Journal of Clinical and Aesthetic Dermatology • Hot Topics in Squamous Cell Carcinoma July 2024 • Original Research

Sarah T. Arron, Javier Cañueto, Jennifer Siegel, Alison Fitzgerald, Anesh Prasai, Shlomo A. Koyfman, Sue S. Yom 2024-06-26 06:19:35

Association of a 40-gene Expression Profile with Risk of Metastatic Disease Progression of Cutaneous Squamous Cell Carcinoma (cSCC) and Specification of Benefit of Adjuvant Radiation Therapy

Dr. Arron is with Peninsula Dermatology in Burlingame, California. Dr. Cañueto is with Department of Dermatology, Universitario de Salamanca in Salamanca, Spain. Drs. Siegel, Fitzgerald, and Prasai are with Castle Biosciences in Friendswood, Texas. Dr. Koyfman is with Department of Radiation Oncology, Taussig Cancer Institute, Cleveland Clinic in Cleveland, Ohio. Dr. Yom is with Department of Radiation Oncology, University of California San Francisco, San Francisco, California.

Reprinted with permission. Int J Radiat Oncol Biol Phys. 2024:S0360-3016(24)00685-0. Epud ahead of print.*

Purpose: Adjuvant radiation therapy (ART) for cutaneous squamous cell carcinoma (cSCC) is recommended based on a number of wide-ranging clinicopathologic features, which encompass a broad array of patients. The 40-gene expression profile (40-GEP) test classifies cSCC tumors into low (Class 1), higher (Class 2A), or highest (Class 2B) risk of nodal and/or distant metastasis. This study’s hypotheses are 1) local recurrence is associated with metastatic disease progression and 2) 40-GEP, by identifying high risk for metastasis, could predict a metastasis-specific benefit from ART. Methods: Samples were obtained from 920 patients (ART-untreated: 496 Class 1, 335 Class 2A, 33 Class 2B; ART-treated: 11 Class 1, 35 Class 2A, 10 Class 2B) who were matched on clinical risk factors and stratified by ART status to create 49 matched patient strata. To control for the variety of characteristics and treatment selection bias, randomly sampled pairs of matched ART and non-ART patients comprising 10,000 resampled cohorts were each analyzed for five-year metastasis-free survival and predicted time to metastatic event. Results: Of 96 patients experiencing local recurrence, 56.3 percent experienced metastasis; of those experiencing both, 88.9 percent had local recurrence before (75.9%) or concurrently (13.0%) with metastasis. After matching for clinicopathological risk, median five-year disease progression rates for resampled cohorts demonstrated approximately 50-percent improvement for Class 2B ART-treated cohorts as compared to ART-untreated cohorts. Class 2B ART-treated cohorts had a five-fold delay in predicted time to metastatic event and deceleration of disease progression as compared to ART-untreated cohorts (Kolmogorov-Smirnov test, p<0.01); this was not observed for Class 1 or 2A patients (p>0.05 for each). No risk factor or staging system combined with ART status identified groups that would benefit from ART as well as 40-GEP. Conclusion: 40-GEP identifies patients at highest risk of nodal/distant metastasis who may derive the greatest benefit from ART, as well as patients who may have clinical indications for ART but are at low risk of metastasis. Compared to current guidelines, 40-GEP could provide greater specificity concerning the benefit of ART in individual patients.

KEYWORDS: Cutaneous squamous cell carcinoma, skin cancer, gene expression profiling, prognostic, predictive, nodal metastasis, metastasis risk reduction

FUNDING: Sequencing and data analysis support provided by Castle Biosciences, Inc.

