This test sits between "genuinely life-saving" for a handful of drugs and "overhyped" for nearly everything else. For a child starting leukemia treatment, a cancer patient about to begin chemo with 5-FU or capecitabine, or a baby whose breastfeeding mother is offered codeine, it catches the rare deficiency that turns a normal dose into a lethal one. For most readers it is a card in their records — useful if they end up on the right drug, easy to miss when it would help. Self-pay runs about $300 to $500; Medicare, Medicaid, and many private insurers cover it when a specific drug is on the table.
Your liver breaks down most drugs with a family of enzymes called cytochrome P450 — CYP2D6, CYP2C19, CYP2C9 and a few others. Some of these come in versions you inherit. One inherited version of CYP2D6 processes drugs at three to five times the normal speed; another barely processes them at all. Roughly one person in twenty in European populations is a poor metabolizer for CYP2D6; up to one in five in East Asian populations is a poor metabolizer for CYP2C19.
What this means in plain terms: a standard dose of codeine in someone whose CYP2D6 runs hot turns into morphine far faster than the label expects, and that person can stop breathing — infants of breastfeeding mothers with this variant have died FDA 2017. A standard dose of the chemotherapy drug 5-fluorouracil in someone missing the enzyme that detoxifies it (the DPYD gene) overwhelms the bone marrow and gut lining; about one in a hundred such patients dies from the first cycle Amstutz et al. 2018. The test simply tells the prescriber which version of each enzyme you carry, so the drug or the dose can be chosen accordingly.
Where the evidence is strong, and where it isn't
The honest picture splits cleanly. For a short list of drug-gene pairs, the test prevents serious harm and major guidelines now expect it. For a much larger list, the test produces a real but small improvement that experts still argue about.
The settled list is the one to memorise. Before 5-FU or capecitabine chemotherapy, test DPYD — the European Medicines Agency made this standard in 2020, the FDA added a boxed warning to capecitabine in late 2025 and to 5-FU in early 2026 FDA 2025. Before the leukemia and inflammatory-bowel drug 6-mercaptopurine or its cousin azathioprine, test TPMT and NUDT15 — homozygous-deficient patients tolerate less than a tenth of the normal dose without life-threatening bone-marrow failure Relling et al. 2019. Before the HIV drug abacavir, test HLA-B*57:01 — universal screening eliminated the severe hypersensitivity reaction in a 1,956-patient trial Mallal et al. 2008. Before carbamazepine for seizures in a patient of East or Southeast Asian descent, test HLA-B*15:02 — a positive result means a real risk of Stevens-Johnson syndrome. And codeine and tramadol are off-limits for children under twelve and for breastfeeding mothers after several infant deaths in CYP2D6 ultra-rapid metabolizers FDA 2017.
The contested list is where most readers actually encounter this test. Pharmacogenomic panels for antidepressants are aggressively marketed, but the big trials are mixed. The 1,167-patient GUIDED trial missed its primary endpoint of depression-symptom improvement; a pre-defined subgroup of patients already on a drug with a known gene-interaction problem did better Greden et al. 2019. The 1,944-patient PRIME Care trial in U.S. veterans did show that test-guided prescribing led to fewer high-risk drug-gene mismatches and a small, real lift in remission over six months — the lead author called the overall effect "small" Oslin et al. 2022. The American Psychiatric Association does not yet recommend the test for someone starting their first antidepressant.
Cardiology is closer to settled. The TAILOR-PCI trial (5,302 patients after stent placement) narrowly missed its main endpoint but showed a clear benefit in the first three months for patients with a CYP2C19 loss-of-function variant who were switched off clopidogrel onto ticagrelor Pereira et al. 2020. The American Heart Association, in a 2024 scientific statement, endorses the test after stent procedures for that reason AHA 2024. Warfarin is the cautionary tale: a European trial showed clear benefit from genotype-guided starting doses, an American trial in a more diverse population showed no benefit and a hint of harm in African American patients — because the panel did not include the variants that drive dose response in African ancestry Pirmohamed et al. 2013 Kimmel et al. 2013.
Who this actually saves
The test is not a wellness test. It earns its keep in a small number of moments, and in those moments it is the difference between a normal week and an intensive-care admission.
The first moment is the day a colorectal-cancer patient is offered first-cycle 5-FU or capecitabine. Around three to seven percent of patients carry a DPYD variant that slows the enzyme that clears these drugs. At full dose, that group has a roughly two-in-three chance of severe early toxicity — diarrhea bad enough to require hospital fluids, a white-cell count low enough to require isolation, mouth ulcers that make eating impossible. Death rates run about one percent of treated DPD-deficient patients. With a 50% starting-dose reduction, hospitalization for first-cycle toxicity drops from around two-thirds of carriers to one-quarter Amstutz et al. 2018.
The second moment is a child starting maintenance therapy for acute lymphoblastic leukemia. 6-mercaptopurine is the backbone drug for the longest phase of treatment, and the gene TPMT — plus, in East Asian, South Asian, and Hispanic children, the gene NUDT15 — decides whether the child tolerates that drug or collapses into life-threatening neutropenia. A child with two non-functional copies tolerates less than a tenth of the protocol dose Relling et al. 2019.
The third moment is harder to picture but the FDA action was specific. A new mother is offered codeine for postpartum pain. If she carries CYP2D6 ultra-rapid metabolizer variants — common in some Mediterranean and Northeast African populations — the codeine is converted to morphine in her milk at higher concentrations than the label assumes, and her breastfed infant can stop breathing. The 2017 FDA contraindication closed that door for children under twelve and for breastfeeding mothers regardless of genotype, but the test was the evidence that opened the conversation FDA 2017.
