Fax: (215) 573-4865
Predicting sentinel node status in AJCC stage I/II primary cutaneous melanoma†
Article first published online: 20 OCT 2006
Copyright © 2006 American Cancer Society
Volume 107, Issue 10, pages 2436–2445, 15 November 2006
How to Cite
Kruper, L. L., Spitz, F. R., Czerniecki, B. J., Fraker, D. L., Blackwood-Chirchir, A., Ming, M. E., Elder, D. E., Elenitsas, R., Guerry, D. and Gimotty, P. A. (2006), Predicting sentinel node status in AJCC stage I/II primary cutaneous melanoma. Cancer, 107: 2436–2445. doi: 10.1002/cncr.22295
Presented in part at the 41st Annual Meeting of the American Society of Clinical Oncology, Orlando, FL, May 13–17, 2005.
- Issue published online: 8 NOV 2006
- Article first published online: 20 OCT 2006
- Manuscript Accepted: 7 SEP 2006
- Manuscript Revised: 1 SEP 2006
- Manuscript Received: 24 FEB 2006
- SPORE on Skin Cancer. Grant Number: CA-093372
- NRSA Cancer Epidemiology Training. Grant Number: T32 CA-009679
- sentinel node;
- tumor infiltrating lymphocytes
Sentinel lymph node (SLN) status is an important prognostic factor for survival for patients with primary cutaneous melanoma. To address the issue of selecting patients at high and low risk for a positive SLN, prognostic factors were sought that predict SLN involvement by examining characteristics of both the primary tumor and the patient within the context of a biological model of melanoma progression.
The study included 682 patients with primary vertical growth phase (VGP) melanoma and no clinical evidence of metastatic disease who underwent SLN biopsy (1995–2003). Logistic regression and classification tree analyses were used to investigate the association between SLN positivity and Breslow thickness, Clark level, tumor infiltrating lymphocytes (TIL), ulceration, mitotic rate (MR), lesion site, gender, and age.
In all, 88 of the 682 patients had ≥1 positive SLN (12.9%). In the multivariate analysis, MR, TIL, and thickness were found to be independent prognostic factors for SLN positivity. In the classification tree, four different risk groups were defined, ranging from minimal risk (2.1%) to high risk (40.4%). In lesions ≤2.0 mm, MR was important in risk-stratifying patients, and in lesions >2.0 mm TIL was important.
By incorporating biologically based variables such as VGP, TIL, and MR along with thickness into a prognostic model, both patients at high risk and minimal risk for SLN positivity can be identified. If validated, this model can be used in patient management and trial design to select patients to undergo or be spared SLN biopsy. Cancer 2006. © 2006 American Cancer Society.
Lymphatic mapping and sentinel node biopsy (LM/SNB) is an important advance in the management of primary cutaneous melanoma.1, 2 As pioneered by Morton et al.,2 this technique has allowed surgeons to identify patients who harbor lymph node micrometastasis via a minimally invasive procedure. At most institutions, this procedure is targeted to patients with melanomas >1.0 mm in thickness.3 The detection of subclinical nodal involvement defines a subset of patients who may benefit from a completion lymphadenectomy and who are candidates for trials of adjuvant therapy. For those patients with sentinel lymph node (SLN) involvement, a completion lymphadenectomy provides effective prophylaxis against regional nodal failure, whereas its impact on overall survival remains incompletely defined.4 Those without nodal involvement are spared the expense and potential morbidity of further interventions. In addition to guiding therapy, the status of the SLN is an important prognostic factor for survival.2, 5, 6
Several studies have evaluated predictors of SLN status. Thickness has consistently been shown to predict SLN status.5, 7–14 Other factors, such as ulceration, Clark level IV/V, vertical growth phase (VGP), higher dermal mitotic rate (MR), blood and lymphatic vessel invasion, axial location, and younger age have all been cited with conflicting results.5, 7–20 Our group evaluated prognostic factors within the context of a biological model of melanoma progression in which melanomas are hypothesized to evolve from cell populations incapable of metastasis (the in situ and invasive radial growth phase [RGP]) to those with variable metastatic potential (the VGP).21–24 Histologic features such as VGP (defined below) and dermal mitotic rate are taken to indicate that a lesion may have gained the capacity to metastasize.21–24 These 2 factors have been shown to be associated with SLN metastasis in lesions ≤1.0 mm.19, 25 Clark et al.,21 as well as other investigators,26–29 have also found that markers of antitumor immunity predict the likelihood of metastasis: the presence and density of tumor infiltrating lymphocytes (TIL) independently associate with survival.
