Racial and Ethnic Differences in Sarcoma Incidence Are Independent of Census-Tract Socioeconomic Status

Sarcomas, which can be broadly categorized as soft-tissue sarcomas (STS) or bone sarcomas (BS), comprise a group of over 50 exceedingly rare and heterogeneous mesenchymal neoplasms (1). The overall incidence of STS and BS is fewer than 5 cases per 100,000 persons per year, although substantial variation in incidence by subtype, age, sex, and race has been described (2–5). Several factors are associated with sarcoma risk (6). These include genetic predisposition from both rare variants (7–9), some of which underlie cancer predisposition syndromes (10), and common variants (11–13). Other studies have observed sarcoma development to be associated with environmental [e.g., phenoxy herbicide exposure (14, 15), radiotherapy (16, 17)], infectious (18), and perinatal factors [e.g., birth weight (19), parental age (20)]. However, the etiology for the vast majority of sarcomas remains unknown, largely because the rarity and heterogeneity of the disease preclude sufficiently powered etiologic studies (6).

Evaluating socioeconomic status (SES) and racial/ethnic disparities in incidence across sarcoma subtypes may provide clues into their etiology. SES can serve as a proxy for locally varying environmental and lifestyle factors that influence health (21), whereas racial/ethnic disparities that are independent of SES may indicate that genetic variants associated with ancestry contribute to tumor development. However, prior analyses of SES and sarcoma incidence are either limited to analyses of childhood cases or to those that combined distinct subtypes into a single group (22–25), and few analyses of race-specific incidence patterns considered probable confounding by SES (2, 5). It is therefore difficult to properly attribute differences observed to environment or to genetic variants associated with ancestry.

In this study, we sought to utilize the population-based Surveillance, Epidemiology, and End Results (SEER) program to investigate the independent associations of census tract-level SES (CT-SES) and race/ethnicity with the incidence rate of individually rare subtypes of sarcoma diagnosed across the age span.

Sarcoma cases were identified from the SEER Census Tract-Level SES database, a specialized database that includes cancer cases diagnosed between 2000 and 2015 in the catchment area of 16 SEER registries (26). The Alaska Native and Louisiana tumor registries are excluded by SEER due to confidentiality concerns and the impact of Hurricane Katrina on population estimates, respectively (27). We identified a sarcoma as any microscopically confirmed, initial primary malignant tumor with an International Classification of Disease for Oncology, 3rd Edition (ICD-O-3) histology code recognized by the 2013 World Health Organization's (WHO) Classification of Tumours of Soft Tissue and Bone (28). We then categorized sarcomas into subtypes according to WHO classifications and the recommendations of an expert sarcoma oncologist (B.J. Weigel) and pathologist (P. Murugan; Supplementary Table S1). Kaposi sarcoma was excluded because it primarily arises in the setting of HIV infection (18). In addition, we classified sarcoma subtypes by their somatic genome as either fusion-positive (F+) or fusion-negative (F−) according to published data (Supplementary Table S2; refs. 29, 30). The (F+) category included subtypes with somatic genomes characterized by simple chromosomal translocations, whereas the (F−) category included subtypes with somatic genomes characterized by complex chromosomal rearrangements (31). We opted to categorize sarcoma subtypes by their presumed fusion status because evidence suggests that (F−) sarcomas are more likely than (F+) sarcomas to arise from genetic susceptibility (29, 32–34). Cases diagnosed in the year 2000 were excluded because they were not coded under the ICD-O-3 guidelines introduced in 2001 (35).

SES was assessed using a composite index of SES measured at the level of the census tract, defined as a small geographic unit (∼4,000 people) designed to be homogeneous with respect to population characteristics, economic status, and living conditions (27). The CT-SES index was generated from a factor analysis of seven SES characteristics identified by Yost and colleagues (36). Cases were mapped to a census tract based upon their address at diagnosis (36). CT-SES indices assigned to each census tract were then categorized by SEER into quintiles of equal population size, with the first quintile (Q1) representing the lowest CT-SES and the fifth quintile (Q5) representing the highest (26).

