Pharmacogenetics in oncology: DPYD, TPMT, UGT1A1 and safe dosing

The same cancer drug, at the same dose, behaves very differently from one patient to the next. Part of the difference is genetic: inherited variants in the enzymes that break chemotherapy down. Pharmacogenetics identifies them before the first dose. This article explains what that means for a hospital: which gene-drug pairs have solid evidence, what guidelines and regulators say, what trials show about toxicity and cost, where the software stops, and what the law requires for genetic data.

What pharmacogenetic testing is and why it matters in oncology

Pharmacogenetics studies how a person's genetic variants influence the response to medicines. In oncology, the reference case is dihydropyrimidine dehydrogenase (DPD), the enzyme that breaks down 5-fluorouracil (5-FU) and its prodrugs capecitabine and tegafur. The gene that encodes it is called DPYD. When DPD activity is reduced, the drug accumulates and the risk of severe toxicity rises.

According to the July 2020 direct healthcare professional communication issued in Romania by the National Agency for Medicines and Medical Devices (ANMDMR) in agreement with the European Medicines Agency (EMA), complete DPD deficiency is rare, 0.01-0.5% of the Caucasian population, while partial deficiency affects 3-9%.

The test can be done in two ways, both accepted by EMA: genotyping (looking for DPYD variants in a blood sample) or phenotyping (measuring uracil in blood). This article deals with genotyping, where interpretation software comes in.

Three gene-drug pairs with solid evidence

Gene	Drugs	Guideline	What the guideline recommends when enzyme function is reduced
DPYD	5-FU, capecitabine, tegafur	CPIC 2017 update, EMA 2020	activity score 1: dose reduced by 50%; score 1.5: by 25-50%; score 0-0.5: avoid fluoropyrimidines
TPMT, NUDT15	azathioprine, mercaptopurine, thioguanine	CPIC 2018 update	intermediate metabolizer: 30-80% of the dose; TPMT poor metabolizer, in malignancy: dose reduced 10-fold, given three times weekly
UGT1A1	irinotecan	DPWG 2022/2023	poor metabolizer: 70% of the dose, then escalation guided by neutrophil count; intermediate: no action

CPIC is the Clinical Pharmacogenetics Implementation Consortium, a US consortium (Amstutz et al. 2018; Relling et al. 2019). DPWG is the Dutch Pharmacogenetics Working Group, whose recommendations were also used in the European PREPARE study (Hulshof et al. for UGT1A1).

DPYD and fluoropyrimidines

The CPIC guideline works with a DPYD activity score from 0 to 2. Score 2 means a normal metabolizer and standard dose. Score 1 or 1.5 means an intermediate metabolizer: a starting dose reduced by 50% (strong recommendation) or by 25-50% (moderate recommendation). Score 0 or 0.5 means a poor metabolizer: avoid 5-FU-based regimens, and if there is no alternative, a strongly reduced dose with early therapeutic drug monitoring. According to the same guideline, about 7% of Europeans carry at least one decreased-function DPYD variant.

TPMT, NUDT15 and thiopurines

Thiopurines are used in acute leukaemias and in autoimmune disease. According to the 2018 CPIC guideline, TPMT poor metabolizers are about 0.3% of Europeans and intermediate metabolizers about 10%; the NUDT15 poor metabolizer phenotype occurs in about 1 in 50 people of East Asian descent. Intermediate metabolizers of either gene start at 30-80% of the normal dose. For TPMT poor metabolizers treated for malignancy, the guideline recommends a daily dose reduced 10-fold, given three times a week.

UGT1A1 and irinotecan

Irinotecan is inactivated by the enzyme UGT1A1. The DPWG guideline recommends for poor metabolizers (for example the 28/28 genotype) a starting dose of 70%, with later escalation guided by neutrophil count. Intermediate metabolizers need no action. The *28 allele has a prevalence of 22-39% in European populations. DPWG gave the interaction a score of 8 out of 10 and rated genotyping "essential", meaning it should be done before therapy starts.

What regulators say: EMA 2020 and the Romanian communication

On 30 April 2020, EMA recommended that patients be tested for DPD deficiency before starting treatment with injectable fluorouracil, capecitabine or tegafur, following a review started by the Pharmacovigilance Risk Assessment Committee (PRAC) and adopted by the Committee for Medicinal Products for Human Use (CHMP). According to EMA, up to 9% of the Caucasian population has low levels of the enzyme and 0.5% lack it entirely. With complete deficiency, treatment is contraindicated; with partial deficiency, a reduced starting dose should be considered.

