- Top of page
- Conflict of interest
Objective: To identify lymph node density thresholds and their prognostic role in patients who underwent radical cystectomy and pelvic lymph node dissection, and to validate findings in an external series.
Methods: Between May 2001 and September 2009, data from 750 radical cystectomies carried out at “Regina Elena” National Cancer Institute (Rome, Italy) were collected in a prospectively-maintained database. Once patients who had undergone neoadjuvant treatments and those who had undergone salvage radical cystectomy were excluded from the 210 pN+ patients, 156 patients with urothelial carcinoma were selected for analysis. Optimal cut-off points for age, lymph node count and lymph node density were identified by considering these variables as continuous. External validation of findings was carried out by using data of 154 pN+ patients selected from two prospective series.
Results: The optimal identified cut-off points were 11% and 30% for lymph node density, nine and 30 nodes for lymph node count, and 73 years for age. Median cancer-specific survival of patients were significantly different in patients with lymph node density <12%, between 12% and 30%, and >30% (71 months, 24 months and 11 months, respectively; P < 0.001). Cancer-specific survival was independently predicted by lymph node density cut-off points (12–30% vs <12%: hazard ratio 1.51, P = 0.047; >30% vs <12%: hazard ratio 2.89, P < 0.001). In the external series, the prognostic effect of lymph node density according to tertiary distribution of risk based on these lymph node density cut-off points was confirmed at Cox multivariable analysis (12–30% vs <12%: hazard ratio 1.5, P = 0.048; >30% vs <12%: hazard ratio 2.5, P = 0.004).
Conclusions: Lymph node density is the strongest predictor of cancer-specific survival. Identified lymph node density thresholds have shown to be independent predictors of cancer-specific survival in the external validation series.
- Top of page
- Conflict of interest
RC and pelvic LND are the standard treatments for muscle-invasive bladder cancer.
Although the optimal extent of LND remains to be defined, progression-free survival and overall survival have been correlated with the number of nodes removed during surgery.1
The 2009 edition of the TNM staging system considers pN1 as metastasis in a single node in the true pelvis (hypogastric, obturator, external iliac or presacral), pN2 as metastasis in multiple nodes in the true pelvis and pN3 as metastasis in a common iliac node.2
In a recent prospective nonrandomized study aimed at identifying the therapeutic role of LND extent, Abol-Enein et al. found a significant survival benefit in node-positive patients who underwent super-extended LND versus those who underwent a standard LND. The authors have also shown prospectively that positive nodes in bladder cancer patients are not found outside the pelvis if the pelvic nodes are free of tumor.3
In 2003, Herr4 and Stein et al.,5 in two different reports, had already highlighted the superiority of LN-d, the ratio of positive lymph nodes to the total number of lymph nodes removed, over the pN stage in predicting CSS. This superiority is based on the “quality” and on the extent of LND, expressed by the LN-c and by the number of positive nodes, two variables that are missing from the TNM staging system and that proved to have a prognostic role.6
To date, no evidence exists about the minimum LN-c, with reports suggesting a range from eight to 16 lymph nodes.7
Despite the lack of strong evidence, Karl et al. recommended removal of 20 lymph nodes to stage tumors appropriately.8
With regard to LN-d, a threshold was not clearly defined until now, with very variable cut-off points ranging from 4% to 25%.9,10
The purpose of the present study was to identify optimal LN-d cut-off points in a single center prospective series of 156 node-positive patients, to identify independent predictors of CSS and to validate findings in an external cohort.
- Top of page
- Conflict of interest
Internal and external series were comparable for LND template, indications to adjuvant treatments, demographic and all pathological data, except for LN-c (Table 1).