Cases were identified as having Hispanic ethnicity by SEER based upon the North American Association for Central Cancer Registries Hispanic-Latino identification algorithm (37). Those without a Hispanic ethnicity were categorized into mutually exclusive race categories as either non-Hispanic (NH)-White, NH-Black, or American Indian/Alaskan Native or Asian Pacific Islander (AIAN/API); the AIAN and API race categories were combined due to sample size constraints. Age at diagnosis was categorized into nine age groups (0–9 years, 10–19 years, 20–29 years, 30–39 years, 40–49 years, 50–64 years, 65–74 years, 75–84 years, and 85+).

The population denominator estimates provided by SEER were stratified by single calendar year, sex, age, race, and Hispanic ethnicity. SEER allocates multiracial populations into one of the four single race categories with a probability proportional to the size of that race in the population (26).

After excluding cases with missing CT-SES (n = 914) or race/ethnicity (n = 775) variables, we analyzed 97% of the starting dataset (55,415 of 57,044 cases).

Incidence rate ratios (IRR) and 99% confidence intervals (CI) for CT-SES and race/ethnicity, adjusted for age at diagnosis, sex, and year of diagnosis, were estimated from quasi-Poisson models, which handle overdispersion by assuming the variance is a linear function of the mean (38). In all models, log-transformed population denominator estimates served as the offset term. Analyses were stratified by broad age group categories (pediatric <20 years; adult 20–65 years; older adult >65 years) to account for possible differences in sarcoma etiology. Within each age group strata, associations were assessed for sarcomas classified by fusion status [either (F+) or (F−)], as well as for any subtype with more than 200 total cases, a cutoff that was set a priori to maintain adequate power. For each sarcoma subtype, CT-SES was assessed as either a categorical or ordinal variable, with the P value from the ordinal variable used as a test for trend. For ease of interpretation, we have focused our reporting on the results obtained from analyzing CT-SES as an ordinal variable (Fig. 1). In some instances, these results appeared to obscure a nonlinear association apparent in the analysis of CT-SES as a categorical variable. We, therefore, present results from analyses of CT-SES as a categorical variable in Supplementary Table S3.

A P value less than 0.01 was considered statistically significant. This threshold adjusted for the multiple primary hypothesis tests evaluated within each sarcoma subtype and age group strata (i.e., the ordinal CT-SES and 3 race/ethnicity comparisons) and controls the overall Type I error rate at approximately 0.05. All reported P values are two-sided. Datasets were created using SEER*Stat 8.3.6 (39), and analyses were performed in R version 3.4.4 (40).

Table 1 presents the distribution of sarcoma cases by demographic and clinical characteristics. Overall, 55,415 initial primary malignant sarcoma cases were identified (48,348 STS and 7,067 BS). Among all sarcoma cases, 64% were NH-White, 12% were NH-Black, 16% were of Hispanic ethnicity, and 9% were AIAN/API. The distribution of sarcoma cases from lowest to highest SES quintile was 17%, 18%, 20%, 22%, and 24%, respectively.

Figure 1 shows the IRRs for CT-SES, stratified by sarcoma category and age group strata (values are given in Supplementary Table S3). In the pediatric age group, a positive trend in the incidence of fibroblastic/myofibroblastic (F/MF) tumors was observed across CT-SES quintiles (IRR, 1.09; 99% CI, 1.00–1.18; P value = 0.007). However, the results from CT-SES evaluated as an ordinal variable may have obscured a potential nonlinear association in the incidence of F/MF tumors; the IRR comparing CT-SES Q2 versus Q1 (IRR, 1.37; 99% CI, 0.97–1.94) appeared higher than the IRR comparing CT-SES Q3 versus Q1 (IRR, 1.04; 99% CI, 0.71–1.52; Supplementary Table S3). The incidence rates for other sarcoma subtypes evaluated in the pediatric age group were not associated with CT-SES.

