An interpretable LASSO-Boruta nomogram for non-muscle-invasive bladder cancer recurrence after intravesical instillation

Introduction

Non-muscle-invasive bladder cancer (NMIBC) represents about 70–75% of newly diagnosed bladder cancers (1,2). Most patients are treated with transurethral resection of bladder tumor (TURBT) and then receive intravesical therapy. Recurrence remains common after treatment (3). In an analysis of 2,596 patients from seven EORTC trials, the estimated 1-year recurrence probabilities ranged from 15% to 61% across risk groups (4). When recurrence occurs, patients often need another operation and face higher costs. It may also increase progression risk and affect long-term survival (5). Current recurrence assessment mainly uses European Organisation for Research and Treatment of Cancer (EORTC), Club Urológico Español de Tratamiento Oncológico (CUETO), and European Association of Urology (EAU) risk systems (6,7). However, many of these models were derived from cohorts built 10–20 years ago and used only a limited set of pathological variables (8). Recent studies suggest that, in the era of wider Bacillus Calmette-Guérin (BCG) maintenance therapy, these conventional models may overestimate NMIBC recurrence risk and show limited discrimination (8,9). Gontero et al. (10) noted in the latest EAU guidelines that new biomarkers should be incorporated to improve post-instillation recurrence prediction in NMIBC. Artificial intelligence methods have also been tested for this task. Jiang et al. (11) reported a whole-slide image deep learning model for early NMIBC recurrence. The model outperformed traditional clinical models. Bardowska et al. (12) combined plasma markers of the plasminogen activation system with machine learning in a prospective study, which outperformed the standalone EORTC risk score. These reports show that artificial intelligence can be useful. However, many models still require costly imaging, pathology images, or digital pathology resources. Such data are not always easy to obtain, especially in primary medical settings. In addition, preoperative baseline features and laboratory tests are still not fully used, and many models are hard to interpret in daily practice. A practical model should be based on accessible data. It should also be interpretable and accurate enough for risk stratification, follow-up optimization, and individualized intervention.

We therefore combined patient information with preoperative laboratory indicators to estimate NMIBC recurrence risk. We also added an interactive system and interpretability analysis to show how the key predictors contributed to recurrence after intravesical instillation. We present this article in accordance with the TRIPOD reporting checklist (available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0557/rc).

Methods

Study design and participants

This retrospective cohort enrolled consecutive NMIBC patients who received transurethral bladder tumor resection and postoperative intravesical instillation in the Department of Urology, Affiliated Hospital of Xuzhou Medical University, between January 2020 and December 2020. Patients were eligible if they were at least 18 years old, had postoperative pathological confirmation of NMIBC, had negative surgical margins, started intravesical instillation within 4 weeks after surgery, and had complete medical records. We excluded patients with distant metastasis at diagnosis, immediate radical cystectomy, prior bladder cancer or pelvic radiotherapy, or another malignancy. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of The Affiliated Hospital of Xuzhou Medical University (No. XYFY2025-K4086-01) and informed consent was obtained from all participants for data collection and publication of this study.

Data collection and variable definitions

Electronic medical records provided the baseline information, including age, sex, and other demographic variables. Laboratory tests within 1 week before surgery included routine blood tests and blood biochemical parameters. The preoperative Self-Rating Depression Scale (SDS) score was available in routine records and was included only as an exploratory psychosocial covariate. We also recorded tumor stage, tumor count, and other clinicopathological variables. Recurrence was defined as a new finding of urothelial carcinoma in the bladder or urethra after the first TURBT, with detection by follow-up cystoscopy, urinary cytology, or imaging and confirmation by pathology. The endpoint was recurrence during follow-up.The study workflow is summarized in Figure 1.

Figure 1 Technical roadmap. CIC, clinical impact curve; DCA, decision curve analysis; NMIBC, non-muscle-invasive bladder cancer; ROC, receiver operating characteristic; SHAP, Shapley additive explanations; TURBT, transurethral resection of bladder tumor.

