Early (sentinel lymph node biopsy-guided) versus delayed lymphadenectomy in melanoma patients with lymph node metastases

Personal Experience and Literature Meta-Analysis


  • See editorial on pages 1150-2, this issue.

  • We thank Dr. Loris Bertazza for his help with figures.



It is debated whether patients with melanoma who undergo lymphadenectomy after a positive sentinel lymph node (SN) biopsy (SNB) have a better prognosis compared with patients who are treated for clinically evident disease.


The records of 190 patients with cutaneous melanoma who underwent radical lymph node dissection after a positive SNB (completion lymph node dissection [CLND]; n = 100) or who had clinically evident lymph node metastasis (therapeutic lymph node dissection [TLND]; n = 90) were analyzed. Moreover, the MEDLINE, EMBASE, and Cochrane databases were searched for studies that investigated the survival impact of SNB-guided CLND compared with TLND for clinically evident disease. Standard meta-analysis methods were used to calculate the overall treatment effect across eligible studies.


In the authors' series, tumor characteristics did not differ significantly between patients who underwent CLND and those who underwent TLND. After a median follow-up of 52.6 months, the 5-year overall survival rate did not differ significantly between CLND patients and TLND patients (68.9% vs 50.4%, respectively; log-rank test; P = .17). In contrast, a meta-analysis of 6 studies (n = 2633) that addressed this issue (including the authors' own series) indicated that there was a significantly higher risk of death for patients who underwent TLND compared with that for patients who underwent CLND (hazard ratio, 1.60; 95% confidence interval, 1.28-2.00; P < .0001).


Although no significant survival difference was observed in either series, the pooling of summary data from all the studies that dealt with this issue suggested that SNB-guided CLND is associated with a significantly better outcome compared with TLND for clinically evident lymph node disease. Cancer 2010. © 2010 American Cancer Society.

Although regional lymph node status is the most important prognostic factor in patients with cutaneous melanoma, the impact of the timing of radical lymph node dissection (RLND) on survival still is debated.1, 2 Results from randomized controlled trials (RCTs) of prophylactic versus therapeutic RLND have failed to demonstrate a clear survival advantage of either approach in patients with clinical stage I and II melanoma.3, 4

Since its introduction into clinical practice,5 sentinel lymph node (SN) biopsy (SNB) has become a widely accepted procedure for predicting the status of regional lymph nodes.2, 6 This minimally invasive procedure is characterized by a high negative predictive value (when the SN is negative, the lymph node basin is truly negative in approximately 95% of patients).7 Consequently, after a positive SNB, surgeons can perform completion RLND selectively in patients who harbor metastatic disease in the regional lymphatic basin while sparing SN-negative patients the morbidity associated with lymphadenectomy. Moreover, SNB can detect subclinical metastatic deposits, which allows them to undergo RLND earlier in the natural history of the disease (on average, 16 months before clinically evident metastatic disease8). However, the question of whether this anticipated therapeutic intervention translates into a significant survival benefit remains unanswered. In this regard, the findings of retrospective series are conflicting.9-18 Moreover, the third interim analysis of Multicenter Selective Lymphadenectomy Trial 1 (MSLT-1),8 which, to our knowledge, is the only RCT on this subject for which the results are available, failed to demonstrate an overall survival (OS) advantage for patients who underwent SNB compared with those who underwent observation of the lymph node basins (only a disease-free survival benefit was observed). It is noteworthy that, in a subgroup analysis, the OS for patients who had a positive SN and underwent subsequent early RLND (completion lymph node dissection [CLND]) was significantly longer than that for patients in the observation arm who developed clinically detectable lymph node metastases and, thus, underwent delayed RLND (therapeutic lymph node dissection [TLND]).

In the current study, we retrospectively compared the OS of our patients who underwent either CLND after a positive SNB or TLND for clinically evident disease in regional lymph nodes. Our objective was to compare the therapeutic effect of early lymphadenectomy versus delayed lymphadenectomy on the natural history of the disease by analyzing the survival of patients who underwent treatment (ie, radical lymph node dissection) early (ie, when the disease was microscopic; that is, clinically undetectable and detectable only by SNB) versus the survival of patients who underwent the same treatment later (ie, when the disease was macroscopic, as in of patients who developed clinically detectable disease either without undergoing SNB or after a false-negative SNB). Moreover, we conducted a systematic review of published series that addressed this issue and performed a meta-analysis of available data to investigate whether patients with lymph node metastases who underwent early lymphadenectomy after SNB had different a clinical outcome compared with patients who underwent TLND.