These considerations suggest that biologically based prognostic factors, particularly VGP, mitogenicity (any tumor cell mitoses in the dermis), and TIL, might better predict SLN status than AJCC stage, with its limited range of variables (thickness, level in lesions ≤1 mm, and ulceration). In this study the primary objective was to determine the clinical and histopathologic factors that predict SLN positivity in a large cohort of 682 patients with primary cutaneous VGP melanomas by developing a multivariable prognostic model for more accurate risk prediction. A second objective was to develop a classification scheme of potential clinical utility to further identify patients with statistically different levels of risk for a positive SLN.
MATERIALS AND METHODS
Patients eligible for this study were identified through the information management system of the University of Pennsylvania's Pigmented Lesion Clinic (PLC) and the electronic patient tracking system of the Division of Surgical Oncology. This retrospective cohort included all patients (n = 791) with clinical Stage I and II cutaneous melanoma who underwent an SLN biopsy from January 1995 to June 2003. Charts of those patients not seen in the PLC were abstracted. Patients were eligible for inclusion in this study if they had a diagnosis of VGP primary melanoma and no clinical evidence of regional or distant metastasis at the time of diagnosis. This study was approved by the University of Pennsylvania Institutional Review Board.
Clinical and Histologic Features
Clinical factors evaluated were the age of the patient, gender, and site of lesion. The anatomic site was classified into 1 of 2 categories: axial (comprised of truncal, head/neck, volar, and subungual lesions) or extremity. Volar and subungual melanomas were coded as “axial” because, as in melanomas of the head, neck, and trunk, these sites were found previously to be adverse prognostic factors for survival.15 Of note, melanomas of these acral sites have been found to have specific genetic abnormalities that distinguish them from those located elsewhere.30 Because in situ and RGP lesions are apparently incapable of metastasis,21–24 our study focused on lesions with VGP, a marker of dermal proliferation defined as the presence of tumor cell mitoses in the dermis and/or clusters or nests of melanoma cells in the dermis that are larger than those in the epidermis.21–23 Candidate histopathologic prognostic factors were Breslow thickness (measured in millimeters), level, mitotic rate (MR: number of mitoses per square millimeter), TIL (measured as lymphocytes infiltrating the VGP and disrupting the melanoma cells, see Fig. 1), ulceration, and regression in association with the adjacent RGP (as recently described in detail23).
Lymphatic Mapping and Sentinel Node Biopsy
In general, patients with lesions >1.0 mm in thickness were offered LM/SNB. Patients with VGP lesions ≤1.0 mm were offered LM/SNB if adverse histopathologic features of the primary tumor were present, such as ulceration, regression, or Level IV/V. Patients with thin melanomas with VGP but without adverse features were also counseled about the potential risks and benefits of LM/SNB.19 The decision to proceed with LM/SNB was made on an individual basis.
Preoperative lymphoscintigraphy was performed on the same day of surgery in all patients. The primary lesion or biopsy site was injected with 99mTc-labeled serum albumin (Amersham Health, Princeton, NJ) or 99 mTc-labeled sulfur colloid (Tyco International, Mallinckrodt, Phillipsburg, NJ) intradermally to identify lymphatic basins at risk. In cases where lymphoscintigraphy failed to identify any draining nodes after 2 hours, attempts were made intraoperatively to locate the SLN with the use of the gamma-probe (Tyco Healthcare, US Surgical, Norwalk, CT) and blue dye.
After the induction of anesthesia, 1.0 mL of 1% isosulfan blue dye (Lymphazurin, Tyco Healthcare, US Surgical) was injected intradermally at the site of the primary melanoma or biopsy site. In the SLN procedure, all blue nodes and nodes with radioactive counts measuring >10% of the ex vivo counts from the “hottest” node were removed.31 All SLN were serially sectioned and stained with hematoxylin and eosin (H&E) and had immunohistochemical staining with S-100 and HMB-45 with appropriate controls.