The fourth moment is a person of East or Southeast Asian descent starting carbamazepine for seizures or trigeminal neuralgia. Up to one in ten Han Chinese, Thai, Malay, and Filipino patients carries HLA-B*15:02; that variant is the strongest known predictor of Stevens-Johnson syndrome and toxic epidermal necrolysis, conditions that strip the skin like a severe burn and can kill. Test before starting; if positive, pick a different anticonvulsant.
If none of those four moments is on the horizon for you or someone you care for, the test's payoff is much smaller and depends on the rest of this article.
How to actually get the test, and when
There are two ways into this, with different rhythms.
The first is reactive: you or your doctor know a specific drug is about to be prescribed where genotype changes the call. The doctor orders the panel that covers the relevant gene, the lab returns a result in three to ten days, and the prescription waits for the answer. This is the model for chemotherapy, for clopidogrel after a stent, for thiopurines before maintenance leukemia therapy. It is also the model that insurance pays for most reliably — coverage hangs on whether the drug-gene pair has a well-evidenced action from the CPIC consortium or the FDA label.
The second is pre-emptive: a single panel of ten to twenty genes, run once, stored in your electronic record, available the next time any drug on the list comes up. The UK's National Health Service is building this. A handful of US academic centres (Mayo, Vanderbilt, the University of Florida, St. Jude) already do it. The advantage is that the result is there when needed; the catch is that the result has to be findable in the chart years later, by a clinician who knows what to do with it.
Cost, coverage, and what actually shows up on the bill
Self-pay panels from the major US labs (GeneSight, Genomind, OneOme RightMed) cluster between $300 and $500. Medicare Part B covers the test at no out-of-pocket cost when the doctor documents that a specific drug-gene pair on the CPIC level A or B list is in play. Medicaid generally covers it. Private insurance is where the road forks: a 2024 review of major US insurers found one Medicare contractor (MolDX) covering all 65 reviewed drug-gene pairs, several private insurers covering ten or fewer, and the test for abacavir and carbamazepine in Asian-ancestry patients covered universally.
Two practicalities that catch people out. First, the test result needs to live somewhere the next prescribing doctor will actually see it — ideally directly in your electronic health record, not in a paper report tucked into your home filing cabinet. Ask the lab to send results to your primary care provider's EHR. Second, the panel matters: many older panels were built around European-descended populations and miss variants that drive response in African American, East Asian, and South Asian patients. If your ancestry includes a population where the panel was not designed for you, ask the ordering clinician what variants are actually tested — not just which genes.
The test itself is a cheek swab or a single blood draw. No fasting, no preparation. The result is genotype-stable for life: you do this once.
Why people get the test and still don't benefit
Three common failures, in roughly descending frequency.
The result is in your record, but the prescriber doesn't see it. A test ordered by a psychiatrist in 2022 doesn't surface in the emergency department in 2027 when a new clinician picks an opioid, unless the result is in a structured field of the shared electronic record. PREPARE worked partly because the result went into a flagged section of the chart that fired an alert at prescribing time Swen et al. 2023. Reactive single-gene testing, ordered once and forgotten, routinely fails this step.
The panel didn't cover the variants that matter for you. The COAG warfarin trial is the textbook case: a standard panel tested CYP2C9*2 and *3, the dominant variants in European-ancestry populations, and missed *5, *6, *8, *11, and a regulatory variant called rs12777823, which together drive most warfarin dose variation in African American patients. The pharmacogenetic arm did worse than the clinical arm in that subgroup Kimmel et al. 2013. The same issue affects NUDT15 testing in East Asian populations if only TPMT is checked, and DPYD testing in West African ancestry if only the four common European variants are covered.
Other drugs you take are blocking the enzyme anyway. Several common medications — paroxetine, fluoxetine, bupropion — are potent inhibitors of CYP2D6. If you are on one of those and you start a second CYP2D6-handled drug, your genotype is functionally a poor metabolizer regardless of what the test says. Clinicians call this phenoconversion. Most decision-support tools don't yet weight it well; ask your prescriber to consider the full drug list, not just the genotype card.
What this test is not
It is not "the personalised medicine test." It personalises a narrow subset of prescriptions — the ones where a specific gene-drug pair is well studied. Most drugs you'll be prescribed in your life will not have a panel-relevant pair.
It does not tell you which antidepressant will work — it tells you which ones are more likely to be cleared abnormally fast or slow. Whether a given antidepressant lifts your depression depends on receptor binding, your specific symptom profile, your other medications, and a dozen other things the test cannot see. The published trials in depression show modest, real benefit when used after a first failure; they do not show a transformation.
It is not the same thing as the genetic-health reports that come bundled with consumer ancestry kits. The 23andMe pharmacogenetic report is FDA-authorised and analytically accurate, but its panel is narrow and its label carries a mandatory warning not to change medications based on the result without a clinician. For clinical decisions, a clinical-grade panel ordered by a doctor is the path.
What you actually get
If you fall in one of the four high-stakes moments — chemo, leukemia maintenance, abacavir, carbamazepine in an Asian-ancestry patient — the payoff is immediate and concrete: you avoid a near-fatal first-cycle event. There is no waiting period. The clinician sees the result, the dose or the drug changes, and the harm that would have happened doesn't.
If you fall in one of the contested middle cases — a stent, a treatment-resistant depression — the payoff is real but modest. A correctly genotype-matched antidepressant gets you to remission a few weeks sooner, by the average of the trials. A correctly chosen blood-thinner after stent placement reduces your odds of a heart attack or stroke in the first three months by enough to matter, even if the trial's main endpoint didn't quite clear statistical significance.
