Long-term survival prognosis and risk stratification after radical cystectomy with Mainz Pouch II urinary diversion: a retrospective cohort study and prognostic model construction

Introduction

Background and objective

Bladder cancer is the tenth most common malignancy globally and imposes a significant burden, with a notable gender disparity, affecting males approximately four times more often than females. Epidemiological studies have established smoking as the primary risk factor, contributing to nearly 50% of cases (1). Clinically, approximately 75% of cases are classified as non-muscle invasive bladder cancer [NMIBC, including carcinoma in situ (CIS), Ta, and T1 stages] (2). Despite the prevalence of early-stage cases, the biological behavior of bladder cancer remains highly heterogeneous; over 50% of patients experience recurrence within two years post-surgery, and 10–30% progress to the more aggressive muscle invasive bladder cancer (MIBC, T2–T4 stages). This progression leads to a sharp decline in survival rates for advanced patients, with 5-year survival rates dropping to approximately 60–70% for T2, 40–50% for T3, and less than 30% for T4/N+ stages, respectively (2).

Radical cystectomy (RC) combined with pelvic lymph node dissection is the standard treatment for MIBC (3). To restore urinary function post-surgery, urinary diversion is required. Among the options, the Mainz Pouch II technique has become a significant surgical choice due to continence rates reaching 93–100% (4,5). Existing studies indicate that tumor recurrence and survival outcomes after RC are regulated by multiple factors, including traditional pathological indicators such as tumor stage, grade, lymph node metastasis (LNM), and urinary tract obstruction (2,3).

Although the Mainz Pouch II is a mature technique with good functional outcomes, patients face unique long-term risks. Survival is determined by both the tumor’s aggressiveness and the procedure-specific metabolic and infectious complications. Specific long-term oncological prognostic factors for this procedure have not yet been systematically established; in particular, there is a lack of data on how variables such as age, gender, tumor multiplicity, recurrence frequency, and degree of differentiation affect outcomes in this specific context.

General prognostic models, derived largely from ileal conduit cohorts, often fail to capture the specific physiological toll of the Mainz Pouch II diversion. Therefore, this study aims to conduct a long-term, single-center retrospective cohort analysis. By utilizing robust statistical methods, including Elastic Net, we aim to identify the main risk factors determining prognosis for this procedure from complex clinical features, helping to improve patient survival. We present this article in accordance with the TRIPOD reporting checklist (available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0442/rc).

Methods

Study design and data source

We conducted a retrospective single-center cohort study using medical records from the Department of Urology, Lanzhou University Second Hospital. The cohort included 309 patients diagnosed with bladder cancer who underwent “Radical Cystectomy + Mainz Pouch II Urinary Diversion” without receiving chemotherapy or radiotherapy between January 2004 and September 2023. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Medical Ethics Committee of Lanzhou University Second Hospital (Approval No. 2025A-1301). The requirement for informed consent was waived by the Medical Ethics Committee due to the retrospective nature of the study.

Patient cohort selection

Inclusion criteria: diagnosis of bladder cancer confirmed between January 2004 and September 2023. Histopathological confirmation via cystoscopic biopsy or transurethral resection prior to RC. Postoperative diagnosis confirmed by two urogenital pathologists based on surgical specimens. No history of preoperative chemotherapy or radiotherapy. Underwent RC and Mainz Pouch II reconstruction with concurrent standard pelvic lymph node dissection (PLND, anatomically defined as the routine clearance of bilateral obturator, external iliac, and internal iliac lymph nodes, with the cranial limit at the common iliac bifurcation). Complete medical and follow-up data. Adult patients.

Exclusion criteria: patients with incomplete data regarding the primary endpoint or key predictive factors. Survival time of less than one month. Patients who underwent non-standard surgical procedures. The detailed patient screening and selection process, including the specific reasons for exclusion and corresponding attrition counts, is illustrated in the study flowchart (Figure 1).

Figure 1 Flowchart of patient cohort selection. Between 2004 and 2023, a total of 844 bladder cancer patients underwent radical cystectomy at our institution. Following the strict application of treatment-related exclusion criteria (n=313; including non-Mainz Pouch II diversions and prior neoadjuvant therapy) and data/outcome-related exclusion criteria (n=222; including incomplete follow-up and perioperative mortality), a final cohort of 309 eligible patients was included in the Elastic Net prognostic modeling.

Feature selection and data collection

Baseline demographics (age, gender, smoking history) and prognosis-related clinicopathological parameters were retrospectively collected for each patient. The expanded variable pool included pathological T stage, tumor grade, histological differentiation, LNM status, lymph node density (LND), lymphovascular invasion (LVI), surgical margin status, tumor frequency, tumor multiplicity, and surgical eras (to adjust for temporal shifts in care protocols). Survival status and tumor recurrence were also analyzed. Tumor staging and grading were defined according to the American Joint Committee on Cancer (AJCC)and World Health Organization (WHO) standards. Tumor differentiation, depth of invasion, and LNM were confirmed pathologically. Chi-squared tests were used to analyze associations between categorical features. To assess data redundancy and multicollinearity, a correlation matrix was generated. As illustrated in Figure 2, while most clinicopathological features exhibit weak associations, significant multicollinearity was observed between inherently linked clinical variables, most notably between LNM status and LND (r=0.82), as well as moderate correlations between T stage and surgical margin status (r=0.42).

Figure 2 Correlation matrix heatmap of clinicopathological features. The matrix visualizes the pairwise Pearson/Spearman correlation coefficients across all baseline and clinicopathological variables to assess potential multicollinearity prior to multivariable modeling. Red gradients indicate positive correlations, while blue gradients indicate negative correlations. Notably, severe structural multicollinearity was identified between LNM status and lymph node density (r=0.82), alongside strong artifactual correlation between continuous age and categorized age group (r=0.77). LNM, lymph node metastasis.

