Publications

Background and Aims: Up to half of patients switch or discontinue antihypertensive medications within the first year, but underlying mechanisms remain elusive. This study aimed to identify genetic predictors of antihypertensive medication use trajectories within the first year.

Methods: Using longitudinal medication data from >400 000 genotyped antihypertensive medication users across three cohorts (FinnGen, the UK Biobank, and the Estonian Biobank), short-term antihypertensive medication use trajectories were classified as Continue, Switch, or Discontinue. Genome-wide association studies were performed across five medication classes.

Results: In total, 14 genome-wide significant loci were identified for switching from angiotensin-converting enzyme inhibitors (ACEI) and dihydropyridine calcium channel blockers (dCCB) to other antihypertensive medications. For ACEI switching, evidence converged on the neurotensin-NTSR1 pathway, including a 320-fold Finnish-enriched protective missense variant in the neurotensin receptor gene NTSR1 (rs148569146 [G301R], odds ratio [OR] 0.49, P = 3.3 × 10−43) and a variant near RASSF9 (rs181941187, OR = 0.74, P = 1.2 × 10−49) tagging the neurotensin gene NTS. In drug–gene interaction analyses, NTSR1 G301R was associated with reduced ACEI-induced cough risk (OR 0.39, P = 8.1 × 10−4). The dCCB switching locus at CYP3A43 was in near-complete linkage (r2 = 0.99) with the functional CYP3A4*22 allele (rs35599367, OR 1.23, P = 6.1 × 10−10). A polygenic risk score (PRS) for ACEI switching predicted two-fold ACEI cough risk in the top 10% PRS compared with the middle 20% in an independent sample of the Estonian Biobank.

Conclusions: These findings extend the bradykinin hypothesis of ACEI-induced cough by implicating neurotensin-NTSR1 signalling, pinpoint CYP3A4*22 as a novel functional predictor of dCCB switching with potential for genotype-guided prescribing, and validate medication use trajectories as a framework for pharmacogenetic discovery.

Purpose To investigate the role of CYP2D6 and its pharmacogenetic variability in the association between anticholinergic burden and cognitive function in older adults.

Methods We conducted a cross-sectional analysis of 872 community-dwelling adults ≥ 65 years from the ActiFE-Ulm cohort. Cognitive performance was assessed using the Mini-Mental State Examination (MMSE). Anticholinergic burden scores deriving from CYP2D6-metabolised drugs (ABS2D6) were calculated for all regularly scheduled medications. CYP2D6 metaboliser status was determined via genotyping and categorised as poor (PM), intermediate (IM), normal (NM), or ultrarapid (UM). The 25th, 50th, and 75th percentiles of the MMSE score distribution were studied using quantile regression to examine the association between ABS2D6 and cognition, including interaction and stratified analyses by metaboliser status.

Results ABS2D6 was significantly associated with lower MMSE scores at the 50th and 75th percentiles in early models, but not after additional adjustment for age and sex. A consistent effect modification by IM status was observed at the 75th percentile, with IMs showing a significant negative association between ABS2D6 and MMSE in stratified analyses (β = -0.34 [95% CI: -0.64, -0.05]). After excluding participants with use of CYP2D6 inhibitors, this effect became more pronounced (β = -0.50 [95% CI: -0.82, -0.18]). Findings among PMs remained uncertain due to limited subgroup size.

Conclusion These exploratory findings suggest that CYP2D6 metaboliser status may contribute to interindividual susceptibility to anticholinergic burden-related lower cognitive performance. If confirmed in larger longitudinal studies, pharmacogenetic information may support more individualised anticholinergic risk assessment and safer prescribing in older adults.

Introduction: Individual CYP2D6 activity can be assessed with minimal invasiveness by genotyping and solanidine-derived phenotyping. In this study, we aimed to compare the two methods to predict the metabolism of the CYP2D6 substrate metoprolol in a cohort of the “All-comer” Registry for ImmunocArdiology aNd cardiometabolic disease Aachen (ARIANA) study (DRKS00025716; https://drks.de/search/de/trial/DRKS00025716) with a stable metoprolol regimen.

Methods: In total, N = 48 patients were clinically assessed, and their medical records were analyzed for CYP2D6-interfering drugs. Plasma samples collected in a fasted state were analyzed using liquid chromatography–tandem mass spectrometry (LC–MS). Genotyping was performed using the Infinium Global Diversity Array with Enhanced PGx. CYP2D6 activity scores and metabolic ratios (MRs) of solanidine and metoprolol were calculated.