If you don't fall in any of those, the payoff is a card in your medical record that may or may not be checked the next time a doctor reaches for the prescription pad. Across a population in their seventies on five or more chronic medications, that card cuts the rate of meaningful adverse drug reactions by about a third Swen et al. 2023. Across a healthy thirty-year-old, the same card mostly sits unused — until, one day, it doesn't.
Adjacent topics
If this entry was useful, a few neighbours worth knowing exist. Tumor genotyping — testing the cancer itself for variants that pick the right targeted drug — is a different test from the one described here, and the rules are different too; that's the test that picks BRCA-targeted, EGFR-targeted, or BRAF-targeted oncology agents. Drug-drug interaction checking — your pharmacist's job, and the most underused safety tool in medicine — catches problems the genetic test cannot see. Therapeutic drug monitoring, the cousin of pharmacogenomics, measures the actual blood level of a drug rather than predicting it from genetics; it is standard for several chemotherapies, the immunosuppressant tacrolimus, and a few epilepsy drugs. And direct oral anticoagulants (apixaban, rivaroxaban, edoxaban, dabigatran) have largely replaced warfarin in the settings where warfarin genotyping might have helped — worth knowing before you ask for a warfarin panel.
- — CYP1A2 decides whether coffee protects your heart or strains it. One of the cleaner real-world uses of the test.
- — G6PD is one of the genes that decides which drugs harm you — exactly what a pharmacogenomic panel reads.
- — This is the drug-metabolism slice of the bigger genetic-testing picture — narrow but occasionally life-saving.
- — The abacavir gene test is the textbook pharmacogenomic win — one gene checked before one drug.
- — Caffeine clearance is a familiar example of the metaboliser variation pharmacogenomics looks at.
- — Your genes set the baseline for how you clear a drug; supplements and foods can shove that same metabolism around on top.
- — Your CYP genes set baseline drug metabolism; grapefruit and St John's wort then push it around.
- — MTHFR is the over-hyped cousin of real pharmacogenomics — heavily marketed, rarely changing what anyone should actually do.
- — A pharmacogenomic result is a card for your records — put it on the health summary you bring to any prescriber.
- — Genes are one driver of how a drug lands; for a few medications the manufacturer is another.
Substance + claimed effects
Pharmacogenomic (PGx) testing is a clinical lab test that genotypes one or more drug-metabolizing or drug-target genes — most commonly the cytochrome P450 enzymes CYP2D6, CYP2C19, CYP2C9, and CYP2B6, plus a handful of non-CYP genes critical to specific drugs (DPYD for fluoropyrimidine chemotherapy, TPMT and NUDT15 for thiopurines, VKORC1 for warfarin, HLA-B*57:01 for abacavir, HLA-B*15:02 for carbamazepine, SLCO1B1 for statins, UGT1A1 for irinotecan). The result classifies a person as poor / intermediate / normal / rapid / ultra-rapid metabolizer for each enzyme, and clinicians use that classification — paired with curated dosing guidance from CPIC (Clinical Pharmacogenetics Implementation Consortium) and the FDA's table of pharmacogenomic biomarkers FDA 2024 — to pick a drug, adjust a starting dose, or switch agents. Claimed effects fall into four real buckets: (a) avoiding life-threatening toxicity from drugs that the body activates or clears via a polymorphic enzyme (fluoropyrimidines in DPD-deficient patients, codeine in CYP2D6 ultra-rapid metabolizers, thiopurines in TPMT/NUDT15-deficient patients, abacavir in HLA-B*57:01 carriers) — the strongest evidence; (b) reducing failure on antidepressants and antiplatelet drugs by avoiding gene-drug mismatches at first prescription (PRIME Care, PREPARE, GUIDED, TAILOR-PCI); (c) right-sizing initial warfarin dose to reduce out-of-range INRs in the first weeks of therapy (EU-PACT positive, COAG null); (d) cutting overall adverse drug reactions across a 12-gene panel when tested pre-emptively for a population taking many drugs (PREPARE, 30% reduction Swen et al. 2023). This entry covers all of those.
Evidence by addressing question
mechanism
Most prescription drugs are activated, inactivated, or cleared by cytochrome P450 enzymes in the liver. The CYP family is highly polymorphic — common loss-of-function or gain-of-function variants change enzyme activity by orders of magnitude in a meaningful fraction of any population. CYP2D6 alone metabolizes roughly 25% of all prescribed drugs, including codeine, tramadol, oxycodone, hydrocodone, many SSRIs and SNRIs, tricyclics, antipsychotics, tamoxifen, and beta-blockers. More than 100 CYP2D6 alleles are catalogued; *3, *4, and *5 account for ~95% of poor-metabolizer phenotype in European-ancestry populations, while *10 and *17 dominate the intermediate-metabolizer phenotype in East Asian and African populations respectively. Activity-score arithmetic (CPIC-standardised since 2019) maps a diplotype to a phenotype: 0 = poor metabolizer, 0.25–1 = intermediate, 1.25–2.25 = normal, >2.25 = ultra-rapid Bousman et al. 2023.
