Long-term Outcomes for Extraskeletal Myxoid Chondrosarcoma: A SEER Database Analysis

Extraskeletal myxoid chondrosarcoma (EMCS) is a rare mesenchymal tumor characterized by one of several chromosomal rearrangements. The most common is a fusion of EWSR1 (formerly called EWS) to NRA4A3 (formerly called CHN and TEC; ref. 1). Less common fusions include RBP56-NR4A3 (2, 3), TCF12-NR4A3 (4), and HSPA8-NR4A3 (5).

EMCS typically has an indolent course. Although long-term overall survival (OS) is relatively good versus other cancers, there is a relatively high rate of local and distant recurrence (6). In one large retrospective series, patients presenting with localized disease had a 49% chance of either local or distant recurrence, with 5- and 10-year relapse-free survival (RFS) of 42% and 21%, respectively (7). Multimodal treatment strategies with surgery and radiotherapy reputedly improve local control (8).

Metastatic EMCS is generally unresponsive to cytotoxic chemotherapy, with no observed responses in the largest series of 21 patients treated with chemotherapy (7). More recently, antiangiogenic tyrosine kinase inhibitors have demonstrated clinical activity (9–11). Long-term results with targeted agents have not been reported.

Anticipated long-term outcomes with current therapies must be characterized to serve as benchmarks for future studies of new therapies. Here, we assess the Surveillance, Epidemiology, and End Results (SEER) database to describe long-term OS outcomes in patients with EMCS. SEER is a cancer registry funded by the National Cancer Institute. It includes cases representing approximately one third of the U.S. population. Data is derived from cancer registries based at cancer treatment facilities across the U.S. SEER data are deidentified and publicly accessible (12, 13). This analysis can guide clinical discussions regarding prognosis and long-term outcomes for patients with EMCS and serve as a benchmark for survival in future interventional studies in this rare histology.

We queried the 1973–2016 SEER 18 database (12, 14) for patients with primary tumors diagnosed as myxoid chondrosarcoma (IDC-O-3: 9231). Tumors of the bones and joints were excluded from this analysis (variable: site recode ICD-O-3/WHO 2008 coded as “Bones and Joints”). Cases with unknown primary tumor location, surgery status, or survival months were excluded.

Cases were classified anatomically as “localized,” “regional,” or “distant.” The schema used to classify cases into these three groups from SEER data is shown in supplementary materials (Supplementary Table S1). Because of relatively small numbers of cases in certain subgroups, and both local and regional disease can be treated with curative intent, cases that were classified as “localized” and “regional” were combined into an analytic group referred to as “locoregional disease” (LRD). Cases for which no staging information was available were excluded.

The following clinical and demographic characteristics were evaluated: sex, age at diagnosis, year of diagnosis, primary tumor location, tumor grade, and tumor size. Age at diagnosis was grouped as: age 0–39, age 40–64, and age ≥65. Year of diagnosis was dichotomized into two categories: cases diagnosed between 1987 and 2007 or between 2008 and 2016. Primary tumor site was grouped by anatomic region as lower extremity, upper extremity, abdomen/trunk, and head/neck. Tumor size was categorized as either ≤5 centimeters or >5 centimeters. Treatment modalities, including receipt of chemotherapy, surgery, or radiotherapy, were also assessed.

Fisher's exact test was used for categorical variables to assess the distribution of demographic and clinical characteristics by LRD and distant disease. Logistic models were fitted to assess the association between demographic/clinical/treatment characteristics and receipt of surgical treatment. Pearson correlation with Bonferroni correction for multiple comparisons was used to assess patterns of missing data with respect to other variables.

The Kaplan–Meier method and Cox proportional hazard models were used to assess the impact on OS of patient demographic and clinical characteristics. SEER vital status variable was used to define OS; a case was considered a failure event if coded as “dead,” otherwise the case was censored. Survival time periods were calculated from time of diagnosis or censoring. One-, 5-, and 10-year OS and median OS, and their respective 95% confidence intervals (CI) by extent of disease, were evaluated. OS by extent of disease was compared using a pairwise log-rank test. In a subanalysis, we evaluated OS by primary tumor location in LRD cases who had received surgical treatment and assessed the difference in OS between the different primary tumor location groups using pairwise log-rank tests.