DISCLOSURES: STA reports being a paid consultant for Enspectra Health, Castle Biosciences, and WorldCare Clinical; paid speaker bureaus for Regeneron and Castle Biosciences; paid expert testimony from Forensis, Inc, and Muro & Lampe; travel expenses paid by Castle Biosciences; paid participation on a data safety monitoring board for Replimune; paid advisory board participation for Regeneron, Dermatology Times/Multimedia Medical, and Matrix Medical Communications; leadership in the International Transplant Skin Cancer Collaborative; and stock holding in Rakuten Medical and Enspectra Health, with stock held by spouse in Genentech and 23andMe. JS, AF, and AP are employees and stockholders of Castle Biosciences, Inc. SAK reports being a paid consultant for Merck, Regeneron, Bristol-Myers Squibb, and Galera Therapeutics, and had travel expenses paid by Castle Biosciences, Inc.; he received research support from Merck, Bristol-Myers Squibb, Regeneron, and Castle Biosciences, Inc. and honoraria from UpToDate. SSY reports research funding from Bristol-Myers Squibb, EMD Serono, and Nanobiotix, personal payments of book chapter royalties from UpToDate and Springer, and editorial honoraria from Elsevier and ASTRO.

*COPYRIGHT: © 2024 by the authors. Published by Elsevier, Inc. This article is an open-access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

Of nearly two million cases of cutaneous squamous cell carcinoma (cSCC) diagnosed each year,1–3 approximately 74,000 (3.7–5.2%4) will progress to nodal or distant metastasis. For patients deemed to be at high risk of recurrence, current management guidelines typically recommend definitive surgical resection where possible, followed by consideration of adjuvant radiation therapy (ART), possibly in combination with systemic therapies. The decision to prescribe ART is challenging,5–7 given the large number and extent of recommended clinicopathologic factors that inform risk assessment and indications for ART.8–10 These decisions are further complicated as these clinicopathological factors have not been consistently independently demonstrated to predict benefit from ART.5,11

The 40-gene expression profile (40-GEP) test is a prognostic tool designed to assess the risk of metastasis (nodal and/or distant metastasis) in patients diagnosed with cSCC and having one or more clinicopathologic risk factors. Results are reported as low (Class 1), higher (Class 2A), or highest (Class 2B) metastatic risk. The test determines the risk of a future metastatic event based on the pattern of gene expression observed for a specific set of genes in malignant and non-malignant cells located in the tumor microenvironment; both analytical and clinical validity for risk stratification in patients with high-risk cSCC have been described.12–14 Validation studies show that the test adds prognostic value independent of current risk assessment methods13–15 and could positively influence treatment decisions for patients with high-risk cSCC,16,17 based on more personalized management.18–20

The goal of this study was to determine whether the 40-GEP could be used to identify patients with high-risk cSCC who achieve benefit from ART through preventing metastatic disease progression. Specifically, the hypotheses were that 1) a temporal relationship could be identified between local recurrence and metastatic progression and 2) application of 40-GEP testing, by identifying patients at high risk for metastasis, could predict an increase (improvement) in metastasis-free survival (MFS) after receiving ART, with prolongation in the estimated time to disease progression.

Patient selection and matching on clinicopathologic risk factors. This study was conducted under an institutional review board approval including waiver of patient consent. As previously described,14 archival formalin-fixed, paraffin-embedded (FFPE) cSCC tumor tissue and associated clinicopathologic data were obtained for patients diagnosed with cSCC from February 2006 to October 2020 at 59 participating institutions. Tissue from one single primary tumor for each patient could be enrolled in the study, and in patients with multiple cSCCs, clinicians were asked to submit the tumor thought to have the greatest risk of disease progression. To be accepted into the study, cases were required to have availability of the original paraffin-embedded tissue block from the primary tumor and a minimum of three years of documented follow-up or report of a non-local metastatic event (regional or distant). Selection of samples was without consideration or stipulation regarding treatments, and 40-GEP test results were unknown to sites. Inclusion/exclusion criteria have been previously described;14 of note, cases with a prior history of cSCC, cutaneous basal cell carcinoma, or melanoma in situ were accepted if these prior malignancies were considered cured. Only patients who would qualify for 40-GEP testing on the basis of having at least one qualifying risk feature were included. Qualifying features were those listed in the National Comprehensive Cancer Network (NCCN) risk assessment categories of “high” or “very high” risk, plus one additional qualifying feature of tumor diameter of 2cm or greater (1.7% of this cohort). Eligibility for this present analysis is shown in the flow diagram in Figure 1, and demographics of the cohort are presented. Importantly, the 40-GEP risk status of the included patients was unknown by clinicians at the time of diagnosis and treatment decisions.

