1. Introduction 

Radical cystectomy (RC) with pelvic lymph node dissection (PLND) remains the gold standard treatment for patients with muscle-invasive bladder cancer (MIBC) ^1,2.  Although there is a general agreement that PLND should be performed with radical cystectomy, controversy remains on the optimal extent of the procedure ^3,4. Even though the majority of the existing literature on this topic is retrospective in nature, yielding conflicting results ^5–8, two randomized controlled trials have been recently conducted with the aim of evaluating potential oncological benefits associated with extended lymph node dissection (ELND) compared to standard lymph node dissection (SLND). The first randomized trial, conducted in Europe (Germany) by Gschwend et al., included 401 patients with T1G3 non-muscle-invasive (NMIBC) or T2–T4a bladder cancer (BCa), without demonstrating a significant advantage for ELND over SLND in recurrence-free survival, cancer specific survival, or overall survival⁹. However, the design of that study was widely criticized due to several methodological limitations that may have hindered the detection of statistically significant survival differences between the two groups, including: small sample size, inclusion of patients with NMIBC, a relatively low proportion of patients with pT3–4 disease, and a high lymph node yield in both LND templates ⁹. Moreover, the exclusion of patients who had received neoadjuvant treatment limits the generalizability of the findings to current clinical practice. On the contrary, the results of the SWOG S1011 trial, whose design overcomes the aforementioned methodological limitations, were recently published ¹⁰. Lerner et al. included 592 patients diagnosed with cT2–4a N0–2 BCa and randomized them to undergo RC with SLND or ELND ¹⁰, with the aim of examining whether ELND improves disease-free survival and overall survival in this setting. According to their findings, performing an ELND was not associated with any survival benefit compared to SLND in patients with MIBC treated with RC ¹⁰. Although the SWOG S1011 trial provides high-level evidence, its external validity has not yet been assessed in a population-based setting. For a randomized controlled trial to be generalizable, the study population should adequately reflect patients encountered in routine clinical practice. Therefore, the aim of the present study was to evaluate the external validity of this trial within a large, real-world North American cohort using a nationwide database.

2. Materials & Methods 

2.1 Data Sources 

To contextualize trial findings within routine practice, we recreated the trial eligibility criteria using data from the National Cancer Database (NCDB). The NCDB captures approximately 72% of newly diagnosed cancer cases in the United States and includes data submitted by Commission on Cancer–accredited facilities across more than 1,500 programs ^11,12. We identified adult patients (≥18 years at diagnosis) with clinical T2–T4a, N0–2 bladder cancer who underwent radical cystectomy with lymph node dissection between 2010 and 2022 and met the key eligibility criteria of the trial ¹⁰. Although detailed comorbidity data are not available in the NCDB, the Charlson Comorbidity Index (CCI) is reported. Given that the original trial required a Zubrod performance status of 0–2 ¹³, we restricted our cohort to patients with a CCI of 0–1. In line with the trial protocol, patients who received prior pelvic radiotherapy were excluded. Patients who were missing cT, cN, pT, pN, type of surgery (e. g., robotic, open), or did not have adequate information about neoadjuvant systemic treatment were excluded. Our selection criteria, as detailed in Figure 1, yielded a total of 26,183 patients. Notably, the original trial did not allow robot-assisted radical cystectomy. However, given that the use of RARC has become increasingly common and currently accounts for approximately 40% of all radical cystectomies ^14,15, we decided to include both open and robotic approaches in our analysis. Under this premise, we conducted separate analyses for open RC and RARC, treating them as distinct entities. Patients undergoing laparoscopic radical cystectomy were excluded due to their limited representation in the NCDB and to maintain focus on the two predominant and clinically relevant surgical approaches (open and robotic) in the current clinical practice.