Univariable (unadjusted) and multivariable (adjusted) Cox proportional hazards regression models were fitted to assess outcomes in patients with LRD, and in those with LRD who received surgical treatment. Among those with LRD, all variables except surgical treatment satisfied the proportional hazards assumption. For adjusted models, face validity was the primary criterion for inclusion of parameters (demographic, clinical, pathologic, and treatment) in the initial adjusted models (15). Adjusted models were then assessed by the likelihood ratio (LR) test to compare the full model with reduced models in which all statistically nonsignificant variables were omitted. Variables in the full adjusted model were statistically insignificant if the HR 95% CI for the variable overlapped HR = 1.00. For variables containing multiple categories, they were only omitted in the reduced model if the HR 95% CI overlapped 1.00 in all subcategories. The primary analyses were complete case analyses.

Because of significant missing data in tumor grade and tumor size, which were both judged to be important prognostic variables in those with LRD, sensitivity analyses were conducted in which the adjusted Cox analyses were repeated using multiple imputation (MI; refs. 15, 16). This was conducted using the MI function (“mi”) of the STATA 12.1 statistical package. Multinomial logistic regression was used to impute tumor grade, an ordinal variable, while logistic regression was used to impute the binary categorical variable tumor size. Because approximately 50% of cases were missing data in one or both of these variables (all other variables had complete data for all cases), we undertook 50 imputations in the MI analysis (17). The datasets were imputed on the basis of the following variables: sex, age group, year of diagnosis, primary tumor location, extent of disease, chemotherapy status, surgery status, and radiation status.

Statistical significance was reported at P ≤ 0.05. Stata version 12.1 was used for all statistical analyses (StataCorp).

We identified 791 cases that were diagnosed as myxoid chondrosarcoma. A total of 439 cases met inclusion criteria for the study (Supplementary Fig. S1). A total of 295 cases were excluded due to a primary skeletal site. In addition, 11 cases were excluded due to incomplete survival data. We were able to neither derive stage in 29 cases nor classify 6 cases into primary tumor location groups. Ten cases had duplicate SEER ID entries and 1 case had unknown surgical status. Only 3 of 791 cases were diagnosed prior to 1987. Only cases from the 1987–2016 era met eligibility criteria. There were 373 cases (85%) that had LRD and 66 cases (15%) had distant disease.

Patient and tumor characteristics are outlined in Table 1. The study population was predominately male, and most cases were diagnosed after the age of 39 (84%). Those older than 65 years were more likely to be diagnosed with metastatic disease. The lower extremity was the most common primary site, representing 57% among those with LRD and 56% among those with distant disease. Among cases with available tumor grade data, 75% of those with LRD had grade I or II disease and 59% of those with distant disease had grade I or II disease. Among cases with tumor size data, 64% of those with LRD had tumor size greater than 5 centimeters and 82% of those with distant disease had primary tumor size greater than 5 centimeters.

Missing data were only present in tumor grade (n = 204 out of 439 cases; 46.5%) and primary tumor size (n = 56 out of 439 cases; 12.8%). For both variables, a higher proportion of data were missing among those with distant disease than with LRD (Table 1; grade P = 0.033; size P = 0.015). Missing tumor grade data correlated positively with missing primary tumor size (r = 0.164, P = 0.006). Both missing tumor grade and missing primary tumor size data correlated negatively with receipt of surgical treatment (r = −0.155, P = 0.0011 and r = −0.230, P < 0.0001, respectively). There was no significant correlation of missing grade or size data with any other variable.

Median OS was 262 (95% CI, 167–280), 127 (95% CI, 82–218), and 48 (95% CI, 18–61) months for cases presenting with localized, locally advanced/SEER “regional,” and distant disease, respectively (Fig. 1). The OS at 1, 5, and 10 years by extent of disease is presented in Table 2. Cases with local and regional disease were aggregated for subsequent analysis of LRD, as this was judged to more closely reflect clinical practice. In cases with LRD receiving surgical treatment, median OS was 218 months (95% CI, 152–274). Six-month, 1-year, and 5-year OS were 99% (95% CI, 97%–99%), 97% (95% CI, 95%–98%), and 81% (95% CI, 76%–85%), respectively.