Personal Series

We reviewed the medical records of patients with cutaneous melanoma who underwent RLND at our institution from 1990 through 2006. We identified 2 patient groups.

In the early RLND (CLND) group, patients had negative lymph node status after both a physical examination and an ultrasound (US) scan at the time of their primary melanoma diagnosis; moreover, these patients had no distant metastases detected with chest x-ray or abdominal US studies. Patients who had a positive SNB underwent CLND to complete sentinel lymphadenectomy.

In the delayed RLND (TLND) group, patients were lymph node negative after both a physical examination and an US scan at the time of their primary melanoma diagnosis; moreover, these patients had no distant metastases detected with chest x-ray or abdominal US studies. Patients were followed with physical examinations and US scans of the regional lymph node field, and fine-needle aspiration cytology was performed when palpable lymph nodes or suspicious US patterns were detected. In this case, if lymph node metastases were confirmed cytologically, then the patients underwent RLND. Patients were considered eligible for the study if the anthropometric variables and the pathologic slides of primary melanoma and SN specimen were available.

Lymphatic mapping, surgical procedures, and pathologic procedures are described in detail elsewhere.19, 20 In particular, with regard to pathologic evaluations, SNs were cut through their longest axis into 2-mm to 3-mm thick slices, embedded entirely in paraffin blocks, and 10 consecutive sections were prepared from each block and stained with hematoxylin and eosin. Immunohistochemistry for S100 and HMB45 was performed in all specimens. Surgical specimens from completion lymphadenectomy were processed routinely: Lymph nodes with a greatest dimension ≤4 mm were totally embedded, lymph nodes that measured >4 mm in greatest dimension were cut into 3-mm to 4-mm thick slices, and these slices were embedded entirely in paraffin blocks. From each paraffin block, 2 levels were obtained for hematoxylin and eosin staining; and, in suspicious cases, additional immunohistochemical staining was performed.

Systematic Review

Data were extracted by 2 investigators (S.M. and S.P.) to ensure homogeneity of data collection and to rule out the effect of subjectivity in data gathering and entry. Disagreements were resolved by iteration, discussion, and consensus. To unravel potential systematic biases, a third investigator (C.R.R.) conducted a concordance study by independently reviewing the eligible studies: Complete concordance was reached for all variables that were assessed.

A systematic review of original articles that analyzed the therapeutic value of early RLND in patients with melanoma who had lymph node metastases was done by searching the MEDLINE, EMBASE, and Cochrane databases. The search strategy included the following keywords variously combined: “melanoma,” “sentinel lymph node,” “survival,” “completion lymph node dissection,” and “therapeutic lymph node dissection.” We sought original and review articles that were published up to February 2009, because we considered the latter as an additional source of original works that had been overlooked otherwise. When appropriate, cited references from original and review articles also were used as an additional source of relevant studies.

Inclusion criteria were the following: 1) only patients with lymph node metastases could be enrolled; 2) patients must have undergone CLND and TLND after a positive SNB or after they were diagnosed with clinically detectable regional lymphatic disease, respectively; and 3) the survival data necessary for meta-analyses (hazard ratios [HR] and confidence intervals [CI] or Kaplan-Meier survival curves) had to be reported. Then, the quality of each eligible study was assessed according to Methodological Index for Non-Randomized Studies (MINORS) criteria21: according to this method, the studies were scored based on 12 quality-related parameters, and the total score ranges between 0 and 24.

Statistical Analysis

In our series, the following variables were compared using the chi-square test or the Fisher exact test for categorical variables and the Mann-Whitney U test for continuous variables to asses the comparability of the CLND group and the TLND group: age at primary melanoma diagnosis (continuous variable), sex (men and women), tumor site (extremities, trunk, and head and neck), Breslow thickness (continuous variable), Clark level (III, IV, and V), ulceration (absent or present), the number of positive lymph nodes after both SNB and CLND or TLND (continuous variable), and extracapsular extension (absent or present).

