The 31-gene expression profile stratifies recurrence and metastasis risk in patients with cutaneous melanoma

The American Joint Committee on Cancer (AJCC) groups patients with cutaneous melanoma (CM) into stages (I–IV) based on each patient’s tumor characteristics (Breslow thickness and ulceration), sentinel lymph node (SLN) status and distant tumor metastasis [1]. Further, National Comprehensive Cancer Network (NCCN) guidelines classify patients into two general categories (low-risk stage I–IIA and high-risk stage IIB–III), with high-risk patients receiving more intensive patient management than low-risk patients, including more frequent clinical visits, imaging surveillance, and referrals to surgical or medical oncology. However, while survival rates for patients with CM decrease with increasing AJCC stage, outcomes remain heterogeneous within each AJCC stage and NCCN risk categories (e.g., some ‘low-risk’ patients experience a recurrence while some ‘high-risk’ patients do not). Therefore, adjunctive methods that work in conjunction with current guidelines could increase the accuracy of patient prognosis, improve patient management and care, [2] and improve healthcare costs by reducing unnecessary procedures in truly low-risk patients and allowing for more intensive surveillance for earlier recurrence detection in patients with a truly high metastatic risk [3–7].

Gene expression profiling (GEP) uses gene expression levels as a prognostic tool to predict recurrence risk or response to adjuvant therapies. GEP has been used in conjunction with staging guidelines for breast cancer, prostate cancer and uveal melanoma to improve the prognostic accuracy of staging [8,9]. Similarly, the prognostic 31-gene expression profile (31-GEP) test for CM assesses a patient’s risk of regional tumor recurrence, distant metastasis and death in patients with stage I–III CM. The 31-GEP classifies patients as having low (Class 1A), intermediate (Classes 1B and 2A) or high (Class 2B) risk of regional recurrence or distant metastasis and has shown significant, independent value when used in conjunction with staging factors in retrospective and prospective studies and meta-analyses [10–16]. Further, several single center studies have recently reported the clinical validity and utility of the 31-GEP test [12,17], with Kwatra et al. providing general guidelines for incorporating the 31-GEP into clinical practice [18].

Here we report the results of a multicenter study assessing the ability of the 31-GEP to stratify risk in patients with stage I–III CM, for which additional prognostic information could have the greatest impact on patient care decisions. The study’s primary purpose was to show risk stratification by the 31-GEP in patients with stage I–III CM. A secondary end point was to demonstrate the added prognostic value of the 31-GEP when combined with complete AJCC staging (including SLN status) or with SLN status alone, for recurrence detection.

Materials & methods

Study design & patient demographics

A retrospective chart review was performed at one dermatology and two surgical centers in the USA for patients who had primary CM tumors retrospectively and prospectively tested with the 31-GEP as part of routine clinical care between 2014 and 2019. Data collected included the 31-GEP result, date of melanoma diagnosis, sex, age at diagnosis, primary tumor location, Breslow thickness, ulceration, mitotic rate, presence or absence of regression and tumor-infiltrating lymphocytes, melanoma subtype, transected base, SLN status, date and location of metastatic events, treatment, date and cause of death. Inclusion criteria included patients who received the 31-GEP test as part of their routine clinical care and totaled 438 patients. Due to the study’s retrospective nature and the minimal risk of a breach of patient confidentiality, the required patient consent was waived by the institutional review board (IRB).

31-GEP testing

Details of the 31-GEP have been described in detail elsewhere [15]. Briefly, the 31-GEP (DecisionDx-Melanoma, Castle Biosciences, Inc., TX, USA) was used to measure the expression of 28 discriminant gene targets across 27 genes and three control genes from formalin-fixed paraffin-embedded primary tumor tissue. 31-GEP testing was performed in a Clinical Laboratory Improvement Amendments (CLIA)-certified, College of American Pathologists (CAP)-accredited laboratory.

Statistical analysis

Comparisons of patient demographics between 31-GEP Classes were performed using the Wilcoxon test for continuous variables and the Chi-square test for dichotomous variables. Kaplan–Meier (KM) survival analysis and the log-rank test were used to compare the event rates and survival outcomes between 31-GEP classes. Outcomes were defined as follows – recurrence-free survival (RFS) measured the time to any regional metastatic recurrence (occurring more than 4 months after diagnosis) or distant metastases and did not include local recurrences. Distant metastasis-free survival (DMFS) measured the time to distant metastases alone, and melanoma-specific survival (MSS) measured the time to death documented as a result of melanoma. Multivariable Cox regression was used to determine hazard ratios (HR) for the 31-GEP and AJCC stage for predicting RFS. Accuracy metrics were determined by comparing Class 1A to 2B results for GEP and SLN positive to SLN negative for SLN status.