Surgical treatment of LRD was associated with improved OS (HR = 0.27; 95% CI, 0.16–0.47; Table 3). Receipt of chemotherapy or radiation for LRD was associated with inferior OS (HR = 1.90; 95% CI, 1.11–3.27 and HR = 1.45; 95% CI, 1.03–2.06, respectively) in unadjusted univariable analyses. In the subpopulation of cases with surgically treated LRD (n = 351), 13% had primary tumors located in the upper extremities, 58% in the lower extremities, 26% in the abdomen, and 3% in the head and neck region. For patients with LRD who had surgery, there was no difference in OS by primary site (Fig. 2).

We assessed characteristics associated with receipt of surgical therapy in the whole population (Supplementary Table S2). Patients with regional and distant disease had lower odds of receiving surgical treatment (OR = 0.22; 95% CI, 0.09–0.55 and OR = 0.02, 95% CI, 0.01–0.06, respectively) versus those with local disease. Receipt of chemotherapy was also associated with lower odds of surgical treatment (OR: 0.17; 95% CI, 0.09–0.32). Progressively younger patients were statistically more likely to receive surgical treatment than those 65 years of age or older. None of the other variables assessed were associated with receipt of surgical treatment.

A multivariable Cox model was constructed examining outcomes in cases with LRD, adjusting for sex, age at diagnosis, year of diagnosis, primary tumor location, tumor grade, tumor size, and treatment modalities (Table 3). This demonstrated that both higher tumor grade (grades III–IV) and primary tumor size greater than 5 centimeters were associated with inferior OS, while younger age was associated with progressively improved OS. In the adjusted model, treatment modality (surgery, radiation, or chemotherapy) was not significantly associated with OS for those with LRD, although the model only included 7 LRD patients who did not receive surgery. The full model was compared with a reduced model consisting only of age at diagnosis, tumor grade, and tumor size. These were not statistically different, suggesting that the other variables excluded from the reduced model did not contribute significantly in explaining OS (LR χ² = 9.40, 8 degrees of freedom, P = 0.31).

Because oncologic surgical excision is considered standard-of-care for locoregional EMCS, a secondary analysis was performed, including only those cases with LRD treated surgically (Table 4). As in the analysis of all patients with LRD, higher tumor grade (grade IV) and larger tumor size remained associated with inferior OS and younger age was associated with improved OS. Receipt of chemotherapy in this subpopulation was independently associated with worse OS (HR = 2.54; 95% CI, 1.07–6.08). Radiation was not associated with OS in either model. When a reduced model consisting of age at diagnosis, tumor grade, tumor size, and receipt of chemotherapy was compared with the full model, there was again no statistical difference, indicating that the omitted variables did not contribute significantly to explaining OS in this subpopulation (LR χ² = 3.44, 6 degrees of freedom, P = 0.75).

Both tumor grade and tumor size were viewed as potentially important variables associated with outcomes but had high amounts of missing data with 46.5% and 12.8% of cases missing data for these fields, respectively. MI was used in a sensitivity analysis to assess the impact of these missing data. Using a multivariable Cox model with the same variables as above, receipt of chemotherapy, larger tumor size, and head/neck primary site were associated with inferior OS (Supplementary Table S3). Younger age and surgery were associated with superior OS. The analysis allowed inclusion of all 22 LRD patients who received no surgical treatment, explaining the difference in outcome with respect to this variable versus the complete case analysis.

When the same analysis was conducted focusing on those with LRD receiving surgical treatment, head/neck primary site was not significantly associated with OS. The remaining variables had similar associations to LRD patients as a whole. A notable contrast with the complete cases analyses was the lack of association of grade with outcomes when missing data were addressed with MI.