All analyzed patients had definitive surgery, and a subset had ART. ART-treated patients were enrolled by 29 of the 59 contributing sites, with each site contributing 2 to 14 percent of the total number of ART-treated patients in this study (mean: 2.15 patients/site). Treatment volume (available for 54 of the 56 ART-treated patients) indicated that in addition to the primary site, the regional basin was treated in 17 percent of Class 2A patients and 20 percent of Class 2B patients. From responses of 916 out of 920 patients originally screened for this study, only 1.7 percent received adjuvant systemic therapy; given its low prevalence, it was not feasible to control for this feature in addition to ART. Among those patients who received ART, the percentages also receiving adjuvant systemic therapy were 18 percent of Class 1, less than 10 percent of Class 2A, and 20 percent of Class 2B.

Due to the inherent biases that would have guided ART treatment decisions at certain centers, ART-treated and untreated patients were matched according to clinicopathologic risk factors in order to control the analyses. Based on propensity score matching using a general linear model (MatchIt_4.5.3 package in R v4.3.0, using “optimal full” matching for a 1:1 ratio), ART-treated and untreated patients were matched on age, immunosuppression status, tumor location, tumor diameter, tumor thickness, invasion of fat, perineural invasion (PNI), and type of definitive surgical treatment received (Supplementary Figure 1). Propensity score matching resulted in a successful match for all 920 patients into 49 matched strata, in which one or more ART-treated patients were matched to one or more ART-untreated patients. To utilize all possible matched patient pairs without excluding any patients, x10,000 iterations of random pairs of ART-treated and ART-untreated patients from each matched strata were sampled (n=98 patients, 49 pairs), and then bootstrap resampling (with replacement, n=98) was used to allow generalization of the results to the global sample set (Figure 1). Each iterated cohort was analyzed as described below. Note that the number of iterations invoked when using random resampling methods is a function of result stability, such that the greater the number of iterations, the less likely that the results may be different depending on the random seed set at the start; this is different than a “modeling” effort for which overfitting is a concern.

Statistical analysis. Each sample for each iteration formed a cohort. Patients within each sample represented a range of risk levels, matched between ART-treated and untreated patients. Each sample cohort was analyzed using survival analysis (survival_3.5-3 package in R v4.3.0), and using the statistical method of “life expectancy” (substituting time to metastatic event in place of death), estimations for each sampled cohort were calculated for regional or distant metastatic events and recorded for each bootstrapping iteration, for each 40-GEP Class x ART treatment status group (k=6 groups). The primary comparisons were between ART-treated and untreated patients within a given GEP class. The distribution of metastatic event types from this analysis was 66.4 percent regional only, 29.2 percent regional and distant, and 4.4 percent distant only.

The Kolmogorov-Smirnov two-sample test was used to determine statistical differences in sampling distributions of estimated time to metastatic event between ART-treated and untreated patients. However, because the number of observations in each of the tested distributions reflected the number of bootstrap iterations and not the sample number, the critical value (Dcrit) for significance was adjusted to appropriately reflect the n of each sample group (n=98/k=16.33 per group), in for alpha values of 0.05, 0.01, and 0.001. An

observed D value (Dobs) greater than Dcrit indicated a significant difference between the distributions. For this data set, the Dcrit values were 0.48, 0.57, and 0.68, respectively, for the alphas of 0.05, 0.01, and 0.001. Significance was treated as a one-sided test, in which significance would result if the ART-treated group showed a benefit of a longer time to predicted metastatic events. It is very important to note that the time to predicted metastatic event for a given resampled cohort reflects the point at which greater than 50 percent of the cohort is projected to experience an event, with longer event times reflecting a mathematical ceiling on that projection, rather than indicating that a given group will experience actual events at that time point.