Data preprocessing

First, a pattern analysis was conducted on missing data; variables with a missing rate exceeding 20% were excluded. For the remaining missing values, multiple imputation by chained equations (MICE) was used to generate 10 complete datasets. The algorithmic convergence of the imputation process and the distributional consistency between observed and imputed variables were rigorously confirmed (Figure S1A,S1B). To adhere strictly to standard MICE methodology and avoid selection bias, all 10 imputed datasets were retained for downstream analyses rather than selecting a single dataset. For feature selection, a majority-voting strategy was employed across all imputed datasets to robustly account for imputation uncertainty.

All continuous numerical features were Z-score standardized (mean =0, standard deviation =1). Categorical features were labeled and transformed using one-hot encoding to create a data format suitable for multivariate regression analysis, avoiding magnitude-based interference during model training. Given that the incidence of death in this cohort was approximately 39.5% (122/309), the data did not exhibit extreme class imbalance. To avoid calibration bias caused by artificially altering prior probabilities, this study did not employ oversampling techniques such as Synthetic Minority Over-sampling Technique (SMOTE) but modeled directly based on the natural distribution of the data to ensure the external validity of the model in real-world clinical scenarios.

Outcome measures

The primary outcome measure was overall survival (OS), defined as the time span from RC to death from any cause. Dates of death were derived from the initial death records in the Gansu Provincial Center for Disease Control and Prevention system and matched to included patients via social security numbers and birth dates. The follow-up cutoff date was November 1, 2025. All dates of death were ascertained by November 1, 2025.

Statistical analysis

The development and validation of the prognostic models in this study were conducted in strict accordance with the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) statement to ensure the transparency and robustness of the findings. To account for imputation uncertainty where standard Rubin’s rules are inapplicable, a majority-voting strategy was utilized for feature selection. Variables were evaluated across all 10 imputed datasets, retaining only those consistently identified in at least 50% of the imputations for final modeling. To maximize statistical power and prevent unstable estimates caused by random data splitting in small cohorts, the entire dataset of 309 patients was utilized for model development. Robust internal validation was then performed using 1,000 bootstrap resamples with replacement to calculate the optimism-corrected Concordance Index (C-index) for unbiased evaluation of future predictive accuracy. R software (version 4.5.1) was used for statistical analysis of baseline data. Categorical data were compared using the Chi-squared test, and continuous data were compared using the t-test. Survival curves were plotted using the Kaplan-Meier method, and differences in survival outcomes based on various factors were analyzed using the Log-rank test. Univariate OS following RC combined with Mainz Pouch II diversion was calculated using the Kaplan-Meier method and Log-rank test.

Prognostic models were constructed using traditional Cox proportional hazards (PH) regression, extreme gradient boosting (XGBoost), Decision Tree, and Elastic Net models. Parameter optimization for each model was performed on the training set using 10-fold cross-validation and grid search to maximize performance. All hypothesis tests were two-sided, with P<0.05 considered statistically significant.

Model validation and clinical efficacy assessment

We used multiple metrics to evaluate the efficacy of all constructed models. The discrimination of the models was measured by calculating time-dependent area under the curve (AUC) values at 1, 3, and 5 years. The consistency between predicted probabilities and actual observations was verified using calibration curves and Brier scores. Decision curve analysis (DCA) was utilized to assess the clinical net benefit of the models at different thresholds.

Regarding model interpretability, forest plots were drawn for the Cox model to display the hazard ratios (HRs) and 95% confidence intervals (CIs) of independent prognostic factors. For complex nonlinear models and the Elastic Net model, Variable Importance (VIMP) and SHapley Additive exPlanations (SHAP) methods were introduced to quantify feature contributions and explore nonlinear interactions. Furthermore, based on the risk scores from the optimal model, patients were stratified into low-, intermediate-, and high-risk groups. Inter-group survival differences were analyzed using Kaplan-Meier survival curves and Log-rank tests to validate the model’s risk stratification capability. Furthermore, a nomogram was constructed to visualize individual survival predictions. Finally, to rigorously address potential temporal bias introduced by the 20-year study interval, a sensitivity analysis stratified by distinct treatment epochs was conducted to evaluate the stability of the model’s discriminative performance across different historical eras.

Results

Characteristics of the study cohort

Among the 309 patients undergoing RC with Mainz Pouch II diversion, the mean age was 60.8±10.1 years, with 60% (185/309) aged <65 years, and 83.1% (257/309) being male. A positive smoking history was noted in 36.0% (111/309). Over a mean follow-up of 41.38±29.54 months, 39.5% (122/309) of patients died. For temporal evaluation, the cohort was stratified into three surgical eras: 2004–2011 (31.7%, 98/309), 2012–2017 (36.0%, 111/309), and 2018–2023 (32.4%, 100/309).

Pathological evaluation revealed T stages of T1 (16.2%, 50/309), T2 (51.1%, 158/309), T3 (20.7%, 64/309), and T4 (12%, 37/309); notably, T1 patients underwent RC due to large tumor burden, incomplete resection, or refractory hematuria. Most tumors were high-grade (91.9%, 284/309) and urothelial carcinomas (96.4%, 298/309). Of the specified cases, 41.7% (118/283) were recurrent and 61.0% (158/259) were multifocal. Following standard pelvic lymph node dissection, 18.8% (58/309) tested positive for LNM. Consequently, LND was >0 in 18.8% (58/309) of patients (8.4% ≤0.2 and 10.4% >0.2). Additionally, LVI and positive surgical margins were present in 33.7% (104/309) and 11.3% (35/309) of cases, respectively (Table 1).

Screening for prognostic factors

Kaplan-Meier survival analyses and Log-rank tests initially revealed that age, LNM status, pathological T stage, and several other key clinicopathological covariates were significantly associated with OS (Figures 3,4). Patients aged ≥65 years had significantly worse survival than younger counterparts (P<0.001). Advanced T stage and positive LNM status (P<0.001) were also strong predictors of poor survival. Furthermore, higher LND (P<0.001), positive surgical margins (P<0.001), the presence of LVI (P=0.002), and non-urothelial histological types (P=0.003) were identified as significant univariate risk factors (Figure 5). In contrast, tumor grade (P=0.12), smoking history (P=0.81), surgical eras (P=0.69), tumor multiplicity (P=0.78), tumor frequency (P=0.65), and gender (P=0.37) did not exhibit statistically significant impacts on OS in the initial univariate setting (Figures 3-5).

