Secondary Logo

Journal Logo

Original Article

Estimation of Prognosis in Invasive Cutaneous Melanoma: An Independent Study of the Accuracy of a Gene Expression Profile Test

Greenhaw, Bradley N. MD*,†; Zitelli, John A. MD; Brodland, David G. MD

Author Information
doi: 10.1097/DSS.0000000000001588
  • Open


Cutaneous melanoma (CM) accounts for nearly 75% of deaths from skin cancer while representing only 4% of skin cancers diagnosed.1 Two recently published projections forecast an increasing incidence of CM over the next 15 years in white men and women from the United States.2,3 The seventh edition of the American Joint Committee on Cancer (AJCC) Cancer Staging Manual offers valuable prognostic information for patients.4 In particular, for those with localized disease, Breslow depth offers important staging and prognostic information,4–6 and for some, sentinel lymph node biopsy (SLNB) can provide population-based prognostic information.

However, despite this well-regarded practice for assessing CM prognosis, great heterogeneity exists in clinical outcomes for patients with similar clinical and histopathologic features, highlighting the need for improved estimation of prognosis. For example, in the largest prospective SLNB study to date, MSLT-I, intermediate thickness SLNB-negative CMs caused twice as many deaths as intermediate thickness SLNB-positive CMs. In addition, a false-negative rate for regional recurrence of 6% was reported.7 Other examples include 2 studies showing that deaths from AJCC T1 CMs (≤1 mm) outnumbered deaths from both T2 CMs (1.01–2 mm) and T4 CMs (>4 mm), with similar mortality figures for T1 and T3 CMs (2.01–4 mm).8,9 Fortunately, recent research offers hope for improved prognostication.

A commercially available gene expression profile (GEP) test was validated in industry-sponsored studies and shown to add valuable prognostic data to current AJCC staging methods by identifying those CMs that are higher risk factors for metastases.10,11 Using formalin-fixed paraffin-embedded tissue, the GEP test evaluates the expression of 31 genes10 to generate a probability score of 0 to 1. CM tumors are classified as Class 1, low risk, or Class 2, high risk, with 0.50 as the cut-point between the 2 classes. The scores are subdivided as Class 1A “normal confidence,” Class 1B “reduced confidence,” Class 2A “reduced confidence,” and Class 2B “normal confidence.” The “reduced confidence” classes designate those probability scores that fall beyond 1 SD from the median Class 1 and Class 2 scores toward the cut-point of 0.5. The GEP test report provides the expected recurrence-free survival (RFS), distant metastasis-free survival (DMFS) and melanoma-specific survival (MSS) 5 years after the initial diagnosis. The 5-year RFS estimates are 92% for Class 1A, 90% for Class 1B, 77% for Class 2A, and 48% for Class 2B.

After the initial development and validation report,10 a second validation study evaluated the GEP test in a cohort of patients who had undergone SLNB. Although the GEP test provided accurate risk prediction independent of SLNB, prognostication improved when the GEP test was used in conjunction with SLNB.11 In the first study, Stage I CMs represented 42% and 54%, respectively, of the training and validation cohorts, whereas the second study had 28% Stage I CMs.10,11 These proportions are lower than those seen in the general population. For instance, Criscione and Weinstock9 performed a 150,000 case population–based study, which reported that thin CMs (≤1 mm) represented 70% of all CMs.

Because our institution's CM population more closely resembles the CM population encountered by clinicians, the authors performed an independent study to evaluate the test's accuracy and clinical utility in a more representative cohort. In addition, the authors aimed to determine what clinical and histologic features predict high-risk classification and how the area of “reduced confidence” near the cut-point between Class 1 and 2 performs in our patient population.


