Antidepressant Genetic Testing: What It Can and Can't Tell You

Start with the claim, because a version of it is probably why you are here: that roughly a third of people have a gene that stops antidepressants working, and that a test before treatment would spare them two useless courses. The first half is a garbled description of something real. The second half is not supported by the evidence, and tests that promise it have drawn regulatory action, including a formal FDA warning to one laboratory in 2019.

Here is the accurate version. Two liver enzymes, CYP2D6 and CYP2C19, do most of the work of breaking down antidepressants, and the genes coding for them vary a lot between people. In Europe that variation is not a 30% minority: by the reference frequency tables of the Clinical Pharmacogenetics Implementation Consortium (CPIC), around 47% of people of European ancestry are something other than a "normal" CYP2D6 metaboliser and around 60% are something other than a normal CYP2C19 metaboliser. Treat those figures as approximate population estimates: they do not establish any individual's metaboliser status, or the proportion of people who need a treatment change. Genetic variation can influence exposure to particular medicines, but does not measure a person's actual blood concentration or guarantee a particular response. Other metabolic pathways and kidney clearance also matter. And there is no variant that makes someone resistant to almost all antidepressants.

How it actually works, in plain language

Every antidepressant has to be broken down, mostly by the liver. Many of them run through one or both of these two enzymes. Depending on which versions of the genes you inherited, you fall into one of four groups: poor metaboliser (the enzyme barely works), intermediate, normal, or ultrarapid (the enzyme works faster than usual). CYP2C19 adds a fifth, "rapid", between normal and ultrarapid.

The consequences are probabilistic rather than fixed. A poor metaboliser taking a standard dose tends to accumulate more of the drug than the trials assumed, so side effects become more likely at doses that look low on paper, and are easily misread as "this medicine doesn't agree with me". An ultrarapid metaboliser clears the drug faster, so a standard dose may never reach a working concentration and the medicine can look like a failure when it was arguably never properly delivered. That second scenario is the grain of truth inside the "resistance gene" story. It is drug-specific, not universal: your CYP2C19 status matters for escitalopram, citalopram and sertraline (CYP2B6 also contributes to sertraline, with weaker evidence behind the guidance), your CYP2D6 status for paroxetine and the tricyclics, and for some drugs (mirtazapine and CYP2C19, duloxetine and CYP2D6), the Dutch working group found no interaction worth acting on at all.

What the guidelines actually say is worth spelling out, because the advice differs by drug, by metaboliser status and by guideline:

• CYP2C19 and citalopram, escitalopram, sertraline. For poor metabolisers, CPIC's 2023 guideline advises considering a lower starting dose, slower titration and a maintenance dose around 50% lower than standard, or an alternative drug not primarily metabolised by CYP2C19. Rapid and ultrarapid status does not require an increased sertraline starting dose; for citalopram and escitalopram, rapid metabolisers start normally, with adjustment considered if response is inadequate, while ultrarapid metabolisers may need an alternative. There is also a hard safety limit in the product information: the UK Cipramil SmPC recommends an initial 10 mg/day for known CYP2C19 poor metabolisers, with a maximum of 20 mg/day, alongside the usual QT-interval precautions. Observe your local product information.
• CYP2D6 and the newer non-tricyclics: the guidelines diverge by drug. CPIC does not advise routine dose changes for CYP2D6 intermediate metabolisers taking venlafaxine, fluvoxamine or vortioxetine; paroxetine may require a lower starting dose and slower titration. For venlafaxine poor metabolisers, CPIC optionally suggests an alternative. The Dutch DPWG goes further and recommends avoiding venlafaxine in both poor and intermediate metabolisers (Beunk et al., DPWG guideline for non-SSRI/non-TCA antidepressants, 2024). Venlafaxine has an active metabolite, so an altered balance between parent drug and metabolite is not adequately explained by simple drug accumulation, and halving the dose is not an automatic fix.
• Tricyclics. CPIC's tricyclic guidance is the 2016 update, published in 2017 (Hicks et al.). For CYP2D6 poor metabolisers it prefers an alternative antidepressant; if a tricyclic is necessary, consider a 50% reduction of the starting dose with therapeutic drug monitoring. Intermediate metabolisers may need roughly a 25% reduction. These recommendations concern antidepressant dosing (lower doses used for pain are a different situation), CYP2C19 also matters for tertiary amines such as amitriptyline, and not every tricyclic follows the same gene-based rule.
• SLC6A4 and HTR2A, the serotonin transporter and receptor genes, are what most consumer-facing panels lean on when they promise to name the antidepressant that will work for you. Evidence does not currently support using SLC6A4 or HTR2A results to select an antidepressant: CPIC reviewed both in its 2023 guideline and concluded the data do not support their use in prescribing, and there are not even standard genotype-to-phenotype categories, so two labs can report the same DNA differently. Predicting response (pharmacodynamics) is a different and unsolved problem from the established drug-metabolism guidance.