Figure 1. Cohort selection flowchart

Detailed breakdown of the number of participants included in this study based on the inclusion and exclusion criteria used for this study

2.2 Variables 

To account for differences in the examined cohort, we extracted the following variables for each patient: age (continuous), sex (male, female), race (White, Black, Asian, Other/Multi-racial/Unknown), clinical T stage (cT2, cT3-4a), neoadjuvant systemic treatment (yes, no), type of surgery (open, robotic), histological variants (yes, no), pathological T stage (pT0, pTis/pTa/pT1, pT2, pT3-T4), pathological N stage (N0, N1, N2, N3), number of nodes removed (continuous), number of positive nodes (1, 2–5, 6–10, >10). Focusing on the number of nodes removed at surgery, due to the absence of means and standard deviations (SD) in the original trial, we opted to categorize this variable based on the median number of nodes removed reported in the trial. Accordingly, we defined the following categories: (≤ 24, ≥ 25) for SLND, and (≤ 39, ≥ 40) for ELND. Given that the NCDB does not capture the anatomical template of lymph node dissection, patients were categorized based on lymph node count thresholds to approximate the trial-defined groups. These thresholds were derived from the median number of lymph nodes removed in the SLND (24 nodes) and ELND (39 nodes) arms of the SWOG S1011 trial.

2.3 Statistical Analysis 

To detect statistical significance of differences in the categorical variables the Chi-squared test or Fisher’s exact test were used, as appropriate. The Chi-squared test was used when expected cell counts were ≥5 and the distribution across categories was balanced. Fisher’s exact test was applied in cases with small sample sizes, sparse data, or expected frequencies <5 in any cell. The choice of test also considered asymmetry between groups and the number of categories, favoring Fisher’s test in comparisons involving rare subgroups or markedly unbalanced distributions (e. g., race, number of positive nodes). The trial did not provide measures of variance (such as standard deviation) for continuously coded variables, so a comparison test for statistical significance between median age was not possible. As previously mentioned, with the aim of comparing the number of lymph nodes removed at RC for statistical significance, we categorized this variable based on the median value. First, we conducted these comparative analyses between the trial cohorts (SLND and ELND) and a NCDB cohort who underwent open RC. Second, we performed the same analysis, comparing the trial populations with an NCDB cohort who underwent RARC. To complement conventional hypothesis testing and mitigate the impact of large sample size differences between cohorts, measures of effect size were reported. Odds ratios (ORs) with 95% confidence intervals (CIs) were calculated for selected binary variables using contingency tables reconstructed from aggregate data and should be interpreted as unadjusted measures of association. For multi-level categorical variables, global standardized differences were used to quantify the magnitude of imbalance between groups, with values greater than 0.1 indicating meaningful imbalance. The primary objective of this study was not to compare oncological outcomes between different extents of lymph node dissection, but to evaluate the external validity of the SWOG S1011 trial by assessing the comparability between trial participants and real-world patients. Given the lack of detailed information on the anatomical extent of lymph node dissection within the NCDB, no formal comparative effectiveness analyses between standard and extended templates were planned a priori. Exploratory analyses based on lymph node counts as a proxy for dissection extent were not included in the primary study design due to their inherent susceptibility to selection bias and residual confounding, as higher lymph node yield may reflect treatment at high-volume centers, greater surgical expertise, more accurate pathological assessment, or more favorable patient selection. This study was deemed exempt from review by the Henry Ford Institutional Review Board (IRB), as all NCDB data are fully de-identified and do not require individual patient informed consent. All tests were 2-sides with a significance alpha level set at <0.05. All analyses were completed using R-2019 (R Foundation for Statistical Computing, Vienna, Austria).

3. Results 

3.1 Open Radical Cystectomy analysis

Using the NCDB, we identified 26,183 patients who underwent radical cystectomy, of whom 16,814 (64.3%) received an open approach.  Baseline characteristics of the SLND trial cohort (n=300) and the NCDB open RC cohort are summarized in Table 1, with effect size comparisons reported in Table 2. Detailed pairwise comparisons between the SLND trial cohort and the NCDB open RC cohort are provided in Supplementary Table 1. Albeit statistically significantly different, the distribution of race was clinically comparable between groups (standardized difference: 0.13). Specifically, in the NCDB cohort, 1038 (6.2%) were Black patients vs. 12 (4%) in the SLND trial cohort. Focusing on other preoperative and clinical characteristics, patients in the SLND trial cohort were more likely to present with advanced clinical stage disease (≥cT3: 29% vs. 19%; OR 1.76, 95% CI 1.38–2.25), and to receive neoadjuvant systemic therapy (57% vs. 34%; OR 2.57, 95% CI 2.02–3.27), compared to the NCDB open RC cohort. Additionally, NCDB patients had a greater proportion of more advanced disease on final pathology (standardized difference: 0.28). Specifically, patients in the NCDB group had higher proportion of pT≥2 N0 (51% vs. 37%) and any T N+ BCa (27% vs. 24%), compared to SLND trial group. Notably, there was a remarkable disparity in the number of lymph nodes removed at the time of RC between the two cohorts; the median (range) number of nodes removed at was 14 (1–22) vs. 24 (6–61) for NCDB vs. SLND trial cohort, respectively. Similarly, only 20.0% of the patients in the NCDB group had ≥25 nodes removed vs. 50% of the patients in the SLND trial cohort (OR 4.91, 95% CI 3.86–6.25).

