarkers such as circulating tumor DNA and radiomics may further enhance prognostic performance, but such extensions must be evaluated with rigorous external validation to avoid overfitting and inflated optimism. In conclusion, we developed and internally validated multiple machine-learning models to predict 3-year overall survival after radical cystectomy using routinely available clinicopathologic and laboratory variables. The final SHAP-interpretable CatBoost model demonstrated strong overall performance and provides transparent individualized explanations. The model may assist postoperative risk stratification, but it requires external validation and calibration assessment before any clinical use and should not replace clinical judgment. Abbreviations LASSO least absolute shrinkage and selection operator RFECV recursive feature elimination with cross-validation CV cross-validation MSE mean squared error AUC area under the receiver operating characteristic curve λ regularization parameter. Declarations Ethics approval and consent to participate This study was conducted in accordance with the Declaration of Helsinki. This retrospective study was granted an ethics waiver by the Ethics Committee of the First Affiliated Hospital of Xinjiang Medical University. The requirement for informed consent was waived due to the retrospective nature of the study and the use of de-identified data. Consent for publication Not applicable. Trial registration This study did not involve a prospective clinical trial. Trial registration: Not applicable. Availability of data and materials The datasets generated and/or analyzed during the current study are available in the Figshare repository, with the following DOI: 10.6084/m9.figshare.31048870. The data is fully anonymized and includes all variables necessary to replicate the analyses presented in this study. Competing interests The authors declare that they have no competing interests. Author contributions statement YW and AA contributed equally to this work and share first authorship. YW and AA conceived and designed the study, collected and curated the data, and drafted the manuscript. SC contributed to methodology, statistical analysis/model development, and critical revision of the manuscript. WW supervised the study, provided clinical interpretation, and critically revised the manuscript. All authors read and approved the final manuscript. Funding This work was supported by the Tianshan Talent Training Program (Medical and Health High-level Talent Project) of the Health Commission of Xinjiang Uygur Autonomous Region (Grant No. TSYC202401B049). References Bray F, Laversanne M, Sung H, Ferlay J, Siegel RL, Soerjomataram I, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2024;74(3):229-63. Zhang Y, Rumgay H, Li M, Yu H, Pan H, Ni J. The global landscape of bladder cancer incidence and mortality in 2020 and projections to 2040. J Glob Health. 2023;13:04109. van der Heijden AG, Bruins HM, Carrion A, Cathomas R, Compérat E, Dimitropoulos K, et al. European Association of Urology guidelines on muscle-invasive and metastatic bladder cancer: summary of the 2025 guidelines. Eur Urol. 2025;87(5):582-600. Flaig TW, Spiess PE, Abern M, Agarwal N, Bangs R, Buyyounouski MK, et al. NCCN Guidelines® Insights: bladder cancer, version 3.2024. J Natl Compr Canc Netw. 2024;22(4):216-25. Bizzarri FP, Scarciglia E, Russo P, Marino F, Presutti S, Moosavi SK, et al. Elderly and bladder cancer: the role of radical cystectomy and orthotopic urinary diversion. Urologia. 2024;91(3):500-4. Zapała Ł, Ślusarczyk A, Korczak B, Kurzyna P, Leki M, Lipiński P, et al. The view outside of the box: reporting outcomes following radical cystectomy using pentafecta from a multicenter retrospective analysis. Front Oncol. 2022;12:841852. Stein JP, Lieskovsky G, Cote R, Groshen S, Feng AC, Boyd S, et al. Radical cystectomy in the treatment of invasive bladder cancer: long-term results in 1,054 patients. J Clin Oncol. 2001;19(3):666-75. Hinsenveld FJ, Boormans JL, van der Poel HG, van der Schoot DKE, Vis AN, Aben KKH, et al. Intermediate-term survival of robot-assisted versus open radical cystectomy for muscle-invasive and high-risk non-muscle invasive bladder cancer in The Netherlands. Urol Oncol. 2022;40(2):60.e1-60.e9. Shariat SF, Karakiewicz PI, Palapattu GS, Amiel GE, Lotan Y, Rogers CG, et al. Nomograms provide improved accuracy for predicting survival after radical cystectomy. Clin Cancer Res. 2006;12(22):6663-76. Zaak D, Burger M, Otto W, Bastian PJ, Denzinger S, Stief CG, et al. Predicting individual outcomes after radical cystectomy: an external validation of current nomograms. BJU Int. 2010;106(3):342-8. Bochner BH, Kattan MW, Vora KC. Postoperative nomogram predicting risk of recurrence after radical cystectomy for bladder cancer. J Clin Oncol. 2006;24(24):3967-72. Kwon T, Jeong IG, You D, Hong B, Hong JH, Ahn H, et al. Long-term oncologic outcomes after radical cystectomy for bladder cancer at a single institution. J Korean Med Sci. 2014;29(5):669-75. Witjes JA, Bruins HM, Cathomas R, Compérat EM, Cowan NC, Gakis G, et al. European Association of Urology guidelines on muscle-invasive and metastatic bladder cancer: summary of the 2020 guidelines. Eur Urol. 2021;79(1):82-104. Lenis AT, Lec PM, Chamie K. Bladder cancer: a review. JAMA. 2020;324(19):1980-91. Tjoa E, Guan C. A survey on explainable artificial intelligence (XAI): toward medical XAI. IEEE Trans Neural Netw Learn Syst. 2021;32(11):4793-813. Moons KGM, Wolff RF, Riley RD, Whiting PF, Westwood M, Collins GS, et al. PROBAST: a tool to assess risk of bias and applicability of prediction model studies: explanation and elaboration. Ann Intern Med. 2019;170(1):W1-W33. Witjes JA, Lebret T, Compérat EM, Cowan NC, De Santis M, Bruins HM, et al. Updated 2016 EAU guidelines on muscle-invasive and metastatic bladder cancer. Eur Urol. 2017;71(3):462-75. Li X, Ma X, Tang L, Wang B, Chen L, Zhang F, et al. Prognostic value of neutrophil-to-lymphocyte ratio in urothelial carcinoma of the upper urinary tract and bladder: a systematic review and meta-analysis. Oncotarget. 2017;8(37):62681-92. Dobruch J, Oszczudłowski M. Bladder cancer: current challenges and future directions. Medicina (Kaunas). 2021;57(8):749. Gillis C, Wischmeyer PE. Pre-operative nutrition and the elective surgical patient: why, how and what? Anaesthesia. 2019;74(Suppl 1):27-35. Ljungqvist O, Scott M, Fearon KC. Enhanced recovery after surgery: a review. JAMA Surg. 2017;152(3):292-8. Vemulapalli V, Qu J, Garren JM, Rodrigues LO, Kiebish MA, Sarangarajan R, et al. Non-obvious correlations to disease management unraveled by Bayesian artificial intelligence analyses of CMS data. Artif Intell Med. 2016;74:1-8. Collins GS, Reitsma JB, Altman DG, Moons KG. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ. 2015;350:g7594. Van Calster B, McLernon DJ, van Smeden M, Wynants L, Steyerberg EW. Calibration: the Achilles heel of predictive analytics. BMC Med. 2019;17(1):230. Dong W, Jiang H, Li Y, Lv L, Gong Y, Li B, et al. Interpretable machine learning analysis of immunoinflammatory biomarkers for predicting CHD among NAFLD patients. Cardiovasc Diabetol. 2025;24(1):263. Wolff RF, Moons KGM, Riley RD, Whiting PF, Westwood M, Collins GS, et al. PROBAST: a tool to assess the risk of bias and applicability of prediction model studies. Ann Intern Med. 2019;170(1):51-8. Grossman HB, Natale RB, Tangen CM, Speights VO, Vogelzang NJ, Trump DL, et al. Neoadjuvant chemotherapy plus cystectomy compared with cystectomy alone for locally advanced bladder cancer. N Engl J Med. 2003;349(9):859-66. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 30 Mar, 2026 Reviews received at journal 27 Feb, 2026 Reviews received at journal 08 Feb, 2026 Reviewers agreed at journal 08 Feb, 2026 Reviewers agreed at journal 07 Feb, 2026 Reviewers invited by journal 05 Feb, 2026 Editor assigned by journal 05 Feb, 2026 Editor invited by journal 05 Feb, 2026 Submission checks completed at journal 04 Feb, 2026 First submitted to journal 04 Feb, 2026 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-8754027","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":587661729,"identity":"56b936f7-ff86-4f67-a9fc-13158a0be591","order_by":0,"name":"Yunze Wang","email":"","orcid":"","institution":"First Affiliated Hospital of Xinjiang Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yunze","middleName":"","lastName":"Wang","suffix":""},{"id":587661736,"identity":"a4f25f73-4779-493e-8f78-f1a34302687e","order_by":1,"name":"Aikeshanjiang Ailiyaer","email":"","orcid":"","institution":"First Affiliated Hospital of Xinjiang Medical University","correspondingAuthor":false,"prefix":"","firstName":"Aikeshanjiang","middleName":"","lastName":"Ailiyaer","suffix":""},{"id":587661738,"identity":"9ecff820-37f3-45a7-9500-cf0aecacd381","order_by":2,"name":"Shiming Chen","email":"","orcid":"","institution":"First Affiliated Hospital of Xinjiang Medical University","correspondingAuthor":false,"prefix":"","firstName":"Shiming","middleName":"","lastName":"Chen","suffix":""},{"id":587661739,"identity":"e30c43c5-d5e3-404b-87a3-91d04b604882","order_by":3,"name":"Wenguang Wang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAt0lEQVRIiWNgGAWjYJCCAx8MauTY2NsPEK2D8eGMimPGfDxnEojWwmzMc4Y5cZ6EgwFx6g1upF+T5m1jS2+TYEhg+FGxjRgtOWWSc9tkctukGw8w9py5TViL2e2cNIm3bWy5bTIHEpgZ24jVwtvGnM4mkWBArJb0w4ZA7ycQr8X+/htwIBu2AQP5IFF+kew5/gAUlfLy7e0HH/yoIEILAwMPIjoOEKMeCNgfEKlwFIyCUTAKRiwAAPx0QBbSsqbbAAAAAElFTkSuQmCC","orcid":"","institution":"First Affiliated Hospital of Xinjiang Medical University","correspondingAuthor":true,"prefix":"","firstName":"Wenguang","middleName":"","lastName":"Wang","suffix":""}],"badges":[],"createdAt":"2026-02-01 06:09:23","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8754027/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8754027/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":102398218,"identity":"ced3d962-028c-431d-8981-402d1e0dfe91","added_by":"auto","created_at":"2026-02-11 10:21:47","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":82004,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eTwo-stage feature selection using LASSO and RFECV.\u003c/strong\u003e(A) LASSO coefficient paths plotted against log(λ); each curve represents one predictor, and the vertical dashed line indicates the penalty (λ) selected by cross-validation.(B) Cross-validated mean squared error (MSE) versus log(λ) for LASSO; the red point marks the minimum MSE and the vertical dashed line indicates the selected λ, yielding \u003cstrong\u003e16\u003c/strong\u003e predictors with non-zero coefficients.