The primary clinical outcome was having a positive SLN biopsy (one or more positive nodes). Factors were included as either binary (ulceration, age, regression, gender, site, TIL, and Clark level) or categorical (thickness and mitotic rate). Thickness was recoded based on AJCC tumor (T) classifications: ≤1.0 mm (T1), 1.01–2.0 mm (T2), 2.01–4.0 mm (T3), and >4.0 mm (T4).1 Mitotic rate was recoded as: 0, 0.1–5, and >5 mitoses/mm2. Chi-squared analysis and univariate logistic regression models were performed to determine the association between each clinical or histopathologic factor and SLN status. The odds ratio (OR) and 95% Wald-based confidence interval were reported for each variable. All factors were included in the full multivariate model, and nonsignificant factors (P > .05) were removed sequentially with no significant reduction in the likelihood ratio statistic32 (P > .05) factors. Statistical analyses were performed with STATA (College Station, TX).33
A recursive partitioning algorithm was used to develop the classification tree using CART (Salford Systems, San Diego, CA).34 The CART algorithm sequentially identifies the best prognostic factor and a corresponding cutpoint by examining every prognostic factor and all possible cutpoints. For each group it identifies a prognostic factor and cutpoint that defines 2 patient subgroups that are most different based on their rates of SLN positivity, minimizing misclassification. The process is continued for each subgroup until there is no further improvement in homogeneity by splitting the groups.35
A total of 791 patients with VGP lesions were identified, and the 682 patients who had complete data were included in this study. There were no notable differences between those included and excluded in this study except for thickness; excluded patients had thicker melanomas (data not shown). Of the 682 study patients, 36.8% had thin lesions (≤1.0 mm) The mean thickness for the thin lesions was 0.73 mm (median, 0.75 mm), for the lesions >1.0 mm it was 2.7 mm (median, 1.9), and for the entire cohort it was 2.0 mm (median, 1.3 mm; range, 0.24–18 mm). Table 1 shows the clinical and histopathologic factors of the study patients and Figure 2 presents the SLN positivity rates for T1, T2, T3, and T4 melanomas. One or more histologically or immunohistochemically positive lymph nodes were found in 88 (12.9%) of the 682 patients and this rate varied by thickness. The positive SLN rate was 5.2% in patients with thin VGP lesions (13 positive SLN) and was 17.4% (75 positive SLN) in patients with intermediate and thick VGP lesions (>1 mm). The rates of positive SLN for subgroups defined by each of the clinical and histopathologic variables are presented in Table 2.
|≤1.00 mm n = 251 Percent||1.01–2.00 mm n = 228 Percent||2.01–4.00 mm n = 140 Percent||>4.00 mm n = 63 Percent||All n = 682 Percent|
|VGP mitotic rate (per mm2)|
|Variable||≤1.00 mm (n = 251)||1.01–2.00 mm (n = 228)||2.01–4.00 mm (n = 140)||>4.00 mm (n = 63)||All (n = 682)|
|Frequency||Rate (%)||Frequency||Rate (%)||Frequency||Rate (%)||Frequency||Rate (%)||Frequency||Rate (%)|
|VGP mitotic rate, mm2|
Table 3 presents the association of SLN positivity in these VGP lesions (as measured by unadjusted and adjusted odds ratios) and each of the prognostic variables. Increasing thickness, level, mitotic rate, the absence of TIL, and the presence of ulceration were each significantly associated with SLN positivity in univariate logistic regression models. Only thickness, TIL, and MR remained independent prognostic factors in the reduced multivariate logistic regression model. Mitotic rate was also evaluated using the following categories: 0, 0.1–6 mm,2 and >6 mm2 as previously reported21 and as ≥5 mitoses/mm2 based on a recent definition of “high.”8, 14 Neither changed the results (data not shown). Mitotic rate as a continuous variable had an unadjusted OR of 1.07 (P < .001) but the association was no longer significant in the full model (data not shown).
