Beyond staging: toward a layered risk architecture after radical cystectomy for bladder cancer Authors: Christian Gilfrich, Anton Kravchuk, Emily Rinderknecht, Maximilian Haas, Maximilian Burger, Ingmar Wolff, Sabine D Brookman-May, Matthias May Published: 2026 Radical cystectomy (RC) and the unfinished problem of risk prediction RC remains one of the most consequential interventions in urologic oncology: potentially curative, biologically revealing, and accompanied by substantial perioperative morbidity and long-term prognostic uncertainty (1). Once the bladder has been removed and final pathology is available, the central question is no longer whether the patient had high-risk disease, but how multiple signals of risk should be integrated into postoperative decision-making (1,2). Pathologic T stage, nodal involvement, lymphovascular invasion (LVI), surgical margin status, systemic treatment exposure, frailty, comorbidity, inflammatory status, and host reserve are all considered in daily practice (1-3). Yet this integration is often implicit rather than structured, and rarely occurs within a transparent, validated, clinically actionable framework. This is the context in which the systematic review and meta-analysis by Artiles Medina and colleagues should be read (4). The authors did not simply summarize isolated prognostic variables after RC. Rather, they focused on multivariable survival prediction models and synthesized adjusted hazard ratios (HRs) from published Cox proportional hazards models. Their review included 77 studies, comprising models for overall survival (OS), cancer-specific survival (CSS), and recurrence-free survival (RFS) (4). The resulting analysis provides an important overview of clinicopathologic factors that retain independent prognostic relevance across heterogeneous cohorts. The study does not deliver a ready-to-use clinical nomogram. Nor should it be judged by that standard. Its more important contribution is to define an adjusted prognostic architecture after RC. In other words, it shows which variables repeatedly remain informative after multivariable adjustment, and which domains of risk should be carried forward into the next generation of post-cystectomy prediction tools. From isolated predictors to risk architecture A central strength of the meta-analysis is its focus on adjusted effects. Prediction after RC has long been dominated by familiar variables: pathologic stage, nodal status, LVI, surgical margins, and systemic therapy (1-3). However, the clinical question is not whether these variables matter in isolation. The more difficult question is how they should be weighted relative to one another, and whether they can be integrated into a clinically meaningful model. Artiles Medina et al. move the field closer to this goal by synthesizing multivariable evidence (4). This synthesis rested on a substantial but endpoint-specific evidence base, with 36 studies contributing OS models, 44 contributing CSS models, and 26 contributing RFS models (4). For OS, adverse predictors included older age, female sex, higher pathologic T stage, lymph node involvement, LVI, positive surgical margins, metastatic disease, and concomitant carcinoma in situ (CIS). Neoadjuvant chemotherapy, adjuvant chemotherapy, and achievement of Pentafecta criteria were associated with improved OS (4). For CSS, the risk architecture was more disease-centered, with adverse associations for age, female sex, pathologic T stage, lymph node involvement, lymph node density, LVI, positive margins, neutrophil-to-lymphocyte ratio (NLR), hydronephrosis, and sarcopenia, whereas adjuvant chemotherapy was associated with improved outcome (4). These findings are clinically intuitive, but they are not trivial. They place widely recognized variables into a comparative, adjusted framework. They also show that postoperative risk after RC is not a single-layer construct. It is built from several overlapping domains: tumor burden, nodal dissemination, adverse pathology, systemic treatment exposure, surgical quality, host reserve, inflammation, and competing mortality. OS requires a stronger comorbidity and competing-risk layer The distinction between OS and CSS is not merely statistical. It is clinically fundamental. OS is an integrative endpoint. It captures lethal bladder cancer (BC) biology, but also competing non-cancer mortality, surgical fitness, renal function, cardiovascular reserve, frailty, functional status, treatment tolerance, and recovery from major pelvic surgery. CSS is narrower and more directly linked to the biology and control of BC, although even this endpoint is modified by host reserve and access to effective systemic treatment. This distinction reveals one of the most important limitations of the current evidence base. The OS architecture emerging from the review is informative, but incomplete (4). Age is captured, and several source models considered variables such as American Society of Anesthesiologists physical status classification, body