Data Collection

After Institutional Review Board (IRB) exemption in accordance with the principles of the 1975 Declaration of Helsinki, a retrospectively collected cohort of patients who were treated for invasive CM of any Breslow depth within the last 5 years and had had GEP testing performed was identified using our institution's CM registry. Shortly after the commercial availability of the GEP test, it became the institution's protocol to offer testing to all patients with newly diagnosed invasive CMs and to all patients treated within the previous five years of invasive CMs when the patients were either seen in clinic or called per the registry follow-up protocol. Patients younger than 18 years or pregnant at the time of treatment were excluded from the study. Because the test was not validated for locally recurrent CMs, inclusion was limited to primary CMs with no previous treatment. All CM subtypes were included. Deidentified demographic and clinical data were abstracted from the CM registry. Chart reviews were performed in cases of missing data in the registry. Follow-up data for the registry is prospectively obtained every 3 to 12 months by routine office visits or by regular telephone calls made to patients who are followed by other physicians. The follow-up time for each patient was calculated in months from the date of definitive treatment of the primary tumor until the date of the most recent data entry in the CM registry at the time of study initiation. Therefore, no patients were considered lost to follow-up. Separate IRB approval was obtained to test a subcohort of patients with known metastatic disease that developed after the initial excision. Metastatic events were defined by either locoregional (satellite, in transit, or nodal) or distant metastases. In accordance with the second validation of the GEP test,11 a positive SLN was excluded as a metastatic event.

All CMs were scored as either Class 1A, 1B, 2A, or 2B.

Statistical Analysis

Mean values, SDs and errors, and median and interquartile ranges were calculated for normally and non-normally distributed interval variables, respectively. Frequency distributions were produced for nominal and ordinal variables. Group mean values and medians were tested using parametric and nonparametric methods as appropriate. For significant tests, multiple comparisons of mean values were tested using Tukey contrasts or Dunn test setting the familywise error rate at 0.05 with the Bonferroni12 or Holm13 methods. Survival analysis using the Kaplan–Meier (KM) method at 3- and 5-year intervals was performed for MFS and MSS. For nominal variables, proportions were calculated for each group, and statistical association was tested with chi-squared or Fisher exact tests. Kaplan–Meier curves were plotted using 95% confidence intervals for the GEP groups. Survival differences were tested with the log-rank test. The relationship of pathological indicators to GEP score was examined using logistic regression. Statistical analysis was performed using R Version 3.2.4 (Vienna, Austria).



A total of 256 patients were tested. Two hundred fourteen (84%) were Class 1, of which 193 were Class 1A, 21 were Class 1B, and 42 (16%) were Class 2, of which 16 were Class 2A and 26 were Class 2B.

The mean follow-up time for the entire cohort was 23 months, with similar follow-up times for both classes. When comparing traditional prognostic factors, there were no differences in sex or tumor location. Univariate analyses revealed significant differences in age, Breslow depth, mitotic rate, and ulceration status (Table 1). However, with multivariate analysis, only increasing Breslow depth (odds ratio [OR] = 2.4 [1.7–3.6], p < .00001) and the presence of ulceration (OR = 6.7 [2.4–19.4], p = .0003) predicted Class 2 status.

Clinical and Histologic Characteristics

Of the 26 ulcerated tumors, 9 were Class 1, and 17 were Class 2. Seven (41%) of the 17 Class 2 ulcerated tumors metastasized. Only 1 (11%) of the 9 Class 1 ulcerated tumors metastasized, suggesting that ulcerated Class 2 tumors are particularly high risk.

Analysis of the Gene Expression Profile Test's Overall Prognostic Accuracy

Of the 256 tumors tested, 13 developed metastases. The GEP test accurately identified 10 (77%) of these as high risk. As importantly, the test accurately identified low-risk individuals. The negative predictive value (NPV) was 99%, as only 3 of the 214 Class 1 individuals developed metastases. Overall, Class 2 patients were 22 times more likely to develop metastatic disease (Table 2).

Gene Expression Profile Test Performance

Using KM modeling, the 3-year MFS rate was 98% for Class 1 patients and 74% for Class 2 patients, with 5-year MFS rates of 93% and 69%, respectively (p < .00001, Figure 1A). The 5-year MSS rate was 20% higher for Class 1 patients (99%) versus Class 2 patients (79%) (p = .00003, Figure 1B).

Figure 1.:
Kaplan–Meier survival analyses for metastasis-free survival (A) and melanoma-specific survival (B) Blue curves = Class 1A/1B patients and red curves = Class 2A/2B patients.

Analysis of the Gene Expression Profile Test by American Joint Committee on Cancer Stage

At the time of initial CM diagnosis, 219 (86%) of the tested tumors were Stage I. None of the 18 Stage I Class 2 tumors metastasized, whereas 1 (0.5%) of 201 Stage I Class 1 tumors metastasized.

Twenty-four of the 37 Stage II tumors had a Class 2 GEP, and 10 (42%) of these tumors metastasized. Of the 13 Stage II Class 1 tumors, 2 (15%) metastasized.