To determine whether there was a difference in survival between patients who underwent CLND or TLND, OS was considered the interval between the date of primary melanoma excision and the date of last follow-up or death (melanoma-specific death was considered). The Kaplan-Meier method was used to calculate the survival estimates, and log-rank test was used to detect survival differences between the CLND and TLND subgroups. Multivariate survival analysis was performed using the Cox proportional hazards model to assess the impact of patient, tumor, and lymph node features on the type of RLND performed (CLND vs TLND).

Meta-analysis methods, which are described in detail elsewhere,22 were applied to evaluate the overall impact of CLND after a positive SNB on patients' OS.23, 24 TLND was considered the “standard” treatment, whereas CLND was considered the “investigational” treatment; accordingly, the HRs always are expressed as TLND-to-CLND ratios. For studies that did not report HRs from Cox univariate or multivariate analysis, HRs and 95% CIs were calculated from Kaplan-Meier survival curves by adopting a hierarchical series of steps described by Tierney et al and Parmar et al.25, 26 The meta-analysis was performed using a fixed-effects model. The consistency of results (effect sizes) among studies was investigated by means of 2 heterogeneity tests, the 2-based Cochran Q test and the I2 statistic ([Qdf]/Q × 100, where Q is the Cochran statistic, and df is the number of studies −1. In this analysis, we considered that heterogeneity was present when the P value of the Cochran Q test was <0.1. In addition, inconsistency across studies was quantified by means of the I2 statistic, which generally is considered significant (ie, heterogeneity has a significant impact on meta-analysis) for values of 50%. A funnel plot was created to ascertain publication bias according to the Begg rank-correlation test and the Egger regression intercept.

The extent to which the combined risk estimate may have been affected by individual “dominant” studies was assessed by consecutively omitting every study from the meta-analysis (leave-1-out procedure). Furthermore, to identify potential relations between the study quality score and the HR estimate, random-effects metaregression also was implemented.

Results were considered significant with P values <.05. All analyses were performed using the SPSS statistical package (release version 13.0; SPSS Inc., Chicago, Ill) and with Comprehensive Meta-Analysis (release version 2.2.046; Biostat, Englewood, NJ).


Personal Experience

Among 253 patients who underwent RLND at our institution, 190 patients met all the considered inclusion criteria, including 90 patients who underwent TLND for clinically evident disease and 100 patients who underwent CLND after a positive SNB (Fig. 1). Among the patients who underwent TLND, lymph node metastases were detected by physical examination in 59 patients (65.5%), and by US scan in 31 patients (34.5%). Moreover, 63 patients did not undergo SNB (21 patients were treated at our institution before the SNB procedure was available, and the remaining 42 patients underwent primary melanoma excision in other centers where SNB was not performed routinely and came to our attention with lymph node metastases), whereas the remaining 27 patients developed regional lymph node metastases although they had a negative SNB (false-negative SNB).

Figure 1.

This flow chart illustrates patients grouping for the comparison of early dissection (completion lymph node dissection after positive sentinel lymph node biopsy [SNB]) versus delayed dissection (therapeutic lymph node dissection for clinically evident lymph node metastasis). US indicates ultrasound; SN, sentinel lymph node; FNAC, fine-needle aspiration cytology; RLND, radical lymph node dissection.

After they were diagnosed with locoregional lymph node metastasis, all patients underwent RLND. Five patients underwent laterocervical LND, 72 patients underwent 3-level axillary LND, 106 patients underwent iliac-obturatoryinguinal LND, and 7 patients underwent double-basin LND for synchronous metastatic disease to 2 locoregional lymphatic basins.

Patients and the distribution of tumor characteristics in the CLND and TLND groups are reported in Table 1. Among patient characteristics, the median age was significantly older in the TLND group (P = .03); however, the primary tumor characteristics did not differ significantly between the 2 groups. The number of positive lymph nodes (P < .0001) and the presence of extracapsular extension (P < .0001) were more frequent in the TLND group.