The clinical consequence depends on whether the parent drug is the active species or a prodrug. For prodrugs the enzyme activates (codeine→morphine, clopidogrel→active thiol metabolite, tamoxifen→endoxifen), poor metabolizers under-respond and ultra-rapid metabolizers over-respond — sometimes fatally Crews et al. 2014. For drugs the enzyme clears (most SSRIs, tricyclics, antipsychotics, warfarin), poor metabolizers accumulate the drug and over-respond; ultra-rapid metabolizers clear it too fast and under-respond. DPYD, TPMT, and NUDT15 work differently — they're catabolic enzymes that detoxify cytotoxic chemotherapeutics, and a deficient genotype lets the drug accumulate to lethal levels at standard dose Amstutz et al. 2018 Relling et al. 2019. The HLA-B story is different again: certain HLA alleles present specific drugs to T cells as foreign, triggering severe immune-mediated skin reactions (abacavir hypersensitivity, carbamazepine-induced Stevens-Johnson syndrome) — a binding-pocket compatibility problem, not a metabolism problem Mallal et al. 2008.
evidence
Evidence quality is wildly uneven across the drug-gene pairs the same panel reports. The strongest tier: prospective RCT with screening preventing a near-fatal toxicity. HLA-B*57:01 screening before abacavir, in a 1,956-patient double-blind trial across 19 countries, eliminated immunologically confirmed hypersensitivity (0% vs 2.7%, p<0.001), with a negative predictive value of 100% Mallal et al. 2008. DPYD screening before fluoropyrimidine chemotherapy reduces severe early toxicity in carriers from ~73% to ~28% with 50% upfront dose reduction; hospitalization for grade ≥3 toxicity drops from ~64% to ~25%. The EMA mandated DPD evaluation in 2020 EMA 2020; the FDA added boxed warnings to capecitabine in October 2025 and to 5-fluorouracil in February 2026 FDA 2025 Amstutz et al. 2018. TPMT and NUDT15 testing before thiopurine therapy (6-mercaptopurine, azathioprine, thioguanine) is similarly entrenched — homozygous-deficient patients tolerate <10% of standard dose; CPIC recommends 80–90% dose reduction or alternate therapy Relling et al. 2019. CYP2D6 ultra-rapid metabolizers on codeine have died from morphine overdose at standard doses; in 2017 the FDA contraindicated codeine and tramadol in children under 12 and in breastfeeding mothers after multiple infant deaths FDA 2017.
The middle tier: positive panel-level pre-emptive testing. The PREPARE trial — a 6,944-patient cluster-randomised crossover implementation study across seven European countries — used a 12-gene panel to guide prescribing and reduced clinically relevant adverse drug reactions by 30% (OR 0.70, 95% CI 0.54–0.91) Swen et al. 2023. 95% of enrolled patients carried at least one actionable genetic variant; 25% had an actionable interaction with their newly prescribed drug.
The contested tier: psychotropic panel testing. The GUIDED trial (n=1,167, treatment-resistant depression) missed its primary endpoint of symptom improvement on the intent-to-treat analysis (p=0.069); a pre-specified subgroup of patients on a medication with an actionable gene-drug interaction showed significant improvements in remission and response at 8 weeks Greden et al. 2019. PRIME Care, the larger VA trial (n=1,944), met its co-primary outcomes: PGx-guided prescribing reduced antidepressants with substantial gene-drug interactions (10.7% vs 19.7%) and produced a small remission benefit over 24 weeks; the authors called the overall effect "small" Oslin et al. 2022. The American Psychiatric Association does not recommend routine PGx testing for first-line antidepressant selection.
The cardiology tier: CYP2C19 + clopidogrel after PCI. TAILOR-PCI (n=5,302) narrowly missed its primary endpoint at 12 months (HR 0.66, p=0.06) for CYP2C19 loss-of-function carriers escalated to ticagrelor versus universal clopidogrel; pre-specified secondary analyses showed significant benefit in the first 3 months Pereira et al. 2020. The 2024 AHA scientific statement supports CYP2C19-guided antiplatelet therapy after PCI based on the meta-analytic signal across multiple trials, even though TAILOR-PCI was nominally negative AHA 2024.
The mixed tier: warfarin. EU-PACT (n=455, predominantly European) showed genotype-guided dosing produced more time in therapeutic INR over the first 12 weeks (67.4% vs 60.3%, p<0.001) and fewer supratherapeutic INRs Pirmohamed et al. 2013. COAG (n=1,015, 27% African American) compared genotype-guided to clinical-algorithm dosing and found no difference overall and a signal of harm in African American patients, where unmeasured variants (CYP2C9*5, *6, *8, *11 and rs12777823) drive a meaningful share of dose variance Kimmel et al. 2013. CMS issued the only national PGx coverage determination for warfarin in 2009, but coverage is limited to clinical trial enrollment CMS 2009; major cardiology guidelines do not endorse routine genotyping before warfarin initiation, given direct oral anticoagulants have largely displaced warfarin where genotyping might have mattered.
protocol
Operationally, two distinct protocols. Reactive testing: ordered when a specific drug-gene pair is about to matter — DPYD before first cycle of 5-FU or capecitabine, TPMT/NUDT15 before maintenance thiopurines, HLA-B*57:01 before abacavir, CYP2C19 at the time of PCI for clopidogrel selection, CYP2D6/CYP2C19 after one or more failed SSRI/SNRI trials. Cheek swab or blood draw; result in 1–7 days depending on lab; cost $150–$500 self-pay, often covered by Medicare and Medicaid and many private plans when CPIC level A/B drug-gene pair is documented. Pre-emptive panel testing: a single 10–20-gene panel run once, the result stored in the EHR, queried whenever a relevant drug is prescribed thereafter. PREPARE used the latter; it's the model the NHS Genomic Medicine Service and several large US academic medical centres (Mayo, Vanderbilt, St. Jude, U. of Florida) are implementing.
