The study participants were selected from the melanoma databases at Hospital Virgen de la Victoria (HVV) in Malaga, Spain and Instituto Valenciano de Oncología (IVO) in Valencia, Spain. All ethical requirements regarding the use of databases were complied with. The 2 databases have been described in detail in previous studies.14–16

For this particular study, we selected all patients with a single cutaneous melanoma, classified as clinical stage I or II, who had undergone SN biopsy and been prospectively included in the IVO database between January 1, 2000 and December 31, 2012 or in the HVV database between October 1, 2001 and December 31, 2012. Patients without an identifiable SN during the SN biopsy were excluded from the study.

The study was approved by the ethics committees at the HVV and IVO.

The dependent variable was SN status, which was classified as positive in the presence of melanoma cells in the SN and negative otherwise.

The methods used for the pathologic examination of the SN are quite similar in both hospitals and are summarized below:

HVV. The first step is the gross identification and retrieval of the SN, with clearing of surrounding fat where necessary. The SN is then fixed in formalin 10% prior to embedding in a paraffin block. Each block is serially sectioned into 3 slices every 250μm until the whole block is cut. The first and last sections are used for hematoxylin-eosin staining, while the middle section is used for immunohistochemical staining with HMB-45.

IVO. The first stage is the same as at the HVV: identification and retrieval of the SN and clearing of fat where necessary. The specimen is also immersed in formalin 10% and embedded in a paraffin block. Serial sections are taken from the entire block every 250μm. Three cross-sections are taken from each section. One is stained with hematoxylin-eosin, and the other 2 are kept for immunohistochemical staining with S-100 protein and HMB-45. The immunohistochemical study is only performed following confirmation of the absence of metastasis in the hematoxylin-eosin section.

The following clinical and histologic characteristics were selected as independent variables:

Clinical characteristics. Patient sex and age (≤65 vs >65 years) and anatomic location of the primary melanoma. Anatomic location was dichotomized into axial (head, neck, trunk, and volar or subungual sites) and extremities. The system proposed by Kruper et al.8 was used to classify volar and subungual melanomas, as these are associated with a worse prognosis (possibly partly due to genetic alterations) that distinguishes them from other types of melanoma.17

Histologic characteristics. Breslow thickness (categorized according to the AJCC as ≤1.00, 1.01-2, 2.01-4, and >4mm), ulceration (present vs absent), regression (present vs absent), mitotic rate (0, 1-5, or ≥6 mitoses/mm2),10 and tumor infiltrating lymphocytes (TILs) (absent, nonbrisk, and brisk).18

The dependent variable was SN positivity. Independent variables were included as binary variables (sex, age, ulceration, anatomic location, and regression) or categorical variables (Breslow thickness, TILs, and mitotic rate).

In the logistic regression analysis, we first analyzed the association between SN positivity and the study variables using the χ2 test or univariate logistic regression analysis. The Wald test was used to calculate odds ratios with 95% CIs for all variables. Multiple logistic regression was used to identify independent predictors of SN positivity. All possible interactions between variables were also analyzed. Goodness of fit was assessed with the Hosmer-Lemeshow test.

The classification tree was built using the CART method,19 which is used to build decision trees through a set of rules governing decisions taken to assign an output value to each alternative through a process known as binary recursive partitioning. Each parent note is split into 2 homogeneous child nodes by asking questions with a yes and no answer. The CART process involves several stages: a) tree growing; b) application of stopping rules for the tree-growing process (a maximal tree that overadjusts the data used to generate the tree is built); c) pruning to create a simpler tree containing only the most important nodes; and d) selection of tree with the best generalization ability.20

We used the CART tree-growing process in SPSS version 20.0, with analysis of sensitivity based on the Gini criterion20 and the cross-validation system. In the final tree, each group had at least 20 cases (10 for subgroup analyses).

MedCalc statistical software (version 14.8.1) was used to generate receiver operating characteristic (ROC) curves for the CART and logistic regression models and to calculate the respective areas under the curve. The curves were compared using the DeLong method.21

Kaplan-Meier survival curves were used to evaluate disease-free survival (DFS) and melanoma-specific survival (MSS), and the log rank rest was applied to compare groups in the univariate analysis. The Cox proportional hazards model was used for the multivariate analysis. A P value of less than .05 was considered statistically significant. SSPS 20.0 was used for statistical analyses.