One more framing point that matters: CPIC guidelines are written to tell clinicians how to use genotype results that are already available. They deliberately do not recommend whether or when to order testing, and they are not a mandate for universal screening.

What the outcome evidence really shows

There is a difference between "this gene changes drug levels" (well established) and "testing people improves how they get better" (contested). The honest summary of the second question is mixed, with positive and negative findings in the same trials:

GUIDED: primary endpoint missed, secondary remission better. The GUIDED trial (1,167 patients, US, commercial combinatorial test) missed its primary endpoint: symptom improvement at week 8 was 27.2% with test-guided care versus 24.4% with usual care (p=0.107). Its secondary outcomes did favour testing, and the difference in remission was 5.2 percentage points (15.3% versus 10.1%), a statistically significant secondary result and the number the marketing quotes.

PRIME Care: prescribing endpoint met, small remission advantage. The PRIME Care trial (1,944 patients, US Department of Veterans Affairs, published in JAMA in 2022) met its prescribing co-primary endpoint: clinicians given results wrote far fewer prescriptions with a predicted gene-drug interaction. It also found a small average remission advantage across the follow-up period (2.8 percentage points; 95% CI 0.6 to 5.1), without a significant difference at week 24 itself.

ADOPT PGx (2026): no 3-month benefit, higher remission at 6 months. The most recent large trial, ADOPT PGx (Blake et al., JAMA Network Open, 2026), randomised 1,460 participants, including adults and children, to guideline-based CYP2D6/CYP2C19-informed prescribing or usual care; the main analysis concerned the 692 participants with actionable phenotypes. Neither the primary 3-month symptom outcome nor adverse-effect burden improved significantly. At 6 months, PROMIS-defined remission was 48.3% versus 39.4% (a difference of 8.9 percentage points; 95% CI 1.2 to 16.6), a secondary outcome. That suggests possible later benefit, but it does not establish routine testing for everyone, and the trial, like its predecessors, could not blind patients or clinicians, which adds uncertainty to secondary outcomes.

Meta-analysis lands on "modest, uncertain benefits". A 2025 systematic review in BMJ Mental Health pooling 13 randomised trials found higher remission with test-guided treatment at eight weeks (risk ratio 1.37; 95% CI 1.19 to 1.57). At twelve weeks the estimate was imprecise (RR 1.56; 95% CI 0.93 to 2.61), which is not the same as proof that the benefit disappears (Zhang et al., 2025). Comparisons between small and large gene panels were indirect and cannot prove the two equivalent. Almost none of these trials could blind the treating clinician, which is exactly the kind of design weakness that inflates soft outcomes.

Professional bodies are cautious, but not monolithic. An APA research workgroup report found insufficient evidence to support currently available combinatorial tools for selecting depression treatment (Baum et al., Am J Psychiatry, 2024; the journal later issued a correction clarifying that this is a workgroup report, not an APA position statement, and need not represent the whole organisation's views). Other professional statements recognise a possible role for specific gene-drug guidance, particularly after intolerance or inadequate response (ISPG genetic testing statement). Targeted CYP2D6/CYP2C19 genotyping and proprietary combinatorial panels are not the same thing, and the strongest scepticism attaches to the latter. Germany's national depression guideline makes no explicit recommendation for routine genotyping; it allows genotyping to be considered when a measured drug level is outside the expected range and non-genetic causes have been assessed and seem unlikely (NVL Unipolare Depression, version 3.2, sections 4.4.3.1, 4.4.7.1 and 7.1.1).

PREPARE: a safety signal across many medicines, read carefully. The European PREPARE study (nearly 7,000 patients across seven countries, published in The Lancet in 2023) tested a 12-gene panel pre-emptively across 39 drugs. Among participants with an actionable result, clinically relevant adverse drug reactions occurred in 21.0% with genotype-guided care versus 27.7% with usual care: an absolute difference of 6.7 percentage points, or approximately 24% lower observed risk. The reported odds ratio of 0.70 means 30% lower odds, which is not the same as 30% fewer reactions. The study was open-label. PREPARE supports a safety benefit from a testing-and-prescribing programme across multiple medicines. It does not establish the size or certainty of a safety benefit specifically for antidepressant-treated depression.

So who is this test actually for?