mass index, Charlson Comorbidity Index, inflammatory markers, and other measures of patient vulnerability (5-7). However, comorbidity and competing-risk burden were not consistently harmonized into the pooled OS framework (4). This should not be interpreted as evidence that comorbidity is unimportant. On the contrary, it highlights a structural weakness in the post-cystectomy prediction literature (8). RC is often performed in older patients with cardiovascular disease, chronic kidney disease, impaired functional reserve, malnutrition, sarcopenia, smoking exposure, or competing oncologic and non-oncologic risks. Recent perioperative data suggest that smoking should be modeled continuously: greater pack-year exposure increased postoperative morbidity after robot-assisted RC, whereas longer cessation reduced morbidity (9). Two patients with identical pathological tumor (pT) stage, nodal status, and margin status may have markedly different all-cause mortality risks because of differences in cardiopulmonary reserve, renal function, frailty, and treatment tolerance. A model that captures cancer biology but incompletely captures host vulnerability may be informative for CSS, but it is insufficient for OS. Future OS models should therefore treat comorbidity, frailty, renal function, performance status, nutritional reserve, and competing-risk burden as core domains rather than optional covariates. CSS in a changing therapeutic era The CSS architecture identified by Artiles Medina et al. is more tightly anchored to tumor biology (4). Pathologic stage, nodal involvement, lymph node density, LVI, positive surgical margins, hydronephrosis, sarcopenia, and systemic inflammation all help define the risk of BC death (4). This structure is persuasive, but it is also being challenged by rapid therapeutic change. Many historical RC models were developed in periods in which perioperative systemic therapy was inconsistently used, and neoadjuvant therapy was largely conceptualized as cisplatin-based chemotherapy (10,11). That framework is now evolving. Perioperative chemoimmunotherapy and antibody-drug conjugate combinations are changing the treatment landscape of muscle-invasive BC (12,13). Regimens combining cisplatin-based chemotherapy with immune checkpoint inhibition, as well as perioperative enfortumab vedotin plus pembrolizumab, are likely to reshape the prognostic meaning of systemic treatment exposure, pathologic complete response (pCR), residual disease, and postoperative recurrence risk (12-15). Consequently, a binary variable such as “neoadjuvant chemotherapy: yes or no” is becoming increasingly inadequate. Future models will need to distinguish between treatment type, treatment completion, pCR, residual non-muscle-invasive disease, residual muscle-invasive disease, adjuvant continuation, treatment-related toxicity, and molecular residual disease. The prognostic effect of systemic therapy may also influence OS, not only CSS, because more effective treatment can reduce cancer death while introducing new dimensions of treatment selection, toxicity, and survivorship. The next generation of post-cystectomy models must therefore be built for the therapeutic era that is arriving, not only for the era that generated most historical cohorts. RFS remains clinically important but methodologically fragile RFS deserves particular attention. It is central to postoperative surveillance, imaging intensity, early relapse detection, and adjuvant treatment strategy. Yet it was the least amenable to quantitative synthesis in the review by Artiles Medina et al. (4). Although 26 studies included RFS models, heterogeneity in RFS definitions prevented meta-analysis of predictive factors (4). This limitation is not a minor technical detail. RFS can be defined in different ways: local recurrence, distant recurrence, progression, disease-free survival, second malignancy, death, or last disease-free follow-up. These choices directly affect event rates, model coefficients, clinical interpretation, and comparability across studies. Without a consensus definition, RFS models will remain difficult to compare, pool, validate, or translate into surveillance algorithms. This is important in the modern era. If adjuvant immunotherapy, perioperative systemic combinations, circulating tumor DNA (ctDNA), and other molecular residual disease approaches are to be incorporated into post-cystectomy care, RFS must be defined with greater precision (16-18). The field needs endpoint discipline: local recurrence, distant recurrence, progression, cancer-specific death, non-cancer death, and second primary tumors should not be merged without careful justification. Biomarkers and dynamic risk: the missing future layer The review also highlights, by omission, the limited integration of dynamic biomarkers into existing prediction models. Circulating tumor cells (CTCs) and ctDNA were not included in the pooled prediction architecture (18,19). This is understandable, because the underlying model literature