Feature selection

We used a stepwise strategy to reduce overfitting and screen 20 candidate variables. We first used univariate Cox regression to identify recurrence-related variables, and then entered clinically or statistically relevant variables into multivariate Cox regression. In the multivariate Cox model, smoking, hypertension, hemoglobin (Hb), and Serum Na were independent predictors. We then used least absolute shrinkage and selection operator (LASSO)-Cox regression and Boruta analysis to assess feature stability and importance. LASSO-Cox regression used the R “glmnet” package, and cross-validation was used to select the regularization parameter lambda. Variables were kept when their coefficients were nonzero at this lambda value. Boruta compared the original variables with randomly permuted shadow features to identify robust predictors. Final predictors were defined as the union of variables retained by multivariable Cox regression and those jointly selected by LASSO-Cox and Boruta, yielding smoking, hypertension, Hb, and serum Na.

Model development and evaluation

We randomly split the 200 NMIBC patients into training and validation sets at a 7:3 ratio, including 140 and 60 patients, respectively. Based on the selected predictors, smoking, hypertension, Hb, and serum Na, we built a Cox nomogram in the training set to estimate 1-, 3-, and 5-year recurrence risk. We tested the model in both cohorts using time-dependent receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). Shapley additive explanations (SHAP) analysis then showed the contribution of each predictor to the model output. In addition, we developed a dynamic interface for patient-level risk estimation, risk-trajectory display, and local SHAP interpretation.

Statistical analysis

We performed all analyses in R 4.3.1 and Python 3.9. Continuous variables are shown as mean ± standard deviation (SD) or median [interquartile range (IQR)]. Categorical variables are shown as counts. Group comparisons used the t-test or Chi-squared test, as appropriate. Variables with >5% missingness were excluded, whereas those with <5% missingness were handled using multiple imputation. Feature selection in the training cohort used multivariate Cox regression, followed by LASSO-Cox regression and Boruta analysis. We then built the nomogram. A dynamic prediction interface was developed with the R Shiny framework, version 1.8.0. ROC curves, calibration curves, and DCA were used to evaluate model performance. The final model was further internally validated using 1,000 bootstrap resamples of the training cohort. Sensitivity analysis compared the four-variable model with a reduced model containing only Hb and serum Na. Model interpretation used SHAP. A P value <0.05 was considered statistically significant.

Results

Baseline characteristics

We included 200 patients with NMIBC who received intravesical instillation therapy, comprising 110 patients without recurrence and 90 patients with recurrence. The two cohorts were similar for most baseline variables, with fasting blood glucose (FBG) as the only significant exception (Table 1).

Univariate and multivariate Cox regression analysis

Univariate Cox analysis linked hypertension, Hb, serum Na, tumor stage, and tumor count with recurrence. After adjustment, smoking, hypertension, Hb, and serum Na remained independently related to recurrence. Smoking and hypertension were linked to higher risk. Higher Hb and serum Na were linked to lower risk. Therefore, smoking, hypertension, Hb, and serum Na were selected as predictors for the final model (Table 2).

Predictive features via Boruta and LASSO

For predictor screening, we applied multivariate Cox regression first and then LASSO-Cox regression and Boruta analysis. LASSO-Cox regression selected the optimal lambda, with coefficient paths converging for core predictors (Figure 2A) and feature weights reflecting variable contributions (Figure 2B). Boruta confirmed Hb and serum Na as the only important features (Figure 2C,2D). LASSO retained several variables with non-zero coefficients, but the LASSO-Boruta intersection highlighted Hb and serum Na, whereas smoking and hypertension were retained through the multivariable Cox component of the stated union rule (Table S1).

Figure 2 Feature selection using Boruta and LASSO. (A) LASSO coefficient convergence paths. (B) LASSO feature weights. (C) Boruta feature. (D) Boruta top variable. ALB, albumin; BMI, body mass index; Cr, creatinine; FBG, fasting blood glucose; Hb, hemoglobin; LASSO, least absolute shrinkage and selection operator; PLT, platelet count; SDS, Self-Rating Depression Scale; WBC, white blood cell count.