In Romania, the recommendation reached clinicians through the July 2020 direct communication agreed with EMA and ANMDMR. The text names the four DPYD variants associated with increased risk (c.1905+1G>A, also called DPYD*2A, c.1679T>G, c.2846A>T and c.1236G>A/HapB3) and the uracil thresholds for phenotyping: 16-150 ng/ml indicates partial deficiency, above 150 ng/ml complete deficiency. The communication also notes the method's limit: severe toxicity can still occur in patients with a negative test.

The test does not replace clinical monitoring. It moves part of the risk ahead of the first dose, where it can be handled by a dosing decision.

The evidence: less toxicity, costs at worst neutral

The key prospective study was published by Henricks et al. in Lancet Oncology (2018). In 17 Dutch hospitals, between 2015 and 2017, 1,103 evaluable patients were genotyped for the four DPYD variants before treatment. Heterozygous carriers (85 patients, 7.7%) received an initial dose reduced by 25% (c.2846A>T, c.1236G>A) or by 50% (DPYD*2A, c.1679T>G).

The main result: severe toxicity (grade 3 or higher) in 39% of carriers treated with a reduced dose, versus 23% of patients without variants treated with the standard dose. The comparison of relative risks (carriers versus non-carriers) shows the effect per variant against historical cohorts treated with the full dose.

For DPYD2A, the relative risk fell from 2.87 to 1.31. For c.2846A>T, from 3.11 to 2.00. For c.1236G>A, the 25% reduction was insufficient: 1.72 versus 1.69. The authors concluded that for DPYD2A and c.1679T>G the 50% reduction appears adequate, while c.1236G>A and c.2846A>T need a larger reduction.

Costs were analysed separately, in the European Journal of Cancer (Henricks et al., 2019), on data from the same study. Expected total costs were EUR 2,599 per patient with screening versus EUR 2,650 without, a net saving of EUR 51 per patient. The model used a test cost of EUR 100, a ward day of EUR 636 and an intensive care day of EUR 2,015. Sensitivity analyses showed the strategy is very likely cost saving or, at worst, cost neutral.

Evidence at European scale comes from the PREPARE study (Swen et al., The Lancet, 2023): a controlled, cluster-randomised implementation study in 18 hospitals, 9 community health centres and 28 pharmacies in seven countries. It enrolled 6,944 patients, genotyped for 50 variants in 12 genes; those with an actionable result were treated according to DPWG recommendations. Among patients with an actionable result for the index drug, a clinically relevant adverse drug reaction occurred in 21.0% of the genotype-guided group versus 27.7% of controls (odds ratio 0.70); across all patients, 21.5% versus 28.6%. The study covered many drug classes, not only oncology, and shows the model works across different health systems.

How the result reaches the clinician: genotype, phenotype, recommendation

A raw laboratory result (for example "heterozygous c.1236G>A") is not a clinical decision. The standardised chain in the guidelines has three steps:

1. Genotype. The laboratory reports the variants found, in standard nomenclature.
2. Phenotype. The variants are translated into a category: normal, intermediate or poor metabolizer.
3. Dosing recommendation. For each phenotype, the guideline gives a recommendation with a level of evidence: standard dose, percentage reduction, avoidance.

Interpretation software performs steps 2 and 3 in a repeatable, documented way: the same variants, the same phenotype, the same guideline reference, with version and date, in a structured report sent to the oncologist. The decision stays with the physician, who knows the regimen, the comorbidities and the intent of treatment. The ANMDMR communication is explicit: doses may be increased later in the absence of serious toxicity, because the efficacy of a reduced dose has not been established.

Where the software stops: MDR, IVDR and the clinician's decision

Regulation (EU) 2017/745 (MDR), Annex VIII, Rule 11, as quoted in guidance MDCG 2019-11 Rev.1 (June 2025): "Software intended to provide information which is used to take decisions with diagnosis or therapeutic purposes is classified as class IIa", except where such decisions may cause death or an irreversible deterioration of a person's state of health (class III) or a serious deterioration or a surgical intervention (class IIb). The guidance adds that the higher class applies where a decision based on incorrect information from the software is reasonably likely to have such an impact. The MDCG is the Medical Device Coordination Group, made up of Member State representatives; its documents are not legally binding, as the text itself states.