Figure 3 Kaplan-Meier curves of OS stratified by primary tumor pathological features. (A) OS by pathological T stage (T1 to T4, P<0.001); (B) OS by tumor grade (low vs. high, P=0.12); (C) OS by LNM status (negative vs. positive, P<0.001); (D) OS by tumor count (single vs. multiple, P=0.78). LNM, lymph node metastasis; OS, overall survival; T, tumor.

Figure 4 Kaplan-Meier curves of OS stratified by demographic and baseline clinical characteristics. (A) OS by tumor frequency (primary vs. recurrent, P=0.65); (B) OS by histological type (UC vs. non-UC, P=0.003); (C) OS by gender (female vs. male, P=0.37); (D) OS by age group (<65 vs. ≥65 years, P<0.001). OS, overall survival; UC, urothelial carcinoma.

Figure 5 Kaplan-Meier curves of OS stratified by secondary clinicopathological and chronological features. (A) OS by smoking history (P=0.81); (B) OS by lymphovascular invasion (absent vs. present, P=0.002); (C) OS by surgical margin status (negative vs. positive, P<0.001); (D) OS by surgical eras (2004–2011, 2012–2017, and 2018–2023, P=0.69); (E) OS by LND group (0 vs. ≤0.2 vs. >0.2, P<0.001). Notably, panel D confirms that surgical epochs did not introduce severe chronological bias into the survival trajectories. LND, lymph node density; OS, overall survival.

Based on the univariate analysis results and clinical importance, and to further quantify the independent prognostic value of each feature while exploring potential nonlinear interactions, we included all candidate variables in the initial models. All candidate clinicopathological features were simultaneously introduced into traditional multivariate Cox regression, regularized Elastic Net regression, and complex nonlinear models [XGBoost, random survival forest (RSF)] for training and screening. This approach prevented bias caused by premature variable exclusion and utilized the intrinsic feature selection mechanisms of each algorithm to address potential multicollinearity and identify core prognostic factors.

Model construction and testing

Model configuration

To ensure a fair comparison between models, all selected multivariate prediction models utilized identical configurations. To maximize statistical power and avoid unstable estimates inherent to data splitting in a small cohort, the entire dataset (N=309) was utilized for model development and evaluation. To ensure reproducibility, random seeds were fixed across all stochastic processes, including multiple imputation, hyperparameter tuning, cross-validation, and bootstrapping. Model selection was performed via 10-fold cross-validation across the imputed datasets. Internal validation was conducted using 1,000 bootstrap resamples on the full cohort to quantify overfitting bias and calculate the optimism-corrected C-index. Finally, model performance was evaluated using 1-, 3-, and 5-year time-dependent AUC, 5-year Brier scores, 5-year calibration curves, and DCA.

Model results

Given the limited sample size (N=309), the entire cohort was utilized for model development to avoid the unstable estimates associated with data splitting. Internal validation was subsequently performed using 10-fold cross-validation and 1,000 bootstrap resamples. Models were constructed and evaluated using Cox regression, XGBoost [accelerated failure time (AFT)/PH], Elastic Net, and RSF. Overfitted models were excluded, and the best-performing model was selected. To ensure a strictly fair and transparent benchmarking process, all evaluated machine learning models underwent rigorous hyperparameter optimization. The Elastic Net model determined optimal mixing parameters via grid search, dynamically solving the regression coefficient path to balance L1 and L2 regularization. Similarly, the XGBoost models optimized learning rates based on negative log-likelihood loss to prevent premature convergence, while the RSF model selected the optimal mtry and node size to maximize tree diversity and prevent overfitting. The comprehensive hyperparameter tuning profiles, learning curves, and out-of-bag (OOB) error heatmaps for all evaluated algorithms are detailed in Figure S2. Finally, the Cox multivariate regression model utilized bootstrap resampling to calculate the optimism-corrected C-index, adhering to the events per variable (EPV) principle to monitor for overfitting.

Based on the multivariable survival analysis (Table 2), advanced age (HR 1.03, 95% CI: 1.01–1.06, P=0.001, elevated LND (HR 8.13, 95% CI: 1.21–54.8, P=0.03), and advanced pathological T stage (T3: HR 2.75, 95% CI: 1.26–5.98, P=0.01; T4: HR 3.94, 95% CI: 1.72–8.99, P=0.001) were identified as significant independent risk factors for poor prognosis. Notably, while positive lymph node metastasis was a strong predictor in the univariate setting, it exhibited only marginal significance in the multivariable model (HR 1.80, 95% CI: 0.94–3.46, P=0.08). Furthermore, gender, tumor grade, histological type, tumor frequency, tumor count, smoking history, LVI, surgical margin status, and surgical eras did not demonstrate independent statistical significance (all P>0.05). In terms of overall discriminative and clinical performance across distinct time horizons (1-, 3-, and 5-year) (Table 3), the Elastic Net model exhibited a distinct advantage (Apparent AUC: 0.790) compared to the standard Cox regression (Apparent AUC: 0.756) (Figure 6), alongside excellent calibration and superior clinical net benefit (Figures 7,8). More importantly, robust 10-fold cross-validation (Figure S3) and 1,000x bootstrap internal validation (Table 4) confirmed its superior generalization capability by minimizing Brier scores and maintaining stable time-dependent AUCs (Table 3), yielding the highest optimism-corrected C-index (0.712) among all evaluated algorithms. Considering predictive performance, model robustness, and clinical interpretability, the Elastic Net model was the optimal choice for this cohort. We therefore selected the Elastic Net model to construct the prognostic prediction tool and an easily applicable nomogram for patients after RC combined with Mainz Pouch II urinary diversion (Figure 9).