Model performance evaluation and nomogram development

The multivariable Cox model retained smoking, hypertension, Hb, and serum Na, two of which were also jointly supported by LASSO-Cox and Boruta. Using the prespecified union strategy, these four variables were included in the final model. Time-dependent ROC curves showed good discrimination in both cohorts. The 1-, 3-, and 5-year areas under the curves (AUCs) were 0.862 [95% confidence interval (CI): 0.795–0.919], 0.918 (0.870–0.958), and 0.948 (0.913–0.976) in the training set and 0.725 (0.535–0.897), 0.870 (0.769–0.953), and 0.919 (0.834–0.981) in the validation set, respectively (Figure 3A,3B; Table S2). Calibration curves showed agreement between predicted and observed risks (Figure 3C-3E). DCA showed that model-guided surveillance provided greater net benefit than treating all or none, potentially reducing unnecessary interventions; these findings remain exploratory because clinical thresholds were not prespecified (Figure 3F-3H). We built a nomogram with the four predictors to estimate 1-, 3-, and 5-year recurrence risk in NMIBC patients (Figure 3I). Bootstrap-corrected C-index was 0.834 (95% CI: 0.799–0.872), with 1-, 3-, and 5-year AUCs of 0.857, 0.914, and 0.944 and a calibration slope of 0.930 (95% CI: 0.723–1.118) (Table 3). Schoenfeld tests indicated possible nonproportionality for Hb (P=0.02) and the overall model (P=0.040).

Figure 3 Model performance evaluation and nomogram. (A,B) Time-dependent ROC curves. (C-E) Calibration curves. (F-H) DCA. (I) Nomogram. AUC, area under the curve; DCA, decision curve analysis; Hb, hemoglobin; ROC, receiver operating characteristic.

Model interpretability using SHAP

We used SHAP analysis to interpret the final Cox-nomogram model (Figure 4). At the global level, Hb contributed most (mean |SHAP| =0.250), followed by hypertension (0.054), smoking (0.041), and serum Na (0.026). Dependence plots linked higher Hb and serum Na values with lower predicted recurrence risk, whereas smoking and hypertension were associated with higher predicted risk (Figure 4C-4F). The SHAP summary plot displayed the direction and magnitude of each predictor’s contribution for individual patients (Figure 4A), while largely reflecting the linear Cox predictor (Table S3).

Figure 4 Model interpretability using SHAP. (A) SHAP summary plot. (B) Global feature importance. (C,D) SHAP dependence plots for Hb and serum Na. (E,F) SHAP effects of Smoking and hypertension. (G) Predicted 60-month recurrence risk distribution. Hb, hemoglobin; SHAP, Shapley additive explanations.

Sensitivity analysis showed generally comparable performance between the four-variable and two-variable models (Table S4; supplementary sensitivity analysis).

Dynamic interactive prediction interface

For bedside use, we built a dynamic prediction interface from the final Cox-nomogram model (Figure 5). Clinicians can enter smoking and hypertension status with toggle switches and adjust Hb (70–180 g/L) and serum sodium (60–150 mmol/L) via sliders. The interface gives individualized 1-, 3-, and 5-year recurrence-risk estimates, a 60-month risk trajectory curve, a radar profile comparing patient values with the cohort median, and a local SHAP explanation for each predictor. This tool allows real-time, patient-specific risk assessment at the bedside.

Figure 5 Dynamic interactive prediction interface for individualized NMIBC recurrence risk assessment. Hb, hemoglobin; NMIBC, non-muscle-invasive bladder cancer.