has not yet incorporated them consistently. However, this omission is unlikely to remain methodologically acceptable as molecular residual disease testing moves closer to clinical implementation. CTCs may capture hematogenous dissemination and aggressive tumor behavior (18,19). ctDNA may identify molecular residual disease before radiographic relapse and may become central to postoperative risk assignment, adjuvant treatment selection, and surveillance intensity (18-21). These biomarkers could allow risk models to move beyond static postoperative pathology toward dynamic, time-updated prediction: before systemic therapy, after neoadjuvant treatment, after RC, during adjuvant treatment, and throughout surveillance. Still, biomarker enthusiasm must be disciplined. Not every biologically plausible marker should be inserted into a clinical model. Variables should improve calibration, discrimination, decision-curve utility, interpretability, and clinical actionability. The field already has many internally promising but clinically unused models. The next generation must be contemporary, externally validated, recalibratable, transparent, and usable at the point of care. Interpreting the visual risk maps The pooled evidence can be visually organized into relative-effect maps, provided that the purpose and limits of such displays are explicit. Figures 1,2 should be read as variable-level visual syntheses of adjusted prognostic effects, not as patient-level scoring instruments. Each displayed contribution is derived from the pooled multivariable HR for the specified contrast and rescaled to a common visual point axis. The reference category is therefore the comparator used in the corresponding pooled contrast, not an average patient, not an empirically estimated baseline-risk group, and not a calibrated clinical reference population. Meta-analytic relative-effect architecture for OS after RC. Pooled multivariable HRs from Artiles Medina et al. (4) were transformed to the log-hazard scale and displayed on a 0-to-100 visual contribution axis. The figure shows predictor-level relative effects for OS and should not be interpreted as a calibrated survival nomogram or patient-level risk score. CI, confidence interval; HR, hazard ratio; M, metastasis; OS, overall survival; pT, pathological tumor; RC, radical cystectomy. Meta-analytic relative-effect architecture for CSS after RC. Pooled multivariable HRs for CSS from Artiles Medina et al. (4) were transformed to the log-hazard scale and displayed on a 0-to-100 visual contribution axis. The figure shows predictor-level relative effects and does not estimate absolute CSS probability. CI, confidence interval; CSS, cancer-specific survival; HR, hazard ratio; pT, pathological tumor; RC, radical cystectomy. Figure 1 illustrates the OS architecture. It shows that age, advanced pathologic T stage, metastatic disease, lymph node involvement, LVI, positive surgical margins, and the absence of favorable care-related factors carry substantial relative prognostic weight. At the same time, the figure makes visible an important limitation of the available evidence: comorbidity, frailty, renal function, performance status, and competing-risk burden remain insufficiently standardized for integration into a pooled OS framework. This omission is particularly relevant because OS after RC reflects not only lethal BC biology, but also host vulnerability and non-cancer mortality. Figure 2 provides the analogous architecture for CSS. The prominent contributions of pathologic T stage, nodal involvement, lymph node density, positive margins, and LVI are expected and align with established mechanisms of BC progression. However, the additional weight of hydronephrosis, sarcopenia, and systemic inflammation is clinically important. These variables suggest that lethal BC biology is not fully captured by Tumor, Node, Metastasis staging alone. Hydronephrosis may reflect obstructive or locally advanced tumor behavior; sarcopenia may indicate impaired host reserve; and the NLR may capture systemic inflammatory activation. The temptation to convert these pooled estimates directly into a clinical score should be resisted. A higher visual point contribution indicates a larger relative adjusted effect for that predictor, but the display does not provide a cumulative patient score or a validated probability of death at a specified time point. Without a common baseline survival function, the pooled evidence cannot yield 3-year or 5-year survival probabilities. Without individual patient-level data, it cannot be recalibrated across contemporary populations. Without external validation, it cannot guide individual treatment decisions. Its value is conceptual rather than prescriptive: it clarifies the relative architecture of risk, but it does not deliver absolute risk prediction. What future models must include The study by Artiles Medina et al. should be viewed as an important step toward a more mature prediction science after RC (4). It identifies