Pharmacogenomics in Clinical Practice: Evidence and Steps

The same drug at the same dose can work well for one patient, fail in a second and harm a third. Part of that variation is inherited. Genes that encode drug-metabolising enzymes such as CYP2D6, CYP2C19 and CYP2C9 decide how quickly many common medicines are activated or cleared. Pharmacogenomics (PGx) uses that information to choose the drug and dose before a problem occurs. For years the field had strong evidence for individual gene-drug pairs but little proof that panel testing works across a health system. A 2023 trial in The Lancet changed that.

The PREPARE trial: what it tested

PREPARE (Preemptive Pharmacogenomic Testing for Preventing Adverse Drug Reactions) was run by the European Ubiquitous Pharmacogenomics (U-PGx) consortium and funded by the EU's Horizon 2020 programme. It was an open-label, cluster-randomised crossover implementation study in 18 hospitals, nine community health centres and 28 community pharmacies in seven European countries: Austria, Greece, Italy, the Netherlands, Slovenia, Spain and the UK.

Adults starting a drug covered by guidelines of the Dutch Pharmacogenetics Working Group (DPWG) were eligible. In the study arm, patients were genotyped for 50 variants in 12 genes. If the result was actionable, meaning the DPWG recommended a change from standard treatment, clinicians were advised to follow the guideline. The control arm received standard care. Sites switched between arms partway through the study. Local teams were trained, and educational material was made available online.

What it found

Between March 2017 and June 2020, 6,944 patients were enrolled. The primary outcome was a clinically relevant adverse drug reaction within 12 weeks.

• Among the 1,558 patients with an actionable result for their index drug, adverse reactions occurred in 21.0% with genotype-guided treatment versus 27.7% with standard care (odds ratio 0.70).
• Across all patients, the rates were 21.5% versus 28.6% (odds ratio 0.70).

The authors concluded that genotype-guided treatment with a 12-gene panel significantly reduced clinically relevant adverse drug reactions and was feasible across diverse European health-care organisations. The U-PGx consortium summarises this as about 30% fewer serious side effects.

Reading the result carefully

PREPARE is the strongest evidence so far for pre-emptive panel testing, but it has limits that matter when the findings are applied in another country:

• Open-label design. Patients and clinicians knew which arm they were in, which can influence how adverse reactions are reported.
• Short follow-up. Outcomes were measured over 12 weeks. The trial was not designed to show long-term effects on hospital admissions or mortality.
• Loss to follow-up was higher in the study group (11.0%) than in the control group (7.9%).
• Population. 97.7% of participants reported European, Mediterranean or Middle Eastern ethnicity. Allele frequencies differ between populations, so local frequency data are needed.
• Guideline set. Recommendations followed the DPWG. Many health systems also use guidelines from the Clinical Pharmacogenetics Implementation Consortium (CPIC), which overlap substantially but are not identical.

How common are actionable genotypes?

Actionable variants are not rare exceptions. In a 2019 study in Genetics in Medicine, researchers at the Estonian Genome Center translated existing genotype data for more than 44,000 biobank participants into pharmacogenetic recommendations. 99.8% of individuals carried a genotype associated with increased risk for at least one medication. The same study found that genotyping arrays performed similarly to whole-genome sequencing for pharmacogenetic predictions, while exome sequencing was not suitable. That makes array-based pre-emptive testing a cost-effective option.

The practical consequence is clear. If almost everyone carries at least one actionable variant, testing one patient at a time after a problem has occurred misses most of the benefit. Pre-emptive testing stores the result once and makes it available for every future prescription.

How Estonia returns PGx results

Estonia has gone further than most countries in returning pharmacogenomic information at population scale. The Estonian Biobank holds data on about 212,000 participants, around 20% of the adult population. It has returned results in stages: first through targeted recall studies with genetic counselling, and since June 2024 through the MyGenome portal (MinuGeenivaramu).