Discussion

NMIBC often recurs after surgery, and accurate prediction remains difficult (13). Existing EORTC and CUETO scoring systems have shown limited discrimination in the current BCG treatment era and do not include enough multidimensional clinical data (8). Contieri et al. found that the 2021 EAU model was suboptimal in high-grade T1 NMIBC patients treated with repeat TURBT and BCG (14), supporting the need for more individualized models based on clinically accessible variables. Machine learning models are promising for NMIBC recurrence prediction, but many rely on complex imaging data (15). In this study, we combined baseline preoperative characteristics with routine laboratory indicators and built an interpretable nomogram based on feature selection using multivariate Cox regression followed by LASSO-Cox regression and Boruta analysis. We also used SHAP to show how key factors contributed to prediction (16,17). This approach offers a transparent decision-support tool for risk stratification in NMIBC (18,19).

The cohort included 200 NMIBC patients. Most baseline variables were balanced between the training and validation cohorts. FBG was higher in the training cohort. In our cohort, smoking and hypertension were associated with higher NMIBC recurrence risk, whereas higher Hb and serum Na were associated with lower risk. The association between smoking and recurrence agrees with previous evidence linking longer smoking duration and greater pack-years to higher NMIBC recurrence risk (20). Tobacco-related carcinogens can remain in urine and repeatedly contact the urothelium. This exposure may cause DNA damage, oxidative stress, and chronic inflammation, thereby promoting recurrence. Hypertension was also identified as a risk-related factor in our model. Previous epidemiological studies have linked hypertension and metabolic syndrome components to increased bladder cancer risk (21,22). Vascular dysfunction, oxidative stress, chronic inflammation, and metabolic disturbance may partly explain this association, although the exact relationship between hypertension and NMIBC recurrence requires further validation. In this study, higher Hb was linked to lower recurrence risk. This result is consistent with evidence that preoperative anemia is related to recurrence and progression in NMIBC (23). Lower Hb may reflect impaired oxygen delivery, poor nutritional or inflammatory status, and reduced systemic reserve. These conditions may be linked to unfavorable tumor outcomes. Higher serum Na was associated with lower recurrence risk in this cohort (24). However, evidence specific to NMIBC is limited, and serum Na may reflect hydration, comorbidities, medication use, or general systemic status (25). This observational association should not be interpreted as causal and requires external validation.

Multivariable Cox regression was the primary selection method, with LASSO-Cox and Boruta used as complementary stability analyses. Smoking and hypertension were retained because of their independent Cox associations, whereas Hb and serum Na were additionally supported by both algorithms. Sensitivity analysis showed generally comparable performance after excluding smoking and hypertension, suggesting limited incremental predictive value but supporting the robustness of the main findings.

Tang and colleagues developed an externally validated NMIBC recurrence model using tumor, urinary, and hematologic variables (26). Ferro et al. further demonstrated the potential prognostic value of the modified Glasgow Prognostic Score in high-grade T1 NMIBC treated with BCG, highlighting the contribution of accessible systemic inflammatory markers and the need for prospective validation (27). Similarly, our model included smoking, hypertension, Hb, and serum Na, suggesting that systemic host status may complement clinicopathological factors in recurrence assessment. smoking has been associated with increased NMIBC recurrence risk (20), while hypertension and metabolic abnormalities have also been linked to bladder cancer risk (21). Lower Hb may reflect anemia, impaired oxygen delivery, inflammation, or poor nutritional reserve, and routine blood-based inflammatory markers have shown prognostic value in NMIBC (28). Compared with the study by Celik and colleagues, which used urinary flow cytometry to evaluate neutrophil-to-T-cell ratio and myeloid-derived suppressor cells after BCG therapy (29), the predictors in our model are routine, inexpensive, and standardized. They may therefore be easier to use in broader clinical settings.

Our nomogram used four easily available predictors: smoking, hypertension, Hb, and serum Na. The model discriminated well in both cohorts, with AUCs of 0.862, 0.918, and 0.948 at 12, 36, and 60 months in the training cohort, and 0.725, 0.870, and 0.919 in the validation cohort, respectively. Compared with the recurrence prediction model developed by Tang et al. (26) and the 9 deep learning-based pathomics model reported by Wang et al. (30), our model achieved satisfactory performance with routine clinical and laboratory variables only. This may facilitate future implementation after external validation.