In order to address the frequency with which an ART-treated group consistently showed better outcomes within each 40-GEP class for each sampled cohort, the estimated time to metastatic event for the ART-untreated group was subtracted from that of the ART-treated group for each sample iteration. Therefore a positive difference indicated a longer time to metastatic event for the ART-treated group.

Extensive control analyses were performed to test the robustness of the results as well as check for any procedural biases or spurious correlations. One important control was to randomize the GEP result for each sampled cohort, to ensure that the low frequency of the Class 2B designation could not procedurally generate the observed results. Other control analyses tested whether any of the individualmatched risk factors alone could identify an ART benefit as well as 40-GEP, or which in matching might have led to a spurious result falsely attributed to 40-GEP. For the former, the entire analysis was repeated dropping each risk factor from the matching and stratifying that factor with ART treatment status. For the latter, each matching factor was dropped one at a time to determine whether the 40-GEP determinations held. Finally, alternative risk and staging schemas were tested to determine whether any could identify an ART-benefiting group of patients as well as 40-GEP.

Patterns of disease relapse and temporal sequence of progression. In this cohort of 920 patients, 113 experienced a non-local recurrence (regional or distant, 12.3%) and 96 experienced a local recurrence (10.4%). Of the 96 patients reporting a local recurrence, all had received definitive surgical treatment, and 54 (56.3%) progressed to metastasis (nodal or distant). Of these 54 patients who developed metastasis, 98 percent developed nodal disease and 64.8 percent also developed distant metastases. This sequence of disease-associated events supports a mechanism of progression from local recurrence to regional and distant metastatic events. Among patients who had both local recurrence and metastasis, 88.9 percent reported the local recurrence event before or in conjunction with the metastatic event, and 41 patients (75.9%) reported the local event as occurring before the metastatic event. The median time between local recurrence and metastasis was three months (inquartile range [IQR]: 1.2–8 months; Wilcoxon test, p<0.001, Ho: time delta=0), but 21 patients (51.2%) experienced a regional or distant metastasis at greater than three months after local recurrence. Similarly, 47.8 percent (n=54/113) of metastases were associated with a local recurrence, and in the presence of local recurrence, metastasis occurred in an accelerated timeframe, as compared to metastasis in the absence of local recurrence (median 9.3 months from diagnosis to metastatic event vs. 13 months, respectively; Wilcoxon test, p=0.06).

Analysis of 40-GEP testing in relation to metastatic risk. Patients with verified ARTtreated or untreated status were risk-matched on multiple clinical features (Supplementary Figure 1), analyzed for risk of metastatic disease progression (using survival methods) and projected time to metastatic event, and stratified by 40-GEP results and ART status (see Methods; ART-untreated: 496 Class 1, 335 Class 2A, 33 Class 2B; ART-treated: 11 Class 1 [nodal region irradiated in 1/11], 35 Class 2A [nodal region irradiated in 6/35], 10 Class 2B [nodal region irradiated in 2/10]). Median disease progression rates for the resampled cohorts demonstrated a 50-percent decrease on average, specifically for Class 2B, ART-treated patient cohorts as compared to Class 2B, untreated patients, at five years post-diagnosis (Figure 2A). Improvement was also observed for the within-cohort differences in five-year metastatic progression rates only for the Class 2B patients (ART-untreated – ART-treated, 46.2% median difference; Figure 2B). ART decreased the disease progression rate of Class 2B patients to a rate similar to that of Class 2A patients.

The estimated time to metastatic event predictions were generated for each sampled cohort. Note that the time to metastasis is based on cohort probability trajectories as used in traditional life expectancy calculations (see Methods). The observed data for a given sampled cohort (accumulation of events over time) is used to project the point in time where at least half of the cohort would have experienced an event based on the observed trajectory, such that longer times reflect the ceiling of that projection (mathematically). The distributions of results from the 10,000 resampled cohorts are shown in Figure 2C and reveal a substantial difference in projected times to metastatic events between ART-treated and untreated Class 2B patients. For ART-untreated patients, there is a peak rate of metastasis around two years, while for Class 2B ART-treated patient cohorts, there is a nearly five-fold increase in the mathematically projected time to metastatic events. It is critical to understand that the peak timing observed for Class 2B ART-treated patients is near those observed for Class 1 and Class 2A, independent of ART treatment, and thus likely reflects a ceiling in the mathematical projection, not a prediction for event occurrence at that time.

