GLP-1 and Diarrhoea: the Label Says 30%, the Placebo Arm Says 16%, and the Three Trials Behind It Add Up Exactly

Ask an AI "does Ozempic cause diarrhoea?" and you will usually get a percentage from the label and nothing underneath it. For diarrhoea the number underneath is unusually large: in the same table, on the same trials, 16 in every 100 people on placebo reported diarrhoea too — the joint-highest placebo figure anywhere in the label's main adverse-reactions table, tied with nausea.

That does not make the drug effect small. It is one of the biggest and most consistent in this series: in every placebo-controlled trial arm of more than 300 people we could read, the drug arm reports diarrhoea at between about 1.4 and 3.0 times its own placebo arm. But it does mean the headline number on its own is not the answer to the question people are actually asking.

This is the twelfth piece in our evidence series, after which effects have enough evidence to calibrate a model, the corpus-derived pooled rates, what patients themselves report, gallstones end to end, what is inside a personalised risk number, hair loss end to end, why trial vomiting rates run from 6.5% to 58%, constipation read against the placebo arm, headache read against the placebo arm, abdominal pain, where the label and the registry disagree by a factor of two and nausea, where the label's 44% holds up and the largest trial reads 18%. Like those, every figure below is quoted from a primary source with the arm and the number of participants it belongs to, and every one of them was read on 18 September 2026 — from the current US prescribing information, from each trial's own results table on ClinicalTrials.gov, or from our own live production API.

What the label says, and how it reconciles

The current US prescribing information for Wegovy (semaglutide), revised June 2026, lists diarrhoea among the *"most common adverse reactions (incidence ≥5%) in adults or pediatric patients aged 12 years and older"*. Its adverse-reaction tables each give a drug column and a placebo column:

(Quoted from the prescribing information via DailyMed, SPL version 19, effective 18 June 2026, accessed 18 September 2026.)

Two things about the Table 3 row are worth stating precisely, because this series has spent several articles explaining why a label figure often cannot be compared with a registry one.

First, the diarrhoea row groups nothing. In Table 3 the abdominal pain row carries a footnote grouping seven separate adverse-event terms, fatigue groups two and gastritis groups four. Diarrhoea, like nausea, carries no footnote marker at all — so it is the single MedDRA term, directly comparable with the "Diarrhoea" row a trial registry posts.

Second, it reconciles. The label does not name the three trials it pools, but their arm sizes identify them exactly — 2,116 on the drug and 1,261 on placebo is STEP 1 (1,306 and 655) plus STEP 3 (407 and 204) plus STEP 2's 2.4 mg pair (403 and 402), and the label's own hypoglycaemia footnote names its "Study 3" at N=403 and N=402. Add those three trials' own diarrhoea rows as their registry records post them:

The label prints 30% and 16%. Both round exactly. That is a stronger reconciliation than we managed for nausea (44.8% against a printed 44%) and much stronger than abdominal pain, where the grouped row reads 20% against a single-term 9.7%.

Diarrhoea does not appear in Table 4, the 7.2 mg table. That table's own caption states its inclusion rule — *"Adverse Reactions (2% and Greater Than WEGOVY 2.4 mg and Placebo)"* — so its absence says only that diarrhoea at 7.2 mg did not clear that bar. The label publishes no 7.2 mg diarrhoea figure, and we are not going to infer one.

The label also gives the discontinuation figures. Across the placebo-controlled trials, *"6.8% of patients treated with 2.4 mg WEGOVY injection and 3.2% of patients treated with placebo permanently discontinued treatment as a result of adverse reactions"*, and among the reasons, *"diarrhea (0.7% versus 0.1%)"*. So diarrhoea is common and is rarely by itself the reason people stop.

What the trial registries say, arm by arm

Our database cited 23 distinct sources for a stated diarrhoea rate when this article was written at 00:54 BST on 18 September 2026 — 21 trial-registry records and 2 PubMed papers. Eleven of the 21 registry records post a placebo arm of at least 20 people. Every row below was read from that trial's own ClinicalTrials.gov adverse-events module on 18 September 2026. The counts are the registry's; the percentages and the differences are our division and subtraction of them.

