Merkel Cell Carcinoma: Changing Practice Patterns and Impact on Recurrence-Free and Overall Survival at a Single Institution and Nationally

Background Merkel cell carcinoma (MCC) is an aggressive neuroendocrine carcinoma of the skin. Our report describes the evolution of management and characteristics associated with recurrence, disease-specific survival (DSS) and overall survival (OS) in the treatment of MCC. Methods A single institution retrospective review of MCC and SEER data to determine factors associated with RFS, DSS, and OS using a multivariable Cox regression on inverse-probability weighted cohorts. Results One hundred fifty-nine patients were identified with a median age of 75. Of these, 96% were Caucasian and 60% male. Fifty-eight out of 159 (36%) of all patients were deceased with 21/58 (36%) dead from MCC with a median follow-up of 3.1 years. Institutionally, trends over time demonstrated an increased use of immunotherapy with a concomitant decrease in chemotherapy and decreased use of radiotherapy alone. Institutionally and nationally, there has been increased surgical nodal staging. Institutionally, factors associated with shorter DSS included advanced age, active cigarette smoker (p = 0.002), cT2 disease (p = 0.007), and MCC with unknown primary (p < 0.001). Institutionally, factors associated with shorter OS included ages ≥ 75 years (p < 0.001), an immunocompromised state (p < 0.001), truncal primary site (p = 0.002), and cT2 disease (HR 9.59, p < 0.001). Conclusion Changing practice patterns in MCC management have been driven by the adoption of immunotherapy. Our study highlights that competing risks of mortality in MCC patients likely prevents OS from being an accurate surrogate outcome measure to understand factors associated with DSS.

(p \ 0.001). Institutionally, factors associated with shorter OS included ages C 75 years (p \ 0.001), an immunocompromised state (p \ 0.001), truncal primary site (p = 0.002), and cT2 disease (HR 9.59, p \ 0.001). Conclusion. Changing practice patterns in MCC management have been driven by the adoption of immunotherapy. Our study highlights that competing risks of mortality in MCC patients likely prevents OS from being an accurate surrogate outcome measure to understand factors associated with DSS.
In 1972, Toker first described cutaneous trabecular carcinoma (TC), 1 later reclassified as Merkel cell carcinoma (MCC). 2 Although the pathogenesis continues to be ill-defined, two likely pathways, one attributed to UV associated mutagenesis and the other related to the Merkel cell polyomavirus (MCPV) appear to be causitive. 3,4 MCC, which almost exclusively affects patients who are Caucasian, elderly, immunosuppressed, or those with extensive UV exposure, 5 is associated with poor overall survival (OS) ranging from 63 to 81% for stage I to 11-15% for stage IV disease [6][7][8] and can have upwards of 50% recurrence. 9 While MCC is still a rare disease, its incidence continues to rise, and the United States has seen a 95% increase in the incidence from 0.5 cases per 100,000 in 2000 to 0.66 cases per 100,000 in 2016, with a predicted continued increase to 5130 cases/year by 2030. 3,10 Historically, treatment for MCC has been variable and included a combination of surgical resection of the primary with or without sentinel lymph node evaluation, radiation of the primary tumor and/or the associated lymph node basin, and chemotherapy, resulting in generally poor durable responses. 11 Recent clinical trials have demonstrated the efficacy of immunotherapy, while further defining the role of nodal staging and radiotherapy has led to improvements in treatment response. [11][12][13][14][15] Current treatment guidelines recommend wide local excision with sentinel lymph node biopsy and consideration of adjuvant radiation therapy for all N0M0 MCC. 16 However, there is considerable debate concerning what factors may affect recurrence-free survival (RFS), disease-specific survival (DSS) and OS of MCC.
While large databases, such as the National Cancer Database (NCDB) or Surveillance, Epidemiology and End Results (SEER) program of the National Cancer Institute, allow for the study of thousands of patients with MCC, they are unable to provide details concerning diseasespecific or recurrence-free survival. Previous database driven studies have used overall survival as a surrogate for disease-specific outcomes. However, given the predilection of this disease to occur in older and immunosuppressed patient populations, we sought to determine whether such an approach was appropriate, given a likely competing risk of mortality in this population. Additionally, we aimed to both define the changing practice patterns over time for MCC and to determine characteristics associated with recurrence-free, disease-specific, and overall survival through a retrospective review of a single, high-volume institution and the SEER database.