(C) Recursive feature elimination with cross-validation (RFECV) using a random forest model; mean cross-validated AUC is shown across different numbers of retained features, and the red point/vertical dashed line denote the feature subset selected by RFECV (\u003cstrong\u003en = 10\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAbbreviations:\u003c/em\u003eLASSO, least absolute shrinkage and selection operator; RFECV, recursive feature elimination with cross-validation; CV, cross-validation; MSE, mean squared error; AUC, area under the receiver operating characteristic curve; λ, regularization parameter.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8754027/v1/053593381b1618eacb0ef45d.png"},{"id":102375590,"identity":"c30983b8-205d-463d-b28a-20e1295c47f0","added_by":"auto","created_at":"2026-02-11 05:20:57","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":355363,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance comparison of candidate models in the training cohort for 3-year outcome prediction.(A) Receiver operating characteristic (ROC) curves showing discrimination for SVM-RBF (AUC = 0.910), KNN (AUC = 0.894), random forest (AUC = 0.913), XGBoost (AUC = 0.887), LightGBM (AUC = 0.841), CatBoost (AUC = 0.891), and logistic regression (AUC = 0.885); the diagonal dashed line indicates no discrimination.\u003cbr\u003e\n(B) Calibration curves comparing predicted probabilities with observed event proportions; the dashed 45° line denotes perfect calibration.(C) Decision curve analysis (DCA) showing net benefit across threshold probabilities; the dashed horizontal line represents “treat-none,” and the solid black line represents “treat-all.”(D) Precision–recall (PR) curves evaluating event-focused performance; the dashed line indicates the no-skill baseline (event prevalence). Average precision (AP) is shown for each model: SVM-RBF (AP = 0.837), KNN (AP = 0.749), random forest (AP = 0.792), XGBoost (AP = 0.784), LightGBM (AP = 0.734), CatBoost (AP = 0.746), and logistic regression (AP = 0.816).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAbbreviations:\u003c/em\u003eAUC, area under the ROC curve; AP, average precision; DCA, decision curve analysis; KNN, k-nearest neighbors; PR, precision–recall; RBF, radial basis function; SVM, support vector machine.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8754027/v1/363d3d678b66dd2a61573b33.png"},{"id":102375594,"identity":"77a07ccc-9df3-4e07-a3fa-a656c5b57f5b","added_by":"auto","created_at":"2026-02-11 05:20:57","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":345426,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePerformance comparison of candidate models in the validation cohort for 3-year outcome prediction.\u003c/strong\u003e(A) Receiver operating characteristic (ROC) curves showing discrimination for SVM-RBF (AUC = 0.943), KNN (AUC = 0.834), random forest (AUC = 0.950), XGBoost (AUC = 0.882), LightGBM (AUC = 0.859), CatBoost (AUC = 0.931), and logistic regression (AUC = 0.872); the diagonal dashed line indicates no discrimination.(B) Calibration curves comparing predicted probabilities with observed event proportions; the dashed 45° line denotes perfect calibration.(C) Decision curve analysis (DCA) showing net benefit across threshold probabilities; the dashed horizontal line represents “treat-none,” and the solid black line represents “treat-all.”(D) Precision–recall (PR) curves evaluating event-focused performance; the dashed line indicates the no-skill baseline (event prevalence). Average precision (AP) is shown for each model: SVM-RBF (AP = 0.848), KNN (AP = 0.751), random forest (AP = 0.893), XGBoost (AP = 0.817), LightGBM (AP = 0.730), CatBoost (AP = 0.848), and logistic regression (AP = 0.774).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAbbreviations:\u003c/em\u003eAUC, area under the ROC curve; AP, average precision; DCA, decision curve analysis; KNN, k-nearest neighbors; PR, precision–recall; RBF, radial basis function; SVM, support vector machine.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-8754027/v1/d424dc5ccc6c1823d2d25224.png"},{"id":102375596,"identity":"95e928cc-f170-44cf-b796-a24ff10bd339","added_by":"auto","created_at":"2026-02-11 05:20:58","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":107150,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFeature-importance comparison across 3-year prediction models. Radar plot comparing normalized mean absolute SHAP values among CatBoost, random forest, and SVM-RBF models.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAbbreviations: \u003c/em\u003eSHAP, Shapley additive explanations; RBF, radial basis function; SVM, support vector machine.\u003c/p\u003e","description":"","filename":"Fig4.png","url":"https://assets-eu.researchsquare.com/files/rs-8754027/v1/51726cdbc9702a42486e3b83.png"},{"id":102375589,"identity":"b9cc7887-ba72-44d6-b18f-a5d6b3531749","added_by":"auto","created_at":"2026-02-11 05:20:57","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":102315,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eGlobal feature importance and direction of effects using a SHAP summary (beeswarm) plot for the final CatBoost model. Each dot represents one patient; color indicates feature value (low to high), and the x-axis shows SHAP value (impact on the predicted probability of death within 3 years).\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAbbreviations: \u003c/em\u003eSHAP, Shapley additive explanations; BMI, body mass index.\u003c/p\u003e","description":"","filename":"Fig5.png","url":"https://assets-eu.researchsquare.com/files/rs-8754027/v1/b7d549294a8906a56437f9eb.png"},{"id":102375595,"identity":"084523b2-aee7-44f9-a5c9-d106676364c1","added_by":"auto","created_at":"2026-02-11 05:20:57","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":85618,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eIndividual-level model explanation using a SHAP waterfall plot for a representative low-risk prediction (final CatBoost model; f(x) = 0.298). The plot shows how each predictor shifts the baseline risk to the final predicted probability.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAbbreviations: \u003c/em\u003eSHAP, Shapley additive explanations; BMI, body mass index.\u003c/p\u003e","description":"","filename":"fig6.png","url":"https://assets-eu.researchsquare.com/files/rs-8754027/v1/6ef2724e17f6a17f0ee691be.png"},{"id":102404216,"identity":"e8043b92-0e27-49d5-8ffe-59b4866ae5b5","added_by":"auto","created_at":"2026-02-11 11:03:09","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":83702,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eIndividual-level model explanation using a SHAP waterfall plot for a representative high-risk prediction (final CatBoost model; f(x) = 0.728). The plot shows how each predictor shifts the baseline risk to the final predicted probability.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAbbreviations:\u003c/em\u003e SHAP, Shapley additive explanations; BMI, body mass index.\u003c/p\u003e","description":"","filename":"fig7.png","url":"https://assets-eu.researchsquare.com/files/rs-8754027/v1/d022d7254571e7d15a3219ac.png"},{"id":102397671,"identity":"10c96480-4730-4b65-b586-f2bc930b1643","added_by":"auto","created_at":"2026-02-11 10:18:51","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":41026,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSHAP force plot illustrating feature contributions for a representative low-risk prediction from the final CatBoost model. Red features increase the predicted probability, whereas blue features decrease it.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAbbreviations:\u003c/em\u003e SHAP, Shapley additive explanations; BMI, body mass index.\u003c/p\u003e","description":"","filename":"fig8.png","url":"https://assets-eu.researchsquare.com/files/rs-8754027/v1/e418e78f09d9e4fede699b2d.png"},{"id":102375592,"identity":"9ff06638-c88c-450c-be00-d3c2e825ab56","added_by":"auto","created_at":"2026-02-11 05:20:57","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":38496,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSHAP force plot illustrating feature contributions for a representative high-risk prediction from the final CatBoost model. Red features increase the predicted probability, whereas blue features decrease it.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAbbreviations:\u003c/em\u003e SHAP, Shapley additive explanations; BMI, body mass index.\u003c/p\u003e","description":"","filename":"fig9.png","url":"https://assets-eu.researchsquare.com/files/rs-8754027/v1/af4eb9c7caf4da58fbef640d.png"},{"id":102405401,"identity":"f6138778-9dc7-470f-aeb2-f123cf5cc985","added_by":"auto","created_at":"2026-02-11 11:14:33","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2483209,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8754027/v1/759b7175-80a8-41d1-b261-d112774400be.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Development and Internal Validation of an Explainable Machine-Learning Model to Predict 3-Year overall survival rate After Radical Cystectomy","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eBladder cancer is one of the most common malignancies of the urinary tract and its incidence has been increasing worldwide. In 2020, approximately 570,000 new cases were diagnosed globally, accounting for about 3% of all cancers, and by 2022 this number had risen to roughly 610,000 [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. China bears a particularly heavy burden, with about 93,000 new cases and more than 270,000 patients living within 5 years of diagnosis in 2022 [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e].Clinically, bladder cancer is classified into non\u0026ndash;muscle-invasive bladder cancer (NMIBC) and muscle-invasive bladder cancer (MIBC), which differ markedly in prognosis and management [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. For patients with MIBC or high-risk NMIBC, radical cystectomy (RC) with pelvic lymph node dissection remains the standard treatment [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eDespite advances in surgical techniques and perioperative care, long-term outcomes after RC remain unsatisfactory. The 5-year recurrence-free survival rate of patients with invasive bladder cancer undergoing RC is approximately 60\u0026ndash;70%, and overall only about half of patients achieve durable survival [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Consistent with these long-term outcomes, the RAZOR trial reported that the 36-month (3-year) overall survival after radical cystectomy was 73.9% in the robot-assisted group and 68.5% in the open-surgery group [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e].Early identification of patients at high risk of death in the first few years after RC is therefore crucial to guide follow-up intensity and adjuvant therapy.