Results: N = 47 patients with detectable metoprolol plasma concentrations were included in the primary analysis. The MRs ln MR (OH-solanidine/solanidine) and ln MR (3,4-seco solanidine-3,4-dioic acid/solanidine) correlated very strongly (r = 0.907 and r = 0.945) with the ln MR (OH-metoprolol/metoprolol). There was a strong negative correlation of both solanidine-derived MRs with metoprolol trough levels (r = −0.625 and r = −0.703). The genotyping-derived CYP2D6 activity score showed weaker correlations in all comparisons, even when adjusting for covariates.

Discussion: Solanidine-derived phenotyping showed stronger correlations to the measured CYP2D6 activity and metoprolol trough levels than the genotyping-derived activity scores. More research is needed in order to implement solanidine-derived dosing recommendations into clinical practice.

CYP2C19 and CYP2D6 are involved in the hepatic metabolism of approximately 35–40% of clinically used drugs. We conducted an in vivo phenotyping study encompassing 114 Estonian Biobank participants to evaluate the functional impact of rare or novel single-nucleotide and structural variants in the CYP2C19 and CYP2D6 genes using omeprazole and metoprolol as respective probe drugs. Plasma concentrations of these drugs and their metabolites were measured at 10 time points, and parent drug-to-metabolite ratios were calculated to determine enzymatic activity. Long-read sequencing enabled high-resolution star allele calling. Our results provide the first in vivo confirmation that partial gene and intragenic deletions in CYP2C19 (CYP2C1937 and CYP2C1942), enriched in Estonians and Finns, are associated with poor metaboliser phenotypes (P < 1.2 × 10−7). Additionally, we offer in vivo evidence of reduced metabolic activity of the CYP2D6*124 allele and a novel missense variant (c.940C>A) in exon 6 of CYP2D6. Furthermore, we observed that inhibitor exposure was significantly associated with higher metabolic ratios for both CYP2C19 (P = 3.0 × 10−6) and CYP2D6 (P = 0.02). Our findings emphasise the importance of identifying genetic variants in CYP2C19 and CYP2D6 beyond commonly assessed star alleles and that profiling for drug interactions can provide more precise assignments of metabolic phenotypes and improve personalised treatment.

Background: Drug adherence is crucial for chronic disease management, yet treatment discontinuation remains common due to factors such as side effects, inefficacy, or cost. These reasons are often recorded only in free-text clinical notes, making large-scale analysis difficult. While large language models (LLMs) can interpret such unstructured data more effectively than traditional natural language processing methods, few studies have systematically categorized reasons for discontinuation or identified whether the decision was initiated by the patient or the clinician, especially in low-resource languages such as Estonian.

Objective: This study aimed to assess the ability of LLMs to extract and classify reasons for drug discontinuation and identify who initiated it using Estonian electronic health records and characterize the observed discontinuation patterns and initiators for statins and antidiabetic medications.

Methods: We combined prescription data with free-text anamneses from a 10% sample of the Estonian population (2012-2019). LLMs (Llama 3.1-70B and GPT-4o) were applied to extract discontinuation phrases and reasons, classify them into a clinician-developed taxonomy, and identify who discontinued the treatment. Performance was evaluated on 100 randomly chosen cases per drug group.

Results: Extraction yielded 625 antidiabetic drug and 233 statin discontinuation cases. Validation confirmed a precision of 0.93 to 0.98 for extracting phrases and 0.95 to 0.96 for extracting reasons. Classification of discontinuation reasons achieved weighted F1-scores of 0.81 to 0.84, whereas classification of who initiated discontinuation achieved weighted F1-scores of 0.64 to 0.78. Adverse reactions were the most frequent reason overall, accounting for 70% (163/233) of statin discontinuations and 44.8% (280/625) of antidiabetic drug discontinuations. Regarding antidiabetic drugs, treatment inefficacy and contraindications were more common. Patients more often stopped due to adverse reactions or nonmedical reasons, whereas physicians more often initiated discontinuation for contraindications.

Conclusions: LLMs can accurately extract and classify medication discontinuation reasons and show variable performance in identifying discontinuation initiators in Estonian clinical narratives. Both local and proprietary models showed promising results, enabling scalable analyses that complement structured health records. This demonstrates the potential of LLMs to unlock information from clinical notes, turning this underused electronic health record component into a valuable resource for monitoring treatment patterns and detecting adverse event signals.

Pharmacogenetics (PGx) aims to optimize drug treatment outcomes by using a patient's genetic profile for individualized drug and dose selection. Currently, reactive and pretherapeutic single-gene PGx tests are increasingly applied in clinical practice in several countries and institutions. With over 95% of the population carrying at least one actionable PGx variant, and with drugs impacted by these genetic variants being in common use, pretherapeutic or preemptive PGx panel testing appears to be an attractive option for better-informed drug prescribing. Here, we discuss the current state of PGx panel testing and explore the potential for clinical implementation. We conclude that available evidence supports the implementation of pretherapeutic PGx panel testing for drugs covered in the PGx guidelines, yet identification of specific patient populations that benefit most and cost-effectiveness data are necessary to support large-scale implementation.