Patient Selection and Variable Definitions
This was a single institution retrospective review of medical records with patients identified using ICD-9 and ICD-10 codes for MCC diagnosed between 2002 and 2020. In all, 174 patients were identified as having biopsy proven MCC. Fifteen were excluded due to incomplete clinicopathologic data, leaving 159 patients evaluable for analysis. Patient demographics, medical histories, diagnoses, and treatments for MCC were recorded. All staging was based on 8 th Edition of American Committee on Cancer (AJCC). 17 To contextualize changes in management at our institution, we looked at national trends for these parameters using the Surveillance, Epidemiology, and End Results (SEER) Program database. The SEER database was queried for new MCC (histology code 8247) cases diagnosed between 2004 and 2016 for therapy trend analysis, and between 1975 and 2016 for survival analysis. Analysis was performed on the first primary tumor case listed for each patient.
Rates for surgeries performed is accurately documented within the SEER database. However, there remains ambiguity in coding for specific procedures. We documented procedures of biopsy followed by gross excision, Mohs micrographic surgery, wide local excision, and amputations as receipt of surgery. Patients who were coded as having excisional biopsies or local ablations were grouped with patients having received no surgery, as these were deemed likely to not be therapeutic in nature. Patients with unknown surgery to the primary site or LN drainage bed were not included in the proportion analysis. Analysis of surgical management of the primary tumor and lymph node drainage bed was performed on patients with known surgical procedures to the respective sites.

Statistical Methods
Age-adjusted therapy utilization rates were calculated. Trends in rates were analyzed through Joinpoint regression, and proportions were fitted using simple linear regression. Proportions were calculated from software reported rate ratios. Ninety-five percent confidence intervals (CI) are reported.
For institutional data, RFS, DSS, and OS were assessed using multivariable Cox regressions. For SEER data, DSS and OS were assessed using bivariate Cox regressions. To adjust for the potential selection bias associated with surgical candidates theoretically being ''healthier'' than nonsurgical candidates, inverse-probability weighting (IPW) was used in our survival analyses. Cox regression models were performed using backward elimination with p out [ 0.05. DSS analysis using the institutional and SEER datasets was performed using competing event Fine and Gray models to account for competing causes of mortality. Patients with an unknown cause of death were not censored from institutional and SEER datasets, and unknown cause of death was treated as a competing cause of mortality.
Overall survival over time was analyzed in the SEER database comparing all-cause mortality in 1975-1986 with the periods 1987-1996, 1997-2006, and 2007-2016 using a univariate Cox regression stratified by patient age (B 64, 65-74, 75-84, and C 85 years). Trends in DSS was not assessed given differences in rates reporting patient cause of death over time, which may have reflected differences in reporting as opposed to true differences in survival outcomes.

National Treatment Trends
Query of the SEER database revealed 6766 patients with complete data available to allow for the analysis of trends in therapy utilization, whereas a total of 9551 patients were available for survival analysis. When comparing trends in surgical approaches from our experience to those nationally, utilizing patient data from the SEER database, we found that there has been an increase in the utilization of surgery and lymph node examination during the same time period (Fig. 2). The rate of surgical resection increased from 0.422 to 0.578 per 100,000 persons from 2004 to 2016, which corresponded to an average annual percentage change (AAPC) of 2.7% per year (95% CI 1.6, 3.8%) ( Fig. 2A). The rate of no surgical resection or unknown/ unspecified surgery remained unchanged with AAPCs of 0.3% (CI -0.9, 1.5%) and 6.3% (CI -2.1, 15.5%) per year, respectively. In patients with known surgery, surgical resection increased from 61.   (Fig. 2C). Nationally, receipt of radiation was stable between 2004 and 2016 from 44.2 to 46.9%, respectively, while there may have been a decline in utilization of chemotherapy from at least 16.0-8.7% of patients over the same time, respectively (Fig. 2D).
Discrepancies in characteristics associated with RFS, DSS, and OS in the institutional dataset were further evaluated using the SEER database. Analysis of SEER data demonstrated that competing-cause mortality is accentuated with increasing age and increasing time from diagnosis. In patients B 64 years of age, DSS and OS closely mirror each other until approximately 1 year of follow-up, with approximately 83.1% of overall mortality attributable to MCC disease, (Appendices B and C). However, in patients C 85 years, DSS and OS diverge almost immediately, with only 26.6% of overall mortality attributable to MCC disease at 1 year. Mortality associated with MCC remains nearly unchanged after approximately 3 years, and thereafter mortality is almost exclusively a result of death from competing causes, Appendix B.