\u003c/p\u003e \u003cp\u003ePostoperative prognostic assessment currently relies mainly on traditional clinicopathological variables, including pathological T and N stage, tumor grade, tumor size and multiplicity, surgical margins, and lymph node status [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Nomograms derived from these factors can outperform TNM staging but still have limitations in modelling nonlinear relationships and complex interactions, and often omit potentially relevant markers such as systemic inflammation, nutritional status, and perioperative complications [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. More flexible methods are needed to integrate heterogeneous features and refine individual risk stratification.\u003c/p\u003e \u003cp\u003eMachine learning (ML) models, such as random forests, support vector machines, and gradient boosting, have shown superior predictive performance over conventional approaches in various oncologic settings, including bladder cancer after RC [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. However, many ML models function as \u0026ldquo;black boxes\u0026rdquo; and provide little insight into how specific predictors influence the outcome. Shapley additive explanations (SHAP) offer a practical solution by decomposing model predictions into feature-level contributions, thereby quantifying the impact of each variable on the predicted risk at both population and individual levels [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. To date, few studies have combined multiple machine learning algorithms with SHAP to predict medium-term survival, such as 3-year overall survival, after radical cystectomy. Most existing studies focus on a single algorithm, such as logistic regression or random forest, without addressing model interpretability. Additionally, many studies have not validated their models externally, limiting the generalizability of their findings.\u003c/p\u003e \u003cp\u003eTherefore, in this study we retrospectively analysed patients who underwent RC for bladder cancer at our institution and developed several 3-year survival prediction models based on both conventional and ML algorithms. We compared their discrimination, calibration, and clinical usefulness, and used SHAP to interpret the contribution of key predictors. Our aim was to establish an accurate and interpretable tool to support individualized postoperative risk stratification in patients with bladder cancer undergoing RC.\u003c/p\u003e"},{"header":"2 Materials and methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Data Source and Study population\u003c/h2\u003e \u003cp\u003eThis retrospective cohort study was conducted at the First Affiliated Hospital of Xinjiang Medical University. We consecutively identified patients who underwent radical cystectomy for bladder cancer between January 2018 and December 2022 from the hospital electronic medical record system (Jiahe 6.0). Demographic characteristics, preoperative laboratory parameters, perioperative clinical variables, radiologic findings, and postoperative pathological data were retrieved and curated using unique inpatient hospitalization identifiers. Follow-up data were obtained from outpatient records and standardized telephone interviews, and were used to determine survival status and postoperative outcomes.\u003c/p\u003e \u003cp\u003ePatients were included if they met all of the following criteria: (1) underwent radical cystectomy at our institution and had a complete preoperative work-up, including adequate cross-sectional imaging that allowed reliable assessment of tumor location and extent; (2) had postoperative pathological confirmation of urothelial carcinoma of the bladder; (3) had at least 3 years of postoperative follow-up with sufficient documentation to ascertain 3-year survival status.\u003c/p\u003e \u003cp\u003ePatients were excluded if: (1) bladder involvement represented metastasis from another primary malignancy; (2) concomitant upper tract urothelial carcinoma or other synchronous malignancies were present, except for incidentally detected prostate adenocarcinoma in the cystoprostatectomy specimen; (3) the final histology was non-urothelial (e.g., squamous cell carcinoma, adenocarcinoma, or neuroendocrine carcinoma).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Ethics statement\u003c/h2\u003e \u003cp\u003e The study was conducted in accordance with the Declaration of Helsinki. The study protocol was reviewed and approved by the Institutional Review Board of the First Affiliated Hospital of Xinjiang Medical University, and the requirement for written informed consent was waived because of the retrospective nature of the study and the use of de-identified data. All analyses were performed on anonymized datasets. Hospitalization identifiers were used exclusively for data linkage and quality control and did not contain direct personal identifiers.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Data preprocessing and feature selection\u003c/h2\u003e \u003cp\u003eOur study extracted all clinicopathological and perioperative variables from the electronic medical record system. We applied predefined inclusion and exclusion criteria and manually verified data quality. We examined continuous variables for distributional properties. We standardized continuous features to z-scores for scale-sensitive algorithms (logistic regression, support vector machine, and k-nearest neighbours). We left variables on their original scale for tree-based models (random forest, XGBoost, LightGBM, and CatBoost). Categorical variables were one-hot encoded (including T/N/M stages). Binary variables were encoded as 0/1. The primary outcome was 3-year overall survival (OS), defined as survival status at 3 years after radical cystectomy; coded as survivor vs non-survivor. We randomly split the cohort into training and internal validation sets at a 7:3 ratio and stratified the split by 3-year survival status.\u003c/p\u003e \u003cp\u003eThe study conducted feature selection in two stages using only the training set. First, we entered all candidate predictors into a LASSO logistic regression model. We used ten-fold cross-validation to choose the penalty parameter λ that minimized the mean cross-validated error (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eA\u0026ndash;B). At this λ, we retained variables with non-zero coefficients, leaving 16 preliminary features. Second, we applied recursive feature elimination with cross-validation (RFECV) using a random forest classifier to capture potential non-linear effects and interactions. In each iteration, we removed the least important feature based on the mean decrease in Gini impurity and evaluated performance with ten-fold cross-validated AUC. We selected the feature number that maximized the mean AUC (n\u0026thinsp;=\u0026thinsp;10) as the final subset (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eC). We used these 10 variables as inputs for all machine-learning models.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Model establishment and performance\u003c/h2\u003e \u003cp\u003eUsing the final set of 10 predictors derived from the feature-selection procedure, we developed several prognostic models for 3-year overall survival after radical cystectomy. Seven supervised machine-learning algorithms were employed: logistic regression (LR), k-nearest neighbours (KNN), support vector machine with radial basis function kernel (SVM-RBF), random forest (RF), extreme gradient boosting (XGBoost), Light Gradient Boosting Machine (LightGBM) and CatBoost. All models were implemented in Python using the scikit-learn, xgboost, lightgbm and catboost libraries. The endpoint was 3-year overall survival, coded as a binary outcome (0\u0026thinsp;=\u0026thinsp;alive at 3 years, 1\u0026thinsp;=\u0026thinsp;death within 3 years), and each model was trained to output the predicted probability of death within 3 years after radical cystectomy .\u003c/p\u003e \u003cp\u003eHyperparameters for each machine-learning algorithm were tuned in the training cohort by grid search combined with 10-fold cross-validation, with the mean area under the receiver operating characteristic curve (AUC) used as the primary optimisation criterion. For LR, the type and strength of regularisation were optimised; for KNN, the number of neighbours and distance metric; for SVM-RBF, the penalty parameter \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\text{C}\\)\u003c/span\u003e\u003c/span\u003eand kernel coefficient \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\text{\u0026gamma;}\\)\u003c/span\u003e\u003c/span\u003e; for RF, the number of trees, maximum depth and minimum number of samples required for node split; and for XGBoost, LightGBM and CatBoost, key boosting parameters (learning rate, maximum tree depth, number of estimators, subsample ratio and column subsample ratio) were tuned. After selection of the optimal hyperparameters, each model was refitted on the entire training set and then applied to the independent validation set to obtain predicted probabilities.\u003c/p\u003e \u003cp\u003eModel performance was evaluated from multiple perspectives. Discrimination was assessed by ROC curves and corresponding AUC values in both the training and validation cohorts (Figs.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA and \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA). Calibration was examined by plotting observed versus predicted 3-year overall survival (OS) rate across deciles of predicted risk (Figs.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB and \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB). Clinical utility was evaluated using decision-curve analysis (DCA), in which the net benefit of each model was calculated across a range of threshold probabilities and compared with the default strategies of \u0026ldquo;treat none\u0026rdquo; and \u0026ldquo;treat all\u0026rdquo; (Figs.