Results

Patient demographics

Four hundred thirty-eight patients met the inclusion criteria and were included in the final analysis (Figure 1). Most patients received a Class 1A result (56.6%; 248/438), 21.5% (94/438) received a Class1B/2A result, and 21.9% (96/428) received a Class 2B result. Median patient follow-up time was 1.27 years (range: 0.02–5.09 years) for Class 1A, 1.68 years (range: 0.03–6.59 years) for Class 1B/2A, and 1.56 years (range: 0.21–8.26 years) for Class 2B. There were significant differences between groups for age (p = 0.008), Breslow thickness (Class 1A: 0.7 mm vs Class 2B: 2.8 mm; p < 0.001), presence of ulceration (Class 1A: 3.2% vs Class 2B: 57.3%; p < 0.001), number of mitoses/mm² (Class 1A: 1/mm² vs Class 2B: 4/mm^2; p < 0.001), presence of lymphovascular invasion (Class 1A: 0.4% vs Class 2B: 7.3%; p = 0.007), number of patients with transected base (Class 1A: 33.1% vs Class 2B: 54.2%; p = 0.003), a positive SLN (Class 1A: 11.5% vs Class 2B: 36.6%; p < 0.001), presence of tumor-infiltrating lymphocytes (Class 1A: 63.7% vs Class 2B: 55.2%, p = 0.042), and the number of patients in each AJCC stage (p < 0.001). No significant differences were seen for sex (p = 0.08) or the presence of regression (p = 0.312; Table 1).

Survival analysis

First, we assessed the ability of the 31-GEP test to discriminate between low and high-risk populations within the entire cohort (stage I–III). The 31-GEP significantly stratified risk for RFS (p < 0.001), DMFS (p < 0.001) and MSS (p < 0.001). Patients with a Class 1A result had higher 5-year RFS (95.8 vs 60.5%), DMFS (99.2 vs 71.6%) and MSS (100 vs 75.3%) than patients with a Class 2B result. Of the 27 patients who experienced a recurrence, 17 (630.0%) had a Class 2B result and 24 (88.9%) had a GEP result indicating some degree of elevated risk (Figure 2).

To determine whether the 31-GEP provided significant prognostic information independent of AJCC staging, Cox multivariable regression analysis was performed for the end point of disease recurrence using the 31-GEP class result (Classes 2B, 1B/2A vs low-risk Class 1A) and AJCC high-risk groups (stage IIB–III vs low-risk stage I–IIA) as covariates. Both 31-GEP Class 1B/2A (HR: 4.12 [95% CI: 1.05–16.14]; p = 0.042) and 2B (HR: 5.38 [95% CI: 1.40–20.69]; p = 0.014) were significant predictors of recurrence even when accounting for AJCC high-risk stages (IIB–III: HR: 4.53 [95% CI: 1.70–120.05]; p = 0.003) within the multivariable model (Table 2).

Next, we assessed the ability of the 31-GEP to stratify the risk of recurrence in patients with localized (stage I–II) versus regional (stage III) disease. The 31-GEP significantly stratified RFS (p = 0.004), DMFS (p < 0.001) and MSS (p = 0.002). Patients with a Class 1A result had a higher 5-year RFS (95.2 vs 630.0%), DMFS (99.2 vs 73.2%) and MSS (100 vs 73.8%) than patients with a Class 2B result; whereas, patients with stage III CM had an RFS of 77.3%, DMFS of 80.7% and MSS of 93.6% (Figure 3). Of the 16 patients that experienced a recurrence, nine (56.3%) were identified as Class 2B and 13 (81.3%) had a GEP result indicating some degree of elevated risk. In a subset of patients with a pathologically confirmed positive SLN (n = 56), 21 patients received a Class 1 31-GEP result and 35 received a Class 2 result. Notably, all 11 recurrences (100%) in the stage III population occurred in patients with a Class 2 result, with 8/11 (72.7%) occurring in patients with a Class 2B (high-risk) result. Finally, the 31-GEP stratified patient risk when patients with stage I–III CM were separated by 31-GEP main class (Class 1 vs Class 2) for RFS (p < 0.001), DMFS (p < 0.001) and MSS (p < 0.001) or by 31-GEP subclass (Class 1A, 1B, 2A and 2B) for RFS (p < 0.001), DMFS (p < 0.001) and MSS (p < 0.001; Supplementary Figures 1 & 2).

