Projected Prevalence of Actionable Pharmacogenetic Variants and Level A Drugs Prescribed Among US Veterans Health Administration Pharmacy Users

This cross-sectional study uses 1000 Genomes Project data and US Veterans Health Administration demographic characteristics to project the number of veterans who carry an actionable pharmacogenetic variant and who have been prescribed 1 or more level A drugs that have well-established evidence of pharmacogenetic interactions.


Estimating the prevalence of actionable PGx variants among VHA Pharmacy users
The prevalence of PGx variants was calculated assuming that the Hardy-Weinberg law applies to the VHA population as a large, randomly mating population with negligible rates of mutation and migration. Under this law, genotype frequencies are expected to follow the frequencies of p 2 , 2pq and q 2 for p and q being the frequencies of two alleles of a bi-allelic gene. For genes with more than two alleles, we treated variants in the same gene as mutually exclusive, and the frequency of the wild-type allele was calculated as 1 minus the sum of the actionable variant minor allele frequencies (MAFs) reported for a particular racial/ethnic group (eTable 2). This approach is conservative as it ignores existing variants with lower levels of evidence for an abnormal function.
For the gene G6PD that is located on the X chromosome, we estimated the frequency of actionable genotypes separately by sex; the frequency of actionable genotypes among male patients (X*Y) was estimated as the sum of MAF, and as the frequency of homozygote carriers for female patients (X*X*); we weighted separately the gender frequencies for the two ancestry groups to account for the greater number of women Veterans of African ancestry vs European ancestry (16% vs 8%, respectively) (eTable 3).
Estimations using weighted phenotype frequencies allowed us to account for the frequent combinations of genetic variants at those three loci, and the variations in number of copies for CYP2D6. 1

Modeling the population diversity
To account for the diversity of the VHA population we weighted those estimates to produce the number of actionable genotypes among VHA patients with a representation of 15% patients of African ancestry and 85% of patients of European ancestry that are the two predominant groups, and reflected the proportions of VHA Pharmacy users of African ancestry in our sample (eTable 1). 2,3 VHA enrollees with a race/ethnicity that was either unknown or Hispanic, were merged into the European ancestry group. 2, 3 Additionally, we performed sensitivity analyses to model the population diversity, accounting for European admixture among African Americans (eTables 5 & 6).

Estimating the proportion of VHA Pharmacy users who would carry at least one actionable variant
We estimated the proportion of Veterans who would carry at least one actionable variant as 1 minus the product of the probabilities of a wildtype genotype at each locus (compiled in eTable 2, as 1 minus the sum of frequencies for the Level-A variant alleles). Probabilities were treated as independent as all genes included in the study are carried by separate chromosomes, except for CYP2C9 and CYP2C19 located both on chromosome 10. A sensitivity analysis was performed accounting for the linkage of the variant alleles of CYP2C9 with the wildtype allele of CYP2C19 genes which yielded similar estimates.

Identification of patients receiving a new prescription for clopidogrel within 30 days after a percutaneous coronary intervention
In the case of clopidogrel, clinical guidelines are strongest for the impact of PGx testing in the setting of percutaneous coronary intervention (PCI); therefore, we reported the projected number of patients with actionable phenotypes among those patients receiving a new prescription for clopidogrel within 30 days after a PCI, as indicated by the presence of a procedure code [CPT 92928, 92929, 92933, 92934, 92937, 92938, 92941, 92943, 92944, 92980, 92981, C1874-C1877, C9600-C9603]. 4 As many Veterans undergoing PCI have the procedure done at a non-VA medical center, we included patients with a procedure code either from the OMOP Procedure Occurrence table for all procedures performed at VA, and from the CDW Fee Basis table for all procedures performed outside VA and paid by VA, which indicates that PCI was performed in the community.

Projecting the clinical impact of Level-A gene-drug interactions
Using the phenotype data, we identified a subgroup of patients with projected phenotypes putting them at high risk of being exposed to a drug for which they have a high likelihood of 1) non-efficacy and/or 2) drug toxicity and adverse events, and characterized the anticipated nature of the toxicities. We limited our analyses to the medications in Figure 1 with a strong CPIC level-A phenotype-based recommendation that the patient be prescribed alternative or dose-adjusted therapy. For the drug warfarin, we accounted for the combinations of alleles between the two genes assuming independence of genes carried on different chromosomes (CYP2C9 on chromosome 10 and VKORC1 on chromosome 16).These recommendations applied to the following drug-gene interactions: simvastatin-SLCO1B1 "intermediate to low function" carriers; codeine-CYP2D6 ultra-rapid and poor metabolizers; clopidogrel-CYP2C19 poor metabolizers; allopurinol-HLA-B*5801 carriers; paroxetine-CYP2D6 ultra-rapid metabolizers. We graphically depicted the absolute number of patients who were exposed to these high-risk medications by their risk of drug non-efficacy and/or toxicity.  Population-specific prevalence of actionable genotypes were weighted to generate population estimates using the weights of 15% for AFR and 85% for EUR (Model 1 in eTable 5) We assume that the variants were mutually exclusive, and the frequency of wildtype alleles is obtained as 1 minus the sum of frequencies of the variant alleles.
The frequency of homozygote genotypes is calculated as the square of the frequency of the variant (1 copy on each chromosome), and the frequency of heterozygote is estimated as 2 times the product of the variant allele by the other allele. The frequencies of compound heterozygotes combining two variant alleles were calculated in series to avoid doublecounting of combinations, e.g. allele A was combined to alleles B to E, then allele B is combined to allele C to E, allele C is combined to allele D to e, and so on.  The probability of having at least one actionable variant was calculated as 1 minus the probability of a wildtype phenotype at each of the loci analyzed, which is the product of the probabilities of a wildtype phenotype for each gene. For CYP2C19, CYP2D6 and SLCO1B1, a wildtype phenotype can result from carrying 1 or 2 copies of the wildtype allele, and the probability of a wildtype phenotype is the sum of the probabilities of the two genotypes that are mutually exclusive. a For CYP3A5, *1 is the actionable allele, and combinations of other alleles results in a non-actionable phenotype. Medications with a strong CPIC level A recommendation to either avoid or dose-adjust a medication based on available pharmacogenetic test results, genes and phenotypes associated with an increased risk of toxicity and/or adverse drug reaction in response to drug exposure, number of unique drug users for the drug in 2012-2017, projected prevalence and number of patients with specific phenotype at a high risk of adverse drug reaction and/or non-efficacy, total number of patients with an actionable phenotype a The percentage represents the proportion of Veterans prescribed this medication with the projected phenotype or genetic variant. b From eTable 8.
c Warfarin only has a strong recommendation for patients identifying as non-African ancestry. There are additional recommendations for patients with African ancestry, but we limited our analyses to the projected European ancestry proportion of our population. Therefore, 85% of the 205,177 unique warfarin users are projected to be of European ancestry in our model (n=174,400  (eTable 5) AFR 15%, Model 1 with 15% patients of African ancestry and 85% of patients of European ancestry; AFR20%, Model 2, 20% patients of African ancestry and 80% of patients of European ancestry; AFR Models 3 and 4 models the admixture of African American patients with a 21% contribution of European ancestry; the 21% contribution was estimated based on the literature review summarized in eTable4. Model 5 applies the variant frequencies for the African American in Southwest USA in the 1000 Genome Project with weights of 15% ASW/85% EUR. Model 6 tests the effect of accounting the contribution of populations from American origin, and models the population as 15% for AFR, 7% AMR (population from Americas), and 78% EUR.