Figure 6 Full dataset ROC curves evaluating the apparent discriminative performance of the prognostic models. The plot highlights the distinct discriminative advantage of the regularized Elastic Net model (AUC: 0.790) compared to the standard parsimonious Cox regression (AUC: 0.756). While complex non-linear algorithms (XGBoost, RSF) exhibit marginally higher apparent AUCs, they are subsequently associated with pronounced overfitting bias during internal bootstrap validation. AFT, accelerated failure time; AUC, area under the curve; PH, proportional hazards; ROC, receiver operating characteristic; RSF, random survival forest; XGBoost, extreme gradient boosting.

Figure 7 Five-year calibration plots for the evaluated prognostic models. Patients were stratified into five risk quintiles based on their predicted probabilities. The close alignment between the Elastic Net model’s predicted 5-year overall survival risks and the actual observed Kaplan-Meier survival estimates demonstrates its robust calibration and predictive accuracy. AFT, accelerated failure time; PH, proportional hazards.

Figure 8 Five-year DCA of the prognostic models. The curves illustrate the clinical net benefit across a broad range of threshold probabilities. The Elastic Net model consistently provides superior net benefit compared to alternative algorithms, as well as the default “treat-all” or “treat-none” strategies, confirming its practical utility for clinical decision-making. AFT, accelerated failure time; DCA, decision curve analysis; PH, proportional hazards; RSF, random survival forest.

Figure 9 Prognostic nomogram based on the regularized Elastic Net model. This clinical translation tool allows for the individualized prediction of 3- and 5-year OS probabilities for patients undergoing radical cystectomy with Mainz Pouch II diversion. By drawing a vertical line from each patient-specific clinical feature to the “Points” axis, summing the points, and locating the “Total Points” on the bottom axes, clinicians can easily estimate personalized survival risks. LNM, lymph node metastasis; OS, overall survival; T, tumor; UC, urothelial carcinoma.

Model interpretation

Based on the collected feature factors and the key variables identified, the sample was stratified into low-risk, intermediate-risk, and high-risk subgroups according to independent risk factors, and survival analysis for OS was performed respectively. Kaplan-Meier survival curves based on the Elastic Net model’s risk stratification showed extremely significant statistical differences in prognosis among the low-, intermediate-, and high-risk cohorts (P<0.001). The high-risk group exhibited a precipitously pronounced decline in survival probability over time, indicating statistically distinct survival trajectories among the subgroups (Figure 10). However, given the model’s moderate discrimination, this stratification currently serves as an exploratory reference. The Elastic Net model utilized a coefficient distribution plot to display the magnitude and relative ranking of the top 20 selected variables (Figure 11). Notably, this regularized approach explicitly captured the complex biological interactions between baseline features (e.g., T stage interacting with LND), highlighting their precise directional contributions to the predictive risk score. The nomogram transformed the complex regression equation into a simple graphical scoring table. Calibration curves and DCA visually verified the consistency between the model’s predicted probabilities and actual observed events, confirming that utilizing the regularized Elastic Net model to assist clinical decision-making yields a substantial net benefit across a broad range of threshold probabilities (Figures 7,8). To validate the model’s clinical interpretability and demonstrate its utility in personalized medicine, individualized SHAP breakdowns were generated. This local interpretability approach effectively translates the complex global feature interactions into patient-specific risk profiles (Figure S4). Furthermore, Schoenfeld residual plots were utilized to test the PH assumption by assessing whether the partial regression coefficients changed systematically with follow-up time (Figure S5). To rigorously address the potential temporal bias highlighted by the extended two-decade study interval, we performed a stratified sensitivity analysis across three distinct treatment epochs (2004–2011, 2012–2017, and 2018–2023). As illustrated in Figure 12 and detailed in Table 5, the Elastic Net model maintained consistent and robust discriminative performance (Apparent C-index) across all historical sub-cohorts. Specifically, it consistently outperformed the standard Cox model across all epochs and performed comparably to the RSF algorithm, notably demonstrating superior adaptability in the most recent clinical era (2018–2023). This temporal stability objectively confirms that chronological paradigm shifts in perioperative care, surgical protocols, and adjunctive therapies over the 20-year span did not significantly confound the core oncological predictions of our established model.

Figure 10 Kaplan-Meier overall survival curves stratified by the Elastic Net-derived risk groups. Patients were optimally categorized into low-risk, medium-risk, and high-risk cohorts based on predefined 5-year death risk thresholds (low <0.309≤ medium <0.464≤ high). The highly significant statistical divergence among the strata (P<0.001) confirms the robust discriminative utility of the model for clinical risk stratification.

Figure 11 Coefficient distribution plot of the top 20 feature factors selected by the Elastic Net model. The horizontal bar chart displays the magnitude and direction of the regularized coefficients, including key interaction terms (denoted by a colon). Positive coefficients (extending to the right) indicate an increased risk of mortality, whereas negative coefficients represent a protective effect on overall survival. T, tumor.

Figure 12 Stratified sensitivity analysis assessing model stability across distinct treatment epochs. The bar chart compares the apparent discriminative performance (C-index) of standard Cox, Elastic Net, and RSF models across three historical surgical eras (2004–2011, 2012–2017, and 2018–2023). The Elastic Net model demonstrates consistently robust adaptability and outperformance over the standard Cox model across all temporal subdivisions, confirming that chronological paradigm shifts did not severely confound the core oncological predictions. RSF, random survival forest.

Discussion

RC with pelvic lymph node dissection remains the standard treatment for muscle-invasive bladder cancer, as recommended by major international guidelines (6,7). Following bladder removal, urinary diversion is essential to reconstruct the lower urinary tract and maintain physiological voiding or urine storage function (8). The Mainz Pouch II procedure, classified as a continent urinary diversion, involves implantation of bilateral ureters into the sigmoid colon and utilizes the anal sphincter for continence control. This technique eliminates the need for abdominal stoma and external collection bags, thereby preserving the integrity of the patient’s body appearance, alleviating the physical burden associated with stoma care, and significantly promoting postoperative psychological rehabilitation and social reintegration (9,10). Compared with the more complex construction of orthotopic neobladders, Mainz Pouch II offers advantages such as a simplified surgical procedure, shorter operative time, and reduced intraoperative blood loss, making it a preferred option for high-risk patients, including the elderly or those with multiple comorbidities (10-12). Long-term follow-up evidence further indicates that this procedure reduces the risk of upper urinary tract complications through its antireflux mechanism while maintaining stable renal function, achieving a dual balance of survival benefits and quality of life for patients (10,12).