The actionable output is genotype → phenotype → drug-specific dosing. Examples: CYP2D6 ultra-rapid metabolizer on tramadol → switch to non-CYP2D6 opioid (morphine, hydromorphone) Crews et al. 2014; CYP2C19 poor metabolizer on clopidogrel after PCI → switch to ticagrelor or prasugrel AHA 2024; CYP2C19 poor metabolizer on citalopram → 50% dose reduction or alternate SSRI Bousman et al. 2023; DPYD intermediate metabolizer (activity score 1.0 or 1.5) on capecitabine → 50% starting dose with titration Amstutz et al. 2018; TPMT homozygous deficient on 6-mercaptopurine → 90% dose reduction or alternate; NUDT15 should be co-tested for Asian-ancestry patients Relling et al. 2019.
contraindications
The test itself is a buccal swab with no meaningful clinical risk. The hazards are downstream: a result mis-applied by a clinician who doesn't know the gene-drug pair, a panel that doesn't cover the variants most relevant to a non-European-ancestry patient (warfarin in African American patients is the canonical failure mode — major dose-modifying alleles aren't on most panels), and over-extrapolation to drug-gene pairs without CPIC level A/B evidence. Direct-to-consumer reports (23andMe Personal Genome Service Pharmacogenetic Reports, FDA-authorised since 2018) carry a mandatory warning not to act on results without clinician review; analytical accuracy versus Sanger sequencing is >99%, but the test panel is narrow and the clinical interpretation isn't done by the consumer-facing report.
misconceptions
Three persistent misreads. First, that a single PGx test "personalises all your medications" — it personalises a narrow subset of the drugs the panel was designed for, with strong evidence on a small fraction of those. Most prescriptions have no actionable gene-drug pair. Second, that "metabolizer status" predicts efficacy as cleanly as it predicts toxicity — efficacy depends on target binding, downstream pharmacology, drug-drug interactions, adherence, and disease subtype, none of which the panel sees. The GUIDED and PRIME Care trials show real but modest depression-remission benefits in the gene-drug-interaction subgroup, not transformation in the broader population. Third, that pre-emptive testing in healthy young adults pays off the same way it does in older polypharmacy patients — the absolute event rates are far lower, and cost-effectiveness has been demonstrated mainly in older patients on multiple drugs or in patients about to start a CPIC-actionable drug.
alternatives
For oncology: therapeutic drug monitoring (5-FU plasma levels guide dose post-cycle 1) is a complementary tool to DPYD genotyping but doesn't prevent the first-cycle toxicity DPYD does. For warfarin: the rise of direct oral anticoagulants (apixaban, rivaroxaban, edoxaban, dabigatran) has reduced the population that benefits from warfarin genotyping at all; routine INR monitoring remains the dosing tool. For depression: measurement-based care (PHQ-9 tracking, treat-to-remission protocols) and rapid switching are the dominant interventions; PGx is at best adjunctive. For clopidogrel: bedside platelet-function testing (VerifyNow) and choosing prasugrel or ticagrelor empirically are alternative strategies, though more expensive. For codeine in CYP2D6 ultra-rapid metabolizers: any non-CYP2D6 opioid (morphine, hydromorphone) bypasses the problem.
failure-modes
The dossier on real-world failures is thick. (1) Panel coverage doesn't include the variants that matter for the patient's ancestry — CYP2C9 *5/*6/*8/*11 in African American warfarin patients, NUDT15 c.415C>T in East and Southeast Asian thiopurine patients (90% of poor-metabolizer phenotype in those populations is missed by TPMT-only testing) Relling et al. 2019. (2) The result lives in a lab report or fragmented EHR field that the prescribing clinician — months later, a different specialty — doesn't see. PREPARE worked because it embedded results into clinical decision support; reactive single-gene testing routinely fails this step Swen et al. 2023. (3) Clinicians without PGx training make either no change or the wrong change in response to a result; CPIC guideline implementation surveys show meaningful misinterpretation. (4) Direct-to-consumer reports (23andMe) cover a different set of variants than clinical panels and are not designed to drive dosing decisions; the FDA label requires a "do not change medications without a clinician" warning. (5) Drug-drug interactions and phenoconversion — strong CYP2D6 inhibitors (paroxetine, bupropion, fluoxetine) turn a genotypic normal metabolizer into a functional poor metabolizer for any other CYP2D6 substrate, and the genotype alone won't catch it.
practicalities
Cost ranges $0–$2,000 depending on payer and panel. Self-pay multi-gene panels from major US labs (GeneSight, Genomind, OneOme RightMed, RPh Labs) cluster at $300–$500. Medicare Part B covers PGx testing when ordered for a CPIC level A/B drug-gene pair and documented as medically necessary; Medicaid generally covers it; private coverage varies wildly — a February 2024 AJMC review found one major payer (MolDX, the Medicare contractor) covered all 65 reviewed drug-gene pairs, while eight insurers covered ≤10. HLA-B*57:01 (abacavir) and HLA-B*15:02 (carbamazepine) are universally covered. Result turnaround is 3–10 days at most clinical labs; bedside point-of-care CYP2C19 typing (Spartan RX) returns results in ~60 minutes and was used in TAILOR-PCI for periprocedural decisions. The test is genotype-stable for life — done once, theoretically usable across decades of prescribing. The bottleneck isn't the test; it's whether the result is retrievable in the EHR at the next prescribing decision a clinician makes.
stakes
For the small fraction of readers about to start a CPIC-actionable drug, the stakes are concrete. A CYP2D6 ultra-rapid metabolizer prescribed codeine post-surgery is at meaningfully elevated risk of respiratory depression; an infant breastfed by such a mother has died of morphine overdose FDA 2017. A DPYD-deficient patient given full-dose capecitabine for colorectal cancer faces a ~10% mortality from the first cycle Amstutz et al. 2018. A TPMT/NUDT15-deficient child with ALL given full-dose 6-mercaptopurine faces life-threatening myelosuppression. For the much larger fraction of readers — healthy, on no chronic medication, or on drugs that aren't strongly polymorphic — the stakes are abstract: the test sits in your record, may help if you're prescribed something relevant in the next thirty years, may not be retrievable when it's time.