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eC and \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eC). In addition, precision\u0026ndash;recall (PR) curves and the associated average precision (AP) were generated to characterise performance under class imbalance (Figs.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eD and \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eD). Accuracy, sensitivity, specificity, positive predictive value, negative predictive value and F1-score in the validation cohort were also reported. The model that achieved the most favourable balance of discrimination, calibration and net clinical benefit was regarded as the optimal model and was further interpreted using SHAP analysis, as described below. The final model was prespecified to be the algorithm showing the most favorable overall profile across discrimination (AUC), calibration, and decision-curve net benefit in the validation set, rather than the highest AUC alone.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5 Interpretability analysis\u003c/h2\u003e \u003cp\u003eTo improve model transparency, we used Shapley Additive Explanations (SHAP) to quantify the contribution of each predictor to the model output (predicted probability of death within 3 years) for each patient.\u003c/p\u003e \u003cp\u003eSHAP analyses were primarily performed for the final selected CatBoost model. For tree-based models (random forest and CatBoost), SHAP values were computed using TreeExplainer; for SVM-RBF, SHAP values were estimated using KernelExplainer based on predicted probabilities.\u003c/p\u003e \u003cp\u003eGlobal interpretability was evaluated using a SHAP summary (beeswarm) plot and by comparing normalized mean absolute SHAP values across CatBoost, random forest, and SVM-RBF models.\u003c/p\u003e \u003cp\u003eLocal interpretability was assessed using SHAP waterfall and force plots for representative low- and high-risk individuals.\u003c/p\u003e \u003c/div\u003e"},{"header":"3 Results","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Baseline characteristics\u003c/h2\u003e \u003cp\u003eBetween January 2018 and December 2022, 300 patients with pathologically confirmed bladder cancer who underwent radical cystectomy at the First Affiliated Hospital of Xinjiang Medical University were included in this study. According to 3-year overall survival status, 236 patients (78.7%) were classified as survivors and 64 (21.3%) as non-survivors. Baseline clinicopathological and perioperative characteristics are summarised in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eBaseline characteristics of patients stratified by 3-year survival status\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCategory\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSurvival (n\u0026thinsp;=\u0026thinsp;236)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eDeath (n\u0026thinsp;=\u0026thinsp;64)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eP value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge group (years)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2 (0.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1 (1.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e40\u0026ndash;44\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1 (0.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1 (1.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e45\u0026ndash;49\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e7 (3.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1 (1.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e50\u0026ndash;54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e22 (9.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2 (3.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e55\u0026ndash;59\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e36 (15.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e6 (9.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e60\u0026ndash;64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e62 (26.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e7 (10.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e65\u0026ndash;69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e50 (21.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e15 (23.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e70\u0026ndash;74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e36 (15.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e10 (15.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e75\u0026ndash;79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e10 (4.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e7 (10.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e80\u0026ndash;84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e10 (4.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e12 (18.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e85+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2 (3.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSex\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e44 (18.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e11 (17.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.932\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e192 (81.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e53 (82.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePreoperative weight loss\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e170 (72.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e32 (50.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e66 (28.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e32 (50.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePrevious abdominal surgery\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e191 (80.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e48 (75.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.411\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e45 (19.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e16 (25.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNeoadjuvant immunotherapy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e219 (92.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e60 (93.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.000\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e17 (7.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4 (6.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNeoadjuvant radiotherapy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e228 (97.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e62 (96.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.682\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e6 (2.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2 (3.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNeoadjuvant chemotherapy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e176 (74.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e50 (78.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.674\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e60 (25.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e14 (21.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSmoking history\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e174 (73.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e51 (79.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.310\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e62 (26.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e12 (18.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHeart disease\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e215 (91.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e58 (90.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.000\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e21 (8.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e5 (7.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiabetes mellitus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e214 (90.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e59 (92.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.617\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e22 (9.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4 (6.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHypertension\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e190 (80.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e46 (71.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.262\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e46 (19.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e17 (26.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAdjuvant radiotherapy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e221 (93.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e55 (85.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.083\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e15 (6.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e9 (14.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAdjuvant immunotherapy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e215 (91.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e52 (81.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.025\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e19 (8.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e12 (18.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAdjuvant chemotherapy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e183 (77.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e45 (70.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.249\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e51 (21.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e19 (29.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCarcinoma in situ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e213 (90.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e61 (95.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.435\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e21 (8.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3 (4.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTumor grade\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eModerately differentiated\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1 (0.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.047\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLow grade\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e27 (11.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1 (1.