The Mainz Pouch II procedure involves anastomosis between the ureters and sigmoid colon, inherently exposing patients to unique long-term pathophysiological challenges. Specifically, continuous exposure of the colonic mucosa to urine triggers abnormal Cl⁻/HCO₃⁻ exchange, leading to hyperchloremic metabolic acidosis. If uncorrected, this exacerbates bone demineralization and renal impairment over time (13-15). In elderly cohorts, these metabolic tolls can synergize with overall frailty to indirectly shorten survival (16). Furthermore, the reservoir environment facilitates ascending pyelonephritis (10,17) and the bacterial reduction of urinary nitrates into carcinogenic N-nitrosamines, predisposing patients to secondary anastomotic adenocarcinomas after a long latency period (18). These diversion-specific morbidities significantly impact physical performance and long-term survival (12), rendering general prognostic models developed for ileal conduits potentially biased in this setting (19). Because these unique non-oncological morbidities represent substantial mortality risks, a competing-risks analysis isolating cancer-specific survival (CSS) from diversion-related deaths is theoretically mandatory to validate a purely oncological model. However, as our long-term mortality tracking relied on regional demographic registries lacking medically adjudicated causes of death, we were strictly limited to evaluating all-cause OS. Consequently, these severe physiological complications must be explicitly acknowledged as unmeasured competing risks embedded within the OS endpoint. Therefore, rather than isolating surgical complications, our Elastic Net nomogram should be interpreted as an evaluation of baseline clinico-oncological vulnerability interacting with all-cause mortality in this specific demographic.

This study aimed to develop a tailored oncological prognostic model to identify key clinicopathological survival determinants in patients undergoing RC with Mainz Pouch II diversion. The Elastic Net model combines the robustness of regularized regression with the interpretability of traditional statistics. It was ultimately selected to construct the risk stratification tool not only because its L1 and L2 penalty mechanisms mathematically resolved the severe structural multicollinearity (most notably between LNM status and LND, r=0.82, which inherently destabilizes standard Cox regression via variance inflation), but also due to its objectively verified predictive superiority. As demonstrated by the full-dataset ROC analysis (AUC 0.790 vs. Cox 0.756) and 10-fold cross-validation, Elastic Net consistently outperformed standard conventional methodologies. Following rigorous internal validation, it achieved the highest optimism-corrected C-index (0.712), effectively balancing discriminative power with optimal generalization capability. This performance aligns consistently with established international studies on similar bladder cancer cohorts (20,21), reflecting the high biological heterogeneity of the disease while avoiding the pronounced overfitting traps typical of complex nonlinear algorithms.

It has been reported that tumor stage and LNM play important roles in the prognosis of bladder cancer patients (5). In this study, as pathological tumor stage increased from T1 to T4, the 5-year OS gradually declined, and advanced T stage emerged as a highly significant independent predictor in the multivariable analysis. Notably, while positive LNM was a strong predictor univariately, it exhibited only marginal significance in the multivariable model (P=0.08). This phenomenon is primarily due to significant biological and statistical collinearity: advanced T stage intrinsically drives deep invasion and lymphatic spread, leading T stage and continuous LND to mathematically overpower the binary LNM status after regularized adjustment (22). The sample size of this study (n=309) and the complex collinearity among these variables limited the independent statistical power of simple binary LNM. However, this marginal loss of statistical significance does not negate the inherent biological importance of LNM, which, alongside T stage, remains a core stratification variable in contemporary international guidelines and multicenter studies (2,3). A retrospective study reported LNM rates of 24–28% in patients undergoing cystectomy (23). In this study, the LNM positive rate was 18.8%, with a significant negative correlation with OS in the univariate setting (P <0.001); the discrepancy from prior reports may be attributed to selection bias for the Mainz Pouch II procedure, which is preferentially recommended for patients with lighter tumor burden and better performance status, while high-risk cases are often diverted to ileal conduit or other diversions. Although associated with poorer survival trends, tumor grade lost independent prognostic significance in the multivariable model (P>0.05). This is primarily attributable to the overwhelming proportion of high-grade cases (92%) in our cohort (24), which inherently diminishes its statistical discriminative power—a phenomenon consistently observed in large-scale RC cohorts (25,26). Studies by Mitra and Kucuk et al. showed that OS rates in patients aged <65 years were significantly higher than in those ≥65 years (27,28). Consistent with these findings, the univariate Kaplan-Meier analysis in our cohort demonstrated corresponding 5-year OS rates of 64.3% and 45.2%, respectively (P<0.001). Furthermore, to prevent artificial collinearity from categorical thresholding, our regularized multivariable model evaluated age as a continuous variable, strictly confirming advancing age as a significant independent prognostic factor for OS (HR 1.03, P=0.001). Crucially, our multivariable model identified elevated LND as a highly significant independent risk factor (HR 8.13, P=0.03). Unlike binary LNM, LND incorporates both the biological burden of nodal metastasis and surgical adequacy. A higher LND reflects a disproportionate ratio of metastatic to resected nodes, indicating an aggressive tumor phenotype and higher risk of occult micrometastases. For instance, Heck et al. and Stein et al. robustly demonstrated that LND is a superior prognosticator compared to traditional pN staging, with higher LND strongly associated with decreased cancer-specific and OS following RC (29,30). By retaining LND, our Elastic Net model compensated for the marginal statistical power of binary LNM, validating that the quantitative burden of lymphatic spread dictates long-term prognosis. The collective prognostic relevance of these features highlights the critical role of baseline tumor biology in shaping patient outcomes. However, because this retrospective cohort lacked granular data on procedure-specific metabolic or functional metrics, we cannot definitively weigh the prognostic burden of intrinsic tumor biology against the long-term physiological toll of the Mainz Pouch II diversion. Consequently, our nomogram should be interpreted primarily as an objective oncological risk stratification tool tailored to this specific demographic, rather than a comprehensive model encompassing all surgical complications.

