The FDA maintains a public table of pharmacogenetic associations, updated most recently this month, listing the gene and drug pairs where the agency considers the evidence sufficient. It is organized into three tiers, and the tiering is the part worth understanding. Only the first section covers associations where the data support a specific change in how a drug is prescribed. The second describes potential effects on safety or response. The third covers pairs where a genetic variant demonstrably changes how the drug is processed, with no established effect on whether the patient does better or worse.

That structure is the honest shape of this field, and it is almost exactly the opposite of how consumer genetic reports present it.

From the Lab to the Ledger

Most drugs are broken down by a family of liver enzymes called cytochrome P450. The genes encoding them come in common variants, and depending on which pair someone inherits they may clear a given drug quickly, normally or slowly. A slow metabolizer accumulates more of an active drug at a standard dose. Someone who cannot activate a prodrug gets little effect from it at all. None of this is speculative pharmacology.

The best-worked example is clopidogrel, a blood thinner given after stents and certain strokes. It arrives as an inactive compound and needs the enzyme CYP2C19 to become the drug that actually works. People carrying reduced-function variants activate less of it, and in that setting the genotype predicts something that matters: how well the treatment prevents the clot it was prescribed against. Professional guidelines address this pair specifically, and a UK guideline published this year works through who should be tested and what to do with the result.

The distance between that example and a general-purpose test is large. Clopidogrel has a clear activation step, a measurable outcome, an alternative drug to switch to, and trial evidence connecting genotype to events. Most gene and drug pairs have one or two of those, not all four. For a great many medications, knowing someone is an intermediate metabolizer changes a blood level without anyone knowing whether it should change the prescription.

Psychiatry is where this gets most oversold. Panels marketed for antidepressant selection often report on a dozen genes and produce colour-coded lists of drugs to use and avoid. A small number of those associations, mainly involving CYP2D6 and CYP2C19, appear in the FDA's own table and in professional guidance. The rest lean on inference. Antidepressant response is driven far more by diagnosis, dose, adherence and time than by clearance rate, and a report implying otherwise can lead someone to abandon a drug that would have worked.

Bio-Pipeline Ledger

CYP2C19 testing before clopidogrel: the clearest clinical use, addressed by professional guidelines, with a defined alternative when the result is unfavourable.

DPYD testing before certain chemotherapy drugs, and TPMT or NUDT15 before thiopurines: established, and in these cases the test prevents severe toxicity rather than fine-tuning a response. Some health systems now test routinely.

HLA-B*57:01 before abacavir, and HLA-B*15:02 before carbamazepine in certain populations: validated, with the test preventing serious hypersensitivity reactions.

Broad pharmacogenomic panels marketed for antidepressant or pain-medication selection: partly grounded and largely extrapolated. A handful of the reported pairs carry real guidance; the colour-coded verdict on the rest does not.

Direct-to-consumer genetic reports with a medication-response section: generally derived from the same variants, without the clinical context, and sold on a different promise.

The FDA table itself: freely available, tiered by strength of evidence, and explicit that it does not make testing a requirement or a substitute for reading the drug label.

The Clinical Reality Check

What is established is a short, specific and genuinely valuable list. For a handful of drug and gene pairs, mostly in oncology, transplant medicine, infectious disease and antiplatelet therapy, a test before the first dose prevents a serious problem. If you are about to start one of those drugs, this is worth asking about by name.

What is overstated is the framing of pharmacogenomics as a general key to personalized prescribing. Most medications have no actionable genetic guidance, and a panel that returns a result for forty drugs is reporting metabolism for most of them and clinical advice for few. The FDA says as much on the page itself.

The practical approach is to treat a pharmacogenomic result as one input a prescriber holds alongside kidney function, age, other medications and what has actually worked for you before. It is a reason to ask a better question, not a reason to change or stop a medication on your own. If you already have a report from a consumer test, bringing it to the prescriber is reasonable; treating its red and green columns as instructions is not.