The Kolmogorov-Smirnov test was used to determine whether the distributions of time to metastatic event were significantly different between ART-treated and untreated patients for each 40-GEP Class result. Class 2B ART-treated and untreated groups showed a significant difference in the rate of accumulated cohort events (p<0.01, Class 2B Dobs=0.61; Figures 3A and 3B), while differences between ART-treated and untreated patients in Class 2A and Class 1 cohorts were not statistically significant (p>0.05, Class 2A and Class 1 Dobs=0.38 and 0.31, respectively; Figures 3A and 3B). The cumulative probability distributions for each 40-GEP group of ART-treated and untreated patients (a simple transformation of the time event distributions as accumulating over time, reflecting the velocity of distribution shape) demonstrated the underlying functions associated with disease progression in each patient group (Figures 3B and 3C). Class 1 and Class 2A ART-treated and untreated patient cohorts accumulated metastatic events according to a sigmoidal function, in which events began to accumulate at a low rate and then increased monotonically over a 5- to 6-year period until the prediction became asymptotic. In sharp contrast, ART-untreated Class 2B patient cohorts showed an abrupt increasing exponential accumulation of events, suggesting a different underlying process of disease progression that was accelerated relative to Class 1 and Class 2A patients. Interestingly, the disease progression of Class 2B patients treated with ART reverted to a function and time course that mimicked Class 2A patients, reflecting a decreased velocity in the accumulation of metastatic events over time as compared to Class 2B ART-untreated patients.

Of further interest is how often the ARTtreated Class 2B group might show deceleration of projected disease progression relative to the ART-untreated Class 2B patients. Therefore, for each of the 10,000 resampled iterations, the predicted time to metastatic events between the Class 2B ART-treated group was subtracted from that of the ART-untreated Class 2B group (Figure 3D). When estimated in this manner, the Class 2B ART-treated subset demonstrated a longer time to disease progression than the Class 2B ART-untreated subset in 88.3 percent of the resampled cohorts. In contrast, Class 1 and Class 2A patient subsets showed this benefit from ART in only 18.7 and 20.6 percent of resampled cohorts, respectively (Figure 3D).

Control analyses. First, to determine whether the observed results could have been driven by the matching to any single clinicopathologic risk factor, each factor was systematically dropped from the matching scenario. In this control analysis, the same overall result (that a Class 2B result identifies patients who will benefit from ART in terms of metastatic progression) was observed regardless of which factor was not used in matching, including type of definitive surgery used (Table 2). Although not every result was statistically significant, each revealed a benefit for Class 2B ART-treated patient cohorts as a positive difference in median time to metastatic event, which was not observed for Class 2A or Class 1. Second, as the proportion of 40-GEP Class results was not the same within the cohort, with Class 2B being observed less frequently than Class 1 or Class 2A (Table 2), it was tested whether these proportion differences could have influenced the findings. Therefore, 40-GEP results were randomly reassigned to ART-treated and untreated patients after each resampling procedure. The results clearly showed that the proportional differences of 40-GEP class results did not drive a spurious finding, and that the randomization of the test result negated the findings (Table 2). Third, and perhaps most importantly, removing the 40-GEP results from the analysis and stratifying by ART status alone also did not reveal a metastatic prevention benefit from ART in patient cohorts matched by clinical factors (Table 2).

Experimental analyses. Additional analyses tested whether any of the clinicopathologic risk factors used for matching, when combined with ART status, could itself be used to identify patients that would benefit from ART treatment (Table 3; note that the stratifying feature was dropped from matching for this test). No risk factor combined with ART status was able to better identify a patient group that would potentially benefit from prevention of metastatic progression through ART treatment. Finally, a last set of comparisons was to use the NCCN risk category (high or very high risk) or stage (Brigham and Women’s Hospital [BWH] T1/2a and T2b/3 or American Joint Committee on Cancer [AJCC] T1/2 and T3/4 as binary risk levels) in place of the 40-GEP result to determine whether a group of patients who would have prevention of metastatic progression could be identified solely based on these risk assessment criteria; no patient cohort groups could be identified (Table 3).