The other ten registry records are absent from this table for three different reasons. Three are placebo-controlled but split their placebo across parts of 2 to 15 people, too few to read a difference from: a dose-escalation study in Chinese participants (NCT06023095, placebo arm of 4, and the two drug cohorts of 10 report 60.0% and 70.0%), and two ascending-dose studies of orforglipron, in healthy volunteers (NCT03929744, placebo parts of 8, 15 and 2) and in Japanese participants with type 2 diabetes (NCT05086445, placebo parts of 4 and 11). Four post no comparator at all: the long-term safety study ACHIEVE-J (NCT06010004, 9.1%, 11.9% and 11.9% across its three doses), a multiple-dose formulation study (NCT05841238, 18 single-week dosing periods from 0.0% to 8.3%), a tablet-versus-capsule study in 533 adults (NCT06440980, 38 dose-level arms from 0.0% to 12.7%) and a single-dose renal-impairment study (NCT05936138, 0/10, 0/6 and 1/8). And three compare a GLP-1 against another GLP-1 rather than against placebo — they have their own section below.

The difference, and how consistent it is

Of the 25 drug arms in the table above, 24 report more diarrhoea than their own placebo arm. The single exception is retatrutide's lowest dose (1 mg, 6/69 = 8.7%, against 8/70 = 11.4% on placebo) in a Phase 2 trial where a handful of people moves a percentage by double digits.

Restrict to the arms large enough to read — every drug arm of more than 300 people — and the picture is tighter still:

The drug arm sits between about 1.4 and 3.0 times its own placebo arm: lowest is ATTAIN-2's 6 mg orforglipron (21.3% against 15.0%, a ratio of 1.42), highest is SURMOUNT-1's 15 mg tirzepatide (23.7% against 7.9%, a ratio of 2.98).

In absolute terms the gap runs from +6.3 to +15.8 percentage points, and every one of those ten arms is positive.

That is a different shape from the two effects this series read most recently. For headache, every large arm sat within about three points of its placebo and six of nine sources had at least one arm at or below it. For abdominal pain, the gap in large arms was 1.5 to 4.2 points. Diarrhoea's gap is three to five times that, and it never reverses in a large arm. If you want one sentence: diarrhoea is a real and reasonably consistent drug effect, and it is also something that happens to a lot of people who are not on the drug.

Both halves matter for a reader deciding what to expect. Reading Table 3 as a difference rather than a rate: of every 30 people on semaglutide 2.4 mg who reported diarrhoea in those trials, about 16 would have reported it on placebo, and about 14 are what the drug added.

One more thing the table shows, and we are flagging it rather than explaining it: both adolescent datasets sit at the narrow end — 22% against 19% in the label's Table 5 (N=133 and N=67) and 22.7% against 16.7% in the oral-semaglutide adolescent trial (n=66 and n=66), gaps of 3 and 6 points, against roughly 9 to 16 points in the adult weight-management trials. They are not the narrowest rows here — retatrutide's 1 mg arm is 2.7 points *below* its placebo and the Japanese orforglipron trial's 12 mg arm is 3.7 above — and both adolescent placebo arms are small, the two use different formulations, and nothing we have read supports a conclusion about age. We are printing it because it is in the data.

The same molecule, the same dose, 31.5% and 10.6%

The two biggest semaglutide trials in our base both run the drug at the same 2.4 mg maintenance dose, and they disagree by a factor of three.

STEP 1 posts diarrhoea at 411 of 1,306 (31.5%) against 104 of 655 (15.9%) on placebo. SELECT — at 17,604 people, the largest trial in our corpus — posts 933 of 8,803 (10.6%) against 353 of 8,801 (4.0%). Same molecule, same maintenance dose, both placebo-controlled, both posted by the sponsor in the same registry.

The registry itself supplies a large part of the reason, in SELECT's own adverse-events note: *"AEs are reported based on data from in-trial observation period. Not all the AEs were collected systematically and more details on the approach followed for collecting the AEs is mentioned in the section 9.2 of protocol."* STEP 1's note says the opposite kind of thing — *"All AEs mentioned here are TEAE defined as an event that had onset date (or increase in severity) on or after the first day of exposure to randomized treatment and no later than the date of last dose + 7 weeks"* — a systematic treatment-emergent collection over a 75-week window. A cardiovascular-outcomes trial running up to 240 weeks with non-systematic AE collection and a weight-management trial with systematic collection over 68 weeks are not measuring the same quantity, and the placebo columns show it as clearly as the drug columns: 4.0% against 15.9%.

