There are well documented disparities in the incidence of cancer and outcomes of
treatment across patients of different races and ethnicities. This is credited to a variety
of factors which include social, structural, and access to care inequities, as well as
biological differences that may correlate with race/ethnicity. We aimed to measure racial
differences in testing for cancer therapy decision support using real-world data (RWD)
from 100,000 de-identified patients who underwent tumor genomic profiling with the
Tempus xT next-generation sequencing assay (targeting 648 genes). A challenge for
this analysis is that in RWD race/ethnicity is frequently missing from patients’ records.
Instead, we used ancestry-informative markers overlapping assay capture regions to
infer continental ancestry proportions: Africa, Americas, Europe, East Asia, and South
Asia. Our data show that despite a majority of patients being of European descent
(72%), our cohort includes 8 to 12-fold more patients with substantial (>50%)
non-European ancestry when compared to The Cancer Genome Atlas. Recognizing the
complexity of ancestry and race relationships, we imputed several race/ethnicity
categories using ancestry admixture thresholds based on literature and our own
analysis, demonstrating less than 2% error with available race/ethnicity labels. With
imputation, 85% more Black patients were identified within our cohort. Furthermore,
Hispanics/Latinos were substantially overrepresented among those missing ethnicity
metadata; using imputed ethnicity labels increased the numbers of this patient
population by 150%. Patient subpopulation disparities were estimated through
comparison of imputed race/ethnicity distributions with the expected overall cohort-level
distributions. We observed significantly lower than expected percentages of Black
patients with pancreatic (-18%) and urinary tract cancers (-42%), and White patients
with breast (-7%) and colorectal (-5%) cancers, whereas higher than expected numbers
of Hispanic/Latino patients with colorectal (+22%) and thyroid (+48%), and Asian
patients with gallbladder cancer (+32%) were present (p<0.05, chi-squared test). These
disparities are unexpected as compared to the ranking of incidence rates in the SEER
database by race/ethnicity. Our results show that genetic ancestry inference on genomic
data from tumor profiling can partially compensate for the lack of race/ethnicity
information in RWD and allow research on disparities and biological race differences in
cancer etiology and outcomes.