In the adult age group strata, F/MF tumors and liposarcoma showed evidence of having a higher incidence in higher CT-SES quintiles (P value < 0.001 for both), whereas malignant peripheral nerve sheath tumors (MPNST) showed the opposite trend with every increase in CT-SES quintile associated with a 7% lower incidence rate (99% CI, 0.87–0.99; P value = 0.002). The other subtypes evaluated in adults were not associated with CT-SES, including all of the BS subtypes evaluated.

In the older adult age group strata, most subtypes (8 of 11 evaluated) showed a trend of increased incidence with increasing CT-SES quintiles that reached statistical significance, including those subtypes that were not significantly associated with CT-SES in adults. Other rhabdomyosarcoma (RMS), MPNST, and malignant vascular tumors were the only subtypes evaluated in older adults that were not associated with CT-SES.

In all age groups, we observed an increasing incidence of (F+) sarcomas with increasing CT-SES (P value < 0.001 in all age groups). A positive trend was also observed for (F−) sarcomas in the older adult aged strata (P value < 0.001), but the trends in the adult and pediatric age groups were not statistically significant (P values = 0.02 and 0.08, respectively).

Figure 2 shows the IRRs for race and ethnicity, stratified by sarcoma category and age group strata (values are given in Supplementary Table S4). In the pediatric age group, the IRRs observed in NH-Black children compared with NH-White children differed according to sarcoma subtype, with statistically significant IRRs observed for five of the seven subtypes evaluated. Heterogeneous results were also observed among children of Hispanic ethnicity, although we found fewer significant results (three of the seven subtypes evaluated) and smaller effect sizes than those comparing incidence rates in NH-Black to NH-White children. Among AIAN/API children, the IRRs that reached statistical significance (three of seven subtypes) were all less than 1, with some subtypes showing incidence rates that were nearly 50% lower in AIAN/API compared with NH-White children (e.g., Ewing sarcoma of bone, embryonal RMS, and F/MF tumors).

In adults, several STS subtypes had incidence rates in NH-Blacks that were substantially higher (IRR > 1.60) than those observed in NH-Whites, including other RMS, F/MF tumors, desmoplastic small round cell tumor (DSRCT), gastrointestinal stromal tumor (GIST), and leiomyosarcoma. The incidence rates for several subtypes were also significantly different in adults of Hispanic ethnicity compared with NH-White adults (6 of 17 subtypes evaluated), although we again observed weaker associations than those comparing the incidence rates in NH-Black adults with NH-White adults. Among adults identified as AIAN/API, we observed incidence rates that were significantly lower than those observed in NH-White adults for several sarcoma subtypes (9 of 17 subtypes evaluated). A notable exception is GIST, the only subtype evaluated to show a significantly higher incidence rate in AIAN/API compared with NH-White adults (IRR, 1.48; 99% CI, 1.31–1.68).

Several of the results observed in older adults (>65 years at diagnosis) were dissimilar to those in adults. For example, the incidence of undifferentiated pleomorphic sarcoma was lower in older adults identified as NH-Black (IRR, 0.48; 99% CI, 0.35–0.64) and Hispanic (IRR, 0.65; 99% CI, 0.52–0.8) compared with NH-White older adults, but showed no difference in incidence across these racial/ethnic categories in adults. Conversely, results among AIAN/API older adults largely concurred with those observed in younger aged individuals. Incidence rates in AIAN/API older adults that reached statistical significance (5 of 10 subtypes evaluated) were primarily lower than those observed for NH-Whites, except for GIST, which had an incidence rate that was 78% higher in AIAN/API compared with NH-White older adults (99% CI, 1.56–2.04).