SHAP provided global and patient-level visualization of predictor contributions. Because the final Cox model was linear and additive, these SHAP values largely recapitulated the linear predictor and did not identify nonlinear effects or interactions. Unlike the nonlinear modeling approach used by Klinglmair et al. (31), which focused on body composition-related variables in high-risk NMIBC patients treated with BCG, the present SHAP analysis highlighted routine hematologic, electrolyte, exposure-related, and comorbidity-related indicators.

SHAP and the dynamic interface were used to explain and display predictions rather than improve model performance. The interface presents patient-level risks, risk trajectories, and predictor contributions. However, no formal usability or decision-impact testing was performed; therefore, it should be considered a proof-of-concept pending external validation and prospective clinical evaluation (32,33).

Several limitations remain. First, the study was retrospective, single-center, and modest in size. Several variables required for EORTC, CUETO, and EAU stratification—including tumor grade and size, concomitant carcinoma in situ, previous recurrence, lymphovascular invasion, repeat TURBT findings, and maintenance instillation schedules—were incompletely recorded, precluding reliable head-to-head comparison. Future multicenter prospective studies will collect these variables, externally validate the model, and compare it directly with these risk systems before broader clinical application. The web application also lacks formal usability and clinical decision-impact testing. Pinton et al. identified 7,962 BCG-treated patients but modeled 1,524 patients with 56 features; clinically meaningful prediction remained limited by data imbalance (34). This underscores the exploratory nature of our smaller model and the need for richer external datasets. Future models may improve risk stratification by combining digital pathology, urinary cytology, peripheral blood tests, and multiomics data. Dynamic risk-monitoring platforms may also support personalized decision-making in NMIBC management (35,36).

Conclusions

In summary, we built an interpretable nomogram using four readily available predictors. The model showed promising internal performance, but external validation, usability testing, and prospective decision-impact assessment are required before clinical implementation.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0557/rc

Data Sharing Statement: Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0557/dss