|Variable||Univariate model||Multivariate model||Reduced model|
|OR (95% CI)||OR (95% CI)||OR (95% CI)|
|Men||1.3 (0.8–2.0)||1.2 (0.7–2.0)|
|Axial||0.99 (0.6–1.6)||0.8 (0.5–1.4)|
|>60||1.6 (0.99–2.5)||1.1 (0.6–1.8)|
|≤1.00||0.4 (0.2–0.8)||0.6 (0.3–1.3)||0.6 (0.3–1.3)|
|2.01–4.00||1.9 (1.1–3.4)||1.8 (0.97–3.4)||1.9 (1.03–3.4)|
|>4.0 0||4.4 (2.2–8.7)||3.7(1.7–7.9)||3.9 (1.9–8.0)|
|IV/V||2.6 (1.5–4.7)||1.2 (0.6–2.3)|
|Absent||2.0 (1.3–3.2)||2.5 (1.5–4.1)||2.5 (1.6–4.2)|
|Mitotic rate, mm2|
|0.1–5||7.9 (2.4–25.5)||5.2 (1.5–18.1)||5.5 (1.6–19.0)|
|>5||12.5 (3.7–41.9)||5.3 (1.4–20.5)||5.8 (1.5–21.9)|
|Present||2.2 (1.3–3.8)||1.1 (0.6–2.1)|
|Present||0.9 (0.5–1.7)||0.9 (0.4–1.7)|
In Figure 3 we present the distribution of SLN positivity in VGP lesions as a function of MR and TIL. SLN positivity increased with increasing MR, with the differences in SLN positivity in adjacent groups becoming less distinct as mitotic rate increased (Fig. 3, left panel). Tumor infiltrating lymphocytes were important as a predictive factor in lesions >1.0 mm, showing a 2-fold increase in SLN positivity when TIL were absent (Fig. 3, right panel).
The classification tree is presented in Figure 4. Thickness was the first factor selected to split the entire sample of VGP lesions into 2 groups: ≤2.0 mm and >2.0 mm. Thinner lesions (≤2.0 mm) split on MR and thicker lesions (>2.0 mm) split on TIL. The ≤2.0 mm and MR >0 group then split on TIL in a third step. These steps produced 5 groups, with 4 varying levels of risk of SLN positivity from a low risk of 2.1% to a very high risk of 40.4%. The group with an MR >0 and absent TIL was split into 2 groups by thickness: ≤1.0 mm and 1.01–2.0 mm (not shown); in those with lesions ≤1.0 mm (n = 37), there was an SLN positivity rate of 8.1%, compared with a rate of 24.5% in those with lesions 1.01–2.00 mm (n = 53). However, the SLN positivity rates for these 2 risk groups were not significantly different.
Melanomas evolve in stepwise fashion.21, 22, 24 Tumor cells of early invasive disease, the invasive RGP, rarely metastasize and appear to slow their proliferative rate on entering the dermis.22, 24, 36 Metastatic capacity appears with the subsequent resumption and acceleration of proliferation as manifested in the vertical growth phase. Melanoma cells of the VGP may or may not actually metastasize, depending on a combination of tumor and host factors. Among the latter is the host immune response as measured, for example, by TIL. Here we explored the relation of 7 commonly used and 2 more biologically based (mitoses and TIL) prognostic factors within VGP lesions to determine which features are predictive of a positive SLN.
Thickness was significantly associated with SLN positivity in our study and it was identified as the first prognostic factor in our classification tree, with 2.0 mm identified statistically as the optimal cutpoint to create distinct subgroups. The SLN positivity rate for the entire cohort was 12.9%, somewhat lower than that in other published studies evaluating lesions of all thicknesses. This rate is explained in part by the high proportion of thin lesions (≤1 mm) in our cohort (36.8%), with a median thickness of 1.3 mm for the entire group. In 1995, our institution began a study in which all melanoma patients with lesions >1.0 mm were offered SLN biopsies as well as any individual with a lesion ≤1.0 mm with VGP.19 Other studies with lower proportions of thin lesions (4.2%–30.7%) or higher mean thickness values (1.4–2.7 mm) reported higher SLN positivity rates of 14.2% to 23.1%.5, 7–17 Differences among the overall rates may also be a result of selection bias related to different patterns of use of the procedure. In other studies, SLN biopsies were offered to patients with thin lesions only if other adverse features were present such as ulceration,5, 14 regression,14 or Level IV/V.5, 16, 37 In Morton et al.'s16 study the overall positivity rate was 20.7%. Whereas patients with lesions >1.0 mm in Morton's study were regularly offered SLN biopsy, 30.7% of the study patients had lesions ≤1.0 mm, representing either those who had adverse prognostic factors or those who self-selected to have the procedure (the median thickness was 1.43 mm for the entire cohort). Controlling for thickness, the SLN positivity rates in the present study were comparable to those of others5, 7–17, 38 (see Fig. 5). The SLN positivity rate was 5.2% for patients with VGP lesions ≤1.0 mm, similar to other studies5, 7–17, 39, 40 (range, 0–9.5%). For VGP lesions >1.0 mm, the positive SLN rate of 17.4% was within the range of other studies of intermediate and thick lesions (19%–26.5%) with overlapping 95% confidence intervals.
