Sarcomas are malignant tumors arising from connective tissues, including bone, muscle, fat, and cartilage. They are already considered rare cancers, but within this group, the Connective Tissue Oncology Society (CTOS) has further identified a subset classified as "ultra-rare sarcomas" (URS), defined as bone and soft tissue sarcomas with an annual incidence rate of fewer than one case per one million individuals. The WHO classification currently lists 22 bone sarcoma subtypes and 56 soft tissue sarcoma subtypes that meet this threshold, collectively representing roughly 20% of all bone and soft tissue sarcomas.
Clinical challenges of ultra-rarity: The extreme infrequency of URS creates compounding clinical difficulties. Pathologists rarely encounter these tumor types, which contributes to diagnostic error rates of 8%-14% for sarcomas broadly. Clinicians lack sufficient exposure to develop familiarity with disease behavior, and pharmaceutical companies face insurmountable hurdles generating the evidence base needed for drug approvals in populations this small. Patients may wait longer for an accurate diagnosis and have fewer evidence-based treatment options than those with more common sarcoma subtypes.
The knowledge gap: Despite these challenges, no comprehensive epidemiological analysis of URS had been conducted prior to this study. While individual URS subtypes have been described in small case series and single-institution retrospective studies, the aggregate burden, demographic patterns, and prognostic impact of URS as a class had not been characterized at the population level. The recent formalization of the URS classification by CTOS made this kind of analysis both possible and urgent.
This 2025 study in ESMO Open addresses that gap directly, using data from the Bone and Soft Tissue Tumor Registry in Japan, one of the most comprehensive national registries of its kind, to provide the first nationwide analysis of URS epidemiology and prognosis. The study analyzed 22,821 patients over nearly two decades, making it the largest investigation of URS to date.
The study drew on the Bone and Soft Tissue Tumor Registry in Japan, an organ-specific national cancer registry established in the 1950s under the Japanese Orthopaedic Association, with ongoing support from the National Cancer Center Japan. Data from January 2001 to December 2019 were extracted. Starting from 50,550 initially registered malignant tumor patients, the authors applied exclusion criteria to remove cases where treatment occurred elsewhere, pathological diagnosis was absent, the diagnosis was metastatic carcinoma, or the histological classification was inconclusive under current WHO criteria. After exclusions, 22,821 patients were retained: 5,820 bone sarcoma and 17,001 soft tissue sarcoma patients.
URS classification: Histological diagnosis at each facility was assigned by local pathologists according to WHO classification criteria. When diagnostic categories changed during the 2001-2019 period due to WHO classification updates, diagnoses were revised to match the current framework. The URS designation was applied using the histological type list published in the CTOS consensus paper, providing a standardized and externally validated framework for classifying which subtypes qualify as ultra-rare.
Propensity score matching: A critical methodological challenge in comparing URS and non-URS patients is baseline confounding. URS patients differ from non-URS patients in age, stage distribution, and treatment patterns. To address this, the authors used propensity score matching via the R package MatchIt (version 4.5.5), applying the nearest-neighbor method. Variables included in the propensity score calculation were age, sex, TNM stage, histological malignancy grade, resection of the primary tumor, chemotherapy, and radiotherapy. Patients with missing data in any of these variables were excluded from the matched analysis. After matching, 1,258 pairs for bone sarcoma and 2,372 pairs for soft tissue sarcoma were retained for adjusted survival comparisons.
Statistical framework: Overall survival (OS) was defined from the date of first hospital visit to death. Outlier survival times were detected and excluded using the Smirnov-Grubbs test to minimize registration errors. Survival curves were estimated using the Kaplan-Meier method with log-rank testing. Multivariate risk factor analysis used the Cox proportional hazards model. A random survival forest (RSF) analysis, implemented via the R package randomForestSRC (version 3.2.3) with 500 trees, was used to assess variable importance in a fully non-parametric framework, providing a machine learning complement to the classical Cox regression.
URS accounted for 18.9% of all patients in the analysis, with 25.8% of bone sarcoma patients and 16.5% of soft tissue sarcoma patients meeting the ultra-rare classification. This proportion is consistent with prior population-based estimates, but notably this is the first time it has been confirmed using a comprehensive national registry rather than extrapolated from smaller datasets.
