Genotype-predicted drug response phenotypes and their co-occurrence with dispensed medicines among 738,531 participants in the UK Our Future Health study

Technical Summary: Genotype-Predicted Drug Response Phenotypes and Co-occurrence with Dispensed Medicines in the UK Our Future Health Study

Problem Statement
Pharmacogenomics (PGx) holds the potential to improve medication safety and efficacy by tailoring prescriptions to an individual's genetic profile. While international consortia (e.g., CPIC, DPWG) have established evidence-based prescribing guidelines for numerous pharmacogenes, the population-level utility of pre-emptive PGx testing depends on the frequency with which actionable genetic phenotypes co-occur with the actual dispensing of the affected medicines. Previous large-scale studies have established that actionable PGx phenotypes are nearly universal (prevalence >98%), but evidence regarding the co-occurrence of these phenotypes with medication exposure has been limited by small sample sizes, reliance on self-reported medication use, or lack of linkage to primary care dispensing records. Furthermore, existing data is heavily skewed toward European-ancestry populations, leaving gaps in understanding how these dynamics vary across diverse genetic ancestries.

Methodology
This study utilized a cross-sectional design within the "Our Future Health" (OFH) cohort, a new UK national biobank aiming to recruit a diverse population. The analysis included 738,531 participants with imputed genome-wide genotypes and linked primary care dispensing records from England (April 2018 to June 2025).

• Phenotyping: The researchers applied the Pharmacogenomics Clinical Annotation Tool (PharmCAT v3.1.1) to imputed genotypes across 19 pharmacogenes covered by the tool. Phenotypes were translated from star-allele diplotypes. A specific recovery procedure was developed for CYP2C19, as nine positions defining its star alleles were missing from the imputation panel; this allowed for phenotype recovery in 99.8% of participants.
• Actionability Definition: A phenotype was deemed "actionable" if current CPIC or DPWG guidelines advised deviating from standard care (e.g., dose adjustment, drug avoidance, or alternative selection) for at least one medicine.
• Ancestry Stratification: Participants were assigned genetically-inferred ancestry (GIA) groups (European, South Asian, East Asian, African, Middle Eastern/North African, and Admixed American) using principal component analysis, rather than self-reported ethnicity.
• Dispensing Linkage: Medication exposure was ascertained via the NHS Business Services Authority dataset, covering community dispensing in England. The study focused on 80 gene-drug pairs with actionable guidelines, narrowing to 39 pairs (33 distinct medicines) that were actually dispensed in English primary care during the study window.
• Statistical Analysis: Prevalence of actionable phenotypes and co-occurrence rates (having both an actionable phenotype and a dispensed matched medicine) were calculated overall and stratified by GIA and age.

Key Results

• Universal Actionability: Every participant (100%) carried at least one actionable PGx phenotype, with a mean of 6.1 actionable phenotypes per individual (SD 1.3). This saturation persisted even when excluding genes where actionable phenotypes are near-universal (e.g., CFTR, CYP3A5, UGT1A1).
• Ancestry Variations: While the number of actionable phenotypes was similar across ancestry groups (ranging from 5.4 in African ancestry to 6.2 in European ancestry), the specific genes contributing to this count varied significantly. For instance, actionable CYP3A5 phenotypes were present in 99.5% of European participants but only 68.2% of African participants, whereas actionable VKORC1 phenotypes were most common in East Asian (97.8%) and least common in African (15.1%) participants.
• Co-occurrence with Dispensing: Overall, 36.8% (95% CI 36.7–36.9) of participants had been dispensed at least one medicine matching a gene for which they carried an actionable phenotype.
  • This rate varied by ancestry, ranging from 37.6% in European participants to 19.0% in East Asian participants.
  • Co-occurrence increased steeply with age, doubling from the 18–39 age group (12.0–26.6%) to the ≥70 age group (43.7–58.9%).
• Driving Factors: The co-occurrence was concentrated in a few widely prescribed drug classes. The highest rates were observed for proton-pump inhibitors and antidepressants acting through CYP2C19, and statins acting through SLCO1B1. For example, 26.9% of participants with an actionable CYP2C19 phenotype had been dispensed omeprazole.
• Medication Burden: Participants carried actionable phenotypes for a mean of 13.8 of the 33 distinct medicines with PGx guidance dispensed in primary care, but were actually dispensed a mean of only 0.6 of these medicines.

Significance and Claims
The paper claims that this study provides the first large-scale, ancestry-stratified assessment of the co-occurrence between actionable PGx phenotypes and actual primary care dispensing in the UK. By linking imputed genotypes to linked dispensing records in a diverse cohort, the authors demonstrate that:

1. Pre-emptive testing has significant reach: More than one-third of the population carries a genetic profile that could influence the prescribing of a medicine they have recently received, suggesting a substantial opportunity for pre-emptive PGx implementation.
2. Implementation priorities are clear: The co-occurrence is not evenly distributed but is concentrated in specific gene-drug pairs (notably CYP2C19 and SLCO1B1 pathways), identifying where future safety and effectiveness studies should focus.
3. Ancestry matters for implementation strategy: While the overall burden of actionable phenotypes is similar across ancestries, the specific genes and the frequency of medication exposure differ, implying that implementation strategies must account for genetic diversity and local prescribing patterns.
4. Lower bounds: The authors explicitly state that their co-occurrence estimates are likely lower bounds due to the limited seven-year dispensing window and the exclusion of non-primary care medications. As the cohort matures and linkage to full primary care records extends, the observed reach of PGx-guided prescribing is expected to increase.

The study concludes that actionable PGx phenotypes are universal in this national cohort, and the high rate of co-occurrence with dispensed medicines supports the feasibility and potential impact of integrating pre-emptive PGx testing into routine clinical care.