In all age groups, the incidence of (F−) and (F+) sarcomas in NH-Black or Hispanic individuals compared with NH-White individuals were heterogeneous. Conversely, the comparative incidence rates in AIAN/API individuals that reached statistical significance consistently showed lower incidence rates for both (F−) and (F+) sarcomas compared with NH-Whites in all age groups.

In the current study, we found race/ethnicity to be more often associated with sarcoma incidence than CT-SES. Specifically, of the 35 subtype–age group combinations evaluated for an association with CT-SES, 12 were statistically significant, and most occurred in the older age group strata (8 subtypes), with all, except MPNST in adults, having a positive trend in incidence across CT-SES levels. Conversely, nearly every subtype–age group combination displayed racial/ethnic differences in incidence rates that were independent of CT-SES. Notably strong associations included an over two-fold increase in the incidence of GIST among NH-Black compared with NH-White adults. Overall, our findings suggest that genetic variation associated with ancestry may play a stronger role than area-level SES-related factors in the etiology of STS and BS subtypes.

Prior results regarding the association between SES and sarcoma incidence are limited. A large SEER analysis (n = 1,576 STS and 1,113 BS) found that children ages 0 to 19 years living in higher-income counties of the United States had a higher incidence of STS but not BS (22), whereas other case–control studies using maternal or household education as a proxy for SES have found mostly null associations (24, 41). In prior analyses, however, individual subtypes of sarcoma were not separately evaluated. With regards to race/ethnicity, several race-specific incidence patterns have been reported, including the dramatic patterns reported herein among minority populations diagnosed with Ewing sarcoma (42–44), GIST (45, 46), DSRCT (47), and leiomyosarcoma (2). Prior analyses, however, have failed to comprehensively account for likely confounding by SES. The results of our analysis build upon prior work by demonstrating the independent associations of CT-SES and race/ethnicity on the incidence of subtypes of sarcoma throughout the age span.

Area-level SES is speculated to represent complex interactions between environmental, social, and cultural influences on health; therefore, we cannot determine which specific aspect may have driven our observed associations. Several possible risk factors for sarcoma have been identified in prior studies (6), including those we presume to be correlated with both a high SES [e.g., older parental age (20)] and low SES [e.g., residential proximity to industrial facilities (48, 49)], but few of those nominated to date have been definitively established as risk factors (6). Nonetheless, the variation in incidence across CT-SES supports the possibility that exogenous or lifestyle factors contribute to the development of at least some subtypes of sarcoma. Moreover, that results were heterogeneous across subtypes and age groups highlights the likelihood of etiologic differences in the tumors evaluated (i.e., environmental vs. genetic factors).

For example, it is hypothesized that individuals diagnosed with (F−) sarcomas more often harbor rare, pathogenic germline genetic variants than those with (F +) sarcomas, as evidenced by the increased frequency with which (F−) sarcoma cases are diagnosed in cancer predisposition syndromes and as secondary cancers (29, 32–34). That the association with CT-SES appeared weaker in (F−) sarcomas compared with (F+) sarcomas in the pediatric and adult age groups supports this hypothesis, as tumors that are unrelated to SES are perhaps more likely influenced by germline genetics. In the older adult stratum, however, we note dissimilar findings; CT-SES was associated with both classes of tumors and nearly every subtype evaluated (8 of 11). We speculate that CT-SES is unrelated to (F−) sarcomas and other subtypes in younger patients due to the increased frequency with which germline genetic predisposition contributes to sarcoma development early in life (7, 50).