Analysis of “Reduced Confidence” Scores

As hypothesized, Class 1B and 2A tumors were intermediate in risk of metastasis when compared with Class 1A and 2B tumors (Table 3). Moreover, a higher percentage of Class 1B tumors remained metastasis-free when compared with Class 2A tumors, suggesting a progression of risk among these tumors that scored near the probability score cut-point.

Performance of “Reduced Confidence” Classifications Compared With “Normal Confidence” Classifications


Identifying patients with “low-risk” CM who will go on to develop metastatic disease has been a difficult task. A positive SLNB offers staging information and may allow consideration of more aggressive surgical and medical interventions. However, a negative SLNB, which most patients who undergo the procedure have, does not necessarily equate to a “low-risk” CM. In addition, one study found that up to 50% of SLNB eligible patients do not undergo SLNB.14 These findings are troubling because SLNB is the lone prognostic test listed in current National Comprehensive Cancer Network (NCCN) guidelines.15 For patients with CM, who do not meet criteria for SLNB, are unwilling or unable to undergo an additional invasive procedure, or have a negative SLNB, there are no prognostic tests available under current NCCN guidelines to evaluate the metastatic risk.

In this study, the GEP test proved to be a powerful predictor of which individuals were at low risk for metastases. The NPV of 99% offers assurance of a truly low metastatic and mortality risk for Class 1 patients with CM. This knowledge could eliminate anxiety caused by the uncertainty of potential future metastasis and, in turn, could lead to an improved quality of life for many patients with CM. In a confirmatory sense, the NPV also provides value for individuals with a negative SLNB. In the MSLT-I study, two-thirds of all melanoma-specific deaths occurred in the SLNB-negative group, with higher melanoma-specific death rates for false-negative SLNB patients than for positive SLNB patients.7 In the second GEP validation between the GEP test and SLNB, just among the patients who had a negative SLNB, there was a 37% decrease in 5-year DMFS for Class 2 versus Class 1. Five-year overall survival was similarly decreased by 36% for Class 2 patients.11 The GEP test showed the ability to refine prognosis for a subcohort of patients who would otherwise be grouped together using traditional staging methods. Class 2 SLNB-negative patients should receive close clinical surveillance, at minimum, with a low threshold for implementation of imaging modalities.

In our study, the GEP test correctly identified 77% of those who developed metastatic disease as Class 2. In the test's initial validation study, 90% percent of all patients with metastatic disease were correctly identified as Class 2, whereas 80% of AJCC Stage I and IIA patients who developed metastatic disease were Class 210. The slightly higher numbers seen in that study have multiple explanations. First, our patient cohort, largely comprised (86%) of Stage I CMs, did not undergo SLNB. In the study by Gerami and colleagues,10 SLN positivity not only was considered a metastatic event, but also there were only 42% and 54% Stage I CMs in the training and validation cohorts, respectively. Per AJCC staging criteria, that cohort was comprised of higher risk individuals. Because our cohort was mostly thin CMs, it is possible that our improved survival of Class 2 patients resulted from early surgical intervention and cure before any potential metastatic event. Moreover, the median follow-up time was >6 years versus slightly <2 years in our study. These differences explain the smaller number of metastatic events seen in our study, which effectively diminished opportunities for test accuracy among our Class 2 patients.

Among AJCC Stage II patients who later developed metastases in this study, 10 of 12 (83%) were correctly identified as Class 2. In MSLT-I, which omitted data on thin CMs, SLNBs correctly identified 178 of 222 (81%) Stage II CMs that metastasized.7 Despite the relatively short follow-up period of our study, the GEP test compares favorably for Stage II CMs.

This study was limited by its follow-up period and number of metastatic events. Acknowledging the abbreviated follow-up time likely prevented capture of some metastatic events, the authors sought to determine the extent to which the limited follow-up time affected our data. A large population study from Germany, which included 1,078 patients who developed metastasis within 10 years of their initial respective diagnoses, showed that the median time to metastasis was 23 months for men and 25 months for women.16 Through a personal communication with the authors of the original validation study, the authors learned that the median time to recurrence for Class 2 patients in that study was 1.71 years.10 Our Class 2 patients had a median follow-up time of 1.92 years. These data considered collectively suggest that our data likely captured most metastatic events. A KM plot comparing our cohort with the Stage I and II patients from Gerami's validation cohort10 showed similar curves for MFS (Figure 2). The authors expect the curves for Class 2 patients to move closer to one another as the authors gather longer-term follow-up data, which the authors plan to publish once available.