Bone sarcoma demographics: URS bone sarcoma patients were significantly older than their non-URS counterparts, with a mean age of 57.6 years versus 39.2 years (P less than 0.001). The non-URS group showed the characteristic bimodal age distribution of bone sarcomas (peaks at 15 and 70 years, reflecting osteosarcoma in adolescents and chondrosarcoma in older adults), while the URS group displayed a unimodal distribution centered around 70 years. Sex distribution was similar between groups (P = 0.25). The histological composition of ultra-rare bone sarcomas was dominated by chordoma (29.8% of URS bone cases) and undifferentiated pleomorphic sarcoma (25.1%).
Soft tissue sarcoma demographics: The pattern was reversed for soft tissue sarcomas. URS patients were significantly younger than non-URS patients, with a mean age of 49.4 years versus 62.2 years (P less than 0.001). While the non-URS group peaked around age 70 consistent with common adult soft tissue sarcomas such as undifferentiated pleomorphic sarcoma and liposarcoma, the URS group showed a more evenly distributed age pattern without a dominant peak. A slight female predominance was noted in the URS group (46.6% versus 43.6%, P = 0.004). The most prevalent ultra-rare soft tissue sarcoma subtypes were pleomorphic liposarcoma (11.2%) and extraskeletal myxoid chondrosarcoma (9.6%).
Clinical significance of age distributions: Among patients older than 80 years with bone sarcomas, and those younger than 20 years with soft tissue sarcomas, URS and non-URS cases were nearly equally represented. This has direct diagnostic implications: clinicians evaluating elderly patients with bone lesions or young patients with soft tissue masses should not exclude ultra-rare subtypes, as these diagnoses are proportionally more common in those age groups than in the general sarcoma population.
URS and non-URS sarcomas differed substantially in TNM stage at initial presentation, treatment receipt, and recorded disease status at follow-up. Understanding these differences is essential for interpreting the survival data, as they represent confounders requiring statistical adjustment through propensity score matching.
Bone sarcoma treatment patterns: In bone sarcomas, non-URS patients were more frequently diagnosed at stage II (59.9% versus 53.7% for URS), while URS patients were more often at stage I (29.7% versus 23.8%). High-grade histology was more common in non-URS bone sarcomas (75.5% versus 68.2%, P less than 0.001). Primary surgical resection was more frequent in non-URS patients (77.3% versus 57.7%), as was chemotherapy (54.9% versus 30.8%), while radiotherapy was more commonly used in URS patients (36.3% versus 17.0%, P less than 0.001 for all). The lower surgical resection rate in URS may reflect the older age profile of these patients and the tendency for certain ultra-rare subtypes, particularly chordoma, to be located in anatomically challenging sites (sacrum, skull base) where resection is technically difficult or associated with extreme morbidity.
Soft tissue sarcoma treatment patterns: In soft tissue sarcomas, the URS group showed markedly worse stage at presentation, with 26.3% at stage IV compared with only 7.8% for non-URS (P less than 0.001), and far fewer at stage I (19.2% versus 39.1%). High-grade tumors were more prevalent in URS (80.0% versus 60.6%). Despite greater disease burden, URS patients underwent primary surgical resection less frequently (74.0% versus 82.0%) but received chemotherapy (43.6% versus 23.3%) and radiotherapy (28.2% versus 19.3%) more often, consistent with higher rates of inoperable disease. Final disease status reflected these differences: NED (no evidence of disease) was achieved in only 53.5% of URS soft tissue patients versus 70.0% of non-URS patients, while "died of disease" was recorded in 24.0% versus 12.4%, respectively.
Surgical benefit heterogeneity: A notable finding from the multivariate subgroup analysis was that primary surgical resection conferred a significantly lower hazard ratio in non-URS bone sarcomas (HR 0.49, 95% CI 0.41-0.58), whereas no significant benefit was observed in the URS bone sarcoma subgroup (HR 0.97, 95% CI 0.73-1.29). This divergence may reflect the different biology of URS histotypes, some of which are relatively indolent (chordoma) while others have intrinsic resistance to multimodal treatment, or the anatomical constraints on achieving complete resection.
Unadjusted 5-year overall survival was 69.7% (95% CI 68.1%-71.3%) for all bone sarcoma patients and 76.0% (95% CI 75.0%-76.9%) for all soft tissue sarcoma patients. These population-level figures, however, mask important differences between URS and non-URS subgroups and cannot be interpreted without accounting for the substantial baseline differences in age, stage, and treatment between the two groups.