payoff
Payoff scales with the drug being prescribed. In oncology and HIV — DPYD/abacavir/thiopurine settings — payoff is "did not have a near-fatal first-cycle event." In cardiology PCI, payoff is a marginal reduction in 1-year MACE for the LOF-allele subgroup. In psychiatry, payoff is "found a working antidepressant on the second or third trial instead of the fourth or fifth" for the subset with an actionable gene-drug interaction; the GUIDED post-hoc and PRIME Care show small but real remission lifts. In population terms, PREPARE's 30% reduction in clinically relevant ADRs across a broad polypharmacy population is the highest-quality population-level payoff signal we have Swen et al. 2023. Onset latency: zero — payoff materialises at the prescribing decision, not later.
out-of-scope
This entry is bounded to clinical pharmacogenomic testing — germline CYP/HLA/DPYD/TPMT/NUDT15-class panels used to guide drug selection and dosing. Out of scope: somatic tumor genotyping (BRCA, EGFR, KRAS, BRAF) that drives targeted oncology drug selection; nutrigenomics / wellness genetic reports; polygenic risk scores for disease prediction; carrier screening for inherited disease.
Credibility range
The optimist case
For specific high-stakes drug-gene pairs — DPYD/fluoropyrimidines, TPMT/NUDT15/thiopurines, HLA-B*57:01/abacavir, HLA-B*15:02/carbamazepine, CYP2D6/codeine in children — the evidence is settled and the intervention prevents deaths. Regulatory bodies on both sides of the Atlantic (FDA boxed warnings, EMA mandates) have moved from "consider testing" to "test before prescribing." For pre-emptive panel testing in polypharmacy populations, PREPARE provides class-I evidence at the population level: a 30% reduction in clinically relevant ADRs Swen et al. 2023. For clopidogrel after PCI, the AHA 2024 scientific statement endorses genotype-guided antiplatelet selection. Cost-effectiveness analyses across 108 PGx-vs-standard-care studies found roughly three-quarters demonstrated PGx testing was cost-effective or cost-saving. The technology is cheap, the test is one-time, and the result is reusable across a lifetime of prescribing — the marginal cost per actionable decision falls toward zero as panels become routine.
The skeptic case
For the modal drug-gene pair the panel reports, the absolute benefit is small and the trial evidence is thin or null. TAILOR-PCI missed its primary endpoint Pereira et al. 2020; COAG showed no warfarin benefit and a signal of harm in African American patients Kimmel et al. 2013; GUIDED missed its primary endpoint Greden et al. 2019; PRIME Care's effect size was small Oslin et al. 2022. Psychiatry guidelines (APA) do not endorse routine pre-prescription PGx testing for first-line antidepressants. Commercial PGx companies have substantial financial incentive to oversell their panels; combinatorial panel recommendations (proprietary algorithms layered on top of CPIC consensus) often disagree across vendors for the same patient. Panels under-represent variants common in non-European-ancestry populations, encoding the same disparities the test was supposed to fix. The PREPARE 30% ADR reduction has been challenged on methodological grounds in the Lancet correspondence — the implementation-study design, the heterogeneous outcome definition, and the unblinded crossover make causal attribution debatable.
The author's call
The honest read is heterogeneous, not uniform. PGx testing is established-of-care for a short list of drug-gene pairs: DPYD before fluoropyrimidines, TPMT/NUDT15 before thiopurines, HLA-B*57:01 before abacavir, HLA-B*15:02 before carbamazepine in Asian-ancestry patients, CYP2D6 status before codeine in nursing mothers and children. These are essentially "test or you may kill the patient" decisions; the article frames them that way. For psychotropic panel testing, the evidence supports use after failure of at least one antidepressant in a patient with clear interaction-relevant prescribing history — not as a universal first-prescription screen. For clopidogrel post-PCI, the AHA 2024 endorsement is reasonable given the meta-analytic signal, though TAILOR-PCI was nominally negative. For pre-emptive panel testing in any patient about to enter polypharmacy (older adults, transplant patients, complex psychiatric or oncologic care), PREPARE is the strongest population-level evidence and the direction of travel in European health systems. For young healthy patients on no medication, the test pays off if and when a relevant prescription happens — which it often will eventually, but the actionable subset of those prescriptions remains a small fraction. evidence: 3 (heterogeneous, with pockets of 5 and pockets of 1); controversy: 3 (foundational disagreement on routine use, settled consensus on the high-stakes pairs).
Stakeholder + incentive map
- Commercial PGx labs (Myriad/GeneSight, Genomind, OneOme, Admera, RPh Labs, Coriell). Strong incentive to expand indications and pitch combinatorial-algorithm panels beyond CPIC-anchored gene-drug pairs. Have funded the bulk of psychotropic PGx trials.
- CPIC / PharmGKB / DPWG. Academic consortia setting prescribing guidelines anchored to evidence levels (A through D). Generally conservative; only recommend acting on level A and B pairs.
- FDA / EMA / MHRA. Regulatory bodies. EMA was first to mandate DPD testing (2020); FDA followed with boxed warnings (2025 for capecitabine, 2026 for 5-FU). FDA maintains the public table of pharmacogenomic biomarkers in drug labeling (~270 drugs as of 2024) FDA 2024.