Table 1. Baseline, pathological and follow-up data of each center
|Variable||Internal series||External series (center 1)||External series (center 2)|| P-value|
|Mean age (SD)||67.3 (8.3)||65 (9.9)||66.6 (8.3)||0.427|
|Sex (%)|| || || ||0.9|
| Male||129 (82.7)||48 (84)||81 (83.5)|
| Female||27 (17.3)||9 (16)||16 (16.5)|
|Mean follow up, months (SD)||28.9 (28.7)||27.9 (27.4)||29.3 (27.8)||0.9|
|Urinary diversion (%)|| || || ||0.432|
| Ileal conduit||66||28||42|
| Continent cutaneous||3||–||2|
|pT stage (%)|| || || ||0.052|
| 0-a-is-1||11 (7)||12 (21.1)||6 (6.2)|
| 2||18 (11.5)||8 (14)||11 (11.3)|
| 3||82 (52.6)||25 (43.8)||54 (55.7)|
| 4||45 (28.9)||12 (21.1)||26 (26.8)|
|pN (2009) stage (%)|| || || ||0.124|
| 1||35 (22.4)||12 (21)||28 (28.9)|
| 2||85 (54.5)||36 (63.1)||41 (42.2)|
| 3||36 (23.1)||9 (15.8)||28 (28.9)|
|Mean LN-c (SD)||28 (16.1)||22.61 (10.2)||20.6 (11.1)||0.02|
|Mean LN-d (SD)||28% (26.8)||26% (26.2)||27.7% (27.9)||0.89|
|LND template (%)|| || || ||0.416|
| Extended||61 (39.1)||20 (35.1)||30 (30.9)|
| Standard||95 (60.9)||37 (64.9)||67 (60.1)|
|Adjuvant treatments (%)|| || || ||0.836|
| Systemic chemotherapy||43 (28)||24 (40)||26 (27)|
| Radiation therapy||3 (2)||2 (3)||3 (3)|
At a median time to event of 24 months, 93 of 156 patients had died of disease. Five-year CSS was 24%.
Figure 1 shows HR >4 for each LN-d point between 11% and 35%, with the highest value at 23% (HR 5.2). The optimal tertiary distribution of patients at Cox univariable analysis was obtained by grouping patients with LN-d between 12% and 30% in a single risk class.
At log–rank analysis, age (P = 0.006), pT (P = 0.007), LN-d (P < 0.001), LN-c (P = 0.024), pN (P = 0.011) and LND template (P < 0.001) had significant impacts on CSS (Fig. 2).
Figure 2. Univariable analysis on the internal series of 156 pN+ patients. (a) CSS curves stratified by age. (b) CSS curves stratified by pathological tumor (pT) stage. (c) CSS curves stratified by LN-d. (d) CSS curves stratified by LN-c. (e) CSS curves stratified by pathological nodal stage (according to 2009 TNM). (f) CSS curves stratified by LND (LND template). (a) Age , 46–72; , 73–86. (b) pT group: , ≤2; , ≥3. (c) LN-d: , 2–11%; , 12–30%; , 31–100%. (d) LN-c: , 6–9; , 10–30; , 31–90. (e) pN2009: , 1; , 2; , 3. (f) LND Template: , Extended; , Standard.
Download figure to PowerPoint
At multivariable Cox regression analysis, a standard LND (vs extended template: P = 0.016; HR 1.91, 95% CI 1.12–3.23) and LN-d ≥12% (12–30% vs <12%: P = 0.047, HR 1.51, 95% CI 1.05–2.66; >30% vs <12%: P < 0.001, HR 2.89, 95% CI 1.67–5.01) were independent predictors of lower CSS probability.
Results of univariable and multivariable analyses are summarized in Table 2.
