The Genetic Test That Isn't Ready

Tests promising to identify your best GLP-1 are being sold. A 2026 study in Pharmacogenetics and Genomics is titled GLP1R and OCT1 variants modulate semaglutide and metformin response in type 2 diabetes — and its own results report that the GLP1R signal in semaglutide users did not survive correction for multiple comparisons, in an arm of ten people.^[1] The distance between that title and those results is the whole subject of this page.

What the study did

Twenty-seven Bulgarian adults with type 2 diabetes and a BMI of 25 or above, averaging an HbA1c of 8.3%, were given either metformin XR 2000 mg (17 people) or oral semaglutide 14 mg (10 people) and followed for three months. Sanger sequencing genotyped three polymorphisms, and the endpoints were change in weight and change in HbA1c.

The gene list is sensible. GLP1R encodes the receptor semaglutide binds. SLC22A1 — also called OCT1 — and SLC47A1 encode transporters relevant to metformin. Large consortia have linked variants in all three to antidiabetic response, and the stated purpose was to see whether those effects show up in a small real-world clinic.

What it found

The clearest result has nothing to do with genetics. Semaglutide beat metformin for weight loss over three months — −6.5 ± 3.6 kg against −1.6 ± 2.5 kg (95% CI −7.6 to −2.2; P = 0.001), with BMI down 2.0 against 0.3 kg/m². That is a drug comparison, and it is the finding in this paper with the firmest support.

Read the first row against the title. The variant in the receptor these drugs actually bind produced trends that did not reach significance after correction, in ten people. Read the second row and note two things: the association that cleared the bar is about HDL cholesterol in metformin users — not weight, not semaglutide — and it cleared a bar the authors describe as exploratory.

What an exploratory 10% FDR means

When you test many things at once, some associations appear by chance. False-discovery-rate correction accounts for that by asking what proportion of your declared findings you are willing to have be false. The conventional setting is 5%. This analysis used 10% — meaning up to one in ten declared hits is expected to be spurious — and the OCT1 result came in at q = 0.063, which passes at 10% and would not pass at 5%.

A finding that clears a loosened bar by a small margin is a reason to run a bigger study, which is exactly what the authors say.

To their credit they say it plainly: the HDL signal warrants replication, and the data are hypothesis-generating. Nothing in the paper oversells the result. The overselling risk sits in the title, and titles are what circulate.

Why this is not a refutation either

So the claim this particular study supports is narrow: it did not detect an association, in an arm of ten. The authors were right that the consortium data were a reason to keep looking — and a much larger study has since looked and found something. What it found is genuine and small: variants that shift expected weight loss by under a kilogram per copy, and side-effect models that do not reach the usual threshold for a clinical tool. Real associations, not yet a useful test.

What to ask someone selling you a test

A test marketed as identifying your optimal GLP-1 is still making a claim that runs ahead of the evidence, though for a different reason than this pilot suggested. The associations exist. What does not yet exist is an effect large enough to change a decision — and a variant that shifts expected weight loss by under a kilogram cannot tell you which drug to take.

• Which published study supports the specific recommendation being made to me — not the gene, the recommendation.
• How many people were in it, and how many were on the drug I am asking about.
• Did the finding survive correction for multiple comparisons, and at what threshold?
• Was it replicated in a separate group of people?

If the answer is a mechanism story rather than an outcome study, that is the answer. What does predict response, as far as anyone has shown, is early response itself — the first weeks on the drug carry real information about the year ahead, covered in losing fast early. It is a cruder instrument than a gene panel, and it is the one with evidence behind it.