EMCS is a rare tumor with relatively few studies to guide treatment recommendations and care discussions. We used a large database to establish treatment patterns and outcomes for EMCS patients. These data represent the largest report of EMCS patients with long-term follow-up to date.

Prior reports have demonstrated variable local control rates, but high rates of distant metastasis (6–8, 18). Limitations of the SEER database precluded direct assessment of local control. However, surgery significantly correlated with OS in both univariable and multivariable analysis, suggesting that adequate local control is critical for long-term survival. In our analyses, surgery's beneficial impact was only detected in the sensitivity analysis using MI. This was probably due to the exclusion from our primary complete case analyses of those with missing data. While this is a standard approach to missing data in regression analysis, only 7 of 22 LRD patients who did not receive surgical treatment were available for the complete case analysis.

Because of the apparent importance of tumor grade and size in OS outcomes and also the significant number of cases excluded from the primary analyses, sensitivity analyses were conducted in which the adjusted analyses were repeated using MI to address missing data. For tumor size, the missing data were modest (12.8%). Missing data were higher for tumor grade (46.5%). Over half of cases (51%) were missing at least one of these values, meaning that over half of the initial cases would not be included in a complete case analysis. MI provides an approach to systematically replace the missing data with plausible values, allowing regression analyses to be conducted using the entire case set (15–17). This improves the precision of parameter estimates by allowing the use of a large amount of data that would otherwise be discarded.

As the mechanism for missing data is important in the application of MI, we explored this through correlation between the presence of a missing value in one of the variables and the various other variables used in the analysis. Missing values in tumor grade and tumor size correlated with each another, suggesting a common mechanism. Surgical treatment of disease was negatively correlated with missing data in these two variables. This is not surprising; surgical treatment of a tumor should generally yield information reflecting these pathologic parameters. Missing tumor grade or size did not correlate with any other pretreatment or treatment (chemotherapy, radiotherapy) variable. We posit a missing at random mechanism for the missing data, consistent with use of MI.

There is some disagreement regarding the number of individual dataset imputations that must be undertaken to provide reliable analytic results (16, 17). We undertook 50 imputations, consistent with the recommendations outlined in White and colleagues (17). Other workers indicate that as few as 5 imputations can be sufficient. We found no material differences in the analyses conducted using 50 versus 5 imputations (Supplementary Table S4).

In both complete case and MI analyses, patient age and tumor size were significant. Tumor grade was significant in the complete case analyses but did not remain significant in the MI analyses. From the perspective of treatment, chemotherapy was an adverse prognostic factor in patients treated with surgery for LRD, suggesting selection of patients with an adverse prognosis for its receipt. Among all patients with LRD, chemotherapy treatment was associated with adverse outcomes only when the analysis was conducted using MI. It is possible that the cases with missing data in grade and size disproportionately included those who did not undergo surgery, and therefore received palliative chemotherapy.

While the number of cases requiring imputation of tumor size (12.8%) is within the capability of this approach, imputation of tumor grade, missing in about 46.5% of cases, falls toward the upper end of proportions reliably imputable by this technique. As such, MI data served as a sensitivity analysis, rather than the primary analysis. We urge caution in interpretation of the results based on MI, but feel they still have value in this analysis. Our analyses suggest that, among pre-treatment variables, age, primary tumor size, and possibly tumor grade should be considered for stratification purposes.

A prior study of the SEER EMCS cases suggested benefit for radiotherapy, although there was no statistical improvement in OS with inclusion of radiation (19). Bishop and colleagues report a population in which multimodality therapy with radiation and surgery improved local control over surgery alone (8). Although there was no benefit seen with chemotherapy or radiotherapy in those who had surgery for LRD, limitations of the SEER database preclude our drawing conclusions on potential benefits of these modalities (12, 20, 21). The SEER database contains a heterogeneous population; patients with LRD who received radiation or chemotherapy may have had more aggressive disease biology or may have been more likely to have positive surgical margins and thus a higher risk of recurrence, accounting for the worse outcomes found in our study. Multimodal therapy with surgery and radiation is our institutional practice for EMCS patients presenting with large tumors or those for whom adequate surgical margins cannot be obtained with acceptable morbidity.