Table 2. Univariable and multivariable analyses on the internal series
|Variable||Patients (156)||Cancer-related deaths (93)||Log–rank||Cox multivariable|
|Median CSS (months)||95% CI|| P ||HR||95% CI|| P |
|Age|| || || || || || || || |
| ≤72 years||113||62||27||14.9–39||0.006||1.51||0.98–2||0.055|
| ≥73 years||43||31||15||4.1–25.8|
|Sex (%)|| || || || || || || || |
|Urinary diversion|| || || || || || || || |
| Ileal conduit||66||42||23||8.1–73.5||–||–||–|
| Continent pouch||3||2||19||n.e.||–||–||–|
|pT stage|| || || || || || || || |
| pT ≤2||29||10||n.e.||–||0.007||1.45||0.95–1.9||0.12|
| pT >2||127||83||22||16–27.9|
|pN (2009)|| || || || || || || || |
| pN 1||35||21||43||28.6–57.3||0.011||–||–||–|
| pN 2||85||49||26||16.4–35.5||0.56||0.24–1.3||0.18|
| pN 3||36||29||17||10.4–23.5||0.73||0.3–1.8||0.49|
|pN (2002)|| || || || || || || || |
| pN 1||35||16||49||15–82||0.091||–||–||–|
| pN 2||121||77||23||15.9–30||–||–||–|
|LN-c|| || || || || || || || |
|LN-d|| || || || || || || || |
|LND template|| || || || || || || || |
|Adjuvant treatments|| || || || || || || || |
At multivariable analysis on the external validation cohort, pT stage (pT >2 vs pT ≤2: P = 0.009, HR 2.21, 95% CI 1.22–3.99) and the identified 12% and 30% LN-d cut-off points proved to be independent predictors of CSS (LN-d 12–30% vs LN-d ≤12%: P = 0.048, HR 1.5, 95% CI 1.05–2.66; LN-d >30% vs LN-d ≤12%: P = 0.004, HR 2.50, 95% CI 1.34–4.65).
Once patients with a single positive node were excluded, the prognostic role of LN-d was confirmed in 225 patients (pN2-pN3), with 4-year CSS of 4% for patients with LNd >30%, 20% for those having intermediate LN-d and 53.7% for those with LNd ≤11% (log–rank P < 0.001; Fig. 3a).
In contrast, pN1 and pN2 CSS curves of patients with LN-d <12% were mostly overlapping (log–rank P = 0.469; Fig. 3b).
- Top of page
- Conflict of interest
LN-d was introduced as a new concept in 2003 by Herr.4 In the same year, Stein et al. addressed the superiority of LN-d over pN status to stratify prognosis after RC.5 In these two studies,4,5 the authors identified 20% as the best LN-d cut-off, although median LN-c in these two experiences were very different, at 13 and 30, respectively.
In 2005, Fleischmann et al. identified LN-c (less than five nodes vs more than five nodes) and LN-d (cut-off of 20%) as predictors of CSS, but the results were not confirmed by multivariable analysis.15
In a 2007 review including five studies and 979 patients, Herr wondered if LN-d should have been introduced into daily clinical practice and he answered “no”; the author highlighted how LN-d would offer an attractive proxy measure of quality of care, but he faced the lack of multi-institutional validation series.16
Recently, in a single-center series of 181 patients receiving RC and LND up to the inferior mesenteric artery with one or two positive nodes, Bruins et al. found that 4% LN-d was able to independently predict recurrence-free survival.9 The cut-off of 4% is not reproducible by excluding patients with more than two positive nodes; however, Bruins et al. highlighted the potential therapeutic role of an extended LND (up or above the aortic bifurcation) in patients with low-volume metastatic-node disease, but it also shed light on the importance of extended LND in determining the clinical relevance of LN-d.
In confirmation of this concept, Jeong et al. reported 18% LN-d as the only variable with an independent role in predicting CSS of 130 pN+ patients.17 In that series, the median LN-c was 15, and the authors found that when a small number of nodes was removed, pN status was a better predictor than LN-d.
In the series published by Osawa et al., 60 pN+ patients were selected; with a median number of 12 nodes removed, the most informative LN-d cut-off was 25%.18
In 2008, Wright et al. published a population-based study, collecting data from 1260 patients in the Surveillance Epidemiology and End Results database.19
The authors found that the lowest risk of death was in the 0–12.5% LN-d quartile, and they found no interaction among number of lymph nodes, age, LN-c and LN-d. They concluded that the previously published threshold of 20% might be too high, because a 12.5% cut-off strongly predicts survival, even compared with the 12.5–25% interval.