Table 1. Clinicopathologic Features of 190 Patients With Melanoma Who Underwent Radical Lymph Node Dissection Either for a Positive Sentinel Lymph Node (Completion Lymph Node Dissection; n=100) or for Clinically Evident Lymph Node Metastases (Therapeutic Lymph Node Dissection; n=90)
VariableCLND, n=100TLND, n=90P
  1. CLND indicates completion lymph node dissection; TLND, therapeutic lymph node dissection.

Age, y    .03
 Median53 61  
 Range19-76 22-92  
Sex    .56
Site    .53
 Head and neck4433 
Breslow thickness, mm  .43  
Clark level    .77
Ulceration    .08
No. of positive lymph nodes  <.0001  
Extracapsular extension    <.0001

The mean interval between primary tumor excision and CLND or TLND was 3.5 months and 20 months, respectively. The median follow-up was 52.6 months for all patients (range, 7-213 months), 50.1 months for the CLND group (range, 7-159 months), and 56.4 months for the TLND group (range, 12-213 months).

The median survival was 90.5 months (95% CI, 48-133 months) for the entire population, 119.3 months (95% CI, 65.8-172.8 months) for the CLND group, and 62.5 months (95% CI, 30.2-94.8 months) for the TLND group. For the whole series, the 5-year OS rate was 58.3%. Despite the different median OS, 5-year survival rates did not differ significantly between the CLND group and the TLND group (68.9% vs 50.4%, respectively; P = .17; log-rank test) (Fig. 2). Among the patients who had tumors with a Breslow thickness between 1.0 mm and 4.0 mm (the pathologic T2 [pT2] and pT3 subgroups, n = 116; CLND, n = 58; TLND, n = 58), we observed no significant differences in OS between CLND patients and TLND patients (5-year OS: 71% vs 57.8%, respectively; P = .49; log-rank test). Univariate and multivariate Cox survival analyses are reported in Table 2.

Figure 2.

Overall survival curves are shown according to the type of lymphadenectomy performed considering the whole caseload (P = .17). CLND indicates completion lymph node dissection; TLND, therapeutic lymph node dissection.

Table 2. Univariate and Multivariate Cox Regression Analysis of Overall Survival in 190 Patients With Melanoma Who Underwent Radical Lymph Node Dissection Either for a Positive Sentinel Lymph Node (Completion Lymph Node Dissection; n=100) or for Clinically Evident Lymph Node Metastases (Therapeutic Lymph Node Dissection; n=90)
VariableUnivariate AnalysisMultivariate Analysis
  1. HR indicates hazard ratio; CI, confidence interval; CLND, completion lymph node dissection.

Age, y1.011.002-1.03.0231.010.99-1.02.24
Sex: Men1.250.81-
Breslow thickness, mm1.081.04-1.12<.00011.081.02-1.14.004
Clark level      
Ulceration: Absent0.550.36-0.85.0080.660.42-1.04.076
No. of positive lymph nodes1.041.01-1.09\.0291.020.98-1.07.25
Extracapsular extension: Absent0.520.33-0.88.0060.770.41-1.44.48
Type of lymphadenectomy: CLND0.740.48-1.14.1770.930.54-1.59.79

Systematic Review and Meta-Analysis

A literature search with the keywords described above (see Systematic Review) yielded 355 articles; of these, 38 articles that reported on the impact of early lymphadenectomy in patients with melanoma were reviewed in detail (Fig. 3). Excluding reviews and articles that compared SNB with observation, 7 original articles met all inclusion criteria.8-12, 14, 17 Two of those articles were excluded: In 1 study,14 the series partially overlapped the multicenter German report10; and, in the other study, only patients who underwent SNB were considered.17

Figure 3.

This is a systematic flow chart of the current review. CLND indicates completion lymph node dissection; TLND, therapeutic lymph node dissection; SNB, sentinel lymph node biopsy; HR, hazard ratio; CI, confidence interval.

Authors were contacted whenever the data necessary for the meta-analysis were not reported in the original articles. Overall, the data that were included in the meta-analysis were derived from 5 published studies8-12 and from the series described in the current article. None of those 6 studies had a randomized design; in 1 instance,8 we considered data on a subgroup of patients from an RCT (MSLT-1), bearing in mind the nonrandomized nature of the subset.