The distribution of 40-GEP results across BWH T3 patients was 21, 64, and 15 percent for Class 1, 2A, and 2B, respectively, and 33, 53, and 13 percent for BWH T2b patients, respectively. The odds ratio of a T3 tumor receiving a Class 2B was only 1.18 times that of a T2b tumor. The distribution of 40-GEP Class 2B patients across BWH stages was 28, 26, 33, and 14 percent, respectively for T1 to T3, with a T3 patient being twice as likely to receive a Class 2B than Class 2A result, and a T2b patient being 2.7 times as likely to receive a Class 2B versus Class 2A result. Together, these data suggest that there is not a strong relationship between Class 2B 40-GEP results and BWH T3 stage. Similarly, the distribution of 40-GEP results for patients with tumors 6cm or greater in diameter was 22, 59, and 19 percent, respectively, again failing to show a strong relationship between Class 2B and large tumor diameter. Overall, when compared to all other methods of risk assessment, only 40-GEP Class 2B patients were independently identified as benefiting from reduced metastatic risk after ART treatment.

A major challenge in the use of ART for cSCC lies in determining which patients benefit the most from ART and what exactly that benefit entails. Here, we show for the first time that in addition to the 40-GEP’s established stratification of metastatic risk in patients with cSCC who have at least one qualifying factor,13,14 this test can be used to identify patients who are most likely to benefit from prevention of metastatic disease progression as a result of ART. In addition, these results demonstrate a mechanistic association, with the incidence of local recurrence driving accelerated disease progression (nodal or distant metastases).

Although recurrence is not a desirable outcome for any patient, regional disease progressions in the absence of distant metastasis may be surgically salvageable as a means to prevent further progression and death. Nevertheless, avoidance of any recurrence is preferable, and the extent to which control of overall disease progression can be achieved via ART of the tumor bed (RT being administered mostly to the primary site only in this study) has not been previously demonstrated at this level of clarity. Figure 3C highlights the dramatic difference in metastatic disease progression for patient cohorts of Class 1 and Class 2A versus Class 2B during the first five years after diagnosis. It should also be noted that patients presenting with high-risk clinicopathologic features but who have a Class 1 or 2A 40-GEP test result still show a limited degree of metastatic risk-related benefit from ART, but not at a significant level and to a much lesser degree than Class 2B patients (Figure 3B). These data thus provide patient-specific information that more precisely guides discussions about the type and extent of benefit of ART, in a manner superior to clinicopathologically determined risk alone.

For a given patient, final treatment decisions should incorporate all available information, including patient comorbidities and functional status, as well as access to care, in addition to complete clinicopathologic and biologically based risk assessment. Within a framework of multidisciplinary consensus, radiation oncologists are advised to discuss the risks and benefits of radiation with all patients per standard guidelines but could consider a 40-GEP Class 2B result as indicative of a highly specific confirmation of increased nodal and distant risk.21 Furthermore, even acknowledging the inherent uncertainties of diagnostic workup identifying the exact timing of nodal or distant metastases, our study indicated relative proximity of a nodal or distant metastatic event to local recurrence, either simultaneous with or within 1.2 to 8 months after local recurrence. Thus, the Class 2B result might also be regarded as a matter of some potential urgency, indicating a need for promptness in initiating ART.

Within this study, evaluation of clinicopathologic risk factors, including those recommended within the NCCN guidelines, and staging classifications used in the AJCC8 and BWH systems alone did not reveal a patient group as clearly benefiting from prevention of metastatic disease progression through ART as was seen in the 40-GEP Class 2B group (Table 3). Equally interesting, Class 2A patient cohorts did not show an intermediate-level benefit of ART between the low-risk Class 1 and highrisk Class 2B patient groups, even though the overall risk of these three groups was in-line with the monotonicity and distinctiveness reported for the 40-GEP test13,14 (Figures 2 and 3). This observation suggests that the level of metastatic risk may not be the important factor in identifying patients for whom ART will provide prevention of metastasis. Rather, it is possible that a specific pattern of gene expression captured uniquely by the Class 2B result may be driving the identification of higher metastasis-related benefit from ART.