Whereas both level and ulceration were significantly associated with SLN positivity in the univariate models, they were not significant in the multivariate model. Only 17 of several5, 8, 11, 14, 18, 20 studies found level to be an independent prognostic factor. Although ulceration has frequently been shown to be a predictor of a positive SLN,5, 7, 8, 11–13, 18, 20 our study did not confirm this finding and our result is consistent with the recent study by Sondak et al.14
Three studies have shown an association between a positive SLN and a high MR or mitotic index.8, 14, 20 In the current study, we found in multivariate analysis that the presence of any mitoses (MR >0), as well as a high mitotic rate (MR >5), was associated with a higher odds of having a positive SLN. With our classification tree it is important to note that mitogenicity41, 42 (present when MR >0 or absent when MR = 0) was prognostic only in thinner VGP lesions. Within the ≤2.0 mm thickness group, MR was selected to identify patients with lower and higher rates of SLN positivity. However, neither mitogenicity nor the actual mitotic rate was selected to split the >2.0 mm thickness group. One explanation for why mitogenicity did not appear in the classification tree for the thicker VGP lesions (>2.0 mm) was that there was considerable mitotic activity and little variability in mitotic activity among these lesions. As shown in Table 1, few of these lesions had an MR of zero (2.5%), and almost half of the lesions had an MR of >5 mitoses/mm2 (47%). Although increasing MR was associated with a higher rate of SLN positivity (see Fig. 3A) once MR >0, there was less variability in the rates of mitoses.
Our results indicate that there is value in using 2 separate cutpoints for mitosis to differentiate risk levels of a positive SLN: a mitotic rate of zero and an upper level cutpoint. In 2 other studies from our institution focusing on thin lesions (≤1.0 mm), we have shown that once dermal proliferation ensues (as marked by dermal tumorgenicity23 or an MR >0), there is an increased risk for regional and disseminated metastasis,42 as well as for SLN positivity.25 Defining a cutpoint of MR = 0 (“nonmitogenic”) identifies a group at minimal risk of a positive SLN according to our classification tree. In the 3 previous studies mentioned above, the “low” group of MR or mitotic index was <2 mitoses/high- power field (hpf)20 or <1 mitosis/mm2.14 We believe that there is additional information to be gained with the category of MR = 0.
Another useful cutpoint is an upper boundary to define a group at higher risk of a positive SLN. Although MR as a continuous variable was shown to be associated with SLN positivity, indicating that the rate of a positive SLN increased with each incremental increase in mitotic rate, this result was no longer significant in the full model. However, there was an association in our multivariate model between a positive SLN and MR >5 mitoses/mm2, consistent with other studies.8, 14, 20 In the prior studies of MR and SLN positivity, different methods of measuring MR were used. In the study by Wagner et al.,20 the number of mitoses per hpf or mitotic index was recorded, showing that an MR of >5 mitoses/hpf was a significant predictor of a positive SLN. In the studies by Mraz-Gernhard et al.8 and Sondak et al.,14 the number of mitoses per mm2 was measured with the upper level cutpoint of 5 mitoses per mm2. These 2 methods of measuring mitoses are not easily interchangeable.