- Insurers (CMS, MolDX, private). Coverage depends on CPIC level A/B + FDA recognition. MolDX (Palmetto GBA) is the most generous; private payer coverage is fragmented. Cost-effectiveness data favours coverage for actionable pairs.
- Direct-to-consumer (23andMe). FDA-authorised for a narrow set of reports (2018) with mandatory clinician-review warnings. Different incentive: scale, low-cost, customer-curiosity rather than clinical actionability.
- Specialty societies. American Society of Clinical Oncology, NCCN, ESMO endorse DPYD testing. American Heart Association (2024) endorses CYP2C19/clopidogrel testing after PCI. American Psychiatric Association does not endorse routine PGx testing for first-line antidepressants. American Society of Clinical Pharmacy is broadly supportive across the actionable-pair list.
- Health systems implementing pre-emptive testing. NHS Genomic Medicine Service (UK), Mayo Clinic RIGHT, Vanderbilt PREDICT, U. Florida Personalized Medicine Program, St. Jude. These are the populations where panel-then-store-then-query workflows are being tested.
Population variability
Allele frequencies vary substantially across ancestry groups, with direct clinical consequence.
CYP2D6: ultra-rapid metabolizer phenotype prevalence — ~1–2% in Northern European, ~10% in Southern European, up to 29% in some Northeast African and Arabian Peninsula populations. Poor metabolizer phenotype — ~7% in European, ~1% in East Asian. The CYP2D6*10 reduced-function allele is common in East Asian populations (~50% allele frequency).CYP2C19: poor metabolizer ~2% in European, ~15–20% in East Asian. The *17 ultra-rapid allele is common in European and African ancestry but rare in East Asian.CYP2C9: ~35% of Europeans carry at least one variant allele, ~25% of African Americans, ~4% of East Asians. African American-specific dose-modifying variants (*5, *6, *8, *11 plus rs12777823) are commonly missed by panels designed around European-derived data — directly relevant to the COAG warfarin null result Kimmel et al. 2013.NUDT15: c.415C>T variant is rare in European populations but common in East Asian, Hispanic, and South Asian populations (allele frequency up to 10%). TPMT-only testing misses ~90% of thiopurine intolerance in these populations Relling et al. 2019.HLA-B*15:02: prevalence up to 10–15% in Han Chinese, Thai, Malay, and Filipino populations; effectively absent in European-ancestry populations. The FDA recommends carbamazepine screening only in patients with Asian ancestry.HLA-B*57:01: prevalence ~5–8% in European-ancestry, lower elsewhere — but universal pre-abacavir screening is standard regardless of ancestry given the cost-benefit.DPYD: the four variants on most clinical panels (*2A, *13, c.2846A>T, c.1236G>A) capture most European-derived DPD deficiency. African-ancestry patients carry additional variants (e.g., Y186C) that increase toxicity risk and may be missed by Eurocentric panels.
Baseline status interacts with phenotype: a CYP2D6 ultra-rapid metabolizer is at higher risk only when the relevant CYP2D6-activated drug is co-prescribed; pharmacokinetic interactions (CYP2D6 inhibition by paroxetine, fluoxetine, bupropion) can convert a normal-metabolizer phenotype into a functional poor-metabolizer state regardless of genotype — phenoconversion is a significant practical limitation. Pediatric and pregnancy populations have specific actionable pairs (codeine contraindicated in breastfeeding mothers and children <12 since 2017) FDA 2017.
Knowledge gaps
Several genuine open questions. (1) Real-world impact of pre-emptive panel testing in heterogeneous US populations: PREPARE was European, conducted in a polypharmacy-rich elderly sample, with integrated decision support. Generalisability to fragmented US care is untested. (2) Whether the GUIDED/PRIME Care psychiatric signal is large enough to justify routine first-line PGx testing for depression remains contested; APA's current position is conservative, but US payer behavior is heterogeneous. (3) Polygenic and rare-variant contribution to drug response — current panels capture common variants in a short list of pharmacogenes; whole-exome or whole-genome data would catch rare loss-of-function variants but raise interpretive and cost questions. (4) Phenoconversion and drug-drug interaction integration: clinical decision support today is mostly genotype-driven; sophisticated systems weighting concurrent inhibitor/inducer drugs are immature. (5) Cost-effectiveness of pre-emptive testing in young, low-utilization populations — most economic analyses are in older or already-polypharmacy cohorts. (6) Ancestry-representative panel design — the gap between European-derived panel coverage and the genetic architecture of non-European populations is a known equity problem; broader variant sets are being deployed but uneven. (7) Long-term outcome data: the test is genotype-stable, so once-and-stored is the model, but how reliably the result is retrieved at every relevant prescribing decision over decades is an open implementation question.
Scope vs brief. The brief named psychotropics, anticoagulants, pain medications, and oncology agents as the consequence quadrants. The article covers all four end to end: oncology (DPYD, thiopurines) and pain (codeine/tramadol CYP2D6) carry the strongest evidence and the most concrete stakes; psychotropics get a candid mixed-evidence framing (GUIDED missed, PRIME Care small effect, APA does not endorse); anticoagulants are split into the warfarin cautionary tale (COAG vs EU-PACT, with the African-ancestry harm signal) and clopidogrel-post-PCI (TAILOR-PCI nominally negative, AHA 2024 endorsement). No quadrant was silently dropped.