The included 6 studies encompassed 2633 patients who had American Joint Committee on Cancer TNM stage III melanoma (Table 3). In total, 1145 patients (43%) underwent CLND, and 1488 patients (57%) underwent TLND for clinically detectable regional lymph node metastases.

Table 3. Recently Published Series Comparing Overall Survival After Completion Lymph Node Dissection After Positive Sentinel Lymph Node Biopsy and Therapeutic Lymph Node Dissection for Clinically Evident Lymph Node Metastases
StudyStudy DesignMedian FU, moBreslow Thickness, mmNo. of PatientsOS PMINORS ScoreAdjustment by Multivariate Survival AnalysisNotes
  1. FU indicates follow-up; OS, overall survival; MINORS, Methodological Index for Non-Randomized Studies; CLND, completion lymph node dissection; TLND, therapeutic lymph node dissection; RCT, randomized controlled trial; SNB,: sentinel lymph node biopsy; SN, sentinel lymph node.

Morton 20039Retrospective12-360AnyCLND, 287; TLND, 287<.00117NoPatients were selected randomly and matched from a large database
Kretshmer 200410RetrospectiveCLND, 32; TLND, 121AnyCLND, 314; TLND, 623.00219YesMulticentric study
Morton 20068RCT subgroup analysisCLND, 48.4; TLND, 48.41.2-3.5CLND, 122; TLND, 78.00420YesNonrandomized subgroup from an RCT of SNB vs observation
van Akkooi 200711RetrospectiveCLND, 37; TLND, 56AnyCLND, 64; TLND, 124.11517YesExcluding patients (n=15) with SN submicrometastases, the difference in the 5-y OS rate decreased without reaching statistical significance
Nowecki 200812RetrospectiveCLND, 35; TLND, 37AnyCLND, 258; TLND, 286.0418NoFU refers to survivors
   1.01-4.00CLND, 111; TLND, 100.0006   
Current studyRetrospectiveCLND, 50; TLND, 56AnyCLND, 100; TLND, 90.1718YesPatients with false-negative SNB results were included in the TLND group
   1.01-4.0CLND, 58; TLND, 58.49   

The eligibility criteria were the same across the trials; that is, not all patients had clinically evident lymph node or distant metastatic disease at the time of their primary melanoma diagnosis, and the clinical outcome was considered melanoma-specific survival (which was calculated as the time from the date of primary melanoma excision to the date of either the last follow-up or death). All of the retrospective series analyzed the impact of RLND on OS. Taken individually, 4 studies reported a significantly longer median OS in patients who underwent CLND (n = 2255; 86%).8-10, 12 In contrast, 2 studies (van Akkooi et al11 and our personal experience; n = 378 patients; 14%) demonstrated no survival benefit for CLND over TLND.

Between-study heterogeneity was significant (Cochrane Q test, P = .02; I2 = 62%): therefore, a random-effects model was adopted to pool the summary data. The meta-analysis of the 6 studies (n = 2633) indicated that there was a significantly greater risk of death for patients who underwent TLND compared with that for patients who underwent CLND (HR, 1.60; 95% CI, 1.28-2.00; P < .0001) (Fig. 4). No publication bias was reported, because no funnel plot asymmetry was observed (Begg rank-correlation test, P = .70; Egger regression intercept, P = .92). The leave-1-out procedure revealed that no study was able to drive the overall effect of the meta-analysis. Finally, metaregression did not reveal a significant association between the study quality score (MINORS) and the HR estimate (regression slope, 0.02; P = .85).

Figure 4.

This Forest plot illustrates the hazard ratio (HR) (therapeutic lymph node dissection [TLND] vs completion lymph node dissection [CLND]) for overall survival from 6 nonrandomized studies (2633 patients). CI indicates confidence interval.


To date, RCTs have failed to demonstrate a survival advantage for patients with melanoma who undergo early lymph node dissection as part of prophylactic RLND3, 4 or after a positive SNB.8 One possible interpretation of these negative results indicates that these trials may have been underpowered to detect a survival advantage after RLND because of the relative small number of patients who had metastatic lymph nodes (only approximately 20%).27 In fact, subgroup analyses performed in these trials seemed to demonstrate a survival benefit for patients who underwent CLND compared with patients who underwent a delayed dissection.