Regarding the impact of gender, tumor differentiation, tumor frequency, and tumor multiplicity on the survival prognosis of bladder cancer patients undergoing RC combined with Mainz Pouch II diversion, although the constructed model included these for consideration, the results of this study showed their influence to be minor. Several studies suggest that female bladder cancer patients have a worse prognosis compared to males (31); however, compared to previous studies, this discrepancy may be attributed to the smaller number of female patients in this study. Some reports indicate that histological differentiation impacts oncological outcomes in bladder cancer patients after RC (32,33). Marks et al. (32) reported on 517 bladder urothelial carcinoma (BUC) patients treated with RC, where 18.6% (96/517) had variant histology, with squamous and sarcomatoid differentiation being most common. Compared to pure BUC, histological variants of BUC were associated with advanced tumor stage, LNM, and LVI. Kim et al. (33) reported that patients with BUC and squamous and/or glandular differentiation were more likely to present with advanced pT3–T4 tumors and pN+ disease than patients with pure BUC. Previous studies have shown that tumor multiplicity is an important adverse prognostic factor after RC for bladder cancer; multifocality is associated not only with shorter postoperative recurrence-free survival but also with lower cancer-specific survival (7,34). The impact of tumor frequency on oncological outcomes after RC remains a subject of long-standing debate in the academic community. In the latest published meta-analysis indicate that patients with secondary MIBC (sMIBC) have significantly worse survival prognosis after RC than those with primary MIBC (pMIBC) (35). Similarly, our multivariable analysis revealed that smoking history, LVI, surgical margin status, and distinct surgical eras did not demonstrate independent statistical significance. While LVI and positive surgical margins are widely validated as robust adverse prognosticators in general RC populations, their diminished independent impact within our cohort warrants nuanced interpretation (26,36). In the context of a regularized Elastic Net algorithm, the prognostic variance of LVI—which serves as a critical biological precursor to locoregional dissemination—is mathematically penalized and subsumed by more dominant, downstream indicators of tumor burden, notably clinically significant T stage and elevated LND (26). Consequently, baseline oncological vulnerability is overwhelmingly modulated by these primary drivers, rendering the intermediate physiological step of LVI statistically redundant in the final multivariable model. Likewise, the relatively low prevalence of positive margins (11%) reflects the stringent patient selection criteria inherently required for continent Mainz Pouch II diversion, which tends to favor cases with a higher probability of complete local resection and lower extravesical burden (37,38). Furthermore, smoking history, despite its established role in urothelial carcinogenesis, did not independently stratify long-term OS. This reaffirms that once disease progression necessitates extirpative surgery, the pathological burden and intrinsic tumor biology eclipse the initial carcinogenic trigger in determining survival (39,40). Finally, the lack of independent prognostic significance across different surgical eras provides crucial temporal validation. This stability objectively confirms that the chronological evolution of perioperative care, surgical protocols, and adjunctive therapies over the two-decade span did not severely confound the core survival trajectories, solidifying the conclusion that fundamental tumor biology remains the paramount determinant of patient outcomes regardless of the treatment epoch. The inconsistency in study results may be partly attributed to differences in study population characteristics, sample size, and methodology. These results require cautious interpretation and further verification.

Rather than offering immediate actionable clinical pathways, our Elastic Net-based model provides an exploratory framework for understanding postoperative risk profiles. Based on the Kaplan-Meier stratification, we hypothesize that a risk-adapted follow-up strategy could potentially benefit these distinct demographic subgroups. Patients classified as low-risk demonstrated stable long-term survival in our cohort. For this demographic, clinical management might potentially prioritize maintaining Mainz Pouch II-specific metabolic homeostasis. Routine evaluation of serum chloride and bicarbonate levels becomes critical to mitigate chronic bone mineral loss and secondary renal deterioration, a dynamic well-documented in continent urinary diversion cohorts (41). High-risk individuals, characterized predominantly by positive LNM or T4 stage, exhibit a starkly different trajectory. Clinical recurrence mapping indicates a sharp decline in survival probability during the initial 24 to 36 months post-surgery (42). This vulnerability suggests the potential utility of an intensified surveillance protocol. Implementing quarterly cross-sectional imaging like CT or MRI over the first two years aligns with contemporary oncological guidelines (43) and facilitates the early evaluation for adjuvant systemic therapies (44). Early identification of these high-risk phenotypes empowers clinicians to initiate aggressive interventions before clinical relapse becomes overt, ultimately offering a pathway to narrow the survival gap observed in this study.

A key methodological consideration in this study was the exclusion of patients receiving perioperative systemic therapy or radiotherapy. While neoadjuvant chemotherapy (NAC) represents the contemporary standard of care for muscle-invasive bladder cancer, its utilization and specific regimens varied substantially across our 20-year study period. This exclusion criterion was implemented to mitigate significant confounding variables—such as regimen-specific toxicities and heterogeneous tumor downstaging responses—thereby purely isolating the intrinsic prognostic value of baseline clinicopathological characteristics. We acknowledge that this approach introduces selection bias, which may constrain the external validity of the model in modern clinical paradigms where NAC is routinely administered. Therefore, the clinical utility of this nomogram is primarily applicable to the significant subset of patients who are clinically ineligible for cisplatin-based therapy or who decline neoadjuvant interventions.

Limitations of this study include: (I) reliance on data from a single institution, which may limit generalizability; future work should focus on using multicenter international datasets. (II) The study is retrospective rather than prospective. (III) Unverified causes of death in demographic registries restricted our endpoint to all-cause OS, precluding competing-risks analyses for cancer-specific survival. (IV) RC procedures were performed by different surgeons, and specimens were examined by multiple pathologists, which may affect the stability of results. (V) Most included patients were aged <75 years; results should be interpreted with caution for patients aged ≥75 years. (VI) Given its moderate discrimination (corrected C-index: 0.712), strict external validation in independent, multicenter cohorts is required before confirming its clinical utility. (VII) The exclusion of patients receiving perioperative systemic therapy restricts the model’s generalizability in modern cohorts, making it primarily applicable to NAC-ineligible or untreated populations. (VIII) Due to the extended 20-year retrospective timeframe, specific historical data such as the Charlson Comorbidity Index, detailed baseline renal function trajectories, and preoperative hydronephrosis lacked standardized documentation and suffered from substantial missing rates, preventing their inclusion in the regularized modeling.