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigh grade\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e208 (88.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e63 (98.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNumber of tumors\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSingle\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e140 (59.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e42 (65.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.441\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMultiple\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e96 (40.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e22 (34.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTumor size (cm)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e135 (57.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e30 (46.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.053\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026ge;3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e101 (42.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e34 (53.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePathological T stage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eT1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e83 (35.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1 (1.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eT2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e107 (45.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e12 (18.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eT3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e39 (16.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e28 (43.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eT4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e7 (3.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e23 (35.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePathological N stage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eN0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e199 (84.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e28 (43.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eN1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e31 (13.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e15 (23.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eN2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e6 (2.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e21 (32.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePathological M stage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eM0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e230 (97.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e40 (62.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eM1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e6 (2.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e23 (35.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSurgical approach\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOpen\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e151 (64.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e54 (84.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.002\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRobotic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e10 (4.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1 (1.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLaparoscopic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e75 (31.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e7 (10.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUrinary diversion\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOrthotopic neobladder\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e88 (37.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e11 (17.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIleal neobladder\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e109 (46.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e29 (45.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCutaneous ureterostomy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e39 (16.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e24 (37.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePerioperative blood transfusion\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e166 (70.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e35 (54.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.027\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e70 (29.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e29 (45.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePostoperative complications\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e146 (61.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e23 (35.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e90 (38.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e41 (64.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePreoperative creatinine (\u0026micro;mol/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e72.8 (62.2\u0026ndash;87.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e75.6 (62.2\u0026ndash;98.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.140\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlbumin (g/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e37.7 (34.5\u0026ndash;40.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e37.0 (34.8\u0026ndash;40.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.648\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGlobulin (g/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e29.8 (27.1\u0026ndash;32.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e30.8 (27.9\u0026ndash;34.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.039\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlbumin-to-globulin ratio (A/G)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.3 (1.1\u0026ndash;1.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.2 (1.1\u0026ndash;1.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.074\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePlatelet count (\u0026times;10⁹/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e228.5 (180.8\u0026ndash;268.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e237.5 (201.0-296.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.116\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNeutrophils (\u0026times;10⁹/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.7 (2.8\u0026ndash;4.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4.0 (3.2-5.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.102\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLymphocytes (\u0026times;10⁹/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.5 (1.2\u0026ndash;1.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.5 (1.2\u0026ndash;1.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.868\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNeutrophil-to-lymphocyte ratio (NLR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2.3 (1.6\u0026ndash;3.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2.6 (1.9\u0026ndash;3.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.263\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBMI (kg/m\u0026sup2;)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e24.0 (22.0\u0026ndash;26.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e21.0 (20.0\u0026ndash;25.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.002\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLength of stay (days)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e24.0 (19.8\u0026ndash;29.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e24.5 (20.0\u0026ndash;28.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.814\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEstimated blood loss (mL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e300.0 (200.0-500.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e300.0 (200.0-525.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.448\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOperative time, (min)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e430.0 (370.0-505.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e397.5 (338.8\u0026ndash;450.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.011\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTime to ambulation (days)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.0 (3.0\u0026ndash;4.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.0 (3.0\u0026ndash;4.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.384\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTime to flatus, (days)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.0 (3.0\u0026ndash;4.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.0 (3.0\u0026ndash;4.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.574\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTime to oral intake (days)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4.0 (3.0\u0026ndash;4.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4.0 (3.0\u0026ndash;5.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.386\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003eNotes: Survival indicates alive at 3 years after surgery; Death indicates death within 3 years. Categorical variables are presented as n (%), and continuous variables as median (IQR). P values\u0026thinsp;\u0026lt;\u0026thinsp;0.001 are shown as \u0026lt;\u0026thinsp;0.001. BMI, body mass index; NLR, neutrophil-to-lymphocyte ratio; A/G, albumin-to-globulin ratio.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eOverall, the cohort was predominantly male (approximately 80%), and most patients were middle-aged or elderly. Non-survivors tended to be older than survivors, with a markedly higher proportion of patients aged\u0026thinsp;\u0026ge;\u0026thinsp;75 years (32.8% vs 8.4%; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Preoperative weight loss was also more frequent in the non-survivor group (50.0% vs 28.0%; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.001). By contrast, the prevalence of previous abdominal surgery, smoking, cardiovascular disease, diabetes and hypertension did not differ significantly between the two groups (all P\u0026thinsp;\u0026gt;\u0026thinsp;0.05).