Peer Review File: Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0557/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0557/coif). The authors have no 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 Ethics Committee of The Affiliated Hospital of Xuzhou Medical University (No. XYFY2025-K4086-01) and informed consent was obtained from all participants for data collection and publication of this 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. van Hoogstraten LMC, Vrieling A, van der Heijden AG, et al. Global trends in the epidemiology of bladder cancer: challenges for public health and clinical practice. Nat Rev Clin Oncol 2023;20:287-304. [Crossref] [PubMed]
2. Contieri R, Paciotti M, Uleri A, et al. Second Transurethral Resection of Bladder Tumor Can Be Safely Omitted in Selected Patients with T1 Non-muscle-invasive Bladder Cancer: Results from the Prospective HuNIRe Trial. Eur Urol Oncol 2026;9:404-12. [Crossref] [PubMed]
3. Giatsoglou M, Chatzistavrou D, Sopczynski JM, et al. Clinical Outcomes in Bacillus Calmette-Guérin-Exposed Non-Muscle-Invasive Bladder Cancer. JAMA Netw Open 2026;9:e267470. Erratum in: JAMA Netw Open 2026 May 1;9(5):e2618475. [Crossref] [PubMed]
4. Sylvester RJ, van der Meijden AP, Oosterlinck W, et al. Predicting recurrence and progression in individual patients with stage Ta T1 bladder cancer using EORTC risk tables: a combined analysis of 2596 patients from seven EORTC trials. Eur Urol 2006;49:466-5; discussion 475-7. [Crossref] [PubMed]
5. D'Andrea D, Soria F, Rouprêt M, et al. Design and Rationale of the ROGUE-1 Registry: A Multicentric Prospective Registry for Primary T1 Nonmuscle-invasive Bladder Cancer. Eur Urol Oncol 2026;9:705-9. [Crossref] [PubMed]
6. Peyton CC, Dhanda R, Burke T, et al. Psychometric Validation of the QLQ-NMIBC24 in Low Grade, Intermediate Risk Non-Muscle Invasive Bladder Cancer. Res Rep Urol 2026;18:591097. [Crossref] [PubMed]
7. Rubio-Briones J, Guerrero Ramos F, Mercadé Sánchez A, et al. Digital Uromonitor® outperforms quantitative polymerase chain reaction Uromonitor and cytology for non-muscle-invasive bladder cancer surveillance: results from the 'External Validation of Uromonitor as a Biomarker for Optimization of NMIBC Management by the Club Urológico Español de Tratamiento Oncológico Group' (EVALUATION-CUETO) study. BJU Int 2026;138:75-84. [Crossref] [PubMed]
8. Ślusarczyk A, Garbas K, Pustuła P, et al. Assessing the Predictive Accuracy of EORTC, CUETO and EAU Risk Stratification Models for High-Grade Recurrence and Progression after Bacillus Calmette-Guérin Therapy in Non-Muscle-Invasive Bladder Cancer. Cancers (Basel) 2024;16:1684. [Crossref] [PubMed]
9. Xylinas E, Kent M, Kluth L, et al. Accuracy of the EORTC risk tables and of the CUETO scoring model to predict outcomes in non-muscle-invasive urothelial carcinoma of the bladder. Br J Cancer 2013;109:1460-6. [Crossref] [PubMed]
10. Gontero P, Birtle A, Capoun O, et al. European Association of Urology Guidelines on Non-muscle-invasive Bladder Cancer (TaT1 and Carcinoma In Situ)-A Summary of the 2024 Guidelines Update. Eur Urol 2024;86:531-49. [Crossref] [PubMed]
11. Jiang F, Hong G, Zeng H, et al. Deep learning-based model for prediction of early recurrence and therapy response on whole slide images in non-muscle-invasive bladder cancer: a retrospective, multicentre study. EClinicalMedicine 2025;81:103125. [Crossref] [PubMed]
12. Bardowska K, Krajewski W, Kołodziej A, et al. Evaluation of six novel biomarkers for predicting recurrence of non-muscle invasive bladder cancer after endoscopic resection- a prospective observational study. World J Urol 2025;43:114. [Crossref] [PubMed]
13. Hof JP, Kiemeney LALM, Aben KKH, et al. Recurrence Patterns in a Large Contemporary Cohort of Patients With Non-Muscle Invasive Bladder Cancer. Clin Genitourin Cancer 2026;24:102464. [Crossref] [PubMed]
14. Contieri R, Hurle R, Paciotti M, et al. Accuracy of the European Association of Urology (EAU) NMIBC 2021 scoring model in predicting progression in a large cohort of HG T1 NMIBC patients treated with BCG. Minerva Urol Nephrol 2023;75:180-7. [Crossref] [PubMed]
15. Wu S, Wang Y, He J, et al. An explainable machine learning model for predicting bladder tumor aecurrence risk. Front Oncol 2026;16:1728056. [Crossref] [PubMed]
