Background

The standard treatment of cutaneous melanoma has been wide excision of the primary tumor combined with sentinel node (SN) biopsy for staging purposes, the SN status being one of the strongest predictors of prognosis [1,2,3]. For many years, completion lymph node dissection (CLND) has been the standard approach for patients with metastatic SN. However, with publication of the Multicenter selective Lymphadenectomy-2 trial (MSLT2) and the German Dermatologic Cooperative Oncology Group study (DeCOG-SLT) and considering the evolving landscape of adjuvant therapy in melanoma patients, immediate CLND is not recommended in the first course [4,5,6,7,8]. MSLT-2 and DeCOG-STLT showed that CLND increases the rate of regional disease control and provides prognostic information but is not associated with an improved melanoma specific survival [5, 6]. However, these trials, the results of which might lead to abandon the practice of CLND, had some limitations. First, retrospective series produced varied results and were subject to a considerable risk of select bias; next, there were differences in clinic-pathologic features of the patient cohorts between centers; moreover, most patients, enrolled in these studies, had a low-volume nodal tumor burden [5, 7, 9,10,11]. In the end, international guidelines such as those issued by the National Comprehensive Cancer Network, recommend to discuss with the patient the benefit of CLND mainly based on the risk of harboring additional lymph node metastatic disease. In analogy with breast cancer, the current availability of an effective adjuvant therapy (either targeted therapy or immunotherapy) for patients with SN positive melanoma is further pushing against the use of CLND [5, 7, 12, 13]. However, omitting CLND could result in underestimation of patients at high risk progression and so in an impaired selection for adjuvant therapy. In fact, there is an approximately 20% of melanoma patients harboring metastatic disease in non-sentinel nodes [5, 6, 8, 13,14,15,16]. In order to predict the presence of disease in NSN for follow up personalization in clinical practice and for patient risk stratification, Italian Melanoma Intergroup (IMI) built a nomogram for prediction NSN status in melanoma patients with positive SNB [4]. On the other hand, patients with minimal residual disease in the SN often experience a favorable clinical outcome: in this subgroup, adjuvant therapy might represent an overtreatment since surgery is likely to have already completely eradicated the tumor [4, 17].

In this challenging landscape, adjuvant therapy decision making would greatly benefit from the identification independent prognostic factors improving the risk stratification, which ultimately would enable physicians to optimize the management of melanoma patients with positive SNB [1,2,3,4,5,6,7,8,9,10,11,12,13].

In 2017, the 8th edition of American Joint Committee of Cancer (AJCC) melanoma classification was published and one of the major changes was the criteria to allocate patients in stage III. However, no new prognostic factor was introduced [17, 18].

Tumor burden (TB), the maximum diameter of the largest tumor deposit in the SN, has been advocated as a potentially useful prognostic parameter in stage III patients. Although it is not incorporated in the AJCC staging system, the AJCC panel acknowledged the importance of TB and recommended the assessment of tumor load to be performed by every melanoma Center [17,18,19].

Methods

Study design

This is a retrospective study based on information from prospectively maintained databases managed by 13 Italian centers belonging to the Italian Melanoma Intergroup (IMI).

This study aimed to test the hypothesis that TB significantly contributes to predict overall survival (OS) in patients with metastatic melanoma in the sentinel node(s) who underwent CLND. The results were used to build a prognostic nomogram of clinical use.

Data regarding a cohort of 2,086 patients were reviewed. We considered the following covariates: age (as a continuous value), gender (male vs female), melanoma body site (head and neck, trunk, limbs), Breslow thickness (as a continuous value), ulceration (absent vs present), number of positive sentinel nodes (as a continuous value), sentinel node TB (0.01 mm-0.4 mm; 0.41 mm-0.96 mm; 0.97 mm- 3 mm; 3.1 mm-35 mm), non-sentinel lymph node status (positive vs negative).

The main inclusion criteria for SNB was melanoma with Breslow thickness ≥ 1 or melanoma with thinner tumors with adverse prognostic features such as ulceration, a high mitotic rate or Clark level IV o V. The main inclusion criteria for CLND were positive SNB and lack of clinical or radiological evidence of metastatic disease (all patients were M0).

The pathology protocols to assess primary melanoma features, SN and NSN status were shared by all 13 IMI Centers.

T coefficients obtained from the multivariable model were used to set up a nomogram for practical use.

Statistical analysis

OS curve was estimated using the Kaplan Meier method. Univariate and multivariate survival analyses were performed using the Cox proportional hazard regression model.

In order to check for model overfitting we used the bootstrap method (1,000 replications). Briefly, random samples drawn with replacement from the original data set are created with the same size as the original series; the performance index of the model built on the entire cohort is always better than the average of the indices calculated in each replication. The difference between the two is an estimate of the model overfitting (optimism) and the average value of the indices is considered the unbiased estimate of how well the model would perform in future data set. The alpha level of significance was set at 5%.

All analyses were performed using Stata 11.2 SE software (StataCorp LLc, College Station, TX, USA).

Results

Patients diagnosed with cutaneous melanoma between 2000 and 2018 were studied.

Patients and tumor characteristics are reported in Table 1. Univariate and multivariate survival analyses are reported in Table 2.

