Pharmacogenomics in 2026: what the evidence says

Every figure below comes from a primary source we opened on September 28, 2026. Counts that change over time, such as the number of guidelines, are as of that day.

Almost everyone. A study of 487,409 people in the UK Biobank, the largest of its kind when it was published, looked at 14 genes that have CPIC guidelines. It found that 99.5% may have an atypical response to at least one medicine. Nearly 24% had already been prescribed a medicine for which they were predicted to have an atypical response.^[1]

A US study projected the same pattern for 7,769,359 veterans who used VA pharmacies: 99% would carry at least one actionable variant, and 54.8% had received at least one medicine with CPIC level A guidance.^[2]

The exact share depends on which genes and medicines a study counts. It also depends on ancestry. In the UK Biobank study, people in non-European ancestry groups carried more variants predicted to disrupt how these genes work, and many of those variants aren’t captured by the current definitions labs use.^[1]

Three public sources come up throughout this roundup: CPIC, the FDA and the Dutch Pharmacogenetics Working Group. None of them tells you what to take. They tell prescribers how a genetic result can be used when a medicine is being considered.

• CPIC guidelines. The Clinical Pharmacogenetics Implementation Consortium lists 29 guidelines. They are written to help clinicians use genetic results they already have, rather than to decide whether a test should be ordered.^[3]
• CPIC levels. CPIC grades each gene and medicine pair. Only levels A and B have enough evidence for at least one prescribing action; levels C and D carry no prescribing recommendation. Of the 319 pairs in published guidelines, 93 are level A and 15 are level B.^[4]
• FDA drug labels. The FDA keeps a public table of medicines whose labeling includes pharmacogenomic information. It lists more than 400 medicines in 692 entries, and the labeling of some, but not all, of them sets out specific actions based on the biomarker.^[5]
• DPWG recommendations. The Dutch Pharmacogenetics Working Group publishes its own recommendations. The PREPARE trial used them.^[6]

CPIC’s highest level, level A, is spread across a fairly small set of genes, and a few of them, including CYP2C9, CYP2D6 and CYP2C19, each have several level A pairs.^[4] The table gives examples. It says only that guidance exists and what the gene makes. It isn’t advice about any medicine.

Knowing a genotype is one thing. Showing that people do better when prescribers use it takes trials. Here are four of the large ones.

Two things run through all four. The benefits found were modest, and not on every measure. And the tests differ: trials used different genes and different advice, which the pooled analysis named as a source of uncertainty.^[9]

• Guidance covers some medicines, not all. Most of the gene and medicine pairs in CPIC’s published guidelines are rated level C, which carries no prescribing recommendation.^[4]
• Some large insurers still treat multi-gene panels as unproven. UnitedHealthcare’s commercial policy, effective July 1, 2026, calls panels of five or more genes unproven and not medically necessary for any use.^[10] Aetna’s policy lists multi-gene pharmacogenetic panels as experimental, investigational or unproven.^[11]
• Ancestry gaps remain. Current definitions miss many variants carried by people of non-European ancestry,^[1] and 97.7% of PREPARE’s participants reported European, Mediterranean or Middle Eastern ethnicity.^[6]
• Genes are one factor among many. Other medicines, age, and how the liver and kidneys are working can all change how a person responds.
• The prescriber still decides. CPIC writes its guidelines for clinicians deciding how to use a result, not for patients choosing a medicine.^[3]

• Whether a medicine will work for you. A result describes how you are likely to process a medicine, not whether it will help.
• What to take. Choosing a medicine, and how much of it, stays with your prescriber.
• Anything about medicines without guidance. For many medicines there is none yet.
• Rare variants. A genotyping test looks for known variants chosen in advance, so a rare one can be missed.
• Your whole medical picture. Your other medicines, your health and your history matter too.

GenePGx is our pharmacogenomic test: one report from one saliva sample, with four views. Checks selected known variants in 15 genes, plus CYP2D6 copy number, by genotyping. It tests genes named in CPIC prescribing guidelines for 56 medicines (checked September 29, 2026). The lab checks them by genotyping, and the report is written to be read with the prescriber who manages your medicines. The trials above studied other tests, so their results aren’t a promise about this one. Our page on PGx testing explains, in plain words, what a DNA test for medications can and can’t tell you.

• Everyday medicines. Medicines many people take, for the heart and blood pressure, cholesterol, stomach acid and more.
• Pain medicines. Including some prescribed around an operation or dental work.
• Mental-health medicines. Medicines used for depression, anxiety and other conditions.
• Supplements. A separate set of variants, run alongside, studied in how the body handles some vitamins and supplements, for your information. It doesn’t measure your levels or tell you what to take.

The supplement section is never billed to insurance.

New to all this? Start with how pharmacogenomics works. Taking a mental-health medicine? Read what mental health DNA tests can and can’t tell you. To see whether a medicine has published guidance, try the Medication Check.