\u003c/p\u003e \u003cp\u003eRegarding laboratory and nutritional indices, non-survivors had a lower body mass index than survivors (21.0 [20.0\u0026ndash;25.0] vs 24.0 [22.0\u0026ndash;26.0] kg/m\u0026sup2;; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.002) and slightly higher serum globulin levels (30.8 [27.9\u0026ndash;34.9] vs 29.8 [27.1\u0026ndash;32.2] g/L; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.039), whereas preoperative creatinine, albumin, platelet count, neutrophil and lymphocyte counts, and NLR were broadly comparable between groups (all \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.05).\u003c/p\u003e \u003cp\u003eTumour-related and perioperative variables showed more pronounced differences. Non-survivors were more likely to have high-grade disease, advanced pathological T stage (T3\u0026ndash;4: 79.7% vs 19.5%), nodal involvement (N1\u0026ndash;2: 56.3% vs 15.7%) and distant metastasis (M1: 35.9% vs 2.5%; all \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Open radical cystectomy was performed more frequently in non-survivors, whereas minimally invasive approaches were more common among survivors (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.002). With respect to urinary diversion, orthotopic neobladder reconstruction was less frequent and cutaneous ureterostomy more frequent in the non-survivor group (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Non-survivors were also more likely to receive perioperative blood transfusion (45.3% vs 29.7%; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.027), experience postoperative complications (64.1% vs 38.1%; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001), and undergo adjuvant immunotherapy (18.8% vs 8.1%; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.025).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Model performance analysis\u003c/h2\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e and Figs.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e summarize the discrimination and classification performance of the candidate models. Although Random Forest achieved the highest AUC (0.950) in the validation set, CatBoost showed a better overall balance of discrimination, calibration, and clinical utility (accuracy 0.862, specificity 0.951, PPV 0.846) and was therefore selected as the final model for SHAP interpretation. Given the class imbalance in the data (21.3% mortality rate), we also reported precision\u0026ndash;recall curves and average precision (AP) as additional metrics for assessing model performance, as they are more informative under class-imbalanced conditions.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformance of machine learning models\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"13\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c12\" colnum=\"12\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c13\" colnum=\"13\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"6\" nameend=\"c7\" namest=\"c2\"\u003e \u003cp\u003eTraining set\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"6\" nameend=\"c13\" namest=\"c8\"\u003e \u003cp\u003eValidation set\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSensitivity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSpecificity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003ePPV\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNPV\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eAUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e \u003cp\u003eSensitivity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c11\"\u003e \u003cp\u003eSpecificity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c12\"\u003e \u003cp\u003ePPV\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c13\"\u003e \u003cp\u003eNPV\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.913\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.842\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.590\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.947\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.821\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.848\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.950\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.828\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.647\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.902\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.733\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.860\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSVM (RBF)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.910\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.887\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.718\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.957\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.875\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.891\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.943\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.828\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.529\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.951\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.818\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.830\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCatBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.891\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.827\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.513\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.957\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.833\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.826\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.931\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.862\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.647\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.951\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.846\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.867\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eXGBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.887\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.812\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.564\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.915\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.733\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.835\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.882\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.845\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.647\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.927\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.786\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.864\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLogistic Regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.885\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.872\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.692\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.947\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.844\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.881\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.872\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.879\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.706\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.951\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.857\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.886\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLightGBM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.841\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.827\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.590\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.926\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.767\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.845\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.859\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.776\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.412\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.927\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.700\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.792\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.894\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.835\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.564\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.947\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.815\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.840\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.834\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.862\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.588\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.976\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.909\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.851\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"13\"\u003eNotes: AUC, area under the receiver operating characteristic curve; PPV, positive predictive value; NPV, negative predictive value; RBF, radial basis function; KNN, k-nearest neighbors. Classification metrics were calculated using a probability threshold of 0.5 (or the optimal Youden index threshold) in the validation set.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Explanation of risk factors\u003c/h2\u003e \u003cp\u003eGlobal interpretability analyses are presented in Figs.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e. The radar plot (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e) compares normalized mean absolute SHAP values across CatBoost, random forest, and SVM-RBF models, showing consistent prioritization of tumor stage variables (T stage, N stage, and M stage).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe SHAP summary (beeswarm) plot for the final CatBoost model (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e) further illustrates both the magnitude and direction of each feature\u0026rsquo;s effect on the predicted probability of death within 3 years overall survival rate, highlighting the dominant contribution of tumor stage alongside host-related markers (age, BMI, albumin, globulin, lymphocyte count, platelet count, and preoperative creatinine).\u003c/p\u003e \u003cp\u003eLocal SHAP explanations for representative patients are shown in the waterfall plots (Figs.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e\u0026ndash;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e) and force plots (Figs.