The cohort utilized is a reflection of the cSCC population and time period from which it was sampled, during which increased use of ART by clinicians and refinement of practices occurred. For example, perineural invasion quantified as the number of involved nerves has more recently been shown to associate with risk of metastasis, but this was not captured for the cohort, versus caliber of largest involved nerve.22 However, the early-era timing of our sample collection made possible the matching procedure, as patients who might be irradiated now were not irradiated in our cohort.

Limitations. The major limitation of this current study is the infrequently observed Class 2B result and the limited sample size available for analysis. As a result, resampling procedures were enacted after matching to generate the largest amount of information possible from these few cases. This may be criticized, but we note that statistical power is a function of both the number of samples as well as effect size, with larger effects requiring fewer samples for statistical testing. The relatively large effect size revealed here for Class 2B patients, as well as the extensive control analyses performed, reinforce the findings. A subsequent validation study will be important to confirm these effect sizes. Furthermore, the relative lack of metastasis-related benefit for the Class 1 and Class 2A patient groups is supported, as these larger patient groups were powered to detect significant differences, if they had occurred, for at least a modest effect. It remains possible that a metastasis-related benefit of ART could be found in a subset of Class 2A patients with higher-risk clinicopathologic features, but this would require a larger sample of patients to detect such a small-to-modest effect size. Lastly, although this current work focused on the role of ART in prevention of non-local disease progression predicted by the 40-GEP test, local recurrences that do not lead to regional or distant progression are still an important clinical challenge and source of substantial morbidity. Unfortunately, the findings presented here cannot address this clinical need.

Specific details about radiation dosage, volume, and technique were too sparse to be included in analyses. However, a relationship between ART treatment details and GEP result seems unlikely given the independence of the 40-GEP test from schemas such as staging and its unknown status at the time ART was delivered. Another concern may be systematic differences between irradiation of the tumor bed versus the regional basin as related to GEP result. However, the regional basin was treated in 20 percent of Class 2B patients and 17 percent of Class 2A patients, suggesting that the differential benefit seen in Class 2B was not likely due to inclusion of the regional nodes. While this study demonstrates the high value of ART, whether local or locoregional, in Class 2B patients, it cannot answer the question of when to include the draining lymph node basin(s) in the irradiated volume.

Finally, patients receiving adjuvant systemic therapy were neither explicitly excluded nor controlled in analysis. However, only 1.7 percent of patients in the entire cohort had adjuvant systemic therapy, due to the approval of immunotherapy occurring late in the data collection period. As 18 percent of Class 1 patients, less than 10 percent of Class 2A patients, and 20 percent of Class 2B patients received some form of adjuvant therapy, a correlation with adjuvant therapy is unlikely to explain the observed results.

In conclusion, the 40-GEP test was able to identify patients projected to have the highest metastatic-related benefit from ART, which was associated with improved survival and deceleration or abrogation of nodal or distant metastasis. The 40-GEP test also identified patients who have very low metastatic risk (<10% in Class 1) where deployment of ART requires further justification based primarily on the risk of local recurrence. Within the parameters and stated limitations of this investigation, the 40-GEP test discriminated for the metastatic benefit of ART more specifically than guidelines-based clinicopathologic indications, risk assessment criteria, or staging systems in current use.

It is important to note that while ART is most clearly warranted in Class 2B patients based on the reduction of nodal and distant metastasis, there may be an unidentified subset of Class 1 or 2A patients with clinical high-risk features where there would be similar benefit from ART. In addition, there may be a benefit of ART in reducing local recurrence, which this study did not address. However, in this study, an important link was demonstrated between local recurrences and non-local metastases, providing the basis for understanding how the 40-GEP test could be utilized to identify patients who should be most strongly considered for ART based on reduction of metastatic disease progression.