The Supplement That Missed Its Endpoint

A gut bacterium sold in capsules, Akkermansia muciniphila, is widely marketed on the claim that it raises your natural GLP-1. A 142-person randomized trial tested it properly and missed its primary endpoint.^[1] Everything quotable in the paper comes from analyses run afterward — and that sequence is the most reliably misrepresented shape in all of clinical research.

What was tested and what happened

Across sites in Ireland and Germany, 142 adults with metabolic syndrome, some with prediabetes, took daily capsules containing 30 billion pasteurized A. muciniphila cells or placebo for four months, double-blind. The question set in advance was whole-body insulin sensitivity, measured by the Matsuda index, analyzed across everyone randomized.

That endpoint did not differ from placebo.

A prespecified primary endpoint is the question a trial commits to before it can see the answer. It exists precisely so that a study cannot go looking through its data afterward for something that worked. When it comes back null, that is the trial’s result.

What came afterward

The exploratory analyses reported several things: at three months, a liver-specific insulin sensitivity measure improved in participants with prediabetes (12%, p = 0.05) and in those aged 63 and over (p = 0.05), with a larger GLP-1 rise after a glucose drink (p < 0.01).

Then the analysis that carries most of the marketing: participants whose guts contained low levels of Akkermansia to begin with showed improved insulin sensitivity and GLP-1 response (p < 0.05), reduced trunk fat (p < 0.05), and weight reduction at p = 0.06.

Why the subgroup is unusable even if it is real

Suppose the low-Akkermansia finding is genuine — that people short of this bacterium benefit from taking it, and people who already have plenty do not. That is biologically sensible, and the authors say as much.

It still does not help anyone buying a capsule, for a plain reason: the subgroup was identified by deep metagenomic sequencing of participants’ gut microbiota. Nobody purchasing a supplement knows their Akkermansia gene count. A benefit available only to a group you cannot identify yourself is not a benefit you can act on.

And the group was defined using data gathered during the trial, which is what makes it exploratory rather than a result. Splitting a study by a measurement taken inside it, after the main endpoint has failed, is the standard route to a finding that evaporates on replication.

The GLP-1 claim, specifically

The finding is that GLP-1 rose more after a glucose drink in certain subgroups. That is a real measurement and it is a very long way from what the marketing implies.

Semaglutide and tirzepatide are receptor agonists given at doses that hold receptor activation far above anything a meal produces, continuously, for weeks — which is why they change appetite and body weight. A modestly larger natural GLP-1 excursion after a glucose load, in a subgroup, is a different order of thing entirely. The trial measured it and did not find significant weight loss even in the subgroup where it appeared.

The questions to ask about any supplement citing a trial

• Did the trial meet its primary endpoint? If the marketing does not say, assume it did not and check.
• Are the quoted findings from subgroups? And were those subgroups chosen before the data came in, or after?
• Could you tell whether you are in the subgroup? A benefit confined to a group identified by laboratory sequencing is not one you can select yourself into.
• What are the actual P values? A cluster sitting on 0.05 and 0.06, after a failed primary, is the shape of a finding that will not survive.