DISCUSSION
When considering practice patterns, our institutional experience demonstrated changing trends that were consistent with those seen nationally when compared with the SEER database. At our institution, chemotherapy utilization has declined over the past 2 decades from 42 to 4% whereas the rate of use of immunotherapy has increased from 0 to 24%. This is to be expected given that chemotherapy has not improved survival or reduced the rate of distant metastasis or recurrence. 16,18 Our analysis also demonstrated a significant increase in the use of immunotherapy to treat MCC. In 2016, two key clinical trials, the KEYNOTE-017 and JAVELIN demonstrated the effectiveness of systemic immunotherapy in treating metastatic MCC, leading to FDA approval. [11][12][13][14][15]19,20 Current studies including the STAMP trial are underway to better define the use of these treatments in the adjuvant setting (ClinicalTrials.gov NCT03712605). Our institutional data reflects a relatively rapid incorporation of an effective therapy in clinical practice, specifically for patients with in-transit or metastatic disease.
We found that patients who were C 85 years and who had clinically positive lymph nodes were managed in a manner which differed from NCCN guidelines pertaining to surgical excision of the primary tumor and evaluation of the lymph node basin via sentinel lymph node biopsy or lymph node dissection. It is not surprising that older age is associated with deviation from the guidelines and is consistent with findings in patients with breast cancer 21 and melanoma. 22 This is likely attributable to patient comorbidities 23 which may be appropriate in an older population. We and other groups have shown that following national guidelines for staging and management has been associated with improved OS. 24 It is critically important that major centers who treat MCC prospectively record why patients are not receiving guideline compliant therapy. We then turned to our analysis of RFS, DSS, and OS. We failed to find any factors that were associated with improved RFS. The factors associated with lower RFS included age C 85 years, cT2 and cT3 or cT4 disease, lymph node positive disease and ITD at time of diagnosis. Interestingly, the receipt of surgery with lymph node evaluation was not associated with improved RFS which contrasts with previously published work. 25,26 This may be due to the small sample size. While it is appropriate to utilize OS as a surrogate for DSS survival in certain cancers, such as colon cancer, 27 our analysis indicates that this is less informative in MCC. The median age of our cohort was 75 years, which is consistent with the literature. This age group has multiple competing risks 28 for mortality outside of their MCC diagnosis. Our institutional data analysis demonstrated that older age and cT2 disease were both associated with worse DSS and OS. In contrast, current smoker status was associated with worse DSS, while an immunocompromised state and truncal tumor primary location were only associated with worse OS. When there is little consistency between the effects each factor has on DSS and OS, this points to the need for caution when utilizing OS alone as an outcome measure. Based upon our analysis, the competing risk for mortality in this patient population leads to this divergence of factors that are associated with DSS and OS.
This idea is further supported through analysis of the SEER database which demonstrated that as age increases, competing-cause mortality accentuates differences in DSS and OS with age and time. This, combined with the  known therapy versus d with known or unknown therapy. All data from SEER database. Error bars represent 95% confidence intervals inconsistencies we found in factors associated with DSS and OS, demonstrate that OS may be a poor proxy for therapeutic efficacy in retrospective analyses. Indeed, factors associated with OS from large database studies can, and should, be used to pose hypotheses for carefully performed institutional analyses-such as done in this study. However, we demonstrate that without investigation into how such factors associate with RFS or DSS, the realworld application of findings from such OS analyses will remain limited.
This study represents a large academic center's experience with 159 patients compared with national trends in care seen in the SEER database. There are several limitations of this paper. The data collected is from a single, tertiary academic hospital in the northeast of the United States. This may limit the applicability of the findings. Additionally, this was also a retrospective review of medical records, leading to a risk of information and selection bias. Further study is required of MCC; specifically, institutional studies utilizing disease-specific outcomes, given the rarity of this diagnosis. With the notable rise in the use of immunotherapy, further characterizations to look for predictive biomarkers to response will also be an important avenue of continued research. Further study is also required to elucidate how the etiology, whether from UV radiation or MCPV, affects RFS, DSS, and OS. Currently, our institution does not test for MCPV. Thus, we have no data on the effect of MCPV seropositivity on recurrence-free, overall, and disease-specific survival. Finally, we recommend a stronger emphasis on reporting DSS when discussing MCC given the competing risks of the population affected by MCC, making OS a poor indicator of disease-specific outcomes.
The analysis of national treatment trends must be interpreted with caution. Rates of treatment with radiation therapy and chemotherapy may not be accurately coded, especially in the earlier years. Patients with unknown radiation or chemotherapy information were grouped with patients who did not receive such therapies into a ''no/ unknown'' category by the SEER program. Consequently, rates of receipt likely represent the minimum rate for such therapies, since the positive predictive value for receipt of therapy, with a coding of ''yes,'' is high. We, therefore, cannot definitively conclude whether rates of radiation therapy or chemotherapy, or their combinations with surgery, have changed. However, we believe that rates and  proportions are suggestive for national treatment trends when evaluated within the larger scope of the manuscript. Finally, it was not possible to specifically evaluate the effect of immunotherapy on RFS, OS, and DSS given overall small numbers and relatively short follow-up. However, re-evaluation of this subset of patients to understand the effect of immunotherapy will be critical in the future. Furthermore, while we were able to evaluate trends in OS over time with the SEER dataset, DSS was not assessed given variable rates of cause of death reporting. Such variability did not permit the possibility of interpreting results with any degree of confidence.

CONCLUSION
This study demonstrates that the practice patterns in the management of MCC have changed over time both at an institutional and at a national level. These changes appear to match emerging data concerning optimal treatment regimens for MCC, given the significant increase in the utilization of immunotherapy. Further, this study demonstrates that there is a divergence of factors associated with DSS and OS. This indicates that reliance on OS may have limitations in studying MCC outcomes.

APPENDIX A
See Table 6.