Antidepressant side effects are prevalent, leading to significant treatment discontinuity among patients. A deeper understanding of the underlying mechanisms could help identify individuals at risk of side effects and improve treatment outcomes.We aim to investigate the role of genetic variation in CYP2C19 and polygenic scores (PGS) for psychiatric and side effect-related phenotypes in experiencing antidepressant side effects.We pooled Estonian Biobank data from the Mental Health online Survey (N = 86,244), the Adverse Drug Events Questionnaire (N = 49,366) and from unstructured electronic health records using natural language processing (N = 206,066) covering 25 common side effects. The results were meta-analysed with previously published results from the Australian Genetics of Depression Study. Among 13,729 antidepressant users, 52.0% reported side effects. In a subgroup of 9,563 individuals taking antidepressants metabolised by CYP2C19, poor metabolisers had 49% higher odds of reporting a side effect (OR = 1.49, 95%CI = 1.09–2.04), while ultrarapid metabolisers had 17% lower odds (OR = 0.83, 95%CI = 0.70–0.99) compared to normal metabolisers. PGSs for schizophrenia and depression showed the most associations with overall and specific side effects. PGSs for higher body mass index (BMI), anxiety, and systolic blood pressure were associated with respective side effects among any antidepressant and selective serotonin reuptake inhibitor (SSRI) users. Meta-analysis confirmed robust evidence linking a higher BMI PGS and weight gain across nine antidepressants and moderate evidence linking PGS for headache with headache from sertraline. Our findings underscore the role of genetic factors in experiencing antidepressant side effects and have potential implications for personalised medicine approaches to improve antidepressant treatment outcomes.

Conducting clinical studies on drug–drug-gene interactions (DDGIs) and extrapolating the findings into clinical dose recommendations is challenging due to the high complexity of these interactions. Here, physiologically-based pharmacokinetic (PBPK) modeling networks present a new avenue for exploring such complex scenarios, potentially informing clinical guidelines and handling patient-specific DDGIs at the bedside. Moreover, they provide an established framework for drug–drug interaction (DDI) submissions to regulatory agencies. The cytochrome P450 (CYP) 2D6 enzyme is particularly prone to DDGIs due to the high prevalence of genetic variation and common use of CYP2D6 inhibiting drugs. In this study, we present a comprehensive PBPK network covering CYP2D6 drug–gene interactions (DGIs), DDIs, and DDGIs. The network covers sensitive and moderate sensitive substrates, and strong and weak inhibitors of CYP2D6 according to the United States Food and Drug Administration (FDA) guidance. For the analyzed CYP2D6 substrates and inhibitors, DD(G)Is mediated by CYP3A4 and P-glycoprotein were included. Overall, the network comprises 23 compounds and was developed based on 30 DGI, 45 DDI, and seven DDGI studies, covering 32 unique drug combinations. Good predictive performance was demonstrated for all interaction types, as reflected in mean geometric mean fold errors of 1.40, 1.38, and 1.56 for the DD(G)I area under the curve ratios as well as 1.29, 1.43, and 1.60 for DD(G)I maximum plasma concentration ratios. Finally, the presented network was utilized to calculate dose adaptations for CYP2D6 substrates atomoxetine (sensitive) and metoprolol (moderate sensitive) for clinically untested DDGI scenarios, showcasing a potential clinical application of DDGI model networks in the field of model-informed precision dosing.

This study develops methods for detecting Adverse Drug Events (ADEs) from Electronic Health Records (EHR) of participants of the Estonian Biobank to support pharmacogenetics research. It focuses on creating manually annotated datasets and improving ADE detection efficiency by combining rule-based and machine learning (ML) approaches, specifically applied to antidepressants and antipsychotics. To detect potential ADE mentions within free text fields of EHRs, we employed a lexicon-based approach to extract text snippets containing both a drug name and a symptom. We developed a rule-based and an ML-based system to prefilter the extracted text snippets, aiming to reduce the number of non-ADE snippets going into manual annotation. We then applied both systems before manual annotation and assessed their impact on the annotation process. We produced annotated datasets for antidepressants (520 patient–drug pairs) and antipsychotics (1,329 pairs). Our prefiltering method reduced the annotation workload up to 24-fold compared to no filtering, and ML-based filtering outperformed rule-based filtering, requiring only 1.3–1.5 snippets per positive ADE case. Pharmacogenetic validation revealed significant genotype — ADE associations for Escitalopram, Sertraline, and Quetiapine. The implementation of prefiltering methods significantly enhanced the efficiency of manual annotation for dataset creation and pharmacogenetic validation confirmed the datasets’ utility and usability. Therefore, we showed that ADE extraction from free text adds value by expanding the scope and diversity of analyzable cases for discoveries in the genetics of drug response.

