(P-056) Genetic Heterogeneity Patterns of Electronic Medical Record (EMR)-Derived Race Groups in a Deidentified Clinico-genomic Database (CGDB)

Background: Disease incidence, pathology and biomarker prevalence vary based on genetic ancestry. Increased availability of EMR data linked to comprehensive genomic profiling (CGP) data in CGDBs provides the opportunity to leverage genotypes and sociodemographic variables (such as race) in therapeutic and disparities research. Characterizing the relationship between derived race and genetic ancestry from the non-random subset of markers in CGP is an important first step to understand the interplay of genetics and sociodeterminants in patient outcomes.

Objectives: To evaluate the relationship between EMR-derived race and genetic ancestry derived by an admixture model and a multiclass classifier in a CGDB.

Methods: This study used the nationwide (US-based) de-identified Flatiron Health–Foundation Medicine pan-tumor (PT) CGDB containing data from ≈280 US cancer clinics (≈800 sites of care). Genetic ancestry from 5 geographic ancestral groups (AFR: African, AMR: admixed American, EAS: East Asian, EUR: European, and SAS: South Asian) was predicted using a random forest classifier utilizing ancestry-informative markers by CGP. Admixture proportions of these 5 ancestral groups were quantitated using an admixture analysis work flow. Race from EMR records was categorized into Asian, Black/African American [AA], White, and Other (including cases with ≥1 race). Concordance was measured as % agreement between derived race and ancestry classifier calls. Mean (SD) ancestry fractions were calculated for each race group.

Results: Most of the 64,036 patients (pts) were EMR-derived as White (68%), then Black/AA (6%) and Asian (2%). 14% of pts were classified as Other and 9% of pts were missing race information. Ancestry data were complete for all specimens captured from solid tumor tissue (covering 92% of pts). A total of 72% of the pts were predicted to be of EUR ancestry, 8% AFR, 2% EAS, 1% SAS and 7% admixed. Overall agreement between race and expected genetic ancestry was high (86% White:EUR, 98% Black:AFR, 99% Asian:EAS). The mean (SD) of EUR ancestry fraction in White pts was 0.90 (0.11). More genetic ancestry heterogeneity was observed (mean, SD) in Black/AA (0.79; 0.16 AFR fraction) and Asian (0.74; 0.38 EAS fraction) pts.

Conclusions: Genetic ancestry from CGP data was highly correlated with EMR-derived race. However, the proportion of genetic admixture varies across groups and should be considered for studies on ancestry or the impact of genetic predisposition to disease.