Pharmacogenomic Dosing Advisor (CPIC / DPWG / FDA / EMA)

Why use it

Turns a pharmacogenomic report into a prescribing decision at the point of care, across a whole catalogue rather than one drug at a time. It also answers the harder question the reports do not: whether the gene activates the drug or eliminates it. When CYP2C19 loss of function reduces clopidogrel activation, the failure mode is thrombosis, not bleeding, and raising the dose does not fix it — the correct answer is a different antiplatelet. That direction is encoded per gene and is the single most common source of error when reasoning by analogy from one drug to another.

About

A single decision engine over 60 drug–gene pairs spanning 22 germline genes. Select the drug, enter the genotypes the drug actually depends on, and the calculator returns what the guideline says to do: a fraction of the standard dose, a dose ceiling, an alternative agent, or an outright contraindication — with the source, the citation and the reason attached. The architecture is gene-first. The translation from diplotype to phenotype belongs to the gene and is written once, so CYP2D6 activity scoring exists in one place rather than in each of the seventeen drugs that use it. What the phenotype MEANS belongs to the drug, because it genuinely differs: a CYP2D6 poor metaboliser should have nortriptyline reduced to 50% and should not receive codeine at all, and no shared table can express both. Three properties are treated as safety features rather than details. First, percentages are always the fraction of the standard dose that is KEPT — "50%" means start at half, never cut in half again. Second, where CPIC, DPWG and the regulator disagree, the disagreement is shown rather than one of them being silently chosen. Third, for a gene where an ungenotyped patient cannot be assumed negative — HLA-B*57:01, HLA-B*15:02, G6PD, POLG — the calculator refuses to produce a recommendation instead of producing a confident one built on an assumption.

When to use

• A pharmacogenomic panel result has arrived and a prescribing decision depends on it
• Before starting a drug where genotyping is mandated (abacavir, carbamazepine in at-risk ancestries, fluoropyrimidines, mavacamten in the EU)
• Reviewing an existing prescription in a patient newly found to carry an actionable variant
• Explaining to a patient or colleague why a dose is a fraction of the label dose, with the citation to hand
• Choosing between agents in the same class when one is genotype-constrained and another is not

Formula

Pearls & pitfalls

• Percentages are the fraction KEPT, not removed. CPIC writes them that way and the ambiguity is a documented source of dosing error.
• CYP3A5 is inverted relative to the other cytochromes: *1 is the FUNCTIONAL allele, so a tacrolimus expressor needs 1.5–2× MORE drug. Most people are non-expressors, which is why the standard dose suits them.
• For a prodrug the deficient phenotype means treatment failure, not toxicity — clopidogrel, codeine and tamoxifen all need a different drug or a higher dose, never a lower one.
• HLA associations are all-or-nothing immune events. There is no protective dose, which is why every HLA rule is avoidance and why an ungenotyped HLA gene blocks the result rather than defaulting to negative.
• Half-life decides reduction versus avoidance among the CYP2C9 NSAIDs: celecoxib, ibuprofen, flurbiprofen and lornoxicam get a reduced dose; piroxicam and tenoxicam get an alternative, starting already at the intermediate metaboliser. Meloxicam sits on the hinge — reduced in the intermediate metaboliser, avoided in the poor.
• Phenytoin's poor metaboliser rule reduces the MAINTENANCE dose by half and leaves the loading dose untouched, because loading is governed by volume of distribution and not by clearance.
• A composite intermediate phenotype — intermediate in TPMT and NUDT15 at once — needs more reduction than either alone (20–50% versus 30–80%). CPIC added this in 2025; DPWG recognises it without quantifying it.
• Pimozide and atomoxetine are the two drugs where genotype constrains the PACE of titration, not only the dose: 14 days between increases and a plasma level before escalating, respectively.
• CYP2D6 is scored, not looked up: allele activity values are summed and read off a band table, so *1/*4 (score 1) is intermediate while *1/*41 (score 1.5) is also intermediate but nearer normal. A duplication can push a normal metaboliser to ultrarapid.
• An "unknown" genotype for a metabolising enzyme defaults to the normal phenotype and the result is labelled as inferred. That is a convenience for the common case, not evidence — and it is deliberately not extended to HLA, G6PD or POLG.

Next steps

Apply the recommendation as a STARTING point, then titrate on the clinical endpoint the drug is prescribed for. Genotype predicts exposure, not response: it does not replace INR for warfarin, blood counts for thiopurines, plasma concentration for atomoxetine or voriconazole, or symptom scales in psychiatry. Record the genotype in the patient record rather than the recommendation, because guidelines are revised and the genotype is not. Where the result reports a discordance between sources, document which one was followed and why.

Evidence

Rules are transcribed from the primary guideline text, not summarised from it. Sources are CPIC (clinical guidelines with graded strength), the Dutch Pharmacogenetics Working Group of the KNMP, and the FDA and EMA product labels, each recorded per drug with its citation and the date the entry was last verified. Where the sources conflict — and they do, on roughly a fifth of the catalogue — the entry records the conflict explicitly instead of resolving it silently. Notable divergences carried in the data: CPIC advises avoiding tramadol in the CYP2D6 poor metaboliser while DPWG advises monitoring and dose escalation first; CPIC quantifies a fluvoxamine reduction that DPWG states is unnecessary; CPIC switches agent for voriconazole in both extremes while DPWG keeps the drug at 50% and 1.5×; the FDA caps citalopram only in the poor metaboliser at age 60 while DPWG also covers the intermediate metaboliser at 65 and distinguishes tablets from drops; and mavacamten is subject to mandatory pre-prescription genotyping in Europe but not in the United States. Six entries are present with a documented gene–drug association but no dosing rule — paclitaxel, vincristine, doxorubicin, sunitinib, pazopanib and, for its own reasons, tamoxifen. They are included on purpose: "checked, and no dose adjustment is established" is a finding with a source, and it is not the same answer as "not yet catalogued".

References

1. Relling MV, Klein TE. CPIC: Clinical Pharmacogenetics Implementation Consortium. Clin Pharmacol Ther. 2011;89(3):464-467. — link
2. Swen JJ et al. Pharmacogenetics: from bench to byte — an update of guidelines (DPWG). Clin Pharmacol Ther. 2011;89(5):662-673. — link
3. Caudle KE et al. Standardizing terms for clinical pharmacogenetic test results. Genet Med. 2017;19(2):215-223. — link
4. Theken KN et al. CPIC Guideline for CYP2C9 and Nonsteroidal Anti-inflammatory Drugs. Clin Pharmacol Ther. 2020;108(2):191-200. — link
5. Cooper-DeHoff RM et al. CPIC Guideline for SLCO1B1, ABCG2, and CYP2C9 and Statin-Associated Musculoskeletal Symptoms. Clin Pharmacol Ther. 2022;111(5):1007-1021. — link
6. Lee CR et al. CPIC Guideline for CYP2C19 Genotype and Clopidogrel Therapy: 2022 Update. Clin Pharmacol Ther. 2022;112(5):959-967. — link
7. Bousman CA et al. CPIC Guideline for CYP2D6, CYP2C19, CYP2B6, SLC6A4, and HTR2A Genotypes and Serotonin Reuptake Inhibitor Antidepressants. Clin Pharmacol Ther. 2023;114(1):51-68. — link