Reading the evidence honestly, there is no validated rule that identifies who benefits most from testing. What follows is a list of situations in which testing is most worth an individual discussion with your prescriber:

• You have had unusual side effects at low doses, or intolerable side effects on more than one drug in the same family. This is the pattern poor metaboliser status makes more likely, and it is the situation in which even sceptical guidelines and professional statements see a possible role.
• A drug did nothing at all despite good adherence and an adequate dose and duration. Ultrarapid metabolism is one candidate explanation among several, and it is worth checking specifically when the drug in question was one that runs through the enzyme concerned.
• A tricyclic is being considered. Tricyclics have a narrow margin between a working dose and a toxic one, and CPIC's guidance here is more assertive than for the SSRIs: for CYP2D6 poor metabolisers it prefers an alternative, and if a tricyclic is needed, a 50% lower starting dose with blood-level monitoring.
• Your blood level of an antidepressant has come back outside the therapeutic range and non-genetic explanations have been assessed and seem unlikely. This is precisely the situation in which the German guideline says genotyping may be considered.

And the weakest case is the one the marketing leads with: a healthy-curiosity test before your first prescription, in the expectation that it will name the drug that works. That is the claim regulators have objected to, and the trials do not support it, even if ADOPT PGx's 6-month secondary signal keeps the research question open.

How to get tested, country by country

Germany: self-pay, and the guideline reaches for blood levels first when there is an indication. Pharmacogenetic testing is available essentially only as an Individuelle Gesundheitsleistung (IGeL, self-pay); statutory insurance covers a short list of unrelated gene-drug pairs (DPYD before fluorouracil, CYP2C9 before siponimod, UGT1A1 before irinotecan and a couple more) but not psychiatric panels, and laboratory billing information confirms that pharmacogenetic diagnostics are mostly billed directly to patients or referrers (Labor Limbach, billing information). To give the order of magnitude, and these are examples from one laboratory's dated price list, not a national price range: one German laboratory's January 2025 list charges €127.35 for CYP2C19 alone (the relevant test for citalopram or escitalopram), €375.36 for CYP2D6 (paroxetine, fluvoxamine, venlafaxine, vortioxetine), €442.38 for the tricyclic panel and €542.92 for the full SSRI/SNRI panel, plus the doctor's own fee under the GOÄ schedule (MVZ price list, PDF). Individual coverage and current fees need confirmation with the laboratory. Under the Gendiagnostikgesetz, testing to investigate drug effects counts as diagnostic genetic testing: arrange it through a doctor, with the required explanation and written consent, and the doctor should discuss the applicable genetic-counselling requirements (counselling should generally be offered after diagnostic testing; the stricter before-and-after counselling rules belong to predictive testing) (GenDG, sections 3 and 7 to 10). There is no legal do-it-yourself route.

What to say to your psychiatrist, in the order the German guideline itself suggests: "Können wir zuerst den Serumspiegel bestimmen (TDM)? Und falls der Spiegel außerhalb des therapeutischen Bereichs liegt und nicht-genetische Ursachen ausgeschlossen sind: wäre dann eine CYP2D6-/CYP2C19-Genotypisierung sinnvoll?" Therapeutic drug monitoring is covered by statutory insurance and catches non-genetic causes too (interactions, smoking, irregular intake). The guideline recommends TDM for particular clinical indications (such as non-response, tolerability problems, suspected non-adherence, or specific drugs), not routinely for every antidepressant user. Note the practicalities: routine TDM samples are usually taken once you are on a stable dose that has reached steady state, typically as a trough level just before the next dose, with timing, last dose and co-medications recorded according to the drug and laboratory protocol. Urgent assessment of suspected toxicity is a separate situation with its own rules.

United Kingdom: ask locally, and mostly private for now. Antidepressant pharmacogenetic testing is not universally commissioned as routine NHS care. Access can exist through selected specialist or research services, including the PROGRESS programme in England and the PSY-PGx study at South London and Maudsley. Ask locally before paying privately. The NHS has implemented pharmacogenomics narrowly so far (DPYD in oncology, MT-RNR1 in neonatal care, and CYP2C19 before clopidogrel after stroke, per NICE HTG724), but a clopidogrel recommendation says nothing about antidepressant-test availability across the UK nations. Research access is not the same as an established clinical service. If you do buy a private test, expect psychiatric panels to cost several hundred pounds depending on the provider and what is bundled, check that the laboratory is accredited (UKAS/ISO 15189) and that the report cites CPIC or DPWG recommendations rather than only a proprietary algorithm; the NHS Genomics Education CYP2D6 and CYP2C19 pages are the reference your prescriber is most likely to recognise.