Unadjusted survival comparisons: In the unadjusted analysis, there was no statistically significant difference in OS for bone sarcomas between URS and non-URS (P = 0.076). The 5-year survival was 70.2% (95% CI 68.4%-72.1%) for non-ultra-rare bone sarcomas and 68.2% (95% CI 64.9%-71.6%) for ultra-rare bone sarcomas. For soft tissue sarcomas, however, the URS group showed significantly worse unadjusted OS (P less than 0.0001), with a 5-year survival of 64.9% (95% CI 62.5%-67.5%) versus 78.2% (95% CI 77.2%-79.2%) for non-URS, a difference of over 13 percentage points.
Propensity score-matched survival: After applying propensity score matching to account for confounding, the picture shifted meaningfully. In matched bone sarcoma pairs, URS patients actually showed significantly longer survival than non-URS patients (P = 0.022), suggesting that once baseline characteristics are equalized, ultra-rare bone sarcomas have a comparatively favorable biology. In matched soft tissue sarcoma pairs, the survival difference was attenuated and no longer statistically significant (P = 0.052), indicating that much of the observed OS gap in soft tissue sarcoma is attributable to confounders such as the higher proportion of advanced-stage disease in URS.
Age-stratified survival: The most clinically actionable finding emerged from age-stratified propensity-matched analyses. Ultra-rare soft tissue sarcomas were associated with significantly shorter survival in patients younger than 20 years (P less than 0.0001) and in those aged 20-39 years (P = 0.0007), whereas no significant difference was observed in patients older than 40 years. Conversely, ultra-rare bone sarcomas showed significantly better survival in the 20-39-year-old subgroup (P = 0.013). These findings point to a biologically meaningful interaction between URS status and age, particularly for young soft tissue sarcoma patients.
The relative contribution of URS status to prognosis, compared with other clinical variables, was assessed through two complementary approaches: classical Cox proportional hazards regression and random survival forest (RSF) analysis, the latter representing a machine learning method suited to detecting complex, nonlinear relationships between variables and survival outcomes.
Cox proportional hazards results: TNM stage IV at the initial visit had the highest hazard ratio in both groups, reflecting the dominant importance of metastatic disease at presentation. For bone sarcoma, stage IV carried an HR of 8.29 (95% CI 4.65-14.79), while for soft tissue sarcoma the HR was 10.26 (95% CI 6.47-16.26). Surgical resection of the primary tumor was the most protective factor in both groups: HR 0.58 (95% CI 0.50-0.67) for bone sarcoma and HR 0.47 (95% CI 0.43-0.52) for soft tissue sarcoma. URS status itself carried an HR of 0.73 (95% CI 0.63-0.85) in bone sarcoma (indicating modestly better survival) and 1.25 (95% CI 1.13-1.38) in soft tissue sarcoma (indicating modestly worse survival). Both were statistically significant but had HRs close to 1, indicating a relatively small absolute effect compared with stage and surgical treatment.
Random survival forest analysis: RSF is a machine learning technique adapted from standard random forests for survival analysis of right-censored data. Unlike Cox regression, RSF makes no proportional hazards assumption and can capture interaction effects and nonlinearities. The analysis used 500 decision trees. Variable importance was assessed as the degree to which each variable reduced prediction error. TNM stage emerged as the most important variable in both bone sarcoma (importance score 0.27) and soft tissue sarcoma (importance score 0.27). In contrast, URS status ranked among the lowest importance variables: 0.012 for bone sarcoma and 0.005 for soft tissue sarcoma, making it the second least important factor for bone sarcoma prognosis and the least important factor for soft tissue sarcoma prognosis across all variables tested.
Interpreting the hierarchy of prognostic factors: The convergence of Cox regression and RSF results reinforces a consistent message: while URS status does have a statistically detectable association with OS, its clinical magnitude is small relative to TNM stage, surgical resectability, age, and histological grade. This has important implications for clinical counseling. A patient with a stage I ultra-rare sarcoma that is surgically resectable has a far better prognosis than a patient with a stage IV non-ultra-rare sarcoma, despite the "ultra-rare" label of the former.
