Polygenic and pharmacogenomic contributions to medication dosing: a real-world longitudinal biobank study

Abstract

Background: Understanding interindividual variability in medication dosing is central to precision medicine. Despite significant pharmacogenomic (PGx) insights into key biological pathways influencing drug response, the polygenic contribution to dose variability and the potential of electronic health records for maintenance dose estimation remain largely unexplored.

Methods: We leveraged longitudinal drug purchase data linked to the Estonian Biobank (N = 212,000) to derive individual-level daily doses per purchase as well as median and maximum doses as consolidated metrics across purchases for cardiovascular and psychiatric drugs: statins, warfarin, metoprolol, antidepressants, and antipsychotics. Associations with polygenic scores (PGSs) for 16 traits were assessed using linear mixed models and multivariable regression with a forward stepwise approach. Genome-wide association studies (GWAS) were followed by gene set enrichment analyses for known PGx genes.

Results: Sample sizes ranged from 684 (antipsychotics) to 20,642 (statins), with median doses reflecting typical maintenance doses. Trait-specific PGSs were significant for the daily dose of statins (coronary heart disease PGS, β = 0.02, P = 5.9 × 10^-10) and metoprolol (systolic blood pressure PGS, β = 0.03, P = 7.5 × 10^-13). The PGS for body mass index was linked to daily doses of statins (β = 0.02, P = 6.4 × 10^-7), metoprolol (β = 0.03, P = 1.4 × 10^-14), and warfarin (β = 0.03, P = 0.001), whereas the PGS for educational attainment showed opposing associations with statins (β = - 0.01, P = 5.9 × 10^-4) and antidepressants (β = 0.01, P = 0.002). Median and maximum doses yielded similar, though generally weaker, associations. GWAS confirmed signals for metoprolol (CYP2D6, P = 1.1 × 10^-20) and warfarin (CYP2C9, P = 8.9 × 10^-60; VKORC1, P = 4.2 × 10^-148), as well as enrichment of PGx signals for individual statins (P = 0.02 for simvastatin, P = 0.03 for atorvastatin). Associations remained significant after adjusting for disease-specific PGSs, suggesting independent contributions of PGx loci.

Conclusions: These findings illustrate the feasibility and value of leveraging real-world electronic health records to derive pharmacologically meaningful medication dosing phenotypes. Both polygenic and pharmacogenomic signals contribute to dose variability, underscoring their potential utility in personalized prescribing strategies.

Keywords: Biobank; Electronic Health Records; Genome-wide association study; Medication dosing; Pharmacogenomics; Polygenic scores; Real-world health data.

Fig. 1

Overview of the study design. Purchase data were used to derive three dose variables: (i) daily dose at each purchase, (ii) median dose across purchases, and (iii) maximum dose across purchases. PGSs were tested for associations with the dose variables using linear mixed models for daily doses, linear regression for median doses and logistic regression for maximum doses, with GWAS performed for the latter two measures. I(max dose) is the indicator function that is 1 if the person has the max dose, and 0 otherwise. BMI body mass index, HbA1c glycated haemoglobin, LDL low-density lipoprotein, HDL high-density lipoprotein, TG triglycerides, eGFR estimated glomerular filtration rate, ALT alanine aminotransferase, SBP systolic blood pressure, EA educational attainment, EVERSMK ever smoking, CHD coronary heart disease, MDD major depressive disorder, SCZ schizophrenia, VTE venous thromboembolism, AD antidepressant.

Fig. 2

Effect sizes for PGSs significantly associated with derived dose variables. Only PGSs that surpassed Bonferroni correction and were independently associated are shown for A daily doses, B median doses, and C maximum doses. Different colours represent distinct PGSs: red for coronary heart disease (CHD), blue for body mass index (BMI), green for educational attainment (EA), purple for systolic blood pressure (SBP), and yellow for high-density lipoprotein (HDL).

Fig. 3

GWAS results for metoprolol and warfarin. Manhattan plot for metoprolol A median dose and B maximum dose adjusted for birth year, sex, treatment length on supply, and the first ten PCs. Manhattan plot for warfarin C median dose and D maximum dose adjusted for birth year, sex, treatment length on supply, and the first ten PCs. The y-axis represents − log₁₀(P-values) for the association of SNVs with the dose variables, the horizontal dashed line indicates the genome-wide significance threshold (P < 5 × 10⁻⁸), red dots genome-wide significant SNVs, and number 23 on x-axis denotes chromosome X.