A fair question to put to a GP or psychiatrist before spending anything: "If a CYP2D6/CYP2C19 result came back abnormal, would it change what you prescribe or how you dose it?" If the answer is no, the money is better spent elsewhere.

Netherlands: results are wired into the system, but testing is not universal. The Dutch Pharmacogenetics Working Group (DPWG) of the pharmacists' association KNMP writes dosing recommendations per gene-drug pair, and those recommendations are built into the national G-Standaard drug database, meaning that once a result is in your pharmacy record, the software flags a gene-drug interaction automatically the next time something is prescribed (KNMP). That is integration of results into prescribing systems, not universal testing: the Dutch GP guideline (NHG) does not recommend standard testing for everyone taking a CYP450-metabolised medicine (NHG-Standaard Depressie). Testing is ordered through your GP, a hospital or a pharmacy. On cost, Erasmus MC publishes figures of around €86 per determination, with CYP2D6 nearer €191, but labels these as inter-institutional tariffs based on older NZa rates, not quotes to patients; what a GP-ordered test costs you depends on your insurer's agreements (Erasmus MC). Where there is a medical question (side effects or lack of effect), medically indicated testing may be reimbursed, though the cost may come out of your eigen risico; ask for a current estimate before testing. If you are tested, ask for farmacogenetisch onderzoek (CYP2D6/CYP2C19) and make sure the result is recorded at your pharmacy, which is where its value actually accrues.

Elsewhere in Europe. Spain took part in the PREPARE study and testing is being built out through hospital services and regional programmes, but availability differs greatly between autonomous communities. In Poland we could not verify any publicly funded route, which leaves private laboratories; not having verified a funding route is not the same as proof that none exists. For other countries we have not established availability one way or the other: the practical route is hospital psychiatry or clinical pharmacology if you are already in specialist care, and country-specific enquiry otherwise.

The limits: and where the hype lives

A genotype is not a phenotype. Your genes describe the enzyme you were born with; what matters clinically is how fast it is working today. Strong inhibitors change functional enzyme activity: fluoxetine and paroxetine inhibit CYP2D6 strongly enough to make a genetically normal metaboliser function like a poor one (phenoconversion). Smoking and other medicines shift levels too, and missed doses change your drug exposure, not your genetically predicted metaboliser status. This is why blood-level measurement (TDM) and genotyping are complementary rather than competing: one measures what is in your blood now, the other explains one of the reasons why.

Panels are not interchangeable. Different tests interrogate different variants, so the same person can be assigned different predicted phenotypes by different labs, and the "combinatorial" algorithms that convert genotypes into colour-coded drug lists are proprietary, which is why CPIC explicitly places them outside what it can evaluate.

A result is not a verdict on a drug. "Poor metaboliser" does not mean a medicine won't work; it usually means it should be dosed lower. "Normal metaboliser" does not mean the next antidepressant will work; most non-response has nothing to do with these two enzymes. And no result on any panel justifies stopping or changing a medicine on your own: dose changes and switches are prescriber decisions, and abrupt stops cause withdrawal effects.

Be sceptical of anything sold on the promise of skipping the trial-and-error. Regulators have challenged unsupported claims that a test can reliably choose an effective medicine: the FDA's 2019 warning letter to one laboratory (Inova) concerned particular unreviewed tests and unsupported response claims, not a blanket prohibition of pharmacogenetic prescribing guidance, and FDA status does not determine legality in Europe. Validated gene-drug associations can nevertheless inform prescribing. The FDA maintains a table of pharmacogenetic associations it considers supported by evidence, a useful sanity check against any test report that claims far more; an association table does not validate every commercial panel or its algorithm.

If antidepressants haven't worked for you

Genetics is one of several explanations for a treatment that hasn't helped, and rarely the first one worth checking. If two or more antidepressants have failed at an adequate dose and duration in your current episode, that situation has a name and a defined set of next options: switching, augmentation, psychotherapy, rTMS, ECT, and the EU-approved esketamine route. Our guide to what to do when depression treatment isn't working walks through all of them, and our eligibility check will tell you which are realistically open to you where you live.

Frequently asked questions

Is it true that about 30% of people are genetically resistant to antidepressants?

No. The underlying fact is that variants affecting antidepressant metabolism are common: in Europe, more than 30% for either gene, as approximate population estimates. But metaboliser status changes the concentration of a specific drug in your blood, which affects dosing and the likelihood of side effects; it does not confer resistance to antidepressants as a class, and no such "resistance gene" is known.

Will the test tell me which antidepressant will work for me?