In total, 855 patients met the histologic criteria, provided informed consent, and underwent SNB. The SN was identified in 818 patients (95.6%). Of these 818 patients included in the analysis, 434 (53.1%) were women and 384 (46.9%) were men, with a median age of 53 years (interquartile range, 43-66 years). Table 1 summarizes the clinical and histologic characteristics of the study population stratified by SN status.

SN metastasis was associated with increased Breslow thickness, presence of ulceration, absence of regression, a high mitotic rate, and absence of TILs. It was not associated with sex, age, or anatomic location. The association between mitotic rate, increased Breslow thickness, and presence of ulceration is shown graphically in Fig. 1.

Figure 1.

Graph showing association between mitotic rate, Breslow thickness, and ulceration in the study population.

(0.09MB).

The following variables were associated with SN metastasis in the univariate analysis: Breslow thickness, presence of ulceration, absence of regression, high mitotic rate, and absence of TILs (Table 2). In the multiple logistic regression model, only ulceration, Breslow thickness, and high mitotic rate (≥6 mitoses/mm2) were independently associated with SN metastasis.

Six of the 8 variables included initially in the CART analysis were used. Breslow thickness (>2mm vs all other thicknesses) was the most important factor. The variables used for melanomas with a thickness of over 2mm were, by order of importance, inflammatory infiltrate (absent vs brisk/nonbrisk), age, and anatomic location. For thinner melanomas, they were, also in order of importance, mitotic rate (≤6 vs <6 mitoses/mm2), ulceration, and Breslow thickness (Fig. 2). Accordingly, patients with the greatest risk of SN involvement were those aged 65 years or younger with melanomas thicker than 2mm located on the extremities, without TILs (SN positivity 58%, n=60). By contrast, patients with the least risk of SN involvement were those with ulcerated melanomas with a Breslow thickness of 1mm or less and a mitotic rate of under 6 mitoses/mm2 (SN positivity 3.3%, n=181).

Figure 2.

Classification and regression tree with the following variables: Breslow thickness, tumor infiltrating lymphocytes, mitotic rate, age, sex, anatomic location, ulceration, and regression. RR indicates relative risk.

(0.7MB).

The ROC curves for the CART and logistic regression models were similar (P=.7), with an area under the curve of 0.77 (95% CI, 0.73-0.81) and 0.78 (0.74-0.82), respectively (Fig. 3).

Figure 3.

Receiver operating characteristic curve graph comparing classification and regression tree analysis (CART) and multiple logistic regression (MLR).

(0.13MB).

In a median follow-up period of 58 months (range, 1-157 months), 104 patients died of melanoma and 172 experienced some form of recurrence (uncensured data in the Cox proportional hazards model). The most important predictor of both DFS and MSS was SN metastasis.

MSS was 68.1% for patients with SN involvement and 93.1% for those without. The Cox proportional hazards model was adjusted for significant variables in the univariate analysis. SN metastasis was the most important prognostic factor for both DFS (hazard ratio, 2.3; 95% CI, 1.5-3.6) and MSS (hazard ratio, 3.5; 95% CI, 1.9-6.3).

Other significant predictors of DFS (shown in detail in Table 3) were Breslow thickness, age, anatomic location, and mitotic rate (Fig. 4). In the case of MSS (Table 4), mitotic rate was an independent prognostic factor, together with age, anatomic location, and Breslow thickness (Fig. 5).

Figure 4.

Kaplan-Meier survival curves for disease-free survival. A, Comparison of patients with a low mitotic rate (0 mitoses/mm2), a moderate mitotic rate (1-5 mitoses/mm2), and a high mitotic rate (≥6 mitoses/mm2). B, Stratification by sentinel lymph node (SN) involvement.

(0.27MB).

Figure 5.

Kaplan-Meier survival curves for melanoma-specific survival. A, Comparison of patients with a low mitotic rate (0 mitoses/mm2), a moderate mitotic rate (1-5 mitoses/mm2), and a high mitotic rate (≥6 mitoses/mm2). B, Stratification by sentinel lymph node (SN) involvement.

(0.27MB).

The authors declare that no tests were carried out in humans or animals for the purpose of this study.

The authors declare that they have followed their hospital's protocol on the publication of data concerning patients.

The authors declare that no private patient data appear in this article.