Conclusions

Advanced age, elevated LND, and advanced pathological T stage were identified as the primary independent prognostic determinants for patients undergoing Mainz Pouch II diversion. By effectively resolving structural multicollinearity, the Elastic Net-based nomogram outperformed both standard Cox and complex nonlinear models, demonstrating robust temporal stability across a two-decade span. This exploratory tool demonstrates the potential for personalized risk stratification. However, due to its moderate discrimination (C-index: 0.712) and single-center retrospective nature, it cannot yet guide routine practice and strictly requires independent external validation prior to clinical implementation.

Acknowledgments

We would like to thank the staff of the Department of Urology, Lanzhou University Second Hospital, for their support on data collection and management.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0442/rc

Data Sharing Statement: Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0442/dss

Peer Review File: Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0442/prf

Funding: This study was supported by the Gansu Province Healthcare Industry Research Project (No. GSWSQN2024-11) and the Gansu Provincial Department of Education: Youth Doctoral Fund Project (No. 2022QB-012).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0442/coif). All authors report funding from the Gansu Province Healthcare Industry Research Project (No. GSWSQN2024-11) and the Gansu Provincial Department of Education: Youth Doctoral Fund Project (No. 2022QB-012). The authors have no other conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Medical Ethics Committee of Lanzhou University Second Hospital (Approval No.2025A-1301). The requirement for informed consent was waived by the Medical Ethics Committee due to the retrospective nature of the study.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Lobo N, Afferi L, Moschini M, et al. Epidemiology, Screening, and Prevention of Bladder Cancer. Eur Urol Oncol 2022;5:628-39. [Crossref] [PubMed]
2. Compérat E, Amin MB, Cathomas R, et al. Current best practice for bladder cancer: a narrative review of diagnostics and treatments. Lancet 2022;400:1712-21. [Crossref] [PubMed]
3. Hensley PJ, Choudhury A, Khaki AR, et al. The 2023 European Association of Urology Guidelines on Muscle-invasive and Metastatic Bladder Cancer: A Critical Appraisal. Eur Urol 2024;85:32-4. [Crossref] [PubMed]
4. Fisch M, Hohenfellner R. Sigma-rectum pouch (Mainz pouch II). BJU Int 2007;99:945-60. [Crossref] [PubMed]
5. Zhang B, He Y, Zheng D, et al. Efficacy of radical cystectomy plus modified ureterosigmoidostomy (Mainz II) in bladder cancer. Asian J Surg 2023;46:2445-6. [Crossref] [PubMed]
6. Chang SS, Bochner BH, Chou R, et al. Treatment of Non-Metastatic Muscle-Invasive Bladder Cancer: AUA/ASCO/ASTRO/SUO Guideline. J Urol 2017;198:552-9. [Crossref] [PubMed]
7. Alfred Witjes J, Max Bruins H, Carrión A, et al. European Association of Urology Guidelines on Muscle-invasive and Metastatic Bladder Cancer: Summary of the 2023 Guidelines. Eur Urol 2024;85:17-31. [Crossref] [PubMed]
8. Lee RK, Abol-Enein H, Artibani W, et al. Urinary diversion after radical cystectomy for bladder cancer: options, patient selection, and outcomes. BJU Int 2014;113:11-23. [Crossref] [PubMed]
9. Bastian PJ, Albers P, Hanitzsch H, et al. Health-related quality-of-life following modified ureterosigmoidostomy (Mainz Pouch II) as continent urinary diversion. Eur Urol 2004;46:591-7. [Crossref] [PubMed]
10. D’elia G, Pahernik S, Fisch M, et al. Mainz Pouch II technique: 10 years’ experience. BJU Int 2004;93:1037-42.
11. Jing S, Yang E, Luo Z, et al. Perioperative outcomes and continence following robotic-assisted radical cystectomy with mainz pouch II urinary diversion in patients with bladder cancer. BMC Cancer 2024;24:127. [Crossref] [PubMed]
12. Hadzi-Djokic JB, Basic DT. A modified sigma-rectum pouch (Mainz pouch II) technique: analysis of outcomes and complications on 220 patients. BJU Int 2006;97:587-91. [Crossref] [PubMed]
13. Korus J, Szymczak M, Gołębiowski M, et al. Metabolic Acidosis in Patients with Chronic Kidney Disease: Diagnosis, Pathogenesis, and Treatment-A Narrative Review. Diagnostics (Basel) 2025;15:2052. [Crossref] [PubMed]
14. Domínguez García A, Muñoz Rodríguez J, Prats López J, et al. Why do patients with urinary diversions have an increased risk of bone fracture? A systematic review on risk factors for osteoporosis and bone mineral density loss in this group of patients. Actas Urol Esp (Engl Ed) 2024;48:497-511.
15. Smith JA Jr. Modified ureterosigmoidostomy (Mainz II): a long-term follow-up. J Urol 2005;173:885-6. [Crossref] [PubMed]
16. Raphael KL. Metabolic Acidosis in CKD: Core Curriculum 2019. Am J Kidney Dis 2019;74:263-75. [Crossref] [PubMed]
17. Wesson DE, Buysse JM, Bushinsky DA. Mechanisms of Metabolic Acidosis-Induced Kidney Injury in Chronic Kidney Disease. J Am Soc Nephrol 2020;31:469-82. [Crossref] [PubMed]
18. Kälble T, Tricker AR, Möhring K, et al. The role of nitrate, nitrite and N-nitrosamines in carcinogenesis of colon tumours following ureterosigmoidostomy. Urol Res 1990;18:123-9. [Crossref] [PubMed]