Table 1 Patients and tumor characteristic
Full size table
Table 2 Results of univariate and multivariate analysis
Full size table

The 3-year, 5-year and 10-year OS rates were 79%, 70% and 54%, respectively (Fig. 1). At univariate analysis, all variables, except for primary melanoma body site, were found to be statistically significantly associated with patient prognosis. Multivariate Cox regression analysis indicated that age (HR = 1.01; 95% CI: 1.01–1.02, p < 0.0001), male sex (HR = 1.19; 95% CI: 1.00–1.41, p = 0.0005), Breslow (HR = 1.06; 95% CI: 1.04–1.08, p < 0.0001), NSN metastatic status (HR = 2.06; 95% CI: 1.72–2.50, p < 0.0001), ulceration (HR = 1.30; 95% CI: 1.08–1.56, p < 0.0001) and the diameter of sentinel node metastasis (HR = 1.04; 95% CI: 1.03–1.06, p < 0.0001) were independent negative prognostic factors of OS. Number of positive SN did not result a statistically significant predictor of OS.

Fig. 1
figure 1

OS Kaplan Meier curve

Full size image

As regards TB, we grouped SN metastasis diameter in quartiles and so we identified 4 subgroups of patients (Fig. 2). We found that SN TB was an important predictor of OS with a progressive worsening prognosis from first (0.01 mm-0.04 mm) to fourth subgroup (3.1 mm-35 mm). In details, the latest subgroup had a significantly poor prognosis compared to the other subgroups (HR = 3.51; 95% CI: 2.74–4.49, p > 0.0001) (Table 2). A nomogram for clinical and research purposes was built using the coefficients by the multivariate model (Fig. 3).

Fig. 2
figure 2

OS according to TB

Full size image
Fig. 3
figure 3

Prognostic nomogram

Full size image

Discussion

We presented the results of a multicentric study aimed at building a prognostic model in melanoma patients with metastatic melanoma in the sentinel node by combining the information of six clinico-pathological variables including tumor burden (TB). Our findings show that TB significantly contributes to predict the prognosis of these patients.

To the best of our knowledge this is the largest series ever published with this aim, as previous studies have addressed the same issue using data from smaller series of patients. The negative prognostic value of TB, as maximum SN tumor size > 1 mm, was already assessed in terms of non-sentinel node positivity, disease free survival and melanoma specific survival by an international multicentric study of 1,539 SN positive melanoma patients [1].

Another study analyzing 104 SN positive patients who underwent CLND demonstrated that the 5-year melanoma specific survival for CLND-negative patients was 5 years as compared to 3.69 years in CLND positive patients. In this analysis, clinico-pathological parameters such as diameter of tumor deposit, distribution of metastatic focus within the sentinel node, ulceration and number of metastatic melanoma in the sentinel nodes were evaluated and the investigators found that TB > 4 mm and multifocal metastatic disease within the sentinel node were the most important variables that allowed an accurate prognostic stratification of patients [11].

On behalf of the EORTC-DeCOG, some authors developed prediction models for disease recurrence, distant metastasis (DM) and overall mortality analyzing a retrospective cohort of 1,080 patients. The resulting EORTC-DeCOG nomogram included parameters as ulceration, age, sentinel node TB, and Breslow: therefore, this study included only information deriving from the primary melanoma and the SNB, without considering the status of non-sentinel nodes [5].

In 2019, Satzger et al., assessing a total of 736 positive SN melanoma patients, demonstrated that TB, Breslow, ulceration and age are independent prognostic factors of melanoma specific survival. In detail, MSS was significantly better in patients with lower SN TB than in patients with higher SN TB (> 0.5 mm and > 1 mm) [17].

Overall, our results are consistent with the previous studies and confirm the prognostic value of TB. Our findings, along with the already existing literature on this subject, support the hypothesis that prognosis decays continuously with increasing maximum diameter of the larger metastatic deposit within the sentinel node: therefore, accumulating evidence strongly suggest that TB represents an important piece of information while stratifying the risk of these patients, even if the ideal cut-off needs to be determined.

In our opinion, CLND adds prognostic information as the status of non-sentinel nodes is an independent prognostic feature, as we have demonstrated not only in the present analysis but also previously [6, 11]. Therefore, CLND not only improves regional relapse-free survival (as demonstrated by the results of the MSLT-II) but also provides physicians with useful prognostic information [6,7,8].

For practical purposes, we generated a nomogram to easily personalize the prediction of patient prognosis. More precise risk stratification is important for adequate patient information on the severity of the disease and is especially useful for selecting patients who can benefit most from adjuvant therapy. Moreover, our nomogram could enable physicians to personalize the intensity of patients follow up as well as to optimize patient allocation within the frame of clinical trials.

Finally, we recognize that the present study has some limitations. First, it is a retrospective multicentric study with inherent bias: however the analyses were performed on complete cases without missing data. Second, we could not validate our results in an external series of patients, which would improve the assessment of model generalizability. Third, a good proportion of patients received interferon-alpha based adjuvant therapy, which could have potentially influenced the outcome, especially in patients with ulcerated melanoma. Moreover, some patients were treated with modern treatments (e.g., targeted therapy or immunotherapy) after disease recurrence, which could also have affected overall survival. We could not explore the impact of adjuvant therapy on OS because of insufficient data.

In addition, other potential prognostic factor as mitotic rate could not be incorporated in our analysis due to lack of complete data.

As regards the future perspective of survival predictive models, we believe that only the implementation of informative biomarkers will help improve the accuracy of current prognostic tools. Investigation on the molecular mechanisms underlying melanoma progression and aggressiveness has led to the identification of hundreds of potential such biomarkers [4,5,6,7, 18]; unfortunately, none of them has been so far associated with a predictive value independent of conventional clinico-pathological parameters. Therefore, more work is eagerly needed to make further advances in this field of investigation.