Dasatinib is a medication used to treat certain types of leukemia. Its effectiveness can be influenced by other drugs that patients might be taking. Some medications can affect the enzyme (CYP3A4) that breaks down dasatinib in the liver, while others, like antacids, can change stomach acidity and affect how dasatinib is absorbed.
Researchers created a so called physiologically based pharmacokinetic (PBPK) model to predict how dasatinib behaves in the body and interacts with other drugs. This model simulates the human body and helps understand how dasatinib is distributed, broken down, and eliminated over time.
The PBPK model was developed using dasatinib’s properties and clinical study data. It accurately described how dasatinib levels change in the blood. The model was then used to figure out how to adjust dasatinib doses when taken with other medications, for example:
• How the dasatinib dose should be reduced with enzyme blockers (CYP3A4 inhibitors), or
• How the dasatinib dose should be increased with enzyme boosters (CYP3A4 inducers).
This model helps physicians to personalize dasatinib treatment for patients, ensuring they receive the correct dose when taking other medications, thereby improving effectiveness and reducing side effects.

Imatinib is a medication used to treat chronic myeloid leukemia, a type of blood cancer. It is broken down in the body by enzymes called CYP2C8 and CYP3A4. Additionally, a protein called P-glycoprotein pumps imatinib out of cells, affecting its activity. Imatinib can interact with other medications, influencing the enzymes and the protein, which can impact the effectiveness and safety of both imatinib and the other drugs.
Researchers created a computer model, known as a physiologically-based pharmacokinetic (PBPK) model, to predict how imatinib and its byproduct behave in the body and interact with other drugs. This model simulates the human body and helps understand how imatinib is distributed, broken down, and eliminated over time.
The PBPK model was developed using imatinib’s properties, its breakdown by CYP3A4 and CYP2C8, and its interaction with P-glycoprotein. Clinical study data were used to validate the model, which accurately described imatinib levels in the blood over time. The model was then used to explore drug interactions:
• With drugs affecting enzyme activity, like rifampicin, ketoconazole, and gemfibrozil: These simulations showed how these drugs could change imatinib and its byproduct levels.
• With drugs affected by imatinib, like simvastatin and metoprolol: The simulations showed how imatinib could alter the levels of these drugs.
The model successfully predicted these interactions, matching observed data closely. It can help in developing new drugs and optimizing dosages for patients, ensuring safe and effective treatments when multiple medications are used together.

The analysis of substrates of polymorphic cytochrome P450 (CYP) enzymes is important information to enable drug–drug interactions (DDIs) analysis and the relevance of pharmacogenetics in this context in large datasets. Our aim was to compare different approaches to assess the substrate properties of drugs for certain polymorphic CYP2 enzymes. Methods: A standardized manual method and an automatic method were developed and compared to assess the substrate properties for the metabolism of drugs by CYP2D6, 2C9, and 2C19. The automatic method used a matching approach to three freely available resources. We applied the manual and automatic methods to a large real-world dataset deriving from a prospective multicenter study collecting adverse drug reactions in emergency departments in Germany (ADRED). Results: In total, 23,878 medication entries relating to 895 different drugs were analyzed in the real-world dataset. The manual method was able to assess 12.2% (n = 109) of drugs, and the automatic method between 12.1% (n = 109) and 88.9% (n = 796), depending on the resource used. The CYP substrate classifications demonstrated moderate to almost perfect agreements for CYP2D6 and CYP2C19 (Cohen’s Kappa (κ) 0.48–0.90) and fair to moderate agreements for CYP2C9 (κ 0.20–0.48). Conclusion: A closer look at different classifications between methods revealed that both methods are prone to error in different ways. While the automated method excels in time efficiency, completeness, and actuality, the manual method might be better able to identify CYP2 substrates with clinical relevance.

Cardiometabolic medications are prescribed to prevent and treat cardiometabolic diseases such as diabetes, high blood pressure, and high cholesterol. This study investigated whether genes could affect the way in which patients use cardiometabolic medications. The authors combined and analyzed cardiometabolic medication use history and genetic information of 567,671 patients from three large biobank studies: FinnGen, the Estonian Biobank and the UK Biobank. In total, 333 locations across several genes were found to be associated with total medication use, medication switching or treatment discontinuation. This study demonstrates how medication use history can help better understand the underlying biology of cardiometabolic diseases.