Participants log in with their ID card, Mobile-ID or Smart-ID and, under dynamic consent, choose which reports to see. According to a 2025 presentation by the head of the Estonian Biobank, the pharmacogenetics report covers 37 active ingredients, corresponding to about 230 medicinal products. For each one, the report states the gene result, for example CYP2C19 intermediate metaboliser, and summarises the published recommendation, such as a slower titration or a lower maintenance dose for a particular antidepressant. The biobank presents its genetic risk reports as an in-house medical device under Article 5(5) of the EU In Vitro Diagnostic Regulation. The portal also includes questionnaires to evaluate how participants understand and respond to their results.

Estonia is open about what remains. The same presentation asks how to "scale into the clinic", meaning how to move from a portal for biobank participants to decision support at the point of prescribing for the whole population. A Centre of Personalised Medicine, jointly funded by the EU and Estonia, has been set up for that purpose.

Implementation: seven practical steps

1. Choose a guideline base. Adopt DPWG and/or CPIC recommendations and define a national list of gene-drug pairs to start with, focusing on commonly prescribed drugs with strong evidence.
2. Choose the test. A validated array-based or targeted panel covering the main pharmacogenes is enough for pre-emptive use, in line with the Estonian findings. Laboratory accreditation and quality control come first.
3. Store results as structured data. Diplotypes and phenotypes, such as "CYP2C19 poor metaboliser", should be stored as coded entries in the national health record, not as PDF reports, so that software can act on them.
4. Build decision support into prescribing. When a doctor prescribes, or a pharmacist dispenses, a drug affected by a stored genotype, the e-prescription system should show the recommendation. This is where PREPARE's benefit was delivered.
5. Train clinicians and pharmacists. PREPARE included site training and online materials. Short, drug-specific guidance works better than general genetics courses.
6. Decide on returning results to patients. Estonia's portal shows that patients can be given understandable PGx reports with consent options. Results should still reach the prescriber through the record, not only the patient.
7. Measure outcomes locally. Track alert acceptance, prescribing changes and adverse drug reaction reports, and use national dispensing data to check whether the expected effect appears.

What this means for Moldova

Pharmacogenomics links two of the five AI domains in Moldova's National Healthcare Digitalization Program (2025–2030): genetics, which includes screening about 24,000 genomes (around 1% of the population) and pharmacogenomics, and pharmacology and treatment plans. The genomic data generated by screening can provide pre-emptive PGx results for every person tested. A national allele frequency profile from the same data would show which gene-drug pairs matter most for Moldovan patients.

The steps above depend on the wider digitalisation. A coded genotype only helps if it reaches the prescriber, so planned growth in EHR coverage from under 10% to 80% by 2030 and a national e-prescription service are prerequisites, not extras. Genomic data are among the most sensitive data a state holds and must remain in Moldova, so genotype calling, interpretation and decision-support services should run on dedicated computing capacity hosted in the country.

Related reading and solutions

• Genomics: population screening and pharmacogenomics
• Pharmacology and treatment plans
• National genomics programmes in Europe: lessons for Moldova
• How Estonia built a national e-health system
• Sovereign, on-premise AI infrastructure

Planning a pharmacogenomics pilot or integrating PGx into prescribing? Contact Import Medical to discuss testing workflows, data integration and in-country decision support.

pharmacogenomics PREPARE trial adverse drug reactions Estonian Biobank clinical decision support

Sources

1. A 12-gene pharmacogenetic panel to prevent adverse drug reactions: an open-label, multicentre, controlled, cluster-randomised crossover implementation study (PREPARE) — The Lancet
2. Ubiquitous Pharmacogenomics (U-PGx) — U-PGx consortium
3. Translating genotype data of 44,000 biobank participants into clinical pharmacogenetic recommendations: challenges and solutions — Genetics in Medicine
4. Estonian Biobank launched a portal for more than 210,000 biobank participants — Institute of Genomics, University of Tartu
5. Data, trust, and the future of prosperity: lessons from genetics (L. Milani, e-Gov Conference 2025) — e-Governance Conference / University of Tartu
6. CPIC Guidelines — Clinical Pharmacogenetics Implementation Consortium