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e). In a low-risk patient (f(x)\u0026thinsp;=\u0026thinsp;0.298), node-negative and non-metastatic status reduced the predicted probability, whereas higher T stage and adverse host-related features increased risk (Figs.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e and \u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eConversely, in a high-risk patient (f(x)\u0026thinsp;=\u0026thinsp;0.728), metastatic disease, advanced T stage, and markedly elevated preoperative creatinine were the main contributors shifting the prediction toward mortality (Figs.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e and \u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e"},{"header":"4 Discussion","content":"\u003cp\u003eIn this single-center retrospective cohort of patients undergoing radical cystectomy for urothelial carcinoma, we developed and internally validated a machine-learning framework to predict 3-year overall survival using routinely available perioperative variables together with TNM stage. [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e].A two-stage feature-selection strategy (LASSO followed by RFECV) yielded a parsimonious 10-variable signature capturing both tumor burden (pathological T/N/M stage) and host-related status (albumin, lymphocyte count, BMI, creatinine, platelet count, and globulin). Among the evaluated algorithms, CatBoost achieved the most favorable balance of discrimination, calibration, and clinical utility in internal validation and was selected as the final model. To enhance transparency, we used SHAP-based explanations to attribute predicted risk at both cohort and individual levels, supporting clinical interpretability and informing future research on explainable AI-assisted risk stratification in healthcare [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e].Feature selection for prognostic modeling after radical cystectomy used cross-validated LASSO to minimize dimensionality and RFECV to refine predictors, balancing model complexity with clinical applicability while reducing overfitting and information leakage [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eUsing SHAP, We enhanced the interpretability of our machine learning model using SHAP, which provided clinically coherent and biologically plausible insights into 3-year overall survival following radical cystectomy.[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e].Across the SHAP summary plot and radar visualization, pathological T stage, N stage, and M stage were consistently the three most influential predictors, reinforcing the central oncologic principle that anatomical tumor burden underpins prognosis in bladder cancer [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e].Beyond tumor stage, the model highlighted host status\u0026mdash;markers of nutrition (lower BMI and albumin), immune competence (lower lymphocyte count), and renal function (higher creatinine)\u0026mdash;as independent contributors to mortality risk [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e].This pattern aligns with evidence linking systemic inflammation, malnutrition, and frailty-related physiology to adverse outcomes in urothelial carcinoma and major oncologic surgery [\u003cspan additionalcitationids=\"CR19\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e].Consistently, our baseline comparisons showed a lower BMI in non-survivors, supporting the clinical relevance of nutritional vulnerability in this setting [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e].Taken together, these findings suggest that perioperative optimization of modifiable host factors (e.g., nutritional support and renal optimization) may complement oncologic management and merits prospective evaluation [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e].At the individual level, SHAP waterfall and decision plots (Figs.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e\u0026ndash;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e) decomposed risk into patient-specific contributions, improving interpretability for clinicians and patients. For example, a patient with T3 disease but M0 status and preserved host physiology could receive a lower predicted risk because protective host features partially offset anatomical risk. Conversely, a patient with advanced disease (e.g., T4M1) and severe renal dysfunction would show additive risk contributions from both tumor burden and comorbidity, making the rationale for a high-risk classification explicit. Such transparent explanations are particularly relevant for shared decision-making and for increasing clinician trust in model-assisted risk stratification [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAll predictors used by our final CatBoost model are routinely collected in perioperative care, which may support future evaluation of implementation as a postoperative risk-stratification tool. Nevertheless, because this was a single-center retrospective study with internal validation only, external validation and potential recalibration are required across different case-mix, baseline risk, and treatment patterns. In addition, retrospective data collection may introduce selection bias and unmeasured confounding; follow-up ascertainment and missing data mechanisms should be carefully assessed in future prospective, multi-center studies.\u003c/p\u003e \u003cp\u003eOur calibration plots showed jagged curves in some probability ranges, particularly in internal validation, which is consistent with limited event counts, binning effects, and instability in score-to-probability mapping [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e].This issue is especially relevant for models whose native outputs are not probabilities, because additional calibration steps may be required to obtain reliable absolute risks [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e].Therefore, while the model performed strongly overall, raw predicted probabilities should be interpreted cautiously prior to clinical implementation. Before deployment, formal recalibration should be performed and reported using calibration intercept/slope and the Brier score, ideally within a nested resampling framework. External validation is then required to assess transportability under different baseline risks and treatment patterns and to guide model updating if calibration drift is observed. This study is limited by its single-center retrospective design, which may introduce selection bias and restrict generalizability [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e].We modeled a binary 3-year OS endpoint rather than fully leveraging time-to-event information, and future work should extend to survival modeling with time-dependent validation. The relatively limited number of death events likely contributed to mild calibration instability, emphasizing the need for larger cohorts and principled recalibration and model updating. In addition, we did not include granular treatment variables (e.g., chemotherapy regimen details or dose intensity), which could confound observed associations and should be incorporated in subsequent studies. Future research should prospectively validate this model, compare it head-to-head against established clinical nomograms, and test whether model-guided strategies improve patient-centered outcomes [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e].Finally, integrating emerging biomarkers such as circulating tumor DNA and radiomics may further enhance prognostic performance, but such extensions must be evaluated with rigorous external validation to avoid overfitting and inflated optimism.\u003c/p\u003e \u003cp\u003eIn conclusion, we developed and internally validated multiple machine-learning models to predict 3-year overall survival after radical cystectomy using routinely available clinicopathologic and laboratory variables. The final SHAP-interpretable CatBoost model demonstrated strong overall performance and provides transparent individualized explanations. The model may assist postoperative risk stratification, but it requires external validation and calibration assessment before any clinical use and should not replace clinical judgment.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eLASSO\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eleast absolute shrinkage and selection operator\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eRFECV\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003erecursive feature elimination with cross-validation\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCV\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ecross-validation\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMSE\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003emean squared error\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAUC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003earea under the receiver operating characteristic curve\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eλ\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eregularization parameter.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003eEthics approval and consent to participate\u003c/p\u003e\n\u003cp\u003eThis study was conducted in accordance with the Declaration of Helsinki. This retrospective study was granted an ethics waiver by the Ethics Committee of the First Affiliated Hospital of Xinjiang Medical University. The requirement for informed consent was waived due to the retrospective nature of the study and the use of de-identified data.\u003c/p\u003e\n\u003cp\u003eConsent for publication\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003eTrial registration\u003c/p\u003e\n\u003cp\u003eThis study did not involve a prospective clinical trial. Trial registration: Not applicable.\u003c/p\u003e\n\u003cp\u003eAvailability of data and materials\u003c/p\u003e\n\u003cp\u003eThe datasets generated and/or analyzed during the current study are available in the Figshare repository, with the following DOI: 10.6084/m9.figshare.31048870. The data is fully anonymized and includes all variables necessary to replicate the analyses presented in this study.