No, and that is the claim to be most careful about. What a well-conducted test tells your prescriber is which drugs may need a lower or higher dose, and which are best avoided given your enzyme status. Evidence does not currently support using SLC6A4 or HTR2A results (the serotonin transporter and receptor genes on many commercial panels) to select an antidepressant: predicting response is a different problem from the established drug-metabolism guidance, and it is not solved.

Is it covered by insurance?

In Germany, generally not: it is a self-pay IGeL service, while therapeutic drug monitoring is covered when clinically indicated. In the UK it is not universally commissioned as routine NHS care, so it is usually private, though selected specialist and research services exist; ask locally. In the Netherlands insurers may reimburse medically indicated testing (for example after side effects or lack of effect), subject to your eigen risico; get a current estimate first.

Is a consumer DNA kit good enough?

Usually not for this purpose. CYP2D6 in particular involves gene deletions and duplications that many broad consumer arrays do not capture, third-party "interpretation" reports vary in quality and regulatory standing depending on the product and jurisdiction, and in Germany genetic testing legally requires a doctor's order and written consent under the GenDG. If you test, use a clinical laboratory whose report cites CPIC or DPWG.

Do I need to stop my medication before testing?

No. Your genotype does not change, so the sample (usually a cheek swab or blood) can be taken at any time. That is different from a blood-level (TDM) measurement, which has to be taken while you are on the drug, usually at steady state on a stable dose, just before the next dose, with the timing and your other medicines recorded. Never stop or alter a medicine for a test without your prescriber.

How often should I repeat it?

Usually, the same genetic test does not need repeating because your DNA is stable. However, interpretation and prescribing guidance may change, and an older or limited assay may need review or confirmatory or expanded testing. Keep the original report, including the list of variants tested, put it in your pharmacy or GP record, and bring it to any new prescriber: whether it remains sufficient depends on the particular medicine and current guidance.

Sources

1. CPIC guideline (2023) — CYP2D6, CYP2C19, CYP2B6, SLC6A4 and HTR2A and serotonin reuptake inhibitor antidepressants
2. Hicks et al. (2017), Clin Pharmacol Ther — CPIC guideline for CYP2D6, CYP2C19 and tricyclic antidepressants: 2016 update
3. CPIC frequency tables — CYP2C19 and CYP2D6 phenotype frequencies by biogeographical group
4. DPWG (KNMP) guideline — CYP2C19, CYP2D6 and SSRIs
5. Beunk et al. (2024), Eur J Hum Genet — DPWG guideline for CYP2D6, CYP2C19 and non-SSRI/non-TCA antidepressants
6. Greden et al. (2019), J Psychiatr Res — the GUIDED randomised trial
7. Oslin et al. (2022), JAMA — the PRIME Care randomised trial
8. Blake et al. (2026), JAMA Network Open — genotype-guided antidepressant prescribing (ADOPT PGx), randomised clinical trial
9. Swen et al. (2023), The Lancet — 12-gene pharmacogenetic panel (PREPARE)
10. Zhang et al. (2025), BMJ Mental Health — systematic review and meta-analysis, PGx-guided versus unguided antidepressant treatment
11. Baum et al. (2024), American Journal of Psychiatry — APA research workgroup report on pharmacogenomic tools in depression and the journal's correction
12. ISPG — genetic testing statement
13. NVL Unipolare Depression, version 3.2 — German national depression guideline (sections 4.4.3.1, 4.4.7.1, 7.1.1 and Table 37)
14. Cipramil (citalopram) SmPC — sections 4.2 and 4.4, CYP2C19 poor metaboliser dosing
15. McDermott et al. (2025), Br J Clin Pharmacol — pharmacogenomics in the UK NHS, including the PROGRESS programme
16. PSY-PGx clinical study description (2024), Pharmaceuticals — international trial including South London and Maudsley
17. FDA — Table of Pharmacogenetic Associations
18. FDA (2019) — warning letter over a genetic test claiming to predict patients' medication responses
19. KNMP — pharmacogenetics and the G-Standaard (Netherlands)
20. NHG-Standaard Depressie — Dutch GP depression guideline
21. Labor Limbach — billing information for pharmacogenetic diagnostics
22. NHS Genomics Education — CYP2D6 and CYP2C19 knowledge hub

---

This guide is for general information only and is not medical advice, a diagnosis, or a recommendation for or against any test or treatment. Do not start, stop or change a medicine on the basis of a genetic test result: those are decisions for a licensed prescriber who knows your history. Prices and coverage rules were checked on the date above and change; confirm with the laboratory and your insurer. If you are in crisis, contact your local emergency number (112 in the EU) or a crisis line immediately.