Baseline characteristics of the ELND trial cohort (n=292) and the NCDB open RC cohort are summarized in Table 1, with effect size comparisons reported in Table 2. Detailed pairwise comparisons between the ELND trial cohort and the NCDB open RC cohort are provided in Supplementary Table 2. Differences in sociodemographic characteristics were modest, although patients in the ELND trial cohort were more likely to be male (81% vs. 73%; OR 1.59, 95% CI 1.18–2.14). The distribution of race showed moderate imbalance (standardized difference 0.15), with a lower proportion of Black patients in the trial cohort. Consistent with the SLND comparison, substantial imbalance was also observed in pathological stage distribution (standardized difference: 0.30).  Again, in accordance with the previous comparison, a relevant disparity in the number of lymph nodes removed at the time of RC between the two cohorts; the median (range) number of nodes removed at was 14 (1–22) vs. 39 (15–94) for NCDB vs. ELND trial cohort, respectively. Similarly, only 5% of the patients in the NCDB group had ≥40 nodes removed vs. 50% of patients in the ELND trial cohort (OR 18.9, 95% CI 14.4–24.9).

3.2 Robot-assisted Radical Cystectomy analysis

Among the patients we included using the NCDB, 9369 (35.7%) underwent RARC. Baseline characteristics of the SLND trial cohort (n=300) and the NCDB robot RC cohort are summarized in Table 1, with effect size comparisons reported in Table 2. Detailed pairwise comparisons between the SLND trial cohort and the NCDB robotic RC cohort are provided in Supplementary Table 3. No significant differences were detected in terms of sociodemographic variables, including sex (OR:1.26 95% CI: 0.94–1.68) and self-reported race (standardized difference: 0.10), between the two groups. Focusing on other preoperative and clinical characteristics, patients in the SLND trial cohort were more likely to present with advanced clinical stage disease (≥cT3: 29% vs. 16%; OR 2.26, 95% CI 1.76–2.90), and to receive neoadjuvant systemic therapy (57% vs. 45%; OR 1.63, 95% CI 1.27–2.08), compared to the NCDB robotic RC cohort. Additionally, NCDB patients had a greater proportion of more advanced disease on final pathology (standardized difference: 0.18). Specifically, patients in the NCDB group had higher proportion of pT≥2 N0 (47% vs. 37%) and any T N+ BCa (25% vs. 24%), compared to SLND trial group. In line with what we observed for patients undergoing open RC, a remarkable disparity in the number of lymph nodes removed at the time of RC between the two cohorts; the median (range) number of nodes removed at surgery was 17 (1–25) vs. 24 (6–61) for NCDB vs. SLND trial cohort, respectively. Similarly, only 27.0% of the patients in the NCDB group had ≥25 nodes removed vs. 50% of patients in the SLND trial cohort (OR 2.71, 95% CI 2.12–3.46).

Baseline characteristics of the ELND trial cohort (n=292) and the NCDB robot RC cohort are summarized in Table 1, with effect size comparisons reported in Table 2. Pairwise comparisons between the SLND trial cohort and the NCDB robotic RC cohort are detailed in Supplementary Table 4. Differences in sociodemographic characteristics were modest, although patients in the ELND trial cohort were more likely to be male (81% vs. 74%; OR 1.50, 95% CI 1.11–2.03). The distribution of race showed moderate imbalance (standardized difference 0.14), with a lower proportion of Black patients in the trial cohort. Coincident with the SLND comparison, substantial imbalance was also found in pathological stage distribution (standardized difference: 0.20).  Again, in accordance with the previous comparison, a relevant disparity in the number of lymph nodes removed at the time of RC between the two cohorts; the median (range) number of nodes removed at RC was 17 (1–25) vs. 39 (15–94) for NCDB vs. ELND trial cohort, respectively. Similarly, only 6% on the patients in the NCDB group had ≥40 nodes removed vs. 50% of patients in the ELND trial cohort (OR 15.5, 95% CI 11.8–20.4).