In 2008, Wiesner et al. published a study on the extent of pelvic LND and the prognostic value of LN-d. The authors included 46 node-positive patients from a group of 152 patients in the study, and the LN-d cut-off identified was 11%, which correlates with previous considerations, as the median number of lymph nodes removed was 33.20
In a multicenter retrospective study, Kassouf et al. first showed an improved outcome prediction by categorizing LN-d according to a tertiary (<6%, 6–41%, >41%) instead of the usual binary distribution. Despite the large cohort of patients (1038 pN+), the authors acknowledged the lack of a prospective standardized template for LND across surgeons and centers as the main study drawback. In fact, LND was defined as “limited” or “more extensive” when <25 or ≥25 nodes, respectively, were removed. Besides, the identified cut-off points have been validated neither internally nor in an external cohort.21
Based on these reports, we identified two predictive LN-d cut-off points (11% and 30%) that proved to independently predict CSS; in order to minimize limitations of a non-standardized LND, lymph nodes were sent to the pathologist in separate packages and all patients underwent standard pelvic LND (up to iliac bifurcation; six nodal packages) or extended LND (up to aortic bifurcation; nine nodal packages).
This is the first external validation of internally developed LN-d cut-off points in prospective series of patients who underwent RC and standardized LND.
With regard to the optimal extent of LND, in the present series, although extent of LND did not influence prognostic accuracy of LN-d, an extended LND allowed an optimal staging, as shown by superiority of 2009 versus 2002 TNM staging system (P = 0.011 vs P = 0.091), and proved to have an independent therapeutic role in CSS versus a standard LND.
Regarding the clinical reliability of LN-d across pN stages, once pN 1 patients were excluded, where LN-d is directly related to the number of nodes removed, in a subgroup analysis including all 225 pN2 and pN3 patients, LN-d further stratified prognosis with 4-year CSS of 4% for patients with LN-d >30%, 20% for those having intermediate LN-d and 53.7% when LN-d was ≤11% (log–rank, P < 0.001; Fig. 3a).
Removing at least nine negative nodes for each positive node (LN-d <12%) had a clear therapeutic effect on CSS regardless of pN stage. In contrast, CSS of patients with LN-d <12% was comparable regardless of pN status, with the CSS curves of the pN1 and pN2 groups mostly overlapping (log–rank, P = 0.469; Fig. 3b). These subgroup analyses confirmed the higher prognostic power of LN-d over pN stage.
Finally, the need for an adequate LND template to apply LN-d as a predictor of outcome in node-positive patients is still under debate.
The “Will Rogers phenomenon” explains how pN0 patients are potentially untreated pN+ patients and, similarly, how low LN-d in patients who have undergone limited LND might underestimate nodal involvement and determine “stage migration”. This is the reason why a low LN-c and a limited LND impair the prognostic significance of LN-d, an issue highlighted by several of the reports cited earlier.8,9,17
In a recent report by Kassouf et al., LN-d proved to be a stronger prognostic factor in patients with LN-c ≥25 (HR 4.63, P < 0.0001) than in patients with lymph node count <25 (HR 1.62, P = 0.03).21 The lack of information about the extent of LND in that study, as in previous published studies, led the authors to use LN-c as a surrogate of LND extent.
Based on these reports, despite the clear effect of LN-c on LN-d, the question “if LNd could be influenced by LND template” cannot be answered.
The interobserver variability in determining LN-c was confirmed by the present study; despite a homogeneous distribution of baseline and pathological data, as well as comparable extents of LND across centers, LN-c was the only variable not homogeneously distributed (P = 0.02; Table 1).
It reflects the unreliability of LN-c as a surrogate of LND extent; in order to overcome the negative impact of low LN-c on LN-d, in the present study all patients underwent a standardized LND (extended or standard with separate packages technique) and LN-d independently predicted CSS in both the internal series and the external series.
Despite efforts to reduce intrinsic selection biases of non-randomized studies, such as prospectively-maintained databases and a standardized technique of LND (separate packages), the intrinsic intersurgeon variability of LND technique, the arbitrary choice of LND extent and the lack of a central pathologist have to be acknowledged as limitations.
Besides, the use of neoadjuvant treatments, which has progressively become a standard procedure in recent years, will require further validation of these LN-d cut-off points in series of patients who underwent neoadjuvant chemotherapy and radical cystectomy with standardized LND.
LN-d was confirmed to be the strongest variable for outcome prediction in patients with nodal metastases who underwent RC and LND.
The identified 11% and 30% LN-d cut-off points were validated in a prospective external cohort, and can be used to optimize risk stratification and to identify candidates for adjuvant treatments.