The finding of significantly worse survival in patients under 40 years old with ultra-rare soft tissue sarcomas prompted a deeper look at the specific histological subtypes driving this signal. The prognosis of any sarcoma group is heavily influenced by its histological composition, and the URS and non-URS groups in younger patients differ substantially in which tumor types they contain.
Non-URS in young soft tissue sarcoma patients: Among non-URS soft tissue sarcoma patients under 40, the two dominant subtypes are myxoid/round cell liposarcoma (28.9%) and synovial sarcoma (25.8%). The published literature attributes a 5-year survival rate of 79%-91% to myxoid/round cell liposarcoma and approximately 63% to synovial sarcoma. These are tumors with at least partially predictable behavior and, for myxoid liposarcoma, a well-characterized translocation (FUS-DDIT3 or EWSR1-DDIT3) that defines a relatively coherent biological entity.
URS in young soft tissue sarcoma patients: The ultra-rare soft tissue sarcoma group under 40 is dominated by extraskeletal Ewing sarcoma (15.0%), alveolar soft part sarcoma (14.8%), and alveolar rhabdomyosarcoma (13.6%). The 5-year survival rates for these subtypes in published studies are notably lower: 53%-67% for extraskeletal Ewing sarcoma, 68% for alveolar soft part sarcoma, and approximately 47% for alveolar rhabdomyosarcoma. The prevalence of these aggressive subtypes in the young URS population directly explains the survival disadvantage observed.
Implications for bone sarcomas: The opposite pattern, where ultra-rare bone sarcomas had better prognosis than non-URS bone sarcomas after adjustment, may similarly reflect histological composition. Chordoma, the most prevalent ultra-rare bone sarcoma (29.8%), had an overall 5-year survival of 67.5% in this dataset. While not high by absolute standards, chordoma behaves as a locally aggressive but relatively indolent tumor with low metastatic potential. In contrast, dedifferentiated chondrosarcoma, which accounted for 7.9% of ultra-rare bone sarcomas, had a 5-year survival of only 18%, representing the aggressive tail of the URS bone sarcoma distribution. The mix of these subtypes within URS determines the group's aggregate prognosis in ways that simple URS versus non-URS comparisons cannot capture.
Evolving treatment landscape: The registry data span 2001 to 2019, a period during which systemic therapy for sarcomas changed substantially. Pazopanib received regulatory approval for soft tissue sarcoma based on the PALETTE trial. Trabectedin was approved for translocation-related sarcomas. Eribulin was approved for liposarcoma and leiomyosarcoma. Because the registry captured treatments administered but not the line of therapy or sequencing, it is not possible to determine how the introduction of these agents affected outcomes across the study period. The survival estimates may therefore represent a composite of different treatment eras rather than a uniform standard of care.
Missing clinical variables: Eastern Cooperative Oncology Group performance status (ECOG PS) was not collected in the registry, representing a significant gap. Performance status is a strong independent predictor of cancer outcomes and of treatment tolerance, particularly in elderly patients. Its absence from the propensity score and the multivariate analysis means that the prognostic models are incomplete. Similarly, depth of the primary tumor, an important prognostic variable in soft tissue sarcomas, had substantial missing data and was excluded from the analysis. Including both ECOG PS and tumor depth could alter the apparent importance of URS status and other variables.
Diagnostic delay and timeline data: The authors hypothesized that URS might contribute to diagnostic delays given clinician unfamiliarity with these subtypes, which in turn could delay treatment initiation and worsen outcomes. An analysis of the time between first hospital visit and pathological diagnosis was proposed but could not be executed because the date of pathological diagnosis was insufficiently recorded across registry participants. Collecting reliable diagnostic timeline data is identified as a priority for future registry iterations.
Future research priorities: The authors call for analyses stratified by individual histological type within URS rather than treating all ultra-rare subtypes as a single class. Given the enormous biological heterogeneity within the 78 URS subtypes, aggregate analyses necessarily obscure clinically important variation. International collaboration is essential, as the incidence of any single URS subtype is too low for any one national registry to generate definitive evidence. The MASTER KEY platform trial and similar adaptive umbrella designs represent one pathway toward generating prospective evidence for patients with ultra-rare cancers. Specific clinical trials targeting the URS subtypes with the worst outcomes in young patients, particularly alveolar rhabdomyosarcoma and extraskeletal Ewing sarcoma, are needed urgently.