In the pediatric and adult age groups, the incidence of several sarcoma subtypes was unrelated to CT-SES but demonstrated dramatic race-specific incidence patterns. We speculate that these findings may be driven in part by differences in the frequency of genetic variants across ancestries. For example, the binding affinity for EWS-FLI1, the fusion oncoprotein characteristic of Ewing sarcoma (30), depends upon the length of GGAA microsatellite repeats (51), which is polymorphic between populations of primarily European and African ancestries (52). This is consistent with the substantially higher incidence of Ewing sarcoma in NH-White individuals compared with NH-Black individuals reported here and elsewhere (42–44). Whether genetic variants that are associated with ancestry underlie associations with other subtypes is unclear. We note, however, that racial and ethnic categories are imperfect measures of genomic ancestry (53), and that other social, cultural, or lifestyle factors which are unaccounted for by area-level SES may have confounded our observations. The use of admixture mapping (54, 55) may help to resolve the potential mechanisms underlying our observations.

To our knowledge, ours is the largest study to investigate the independent contributions of SES and race/ethnicity on sarcoma incidence. Utilizing the SEER program enabled us to evaluate associations among individually rare subtypes of sarcoma within a racially and ethnically diverse population with reasonably adequate statistical power. In addition, the use of a comprehensive and widely available SES variable measured at the census tract-level allowed us to control analyses for possible environmental and lifestyle factors, and facilitated informed speculations on the relative contribution of environment versus genetics in the development of these rare groups of cancers.

Nevertheless, we acknowledge several limitations. Although we included only microscopically confirmed sarcomas, there may still be misclassification of tumor subsets represented in our study. Accurately diagnosing sarcomas remains a clinical challenge (56, 57), and recording their diagnostic codes into tumor registries is subject to error (58, 59). It is possible that misclassification may be greater among individuals with lower SES, as they may have reduced access to specialty sarcoma centers or clinical trials. Furthermore, we classified subtypes as fusion-positive or -negative according to the predominant somatic mutations in each sarcoma based on prior literature (29, 30), but tumors categorized as (F +) may have lacked a fusion gene, and some (F−) sarcoma subtypes may harbor fusion genes yet to be discovered (60). We also note that area-level SES does not represent individual-level SES, but rather captures the SES characteristics of one's environment that may also influence health (61). SEER includes individual-level insurance status for cases but does not include it in calculations of population denominator estimates. This precluded us from using insurance status as a measure of SES in our analysis of incidence rates. Other individual-level SES characteristics (e.g., income or educational attainment) are not publicly available from SEER. Furthermore, the CT-SES measure was assigned to a case based upon their address at diagnosis. This may not account for potentially long latency periods between exposure and diagnosis, during which an individual may have moved between areas of high or low SES. Furthermore, we note that our analysis could not account for environmental exposures that may be associated with sarcoma development, but which are unrelated to CT-SES. Finally, given this is a broad study of rare cancers across the age spectrum, a large number of statistical tests were performed. Some significant findings may therefore be due to chance.

Through this large and comprehensive analysis, we observed sarcoma incidence rates to be more often associated with race/ethnicity than CT-SES. Although future etiologic analyses—particularly genetic studies—are needed to confirm our findings, our results suggest that genetics play a greater role than environmental factors on the etiology of several subtypes of sarcoma.

B.J. Diessner reports grants from National Institute of Arthritis and Musculoskeletal and Skin Diseases at the National Institutes of Health (T32 AR05938 - Musculoskeletal Training Grant) during the conduct of the study. J.N. Poynter reports grants from NIH during the conduct of the study. No potential conflicts of interest were disclosed by the other authors.

B.J. Diessner: Conceptualization, data curation, software, formal analysis, visualization, methodology, writing–original draft, project administration. B.J. Weigel: Conceptualization, supervision, methodology, writing–review and editing. P. Murugan: Conceptualization, methodology, writing–review and editing. L. Zhang: Conceptualization, supervision, methodology, writing–review and editing. J.N. Poynter: Conceptualization, supervision, methodology, writing–review and editing. L.G. Spector: Conceptualization, supervision, methodology, writing–review and editing.

This work was supported by the National Institute of Arthritis and Musculoskeletal and Skin Diseases at the National Institutes of Health “Musculoskeletal Training Grant” (T32 AR05938; to B.J. Diessner).

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.