4. Discussion

As mentioned earlier, the optimal extent of lymph node dissection in the setting of RC as primary local treatment for MIBC has historically been a matter of debate ^3,4. On one hand, most of the available literature consists of retrospective studies, thus providing heterogeneous and often difficult-to-interpret results ⁵. On the other hand, although a randomized controlled trial published in 2019 showed no oncological benefit of ELND over SLND, its significant methodological limitations reduced its generalizability ⁹. In this context, the recently published results of the SWOG S1011 trial, seem to confirm this trend, providing level I evidence for the lack of difference in disease-free and overall survival between ELND and SLND ¹⁰. Consequently, since the generalizability of the foregoing trial has never been tested in the actual clinical practice, our aim was to test the external validity of the SWOG S1011 trial to a North American population using the NCDB cohort. Several important observations arise from this study. First, among the 26,183 patients included from the NCDB who underwent RC between 2010–2022, 16814 (64.3%) and 9369 (35.7%) underwent open RC and RARC, respectively. Notably, given that the adoption of RARC has increased rapidly since 2010 ¹⁴, the observed rate of patients undergoing robotic surgery for MIBC in our cohort is fully consistent with that reported in the most recent studies, which hovers around 40% ¹⁴. Consequently, it should be emphasized that the inclusion of only patients undergoing open RC in the SWOG S1011 trial may limit its generalizability to current clinical practice, where more than one in three patients with MIBC receive RARC as their primary surgical treatment. Second, focusing on sociodemographic variables, the median observed ages in both the open RC and RARC NCDB cohort were totally comparable with those reported in the trial population. This represents a noteworthy element for which the authors of the SWOG S1011 trial should be commended, as a common issue in many oncological trials is the underrepresentation of older patients, despite their disproportionately higher burden of disease ¹⁶. In addition, albeit statistically significant, race distribution was not clinically significantly different between trial and NCDB cohorts. Specifically, in both cases, White patients accounted for more than 90% of the total study population. Consequently, given the relevance of age and race as key determinants of oncological outcomes among sociodemographic characteristics, the trial succeeded in enrolling a patient population that is virtually representative of those encountered in real-world clinical practice. Third, while the SWOG S1011 investigators were able to include a cohort of patients which is representative of the real-world setting in terms of sociodemographic characteristics, the same does not hold true of the clinical and pathological features of the enrolled population. Indeed, although patients included in the open RC and RARC NCDB cohorts were more likely to present with clinically localized disease at preoperative staging, they ultimately harbored more locally advanced disease on final pathology. In the trial study population (both SLND and ELND subgroups) 29% of patients had cT3–4a BCa, compared to 19% and 16% of patients in the open RC NCDB and RARC NCDB cohorts, respectively. With regard to pathological T stage, only 37% of the entire trial cohort had pT≥2 N0, compared to 51% and 47% in the open RC NCDB and RARC NCDB cohorts, respectively. This discrepancy appears to be largely attributable to the significant difference in the proportion of patients who received neoadjuvant systemic treatment between the real-world cohort and the trial cohort, where the administration of preoperative systemic therapy was considerably more frequent. Specifically, while 57% of the trial population received neoadjuvant treatment, only 34% and 45% of individuals in the open RC NCDB cohort and the RARC NCDB, respectively, underwent systemic therapy before surgery. It is important to highlight that, although the rates of neoadjuvant systemic therapy utilization observed in the NCDB cohorts are higher than those historically reported in the literature ^17–20, they remain significantly lower compared to those in the trial cohort. It is well established that cisplatin-based neoadjuvant chemotherapy yields a 30–40% pathological complete response rate at the time of radical cystectomy ²¹, providing an absolute 5-year overall survival benefit of approximately 5–8% (HR 0.86) compared with surgery alone ²². Nevertheless, despite this level I evidence supporting its use and the recommendations issued by prevailing international guidelines ^1,2, neoadjuvant systemic treatment has historically been underutilized in this setting ^17–20. Moreover, it is also possible that the high rate of neoadjuvant treatment utilization may have limited the ability to detect survival differences between SLND and ELND. Some evidence suggests that patients undergoing NAC, even when diagnosed with clinically positive disease, may derive limited benefit from LND during radical cystectomy ²³. In this context, previous studies have shown that an adequate LND (>10 lymph nodes removed) was associated with an improved overall survival among patients who