variables that future models cannot ignore, but it also exposes the limits of the current evidence base. The next generation of post-cystectomy prediction tools should move beyond a narrow clinicopathologic template toward a layered, contemporary, and clinically interpretable framework. The first layer should remain classic tumor burden: pathologic T stage, nodal status, lymph node density, and metastatic status (1,2,4). These variables continue to define the anatomic foundation of post-cystectomy risk. The second layer should capture adverse histopathology, including LVI, surgical margin status, variant histology where available, and concomitant CIS when clinically relevant (1,4). These features refine stage-based risk by capturing tumor behavior, surgical-pathologic completeness, and biologic aggressiveness. Variant histology should not be treated as secondary: contemporary cystectomy data show subtype-specific differences in cancer-specific outcomes, genomic architecture, chemorefractory risk, and actionable alterations, with implications for treatment selection, trial eligibility, and surveillance intensity (22). Surgical quality also extends beyond composite endpoints: ureteric margin assessment and selective intraoperative frozen section analysis may inform intraoperative decisions and surveillance because distal ureteric margin positivity has been associated with upper tract recurrence and inferior survival (23). A third layer must address host vulnerability (5-7). For OS in particular, comorbidity, frailty, renal function, performance status, sarcopenia, nutritional reserve, and competing-risk burden are not secondary considerations. They are central determinants of whether a patient dies with BC, from BC, or from the cumulative burden of age, treatment exposure, and comorbidity. Their inconsistent integration into current models is one of the clearest gaps exposed by the existing literature (4,6,7). The fourth layer should reflect systemic treatment and response in a way that matches contemporary practice (11-15). A binary distinction between receipt and non-receipt of neoadjuvant chemotherapy is no longer sufficient. Future models will need to distinguish cisplatin-based chemotherapy, chemoimmunotherapy, antibody-drug conjugate combinations, adjuvant immunotherapy, treatment completion, treatment-limiting toxicity, pCR, residual non-muscle-invasive disease, residual muscle-invasive disease, and postoperative residual-risk states. The fifth layer should incorporate dynamic molecular disease signals, including ctDNA and, potentially, CTCs (16-19). These markers may allow risk to be updated over time rather than inferred solely from the cystectomy specimen. Such models cannot be derived from pooled study-level HRs alone (4). They will require individual patient-level data, harmonized endpoint definitions, contemporary treatment cohorts, prespecified modeling strategies, internal validation, external validation, calibration at clinically meaningful time points, and formal assessment of clinical utility (3,4,10). They will also require restraint. The aim should not be maximal variable accumulation, but clinically interpretable precision: a model complex enough to reflect modern BC care, yet transparent enough to inform decisions at the bedside. Conclusions The principal message of the meta-analysis by Artiles Medina et al. is not that RC outcomes can now be reduced to a single score. They cannot. The study’s contribution is more important and more durable: it moves the field from lists of prognostic variables toward a structured risk architecture. That architecture remains incomplete. OS models need a stronger comorbidity and competing-risk layer. CSS models must be updated for contemporary perioperative systemic therapy, treatment response, molecular residual disease, and emerging biomarkers. RFS modeling requires endpoint standardization before pooled prediction can become clinically meaningful. For now, the responsible use of this evidence is not to claim calibrated prediction, but to define the framework in which calibrated prediction should finally be built. The next generation of post-cystectomy risk models should integrate pathology, host vulnerability, treatment response, surgical quality, and molecular residual disease. Artiles Medina et al. provide a rational foundation for that work. The next task is to make prediction not only statistically persuasive, but clinically usable. Supplementary The article’s supplementary files as Acknowledgments None. 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. Footnotes Provenance and Peer Review: This article was commissioned by the editorial office, Translational Andrology and Urology. The article has undergone external peer review. 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-2026-0481/coif). The authors have no conflicts of interest to declare. References Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0481/prf Supplementary Materials The article’s supplementary files as