19. Crozier J, Hennessey D, Sengupta S, et al. A Systematic Review of Ileal Conduit and Neobladder Outcomes in Primary Bladder Cancer. Urology 2016;96:74-9. [Crossref] [PubMed]
20. Sarrió-Sanz P, Martinez-Cayuelas L, Lumbreras B, et al. Mortality prediction models after radical cystectomy for bladder tumour: A systematic review and critical appraisal. Eur J Clin Invest 2022;52:e13822. [Crossref] [PubMed]
21. Meeks JJ, Al-Ahmadie H, Faltas BM, et al. Genomic heterogeneity in bladder cancer: challenges and possible solutions to improve outcomes. Nat Rev Urol 2020;17:259-70. [Crossref] [PubMed]
22. Stein JP, Lieskovsky G, Cote R, et al. Radical cystectomy in the treatment of invasive bladder cancer: long-term results in 1,054 patients. J Clin Oncol 2001;19:666-75. [Crossref] [PubMed]
23. Stein JP, Quek ML, Skinner DG. Lymphadenectomy for invasive bladder cancer: I. historical perspective and contemporary rationale. BJU Int 2006;97:227-31.
24. Humphrey PA, Moch H, Cubilla AL, et al. The 2016 WHO Classification of Tumours of the Urinary System and Male Genital Organs-Part B: Prostate and Bladder Tumours. Eur Urol 2016;70:106-19. [Crossref] [PubMed]
25. International Bladder Cancer Nomogram Consortium. Postoperative nomogram predicting risk of recurrence after radical cystectomy for bladder cancer. J Clin Oncol 2006;24:3967-72.
26. Kluth LA, Black PC, Bochner BH, et al. Prognostic and Prediction Tools in Bladder Cancer: A Comprehensive Review of the Literature. Eur Urol 2015;68:238-53. [Crossref] [PubMed]
27. Mitra AP, Bartsch CC, Bartsch G Jr, et al. Does presence of squamous and glandular differentiation in urothelial carcinoma of the bladder at cystectomy portend poor prognosis? An intensive case-control analysis. Urol Oncol 2014;32:117-27.
28. Kucuk U, Pala EE, Cakır E, et al. Clinical, demographic and histopathological prognostic factors for urothelial carcinoma of the bladder. Cent European J Urol 2015;68:30-6. [Crossref] [PubMed]
29. Heck MM, Lehmann J, Amiel T, et al. Long-term Results from the LEA Randomized Trial: Extended Versus Standard Lymph Node Dissection in Patients with Bladder Cancer Undergoing Radical Cystectomy. Eur Urol 2025;88:566-70. [Crossref] [PubMed]
30. Stein JP, Cai J, Groshen S, et al. Risk factors for patients with pelvic lymph node metastases following radical cystectomy with en bloc pelvic lymphadenectomy: concept of lymph node density. J Urol 2003;170:35-41. [Crossref] [PubMed]
31. Gschliesser T, Eredics K, Berger I, et al. The Impact of Gender on Tumour Stage in In-House Complications and Choice of Urinary Diversion: Results of the Austrian Cystectomy Registry. Urol Int 2017;99:429-35. [Crossref] [PubMed]
32. Marks P, Gild P, Soave A, et al. The impact of variant histological differentiation on extranodal extension and survival in node positive bladder cancer treated with radical cystectomy. Surg Oncol 2019;28:208-13. [Crossref] [PubMed]
33. Kim SP, Frank I, Cheville JC, et al. The impact of squamous and glandular differentiation on survival after radical cystectomy for urothelial carcinoma. J Urol 2012;188:405-9. [Crossref] [PubMed]
34. Sylvester RJ, Rodríguez O, Hernández V, et al. Erratum to "European Association of Urology (EAU) Prognostic Factor Risk Groups for Non-muscle-invasive Bladder Cancer (NMIBC) Incorporating the WHO 2004/2016 and WHO 1973 Classification Systems for Grade: An Update from the EAU NMIBC Guidelines Panel" Eur Urol 2023;83:e140-1. [Eur Urol. 79(4) (2021) 480-488].
35. Zheng X, Qiu S, Yang L, et al. Comparison of survival outcomes between primary and secondary muscle-invasive bladder cancer: An updated meta-analysis. Int J Med Sci 2021;18:505-10. [Crossref] [PubMed]
36. Dotan ZA, Kavanagh K, Yossepowitch O, et al. Positive surgical margins in soft tissue following radical cystectomy for bladder cancer and cancer specific survival. J Urol 2007;178:2308-12; discussion 2313. [Crossref] [PubMed]
37. Almassi N, Bochner BH. Ileal conduit or orthotopic neobladder: selection and contemporary patterns of use. Curr Opin Urol 2020;30:415-20. [Crossref] [PubMed]
38. Marcq G, Afferi L, Neuzillet Y, et al. Oncological Outcomes for Patients Harboring Positive Surgical Margins Following Radical Cystectomy for Muscle-Invasive Bladder Cancer: A Retrospective Multicentric Study on Behalf of the YAU Urothelial Group. Cancers (Basel) 2022;14:5740. [Crossref] [PubMed]
39. Lenis AT, Lec PM, Chamie K, et al. Bladder Cancer: A Review. JAMA 2020;324:1980-91. [Crossref] [PubMed]
40. Tran L, Xiao JF, Agarwal N, et al. Advances in bladder cancer biology and therapy. Nat Rev Cancer 2021;21:104-21. [Crossref] [PubMed]
41. Mills RD, Studer UE. Metabolic consequences of continent urinary diversion. J Urol 1999;161:1057-66.
42. Mari A, Campi R, Tellini R, et al. Patterns and predictors of recurrence after open radical cystectomy for bladder cancer: a comprehensive review of the literature. World J Urol 2018;36:157-70. [Crossref] [PubMed]
43. Flaig TW, Spiess PE, Abern M, et al. NCCN Guidelines® Insights: Bladder Cancer, Version 2.2022. J Natl Compr Canc Netw 2022;20:866-78. [Crossref] [PubMed]
44. Bajorin DF, Witjes JA, Gschwend JE, et al. Adjuvant Nivolumab versus Placebo in Muscle-Invasive Urothelial Carcinoma. N Engl J Med 2021;384:2102-14. [Crossref] [PubMed]