In 1,882 Endometrial Cancers, Race Shapes the Molecular Landscape — but Small Numbers Complicate the Picture

A large Memorial Sloan Kettering cohort maps how tumor mutational burden and molecular subtypes distribute across racial and ethnic groups in uterine cancer, revealing both real biological variation and the statistical limits of studying it.

Endometrial cancer is the most common gynecologic malignancy in the United States, and its incidence is rising. Black women die from it at roughly twice the rate of white women — a disparity that persists even after adjusting for stage at diagnosis and access to care. Something biological appears to be going on, but genomic studies large enough to dissect it by ancestry have been scarce. Most landmark datasets, including The Cancer Genome Atlas, enrolled predominantly white patients, leaving open the question of whether the molecular subtypes that now guide treatment decisions distribute differently across racial groups.

Memorial Sloan Kettering's endometrial cancer ancestry cohort — 1,882 patients sequenced with the MSK-IMPACT targeted gene panels — offers one of the largest opportunities to examine that question directly. The cohort spans the four recognized molecular subtypes of endometrial cancer (POLE-ultramutated, microsatellite-instability-high, copy-number-high, and copy-number-low), multiple histologies (endometrioid, serous, clear cell, and others), and patient-level race and ethnicity annotations that make ancestry-stratified analysis possible. What follows is a walk through its structure, its key molecular distributions, and the quality-control checks required before any of those distributions can be trusted.

Who Is in the Cohort

Any ancestry-stratified analysis is only as credible as the demographic breadth of the cohort behind it. At MSK, a tertiary cancer center in New York City, the patient population skews toward white, non-Hispanic patients — a reflection of referral patterns, insurance access, and geography. The first question to ask of this 1,882-patient dataset is straightforward: are the minority subgroups large enough to support statistically meaningful comparisons?

The self-reported race distribution makes the scale of the challenge visible immediately. White patients dominate the cohort. Black and Asian patients are present in substantial but smaller numbers, and other categories — Native American, Pacific Islander, multiracial — are sparse enough that standalone analyses for those groups would be unreliable. Any racial category with fewer than roughly 50 patients risks producing findings driven more by chance than biology.

Ethnicity and Its Intersections

Race and ethnicity are not the same variable. A patient can be white and Hispanic, Black and non-Hispanic, or Asian and of unreported ethnicity. The Hispanic-versus-non-Hispanic split in this cohort is heavily tilted toward non-Hispanic patients, which is expected for an MSK-based dataset but limits the power of any ethnicity-specific genomic comparison. Whether the Hispanic/Latino group crosses the threshold of roughly 100 patients — a minimum for meaningful subgroup work — is a critical gating question.

The joint race-by-ethnicity cross-tabulation exposes the problem more starkly. Some intersectional cells contain only a handful of patients. Hispanic patients, in particular, are spread across multiple racial categories, complicating any attempt to assign a single ancestry label. Cells with single-digit counts will need to be collapsed or excluded in downstream analyses to avoid over-interpreting noise.

Some intersectional cells contain only a handful of patients. Hispanic patients are spread across multiple racial categories, complicating any attempt to assign a single ancestry label.

The Mutational Landscape, Before Stratification

Tumor mutational burden — the count of nonsynonymous mutations per megabase of sequenced DNA — is the primary molecular readout of this cohort. TMB matters clinically: tumors with high mutational burden are more likely to respond to immune checkpoint inhibitors, and the FDA has approved pembrolizumab for any solid tumor exceeding 10 mutations per megabase.

Across all 1,882 specimens, the TMB distribution is heavily right-skewed. Most tumors cluster at low values, below 10 mutations per megabase. But a long tail stretches into hypermutated territory (above 20) and even ultramutated territory (above 100). That tail is not random — it reflects the known molecular subtype heterogeneity of endometrial cancer, where POLE-mutant and microsatellite-unstable tumors accumulate mutations at dramatically higher rates than their copy-number-driven counterparts.

Molecular Subtypes Separate Cleanly by TMB

When TMB density curves are broken out by molecular subtype, the expected separation holds. POLE-ultramutated tumors form a distinct high-TMB mode, well separated from the rest. MSI-H tumors occupy an intermediate peak. The two copy-number groups — CN-high (associated with aggressive serous histology and TP53 abnormalities) and CN-low (associated with favorable-prognosis endometrioid tumors) — cluster together at low TMB, with substantial overlap between them.

This separation matters for two reasons. First, it validates the subtype classifications in this cohort: if POLE tumors did not show dramatically elevated TMB, something would be wrong with the labeling pipeline. Second, it underscores that TMB alone cannot reliably distinguish CN-high from CN-low tumors — supplementary biomarkers like TP53 mutation status or copy-number profiles are necessary for that boundary.

Histology and TMB: Serous Tumors Cluster Low, Endometrioid Tumors Spread Wide

Histologic type — what the tumor looks like under the microscope — is an older classification system than molecular subtyping, but it still shapes clinical decision-making. Endometrioid carcinomas, the most common histology, show the widest TMB spread because they encompass POLE, MSI-H, and CN-low molecular subtypes. Serous tumors, by contrast, cluster at low TMB, consistent with their predominantly copy-number-high biology.

Any serous specimens appearing in the hypermutated tail deserve scrutiny: they may represent MSI-H tumors initially misclassified as serous on morphology alone, or mixed-histology cases. The histology frequency bar chart confirms that endometrioid carcinoma dominates the cohort, as expected, but serous representation may be elevated compared to population-based registries — a predictable artifact of MSK's role as a tertiary referral center for high-grade disease.

Benchmarking Against TCGA Proportions

The Cancer Genome Atlas established benchmark proportions for endometrial cancer molecular subtypes: roughly 7% POLE-ultramutated, 25–30% MSI-H, 25% CN-low, and 25% CN-high, with a residual unclassified fraction. How this MSK cohort stacks up against those numbers reveals whether referral bias has reshaped the molecular landscape.

An outsized CN-high group, for instance, would signal enrichment for aggressive serous and TP53-abnormal tumors — exactly what a tertiary center attracts. An unclassified or NA bar exceeding 10% would suggest gaps in the molecular classification pipeline. The OncoTree diagnostic codes provide a cross-check: UCEC (uterine corpus endometrial carcinoma) and USC (uterine serous carcinoma) should account for the bulk of specimens, with any codes outside the uterine lineage flagged as possible data-entry errors.

An outsized copy-number-high group would signal enrichment for aggressive serous and TP53-abnormal tumors — exactly what a tertiary cancer center attracts.

The MSK endometrial cancer ancestry cohort does not yet answer the question it was built to ask — whether tumor biology differs meaningfully across racial groups — so much as it establishes the conditions under which that question can be asked rigorously. The demographic imbalances are real but not disqualifying. The molecular subtype classifications behave as expected. The quality-control checks — specimen type, panel version — provide the guardrails against mistaking technical artifacts for biological signal. What comes next is the harder work: ancestry-adjusted models that can separate the contribution of genetics from the contribution of differential access, delayed diagnosis, and treatment history. A cohort of 1,882 patients, painstakingly annotated for race and ethnicity, is the foundation. The disparities in endometrial cancer mortality are too large, and too persistent, to leave the molecular question unresolved.