\u003c/p\u003e\n\u003cp\u003eCompeting interests\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003eAuthor contributions statement\u003c/p\u003e\n\u003cp\u003eYW and AA contributed equally to this work and share first authorship. YW and AA conceived and designed the study, collected and curated the data, and drafted the manuscript. SC contributed to methodology, statistical analysis/model development, and critical revision of the manuscript. WW supervised the study, provided clinical interpretation, and critically revised the manuscript. All authors read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003eFunding\u003c/p\u003e\n\u003cp\u003eThis work was supported by the Tianshan Talent Training Program (Medical and Health High-level Talent Project) of the Health Commission of Xinjiang Uygur Autonomous Region (Grant No. TSYC202401B049).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eBray F, Laversanne M, Sung H, Ferlay J, Siegel RL, Soerjomataram I, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2024;74(3):229-63.\u003c/li\u003e\n \u003cli\u003eZhang Y, Rumgay H, Li M, Yu H, Pan H, Ni J. The global landscape of bladder cancer incidence and mortality in 2020 and projections to 2040. J Glob Health. 2023;13:04109.\u003c/li\u003e\n \u003cli\u003evan der Heijden AG, Bruins HM, Carrion A, Cathomas R, Comp\u0026eacute;rat E, Dimitropoulos K, et al. European Association of Urology guidelines on muscle-invasive and metastatic bladder cancer: summary of the 2025 guidelines. Eur Urol. 2025;87(5):582-600.\u003c/li\u003e\n \u003cli\u003eFlaig TW, Spiess PE, Abern M, Agarwal N, Bangs R, Buyyounouski MK, et al. NCCN Guidelines\u0026reg; Insights: bladder cancer, version 3.2024. J Natl Compr Canc Netw. 2024;22(4):216-25.\u003c/li\u003e\n \u003cli\u003eBizzarri FP, Scarciglia E, Russo P, Marino F, Presutti S, Moosavi SK, et al. Elderly and bladder cancer: the role of radical cystectomy and orthotopic urinary diversion. Urologia. 2024;91(3):500-4.\u003c/li\u003e\n \u003cli\u003eZapała Ł, Ślusarczyk A, Korczak B, Kurzyna P, Leki M, Lipiński P, et al. The view outside of the box: reporting outcomes following radical cystectomy using pentafecta from a multicenter retrospective analysis. Front Oncol. 2022;12:841852.\u003c/li\u003e\n \u003cli\u003eStein JP, Lieskovsky G, Cote R, Groshen S, Feng AC, Boyd S, et al. Radical cystectomy in the treatment of invasive bladder cancer: long-term results in 1,054 patients. J Clin Oncol. 2001;19(3):666-75.\u003c/li\u003e\n \u003cli\u003eHinsenveld FJ, Boormans JL, van der Poel HG, van der Schoot DKE, Vis AN, Aben KKH, et al.\u0026nbsp;Intermediate-term survival of robot-assisted versus open radical cystectomy for muscle-invasive and high-risk non-muscle invasive bladder cancer in The Netherlands. Urol Oncol. 2022;40(2):60.e1-60.e9.\u003c/li\u003e\n \u003cli\u003eShariat SF, Karakiewicz PI, Palapattu GS, Amiel GE, Lotan Y, Rogers CG, et al. Nomograms provide improved accuracy for predicting survival after radical cystectomy. Clin Cancer Res. 2006;12(22):6663-76.\u003c/li\u003e\n \u003cli\u003eZaak D, Burger M, Otto W, Bastian PJ, Denzinger S, Stief CG, et al. Predicting individual outcomes after radical cystectomy: an external validation of current nomograms. BJU Int. 2010;106(3):342-8.\u003c/li\u003e\n \u003cli\u003eBochner BH, Kattan MW, Vora KC.\u0026nbsp;Postoperative nomogram predicting risk of recurrence after radical cystectomy for bladder cancer. J Clin Oncol. 2006;24(24):3967-72.\u003c/li\u003e\n \u003cli\u003eKwon T, Jeong IG, You D, Hong B, Hong JH, Ahn H, et al. Long-term oncologic outcomes after radical cystectomy for bladder cancer at a single institution. J Korean Med Sci. 2014;29(5):669-75.\u003c/li\u003e\n \u003cli\u003eWitjes JA, Bruins HM, Cathomas R, Comp\u0026eacute;rat EM, Cowan NC, Gakis G, et al. European Association of Urology guidelines on muscle-invasive and metastatic bladder cancer: summary of the 2020 guidelines. Eur Urol. 2021;79(1):82-104.\u003c/li\u003e\n \u003cli\u003eLenis AT, Lec PM, Chamie K. Bladder cancer: a review. JAMA. 2020;324(19):1980-91.\u003c/li\u003e\n \u003cli\u003eTjoa E, Guan C. A survey on explainable artificial intelligence (XAI): toward medical XAI. IEEE Trans Neural Netw Learn Syst. 2021;32(11):4793-813.\u003c/li\u003e\n \u003cli\u003eMoons KGM, Wolff RF, Riley RD, Whiting PF, Westwood M, Collins GS, et al. PROBAST: a tool to assess risk of bias and applicability of prediction model studies: explanation and elaboration. Ann Intern Med. 2019;170(1):W1-W33.\u003c/li\u003e\n \u003cli\u003eWitjes JA, Lebret T, Comp\u0026eacute;rat EM, Cowan NC, De Santis M, Bruins HM, et al. Updated 2016 EAU guidelines on muscle-invasive and metastatic bladder cancer.\u0026nbsp;Eur Urol. 2017;71(3):462-75.\u003c/li\u003e\n \u003cli\u003eLi X, Ma X, Tang L, Wang B, Chen L, Zhang F, et al.\u0026nbsp;Prognostic value of neutrophil-to-lymphocyte ratio in urothelial carcinoma of the upper urinary tract and bladder: a systematic review and meta-analysis. Oncotarget. 2017;8(37):62681-92.\u003c/li\u003e\n \u003cli\u003eDobruch J, Oszczudłowski M. Bladder cancer: current challenges and future directions. Medicina (Kaunas). 2021;57(8):749.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eGillis C, Wischmeyer PE. Pre-operative nutrition and the elective surgical patient: why, how and what? Anaesthesia. 2019;74(Suppl 1):27-35.\u003c/li\u003e\n \u003cli\u003eLjungqvist O, Scott M, Fearon KC. Enhanced recovery after surgery: a review. JAMA Surg. 2017;152(3):292-8.\u003c/li\u003e\n \u003cli\u003eVemulapalli V, Qu J, Garren JM, Rodrigues LO, Kiebish MA, Sarangarajan R, et al.\u0026nbsp;Non-obvious correlations to disease management unraveled by Bayesian artificial intelligence analyses of CMS data. Artif Intell Med. 2016;74:1-8.\u003c/li\u003e\n \u003cli\u003eCollins GS, Reitsma JB, Altman DG, Moons KG. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ. 2015;350:g7594.\u003c/li\u003e\n \u003cli\u003eVan Calster B, McLernon DJ, van Smeden M, Wynants L, Steyerberg EW. Calibration: the Achilles heel of predictive analytics. BMC Med. 2019;17(1):230.\u003c/li\u003e\n \u003cli\u003eDong W, Jiang H, Li Y, Lv L, Gong Y, Li B, et al. Interpretable machine learning analysis of immunoinflammatory biomarkers for predicting CHD among NAFLD patients. Cardiovasc Diabetol. 2025;24(1):263.\u003c/li\u003e\n \u003cli\u003eWolff RF, Moons KGM, Riley RD, Whiting PF, Westwood M, Collins GS, et al. PROBAST: a tool to assess the risk of bias and applicability of prediction model studies. Ann Intern Med. 2019;170(1):51-8.\u003c/li\u003e\n \u003cli\u003eGrossman HB, Natale RB, Tangen CM, Speights VO, Vogelzang NJ, Trump DL, et al. Neoadjuvant chemotherapy plus cystectomy compared with cystectomy alone for locally advanced bladder cancer. N Engl J Med. 2003;349(9):859-66.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-cancer","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bcan","sideBox":"Learn more about [BMC Cancer](http://bmccancer.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bcan/default.aspx","title":"BMC Cancer","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Bladder cancer, Radical cystectomy, Machine Learning, SHAP, Predictive Modeling","lastPublishedDoi":"10.21203/rs.3.rs-8754027/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8754027/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eBackground: This study aimed to develop and internally validate an explainable machine-learning model using routinely available clinicopathologic and laboratory variables for predicting 3-year overall survival (OS) after radical cystectomy.\u003cbr\u003e\nMethods: We retrospectively included 300 patients who underwent radical cystectomy between January 2018 and December 2022. Predictors were selected in the training set using LASSO logistic regression followed by random-forest recursive feature elimination. Ten variables were retained. Seven algorithms (logistic regression, KNN, SVM-RBF, random forest, XGBoost, LightGBM, and CatBoost) were trained on a 70% training set and evaluated on a 30% internal validation set. Discrimination, calibration, and clinical utility were assessed, and the final model was interpreted using Shapley additive explanations (SHAP).\u003cbr\u003e\nResults: In internal validation, AUCs ranged from 0.834 to 0.950. CatBoost achieved the best overall classification performance (AUC = 0.931, accuracy = 0.862, sensitivity = 0.647, specificity = 0.951, PPV = 0.846, and NPV = 0.867). SHAP analyses identified tumor stage (T, N, and M stage) as the dominant drivers of predicted risk, with additional contributions from age, BMI, albumin, globulin, lymphocyte count, platelet count, and preoperative creatinine.\u003c/p\u003e\n\u003cp\u003eConclusions: We developed an internally validated, SHAP-interpretable CatBoost model for predicting 3-year overall survival (OS) after radical cystectomy. External validation and recalibration in independent cohorts are required before clinical use.\u003c/p\u003e","manuscriptTitle":"Development and Internal Validation of an Explainable Machine-Learning Model to Predict 3-Year overall survival rate After Radical Cystectomy","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-11 05:20:46","doi":"10.21203/rs.3.rs-8754027/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-03-30T09:52:38+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-02-27T05:31:30+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-02-08T09:20:46+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"54738593393625974610347246621621323136","date":"2026-02-08T09:18:04+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"87915770634444412798002529564521715046","date":"2026-02-08T03:44:23+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-02-05T15:09:34+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-02-05T14:49:51+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-02-05T12:48:52+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-02-05T04:56:54+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Cancer","date":"2026-02-05T04:47:16+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-cancer","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bcan","sideBox":"Learn more about [BMC Cancer](http://bmccancer.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bcan/default.aspx","title":"BMC Cancer","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"ebf78e45-7950-41dd-a0fc-b473347ae3c4","owner":[],"postedDate":"February 11th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-04-30T17:23:21+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-11 05:20:46","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8754027","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8754027","identity":"rs-8754027","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}