Similar results were seen in the subset of 265 patients with a pathological nodal assessment (263 sentinel lymph node biopsy [SLNB] and two lymphadenectomies). Patients with a negative SLN (n = 209) had a recurrence rate of 6.2% (13/209) overall compared with a 15.4% (8/52) rate in patients with a negative SLN and a Class 2B result and 2.0% (2/100) in SLN negative patients with a Class 1A result. On the other hand, patients with a positive SLN had a 19.6% (11/56) recurrence rate overall compared with a 0% (0/13) rate for patients with a positive SLN and a Class 1A result and a 26.7% (8/30) recurrence rate for those with a positive SLN and a Class 2B result (Table 3).

In patients with a pathological SLN assessment, the 31-GEP had high sensitivities for RFS (88.9%), DMFS (92.3%) and MSS (100%), while SLN status had substantially reduced sensitivities for RFS (44.4%), DMFS (46.2%) and MSS (33.3%) by comparison. Combining the 31-GEP with SLN resulted in higher sensitivities for RFS (88.2%), DMFS (92.3%) and MSS (100%) and higher NPV’s for RFS (980.0%), DMFS (99.0%) and MSS (100%) than SLN status alone (Table 4).

Discussion

 The AJCC stages patients with CM according to tumor characteristics (Breslow thickness and ulceration), SLN status and if the tumor has metastasized to a distant location [1]. These tumor characteristics have been shown repeatedly to be prognostic for survival in patients with CM. However, individual patients with similar tumor characteristics or equivalent stages can have widely different recurrence and survival rates. One reason may be that current staging guidelines do not consider the genetic profile of the tumor.

We demonstrated that patients with stage I–III CM who receive the high risk (Class 2B) 31-GEP class result are significantly more likely to experience a recurrence and have lower 5-year RFS, DMFS and MSS rates than patients who receive the lowest risk (Class 1A) result.

SLNB is standard for prognostic staging for patients with CM [19]. However, many patients with a negative SLNB experience tumor recurrence or melanoma-specific death, while some patients with a positive SLNB do not [19]. A recent study by Kangas-Dick et al. showed that in a multivariable regression analysis, neither SLNB nor the 31-GEP were significantly associated with RFS or DMFS [20]. However, those results and interpretations contrast with many previous publications supporting the opposite conclusion and may have been observed for several data sampling or methodological reasons. The current study combines SLNB status with other staging factors such as Breslow thickness and ulceration in the form of established staging as a single variable. This approach allowed us to test the contributions of SLNB and GEP status explicitly while avoiding potential dilution of statistical power.

The 31-GEP was able to separate SLN-positive patients into groups that were less likely to experience a recurrence (Class 1A; 0%) and those more likely to experience a recurrence (Class 2B; 26.7%) than by SLN positivity alone (19.6%). In fact, in the SLN-positive population, 100% (11/11) of the recurrences occurred in the high-risk Class 2 31-GEP group. Further, the 31-GEP stratified patients with stage I–II CM and identified a subgroup of patients (Class 2B) with lower 5-year RFS, DMFS and MSS than patients with stage III CM (Figure 3).

A limitation of the study is the short follow-up time, which may limit the number of recurrences detected. However, it should be noted that accurate risk stratification by the 31-GEP was maintained when comparing results from a published cohort with an interim analysis with 1.5 years of follow-up and a final report of the same cohort with 3.2 years of follow-up [10,21]. Other potential limitations of the study include its retrospective study design that limited the analysis to only those patients with a successful 31-GEP result, a low number of patients with stage III disease that limited analysis to main rather than substage and a relatively low number of distant metastatic events or melanoma-specific deaths that limited multivariable analyses for these event types. Furthermore, while the number of patients who underwent SLNB loosely aligns with NCCN guidelines for performing SLNB, we did not explore why certain patients received an SLNB while others did not.

Combined with the 31-GEP’s ability to independently predict recurrence risk in a multivariable analysis, the data presented in this report support that the 31-GEP can add prognostic information for more personalized and risk-aligned patient management.

Conclusion

This report supports the use of the 31-GEP as adding independent prognostic value to current staging guidelines for CM, adding to the growing body of evidence supporting the use of the 31-GEP in conjunction with current staging factors to identify patients at high and low recurrence or metastasis risk to improve patient management [6,11,12,14,16,22].