Retrospective studies that compared survival after CLND for a positive SNB or after TLND for clinically detected lymph node metastasis reported controversial results (Table 3).9-12 A German multicenter study demonstrated a significant survival benefit (approximately 12% at 5 years) for patients with melanoma who underwent CLND for a positive SN.10 Similar findings were reported in a comparison of patients who were matched for prognostic factors, in which a significant 5-year survival advantage of 22% was observed for early versus delayed RLND.9 A survival advantage also was observed for patients who had tumors that ranged in thickness between 1 mm and 4 mm after a positive SN and underwent subsequent CLND; whereas, in the entire series from the same study, a borderline significant difference was detected.12 Moreover, no survival benefit was described after CLND, particularly when patients who had submicrometastatic disease (a metastatic area <0.1 mm in greatest dimension) were excluded from the analysis.11

In our retrospective series, we observed a nonsignificant 18% difference in the 5-year OS rate between patients who underwent CLND for a positive SN and patients who underwent RLND for clinically evident metastases on physical examination or on an US scan. Similarly, the difference in OS (23%) among patients who had melanoma of intermediate thickness (pT2-pT3) did not reach statistical significance. Although the magnitude of the survival benefit after CLND in our patients was similar to that observed in other series reported in literature, the lack of statistical significance may have been caused by differences in sample size (leading to different statistical power) and/or follow-up length (which can affect the number of events).

We are aware of the retrospective design of our study and, consequently, of the difficulty in drawing any definitive conclusion on the basis of the reported results. However, patients in the 2 subgroups (CLND and TLND) were well balanced in terms of the primary melanoma prognostic features included in the current TNM classification system (ie, tumor thickness, ulceration, and Clark level). Moreover, we believe that the inclusion of patients who developed lymph node metastases as a first recurrence after SNB (false-negative) was justified. In fact, a similar prognosis was reported in patients who underwent TLND8 or who had high tumor burden (ie, >1 positive lymph node, such as patients with positive SN and non-SN).28, 29

Only a prospective RCT with adequate potential to detect a survival advantage among patients with lymph node metastases will yield a definitive answer regarding the therapeutic role of early (SNB-guided) lymphadenectomy in patients with melanoma. In any event, to the best of our knowledge, no findings from such an RCT are available.

Conversely, several retrospective series and a nonrandomized subgroup from an RCT reported conflicting results on the therapeutic utility of CLND when subclinical lymph node metastases have been identified. Therefore, we believe that a formal meta-analysis of these nonrandomized studies is justified to provide a quantitative summary of the available evidence. The main advantage of pooling data from different series is the increased size of the patient population on which the results/conclusions are based. In other words, single studies often are characterized by low statistical power (ie, high type II error of erroneously rejecting the alternative hypothesis that CLND is associated with a lower risk of death) and large CIs (ie, the variability of results makes the conclusions poorly reproducible in other series). In contrast, the large sample size considered in a meta-analysis increases the power of the analysis, which maximizes the likelihood of correctly rejecting the null hypothesis, and decreases the CI of the calculated HR, which minimizes the likelihood that a positive result is caused by chance.

In the current work, although, in 2 series (including ours), there was no survival difference between the groups of interest, the meta-analysis of all 6 studies (n = 2633) that addressed this issue indicated that there is a significant reduction in the risk of death for patients who undergo SNB-guided, early lymphadenectomy compared with patients who undergo therapeutic lymphadenectomy for clinically evident lymph node disease. The lack of a “dominant” study (driving the overall meta-risk) and the absence of publication bias would lend support to the substantial consistency of the results across the eligible series. However, the significant between-study heterogeneity underscores the possibility that the overall HR estimate may not faithfully reproduce the risk in the general population: Consequently, although our findings suggest that SNB-guided lymphadenectomy—and not TLND—provides patients with the best therapeutic option, only the results of dedicated RCTs (such as the final analysis of the MSLT-I trial8 and results from the ongoing MSLT-II trial30) will assess directly and most appropriately whether this is truly the case.


The authors made no disclosures.