Estimating the Frequency of False-Negative Pharmacogenetic Test Results by Self-Reported Ancestry

Abstract

Concerns about the applicability of pharmacogenetic (PGx) testing across diverse ancestry groups have risen from the underrepresentation of non-European populations in PGx research. Current PGx panels may fail to detect relevant variants in non-European populations, increasing the likelihood of false-negative results. To investigate this, we assessed reference allele (*1) and genotype (*1/*1) frequencies by self-reported ancestry in a cohort of 1086 youth aged 6-24 years who underwent PGx testing. Testing included 10 pharmacogenes (CYP2B6, CYP2C19, CYP2C9, CYP2D6, CYP3A4, CYP3A5, NUDT15, SLCO1B1, TPMT, and VKORC1) using a panel covering all Association for Molecular Pathology Tier 1 alleles and 53% of Tier 2 alleles. Compared with Europeans (n = 727), non-Europeans (n = 359) had higher *1 allele frequencies for CYP2C9, CYP2D6, and CYP3A5 (all P < 0.01), while Europeans had higher frequencies for CYP2C19 and VKORC1 (all P < 0.01). Similarly, *1/*1 genotype frequencies were higher in non-Europeans for CYP2C9 and CYP3A4 (all P < 0.01), but higher in Europeans for VKORC1 (P < 0.01). False-negative estimates exceeded 1% for CYP2B6, CYP2D6, CYP2C9, CYP2C19, and SLCO1B1 in at least one ancestry group. These findings support the notion that *1 allele and *1/*1 genotype frequencies are more frequent in non-Europeans for specific genes, but Europeans are also at comparable risk for false-negative results. Expanded allele coverage on PGx panels could mitigate false-negative risks, improving equity in PGx testing across diverse populations.

Figure 1

Forest plots with odds ratios and 95% confidence intervals for (a) *1 allele and (b) *1/*1 genotype assignments in non‐Europeans compared to Europeans. Odds ratios for CYP3A4 and CYP3A5 were not reported because some non‐European subgroups had 100% frequency of CYP3A4*1 or 0% frequency of CYP3A5 *1/*1, leading to 95% confidence intervals ranging from 0 to infinity, making the estimates non‐informative.