Sarcoma pathology is a highly specialized area and change in diagnosis can be seen in a significant proportion of samples after expert second opinion (22). Pathology results in the SEER database are not subject to central review. Thus, diagnostic confirmation depends on the participating institutional pathology review. The presence of the characteristic NRA4A3 fusions could not be confirmed in this dataset. It is possible that some cases may have been misclassified in the registry. Margin status of the resections, likely an important indicator of surgical quality, is also unavailable.

Interestingly, we identify a subset of patients (15% of this study population) who present with metastatic disease and have poor outcomes on par with those of other soft tissue sarcoma subtypes (23). This differs from the general consensus and previous reports characterizing EMCS as an indolent disease (24). Other reports focus primarily on cases with LRD and may therefore underreport this small population with aggressive underlying biology. Improved therapies for this population are sorely needed. The genomic landscape of this subset of patients has not been differentiated from that of EMCS patients as a whole and may reveal underlying factors to explain the more aggressive biology. Additional study is needed to identify whether underlying biological differences or perhaps differing factors in the microenvironment caused this observed difference in metastatic potential. As we found in our multivariate analysis, another possibility is that tumor size and grade are the only significant clinical factors associated with metastasis at presentation.

One of our objectives in this study was to generate benchmarks to be used for future studies. Typical benchmarks for early phase studies, such as objective response rate, local and distant relapse-free survival, and progression-free survival are not available in SEER. OS remains the “gold standard” for oncology study outcomes and is an endpoint available in SEER. We have generated estimated OS outcomes at various time points after diagnosis, along with 95% CI to aid future workers in the design and power calculations for clinical trials (Table 2).

We have included data for the two most common scenarios: those with LRD to be treated surgically, as a benchmark for studies of perioperative therapy, and those with distant disease at diagnosis, representing benchmarks for studies of treatment of advanced unresectable and metastatic disease. Our analyses suggest that age, tumor grade, and tumor size are potentially important stratification factors and covariates to include in adjusted models analyzing the results of clinical trials in the population with surgically treated LRD.

Cytotoxic chemotherapy has minimal activity in metastatic EMCS (25). Targeted agents such as sunitinib and pazopanib have recently demonstrated encouraging results (10, 11). Mechanistic studies of the causative fusion protein have implicated genes involved in axon guidance, neurogenesis, and angiogenesis as potential targets for future clinical development (26). Our data identifies a subpopulation of EMCS patients for whom local control alone is insufficient and improved systemic regimens are sorely needed. Metastatic EMCS carries an overall poor long-term prognosis and patients should be counseled accordingly.

M.J. Wagner reports personal fees and other from Deciphera (clinical trial support to institution), Tempus (clinical trial support to institution), and Adaptimmune (clinical trial support to institution) and other from Incyte (clinical trial support to institution), Athenex (clinical trial support to institution), and GlaxoSmithKline (clinical trial support to institution) outside the submitted work. S.M. Pollack reports grants from Merck, EMD Serono, Incyte, Presage, Janssen, Oncosec, and Juno and personal fees from GlaxoSmithKline, Eli Lilly, Seattle Genetics, Bayer, Tempus, Daiichi Sankyo, and Blueprint Medicine outside the submitted work. L.D. Cranmer reports grants from Eli Lilly (to institution), CBA Pharma (to institution), AdvenChen (to institution), Tracon (to institution), AADi (to institution), Exelixis (to institution), Philogen (to institution), and Iterion (to institution); grants and personal fees from BluePrint Medicines (grant to institution and honoraria); and personal fees from Daiichi Sankyo (honoraria) and Regeneron (honoraria) outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

.

M.J. Wagner: Conceptualization, formal analysis, supervision, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing. B. Chau: Formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing. E.T. Loggers: Writing–review and editing. S.M. Pollack: Writing–review and editing. T.S. Kim: Writing–review and editing. E.Y. Kim: Writing–review and editing. M.J. Thompson: Writing–review and editing. J.L. Harwood: Writing–review and editing. L.D. Cranmer: Conceptualization, formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.