Figure 2.:
Comparison of metastasis-free survival in our cohort versus Gerami's validation cohort10—solid lines represent our study cohort, and dashed lines represent Gerami's cohort. Blue curves = Class 1A/1B patients and red curves = Class 2A/2B patients.

Despite this study's limitations, the authors believe the GEP test demonstrated an ability to stratify patient risk with acceptable accuracy in our independent study and in the test's previous validation studies,10,11 Therefore, our current protocol is to offer the test to all patients with invasive CMs. The GEP score then dictates the appropriate follow-up schedule. Class 1 patients undergo clinical skin and nodal examinations twice yearly for 2 years, then yearly thereafter. Class 2 patients are examined every 3 months for 2 years, then every 6 months for 3 years, then yearly thereafter. Our goal is timely detection of metastases, with earlier institution of salvage therapies.

In conclusion, the GEP test proved to be a prognostic tool with high accuracy for low-risk patients and accuracy approximating SLNB accuracy for high-risk patients. The information it provides has the potential to help direct patient management. Long-term follow-up studies will be needed to further strengthen our findings.


1. Rigel DS. Epidemiology of melanoma. Semin Cutan Med Surg 2010;29:204–9.
2. Guy GP, Thomas CC, Thompson T, Watson M, et al. Vital signs: melanoma incidence and mortality trends and projections—United States, 1982–2030. MMWR Morb Mortal Wkly Rep 2015;64:591–6.
3. Whiteman DC, Green AC, Olsen CM. The growing burden of invasive melanoma: projections of incidence rates and numbers of new cases in six susceptible populations to 2031. J Invest Dermatol 2016;136:1161–71.
4. Balch CM, Gershenwald JE, Soong SJ, Thompson JF, et al. Final version of 2009 AJCC melanoma staging and classification. J Clin Oncol 2009;27:6199–206.
5. Breslow A. Thickness, cross sectional areas and depth of invasion in the prognosis of cutaneous melanoma. Ann Surg 1970;172:902–8.
6. Freeman SR, Gibbs BB, Brodland DG, Zitelli JA. Prognostic value of sentinel lymph node biopsy compared with that of Breslow thickness: implications for informed consent in patients with invasive melanoma. Dermatol Surg 2013;39:1800–12.
7. Morton DL, Thompson JF, Cochran AJ, Mozzillo N, et al. Final trial report of sentinel-node biopsy versus nodal observation in melanoma. N Engl J Med 2014;370:599–609.
8. Whiteman DC, Baade PD, Olsen CM. More people die from thin melanomas (≤1mm) than from thick melanomas (>4mm) in Queensland, Australia. J Invest Dermatol 2015;135:1190–3.
9. Criscione VD, Weinstock MA. Melanoma thickness trends in the United States, 1988-2006. J Invest Dermatol 2010;130:793–7.
10. Gerami P, Cook RW, Wilkinson J, Russell MC, et al. Development of a prognostic genetic signature to predict the metastatic risk associated with cutaneous melanoma. Clin Cancer Res 2015;21:175–83.
11. Gerami P, Cook RW, Russell MC, Wilkinson J, et al. Gene expression profiling for molecular staging of cutaneous melanoma in patients undergoing sentinel lymph node biopsy. J Am Acad Dermatol 2015;72:780–5.
12. Abdi H. Bonferroni and Šidák corrections for multiple comparisons. In: Salkind NJ, editor. Encyclopedia of Measurement and Statistics. Thousand Oaks, CA: Sage; 2007.
13. Holm S. A simple sequential rejective multiple test procedure. Scand J Stat 1979;6:65–70.
14. Bilimoria KY, Balch CM, Wayne JD, Chang DC, et al. Health care system and socioeconomic factors associated with variance in use of sentinel lymph node biopsy for melanoma in the United States. J Clin Oncol 2009;27:1857–63.
15. Coit DG, Andtbacka R, Anker CJ, Bichakjian CK, et al. Melanoma, version 2.2013: featured updates to the NCCN guidelines. J Natl Compr Canc Netw 2013;11:395–407.
16. Mervic L. Time course and pattern of metastasis of cutaneous melanoma differ between men and women. PLoS One 2012;7:e32955.
Copyright © 2018 The Author(s). Published by Wolters Kluwer Health, Inc. on behalf of the American Society for Dermatologic Surgery, Inc.