In our multivariate model, absent TIL were associated with more than a 2-fold increase in the odds of having a positive SLN. TIL were also important in defining groups at different levels of risk for a positive SLN in the classification tree. TIL was the prognostic variable chosen by CART for VGP lesions >2.0 mm as well as for mitogenic VGP lesions ≤2.0 mm. It was not selected as a prognostic factor in nonmitogenic VGP lesions ≤2.0 mm. These observations suggest that the host response as measured by TIL is important later in tumor progression, after the melanoma has begun dermal proliferation. Because the histologic classification of TIL has an acceptable level of reproducibility in routine diagnostic practice,23 this variable should be considered for inclusion in pathology reports.
Once validated, our prognostic model can be used as a decision-making tool in several different ways. It can be used to increase the proportion of patients who could potentially be spared LM/SNB. Currently, patients are routinely offered LM/SNB for melanomas >1.0 mm; however, those with nonmitogenic tumors (MR = 0), including lesions up to 2.0 mm could potentially forego the procedure. Patients with melanomas ≤1.0 mm who are usually recommended for LM/SNB based on other known adverse histologic features4 (such as Level IV/V) could also be spared the procedure if the tumor is nonmitogenic.
Conversely, our model could also be used to identify patients not currently considered for SLN biopsy who are at sufficient risk of nodal metastasis to warrant its consideration. Currently, there remains controversy regarding LM/SNB in patients with lesions ≤1.0 mm given the low risk of nodal metastasis, ranging from 0% to 9.5%.5, 7–17, 39, 40 Many institutions recommend LM/SNB for thin lesions if ulceration or Level IV/V is present.4 Our prediction model suggests broadening the indications for SLN biopsy to include any patient with a mitogenic VGP lesion, particularly if both mitoses are present and TIL are absent; these indications would apply to all VGP lesions, including those ≤1.0 mm.
In summary, we have developed a classification tree that uses thickness, mitogenicity (MR >0), and TIL to identify patients with VGP melanomas with risks of nodal metastasis that range from 2% to 40%. At least in this group of patients, mitogenicity and TIL displaced such traditional factors as level (in thin melanomas) and ulceration. This suggests that the next generation of prognostic models might be based on more robust markers of proliferation (e.g., of activation of the MAP and/or PI3 kinase pathways) and of a tumoricidal or tumor-permissive immune response (e.g., using immunohistochemical measures of cytotoxic or regulatory T-cells). Presently, our model has potential clinical utility. It is biologically based, plausible, uses factors demonstrated in other datasets to be prognostic, is transparent, and is simple to use without calculations. To realize this potential, it requires external validation in other melanoma populations that have been characterized with regard to VGP, mitoses, and TIL.
We thank the patients seen at the Pigmented Lesion Clinic and Surgical Oncology Clinic as well as the investigators (Drs. L.P. Bucky, L.S. Callans, B. Chang, K.T. Flaherty, A.C. Halpern, R. Hamilton, D.M. Hershock, E.J. Kim, D.D. Larossa, S.R. Lessin, D. Low, L.M. Schuchter, P. Van Belle, and J.T. Wolfe) and staff (J. Botbyl, C. Doyle, R. Holmes, S. Hotz, N. Lowden, I. Matozzo, M. Price, M. Synnestvedt, and J. Thompson) for their contributions over the last decade in the collection of melanoma research data on which the current article is based.
We also thank R. Holmes for her dedicated work in helping to coordinate the data used for the current report.
- 4National Comprehensive Cancer Network: melanoma: clinical practice guidelines in oncology. v. 1. 2005. Available at http://www.nccn.org/professionals/physician_gls/PDF/melanoma.pdf
- 22Invasive malignant melanomas lacking competence for metastasis. Am J Dermatopathol. 1984; 6: 55S–61S., , , et al.
- 32Applied logistic regression, 2nd ed. New York: John Wiley & Sons, 2000., .
- 33Stata statistical software: release 8.0. College Station, TX: STATA, 2003.
- 34CART: Tree-structured non-parametric data analysis. San Diego: Salford Systems, 1995.,
- 35Classification and regression trees. Boca Raton, FL: Chapman and Hall, 1984., , , et al.
- 38Surveillance, Epidemiology, and End Results (SEER) Program (www.seer.cancer.gov) Public-Use Data (1973–2003), National Cancer Institute, DCCPS, Surveillance Research Program, Cancer Statistics Branch, released April 2006, based on the November 2005 submission.