did not receive preoperative treatment, but not among those who underwent neoadjuvant therapy ²⁴. Lastly, there is a remarkable disparity in the number of lymph nodes removed at the time of RC between the real-world NCDB cohorts and the trial population. In SLND and ELND cohorts of the SWOG S1011 trial, the median (range) number of nodes removed was 24 (6–61) and 39 (15–94) vs. 14 (1–22) and 17 (1–25) in the open RC NCDB and RARC NCDB cohorts, respectively. This aspect carries several noteworthy implications. Firstly, the number of lymph nodes removed in the SLND group was significantly higher not only compared to that reported in our population-based cohorts, but also in comparison with most previous literature ^7,8,25,26. In a meta-analysis conducted by Bi et al., aimed at evaluating the potential benefit of ELND over SLND in terms of recurrence-free survival, the authors included over 2,800 patients who underwent RC for MIBC across six different studies ²⁷. In the subgroup of patients undergoing SLND, the median number of nodes removed ranged from 8 to 17. Moreover, considering that the previous randomized controlled trial ⁹ on the same topic was criticized due to the high number of lymph nodes removed in both arms (19 vs. 31), which may have hindered the detection of survival differences ⁵, it is reasonable to assume that such a limitation might be even more pronounced in the SWOG S1011 trial. Indeed, despite the retrospective nature and heterogeneity of the available evidence, the majority of previous studies evaluating the optimal extent of LND for MIBC have consistently shown that a survival benefit is most likely when between 9 and 16 lymph nodes are removed ^28–30. Likewise, a recent study by Sodagum et al., including more than 8,000 patients with BCa who underwent LND, demonstrated a significant cancer-specific survival benefit with the removal of at least 11 nodes ⁵. Ultimately, given the markedly high number of nodes removed in the SLND arm of the trial, it seems reasonable to interpret with great caution the absence of significant survival differences according to the extent of LND reported in the SWOG S1011 trial. As such, the generalizability of the trial’s findings warrants further investigation. Finally, a key methodological consideration is that the NCDB does not provide granular information on the anatomical extent of lymph node dissection, thereby precluding a reliable distinction between standard and extended templates. While lymph node counts are sometimes used as a surrogate, this approach is inherently limited and highly susceptible to confounding. As such, comparative survival analyses between SLND and ELND in the NCDB setting would be prone to substantial bias and should be interpreted with extreme caution. Our study is not without limitations. Despite methodology consistent with the literature ^31,32, our study must be interpreted within the limitations of a retrospective design. While every attempt was made to use the same inclusionary and exclusionary criteria, the NCDB cohort may only partially represent the true U.S. population and does not contain all the information used as criteria in the initial study. For instance, only patients with a Zubrod performance status score of 0, 1, or 2 were included in the trial. Although we exclusively included individuals with a CCI of 0–1 in an attempt to exclude patients with poor performance status, this estimation may not be entirely accurate. Moreover, in contrast to the trial, the NCDB does not provide detailed information regarding the specific neoadjuvant agents administered preoperatively. Finally, it should be emphasized that a key limitation of this study is the use of lymph node count as a proxy for dissection extent. The lower node counts observed in the NCDB may reflect differences in surgical extent, pathological processing, or both, thereby limiting direct comparability with trial-defined anatomical templates.

5. Conclusion 

The results of our study suggest that while the SWOG S1011 randomized controlled trial provides high-level evidence on the lack of oncological benefit for ELND compared to SLND in the treatment of MIBC, its external validity to real-world clinical practice is only partially confirmed. In particular, although the trial population was representative in terms of key sociodemographic variables such as age and race, significant differences emerged when comparing clinical, pathological, and treatment-related characteristics with those of a large North American cohort. Notably, patients in the real-world NCDB cohorts were less likely to receive neoadjuvant systemic treatment and to undergo extensive LND, both in the open and robotic-assisted surgery groups. These discrepancies may have influenced the trial’s ability to detect a survival benefit associated with ELND, and they underscore the challenges in applying trial-based evidence to heterogeneous patient populations in routine clinical settings. In light of our findings, caution should be exercised when extrapolating the conclusions of the SWOG S1011 trial to contemporary surgical practice. Further research, potentially incorporating real-world data and prospective registry-based studies, may be warranted to better define the role of ELND in the context of evolving multimodal treatment strategies for MIBC.