Regulation (EU) 2017/746 (IVDR) applies to in vitro diagnostic devices, including software qualified as an IVD, which the MDCG guidance refers to the IVDR rules. Implementing rule 1.4 of Annex VIII says that software which drives a device or influences its use falls in the same class as the device, while independent software is classified in its own right. Rule 3(f) places in class C devices intended to be used as companion diagnostics, defined in Article 2(7) as devices essential for the safe and effective use of a medicinal product, including identifying patients at increased risk of serious adverse reactions. Pharmacogenetic tests for chemotherapy dosing fit this description.

In practice, the classification (MDR or IVDR, and the class) depends on the intended purpose declared by the manufacturer and on the conformity assessment. A hospital can ask for the document that establishes them before the software enters clinical use.

Genetic data: a special category under GDPR

The General Data Protection Regulation (GDPR) names genetic data explicitly in Article 9(1), alongside health data, among the special categories whose processing is prohibited in principle. The exceptions in paragraph 2 relevant to a hospital: point (h) (processing necessary for diagnosis, health care or the management of health services, on the basis of law or a contract with a health professional), point (a) (explicit consent for specified purposes) and point (j) (scientific research, with safeguards). Paragraph 3 requires that data processed under point (h) be handled by or under the responsibility of a professional bound by professional secrecy. Paragraph 4 lets Member States add conditions for genetic, biometric and health data; national law must be checked separately in Romania and in Moldova.

For a pharmacogenetic testing workflow, this means at least: a clear legal basis, limited access to raw results, a contract with the laboratory and the software vendor fixing controller and processor roles, and a retention period set before the first test.

What this means for a hospital in Romania or Moldova

Regulatory situation. In Romania, the EMA recommendation was passed to professionals through the July 2020 direct communication, agreed with ANMDMR. In the public sources we consulted, we did not identify a decision by the National Health Insurance House to reimburse DPYD testing. In Moldova we found no published national recommendation on DPD testing before fluoropyrimidines. In both cases the accurate statement is that reimbursement is not publicly documented, not that it is absent; checking with the insurer and the medicines agency comes first.

Concrete steps, in the order they arise:

1. Volume. How many patients start 5-FU, capecitabine, irinotecan or thiopurines each year. With 7.7% DPYD carriers (Henricks et al., 2018), 500 patients a year means roughly 38 whose starting dose would change; the figure is a worked example, not an estimate for any specific hospital.
2. Laboratory. An accredited laboratory, in-house or partner, reporting at least the four DPYD variants named by EMA. The decisive question is turnaround: the result must exist before the first dose. As a reference point, the DPYD service of Manchester University NHS Foundation Trust states 5 calendar days, on a 5 ml EDTA blood sample.
3. Workflow. Who draws the sample, who orders, where the report lands, how it links to prescribing. A result arriving after treatment has started did not serve the first cycle.
4. Consent and information. A form explaining the purpose of the test, its limits (toxicity can occur even with a negative test), data retention and the legal basis, aligned with Article 9 GDPR and national law.
5. Interpretation. A structured genotype-phenotype-recommendation report, with a reference to the guideline and version used.
6. Measurement. Number of tests, time to result, doses adjusted, grade 3 or higher toxicity before and after.

Questions for a software or laboratory vendor:

• Which variants do you report, which guideline (CPIC, DPWG) do you use to translate genotype into phenotype, and how do you track updates?
• What is the software's intended purpose and class under MDR or IVDR? Where are genetic data stored, for how long, and who has access?

Consdinamic builds software and artificial intelligence to order, with its deepest work in healthcare. Its DNA screening engine starts from pharmacogenetics, with oncology as the first field: it translates genotype into phenotype and into the guideline recommendation, with the source displayed, and leaves the dosing decision to the physician. The company's products work in Romanian, Russian and English, and its sister company Softmed Labs SRL provides the presence in the European Union.

Conclusion

Three gene-drug pairs in oncology have moved from research into guidelines and regulatory recommendations: DPYD with fluoropyrimidines, TPMT/NUDT15 with thiopurines, UGT1A1 with irinotecan. The Dutch prospective study showed that genotype-guided dosing lowers the relative risk of severe toxicity in DPYD*2A carriers from 2.87 to 1.31, at a cost that is at worst neutral. PREPARE confirmed the effect at European scale: 21.0% versus 27.7% clinically relevant adverse drug reactions.

For a hospital, the steps are known: an accredited laboratory, a turnaround compatible with the first dose, a legal basis for genetic data, a structured report and software correctly classified under MDR or IVDR. The software organises the evidence and delivers it on time. The dosing decision stays with the oncologist.