16. Kutlu M, Donmez TB, Freeman C. Machine Learning Interpretability in Diabetes Risk Assessment: A SHAP Analysis. Computers and Electronics in Medicine 2024;1:34-44.
17. Kırboğa KK. Bladder cancer gene expression prediction with explainable algorithms. Neural Computing and Applications 2023;36:1585-97.
18. Smani S, DuBois J, Zhao K, et al. Advancements in the Diagnosis, Treatment, and Risk Stratification of Non-Muscle Invasive Bladder Cancer. Curr Oncol Rep 2025;27:236-46. [Crossref] [PubMed]
19. Guerrero-Ramos F, Subiela JD, Rodríguez-Faba Ó, et al. Predicting Recurrence and Progression in Patients with Non-Muscle-Invasive Bladder Cancer: Systematic Review on the Performance of Risk Stratification Models. Bladder Cancer 2022;8:339-57. [Crossref] [PubMed]
20. Kwan ML, Haque R, Young-Wolff KC, et al. Smoking Behaviors and Prognosis in Patients With Non-Muscle-Invasive Bladder Cancer in the Be-Well Study. JAMA Netw Open 2022;5:e2244430. [Crossref] [PubMed]
21. Ahmadinezhad M, Arshadi M, Hesari E, et al. The relationship between metabolic syndrome and its components with bladder cancer: a systematic review and meta-analysis of cohort studies. Epidemiol Health 2022;44:e2022050. [Crossref] [PubMed]
22. Connaughton M, Dabagh M. Association of Hypertension and Organ-Specific Cancer: A Meta-Analysis. Healthcare (Basel) 2022;10:1074. [Crossref] [PubMed]
23. Soria F, Moschini M, Abufaraj M, et al. Preoperative anemia is associated with disease recurrence and progression in patients with non-muscle-invasive bladder cancer. Urol Oncol 2017;35:113.e9-113.e14. [Crossref] [PubMed]
24. Fibbi B, Marroncini G, Naldi L, et al. Hyponatremia and Cancer: From Bedside to Benchside. Cancers (Basel) 2023;15:1197. [Crossref] [PubMed]
25. Naldi L, Catalano M, Melica ME, et al. Effects of reduced extracellular sodium on proliferation and invasive activity of renal cell carcinoma cell lines. Sci Rep 2025;15:8067. [Crossref] [PubMed]
26. Tang J, Fan L, Huang T, et al. Development and external validation of a model to predict recurrence in patients with non-muscle invasive bladder cancer. Front Immunol 2024;15:1467527. [Crossref] [PubMed]
27. Ferro M, Tătaru OS, Musi G, et al. Modified Glasgow Prognostic Score as a Predictor of Recurrence in Patients with High Grade Non-Muscle Invasive Bladder Cancer Undergoing Intravesical Bacillus Calmette-Guerin Immunotherapy. Diagnostics (Basel) 2022;12:586. [Crossref] [PubMed]
28. Huang J, Lin L, Mao D, et al. Prognostic value of neutrophil-to-lymphocyte ratio in patients with non-muscle-invasive bladder cancer with intravesical Bacillus Calmette-Guérin immunotherapy: a systematic review and meta-analysis. Front Immunol 2024;15:1464635. [Crossref] [PubMed]
29. Celik M, Polat ME, Karaaslan M, et al. Prediction of recurrence using hematologic and urinary markers in intravesical Bacillus calmette Guerin treated bladder cancer. Sci Rep 2025;15:35415. [Crossref] [PubMed]
30. Wang GY, Zhu JF, Wang QC, et al. Prediction of non-muscle invasive bladder cancer recurrence using deep learning of pathology image. Sci Rep 2024;14:18931. [Crossref] [PubMed]
31. Klinglmair G, Lutz M, Telia T, et al. Sarcopenia and recurrence risk in high-risk non-muscle-invasive bladder cancer: a nonlinear modeling approach. J Nutr Health Aging 2026;30:100805. [Crossref] [PubMed]
32. Van Dort BA, Engelsma T, Sneekes P, et al. Explainable AI in hospital clinical decision support systems: A scoping review of healthcare professionals' perspectives. PLOS Digit Health 2026;5:e0001417. [Crossref] [PubMed]
33. Jiang Y, Hu H, Shao X, et al. A novel web-based dynamic prognostic nomogram for gastric signet ring cell carcinoma: a multicenter population-based study. Front Immunol 2024;15:1365834. [Crossref] [PubMed]
34. Pinton P, Kawano H, Patschan O, et al. A machine-learning approach to predict additional treatment after Bacillus Calmette-Guérin induction in non-muscle-invasive bladder cancer. Sci Rep 2026;16:25051. [Crossref] [PubMed]
35. Yoo W. Precision oncology in the age of AI: lessons from AI-driven drug discovery and clinical translation. BJC Rep 2026;4:18. [Crossref] [PubMed]
36. Chang JC, Perez-Londoño A, Kaul S, et al. Dynamic prognostication of non-muscle invasive bladder cancer using conditional recurrence- and progression-free survival: A SEER-Medicare analysis. Urol Oncol 2026;44:119.e7-119.e16. [Crossref] [PubMed]