Hard rating calls.
evidence: 3is the hardest single call. By drug-gene pair, evidence ranges from level 5 (HLA-B*57:01/abacavir, DPYD/fluoropyrimidines, TPMT/thiopurines) to level 1 (panel-based psychotropic prescription for first-line antidepressants). The 3 is a population-weighted average, with the heterogeneity called out explicitly in the justification and in the article body. An alternative defensible call is 4, leaning on PREPARE's class-I population-level result; I landed at 3 because the modal reader's modal use-case is closer to "considering a panel for a chronic prescription with mixed-evidence drug-gene pairs" than to "about to receive 5-FU."longevity: 2not higher despite the genuine death-prevention in narrow subgroups. The scoring framework is holistic across the catalogue's reader — and the average reader's lifetime probability of being in a DPYD/abacavir/carbamazepine-Asian-ancestry/codeine-breastfeeding moment is meaningful but not dominant. The article body carries the lifesaver framing in thestakessection so the score's averaging doesn't bury the high-stakes case.mood: 2reflects the PRIME Care and GUIDED post-hoc signal honestly — real, modest, conditional on having an actionable interaction with an antidepressant being prescribed. Not 3, which would imply clear cognitive/mood-stabilization the trials do not support.controversy: 3rather than 2 because the gap between "APA does not recommend" and aggressive commercial marketing of psychotropic panels is a real, foundational disagreement; not 4 because no one disputes the high-stakes drug-gene pairs.
Excluded or thinly covered.
- Tumor (somatic) genotyping for targeted oncology drug selection — explicitly out of scope, flagged in the
out-of-scopeclosing section. Distinct test class, distinct evidence base, warrants its own entry. - Wellness / nutrigenomics / direct-to-consumer "MTHFR" reports — out of scope. The 23andMe pharmacogenetic report gets a single misconceptions paragraph; broader DTC genetic wellness is a different topic and would dilute this entry.
UGT1A1and irinotecan,SLCO1B1and statin myopathy,IFNL3and hepatitis C therapy,G6PDand oxidative-stress drugs — all real CPIC pairs. Mentioned only in the research dossier; the article body picks the four highest-stakes pairs to anchor the felt-experience framing rather than enumerating the full panel. The choice serves the friend test.- Cost-effectiveness analytics in detail — the article cites the headline ("about three-quarters of 108 cost studies show PGx cost-effective") only in research; the practicalities section in the article gives the reader the price they will actually pay.
Future-link candidates.
- Tumor genotyping / somatic mutation testing — companion topic for an oncology cluster.
- Therapeutic drug monitoring — natural complement; covers the cases pharmacogenomics misses (phenoconversion, drug-drug interactions).
- Direct oral anticoagulants vs warfarin — the existence of DOACs is why warfarin pharmacogenomics matters less; an entry framing the choice would link bidirectionally.
- Drug-drug interaction checking — the more impactful, less glamorous companion intervention.
- Specific drug-class entries (SSRIs, anticoagulants, opioids) would each link back to this entry once written.
Audience and contraindications. Audience left unscoped — the substance applies to anyone who may be prescribed an actionable drug, which is approximately everyone over a lifetime, with disproportionate value to cancer patients, transplant recipients, polypharmacy older adults, and patients with treatment-resistant psychiatric conditions. No contraindications from the closed vocabulary apply — the test itself is a cheek swab with no medical risk.
Voice note. The stakes section anchored on four specific clinical moments rather than abstracting to "drug safety." The four were chosen so the typical reader can picture a person they know — a relative starting chemo, a child with leukemia, a new mother, an Asian-ancestry patient starting an anticonvulsant — and resist the wellness-test framing that "everyone benefits equally." The article's payoff section deliberately tiers the benefit (immediate-and-concrete vs modest vs deferred) rather than averaging it.
Pharmacogenomic Testing
One cheek swab or blood draw, results in 3–10 days, one-time test. Effort is in finding a clinician to order and interpret the result, not in the test itself.
Self-pay multi-gene panels typically $300–$500; $0 out-of-pocket for Medicare/Medicaid beneficiaries with documented medical necessity. Private payer coverage is fragmented; a 2024 AJMC review found eight major insurers covering ≤10 drug-gene pairs each. One-time cost.
Heterogeneous by drug-gene pair. Settled level-5 evidence for HLA-B*57:01/abacavir (Mallal et al. 2008), DPYD/fluoropyrimidines (EMA 2020, FDA 2025 boxed warnings), TPMT/NUDT15/thiopurines. PREPARE delivered class-I evidence for pre-emptive panel testing at population level (Swen et al. 2023). Mixed-to-negative for psychotropic panels (PRIME Care, GUIDED), warfarin (COAG null), and clopidogrel (TAILOR-PCI nominally negative, AHA 2024 still endorses).
Pre-emptive panel testing reduced clinically relevant adverse drug reactions by 30% across a 6,944-patient European cohort (PREPARE, Swen et al. 2023). For the typical reader, the benefit is conditional on starting an actionable drug; for those who do, the short-term safety lift is substantial.
Prevents specific fatal toxicities: DPYD-deficient fluoropyrimidine deaths (~0.5–1% mortality at full dose, Amstutz et al. 2018), codeine respiratory depression in CYP2D6 ultra-rapid metabolizers (FDA 2017), thiopurine myelosuppression in TPMT/NUDT15 deficiency (Relling et al. 2019). Real mortality reduction in narrow subpopulations; small additive effect averaged across the catalogue.
PRIME Care (n=1,944 veterans, Oslin et al. 2022) showed CYP2D6/CYP2C19-guided prescribing reduced antidepressants with substantial gene-drug interactions and produced a small remission lift; GUIDED post-hoc analysis (Greden et al. 2019) showed significant remission/response improvement in patients on actionable-interaction medications. Modest but real effect on depression outcomes in the actionable-interaction subgroup.