This is why we publish an interval and a source count beside every pooled figure rather than a single number. It is also the same finding the nausea article reported for its own effect (44% on the label, 18.1% in SELECT), which makes it a property of the trials rather than of the symptom.

Where the comparator is another GLP-1

Three records in our base compare one GLP-1 with another rather than with placebo. They are not in the placebo table, and they answer a question the label cannot.

Read them together with the placebo table and the same ordering keeps appearing: in a head-to-head, the newer or higher-exposure agent reports more diarrhoea than the older or lower-dose comparator, by 4 to 10 points. Note what these rows are *not*: none is a placebo comparison, so none of these percentages is a drug-attributable rate. Both arms of each row are active GLP-1 treatment, and both sit well above the placebo arms in the table above.

Our own pooled estimate, and why it is not the label's number

Our diarrhoea endpoint published a pooled clinical estimate of 17% (95% CI 5–42%), from 126 stated clinical rates across 23 distinct studies, source diversity "high", when this article was written at 00:54 BST on 18 September 2026. That is the second-highest pooled estimate of the fifteen effects we publish, after nausea at 22.5%, and it is well below the label's 30% for the same reason our nausea, constipation and headache figures are below theirs: the pool is the whole class, not one drug at one dose in one population.

Our pooled figure weights each source by its own stated sample size (a source stating none counts as 1), multiplied by a discount for how confident our extraction was in reading the figure, after collapsing each study to one entry — the unweighted mean of its posted drug arms, carried at the size of its largest arm. Which means the pool is dominated by a handful of entries and dragged in both directions by small ones:

The two heaviest entries pull it down and up at once. SELECT enters at 10.6% with n=8,803 — the single largest weight in the base, and, per the section above, the least systematically collected. SURPASS-CVOT enters at 21.8% (the mean of its tirzepatide and dulaglutide arms) with n=6,647.

Then the weight-management trials: STEP 1 at 31.5% (n=1,306), SURMOUNT-1 at 22.2% (n=636), ACHIEVE-3 at 19.1% (n=426), STEP 2 at 21.6% (n=403), ATTAIN-2 at 24.5% (n=331).

Five Phase 1 orforglipron pharmacology studies sit at the bottom — 0.7%, 2.5%, 3.5%, 3.8% and 4.2% — dosing healthy volunteers or small diabetic cohorts for days or weeks rather than the 40 to 72 weeks of the outcome trials.

And a sixth Phase 1 orforglipron study sits at the top, not the bottom: a Chinese dose-escalation study whose two 10-person drug cohorts report 60.0% and 70.0% enters at 65%, the highest single entry in the base, carrying a weight of 10.

If you want the Wegovy 2.4 mg number, quote the label: 30% against 16% on placebo, N=2,116 and N=1,261, Table 3. If you want the whole class with its heterogeneity visible, quote ours, with its interval and its source count. Neither is the other, and the 5–42% interval is the honest statement of how much those trials disagree.

This base will grow within hours of publication, and we would rather say so than let you find out. STEP 3 was added to our pinned-trial list on 17 September, after the nausea article found that a trial behind the label's own pooled figure had never been collected. Its diarrhoea rows — the 147/407 and 45/204 quoted in the reconciliation above — enter the corpus at the next collection run, so the 126 rates / 23 studies above is a reading with a known, dated expiry. The live endpoint is always the current answer.

What the FDA's adverse-event reports show

A different kind of number sits outside the incidence pool on purpose. FDA FAERS publishes the share of *spontaneous adverse-event reports* for each drug that mention diarrhoea, as read from our own endpoint on 18 September 2026: 6,516 of 82,377 semaglutide reports (7.9%), 6,720 of 122,084 tirzepatide reports (5.5%), 3,345 of 43,057 liraglutide reports (7.8%), 2,743 of 94,688 exenatide reports (2.9%) and 62 of 1,154 orforglipron reports (5.4%). A share of complaints is not a share of patients — it answers "of the people who filed a report about this drug, how many mentioned diarrhoea," not "how many people on the drug got it" — so our eligibility rules report these separately rather than average them into any estimate.

What patients themselves report

Of the 26 distinct community reports in our corpus, 9 mention diarrhoea — a 34.6% reporting frequency. That is the third-highest of the fifteen effects, behind vomiting (11 of 26) and nausea (10 of 26). It is not an incidence rate and we never treat it as one: it is the share of people who chose to post about their GLP-1 experience who mentioned diarrhoea at all, over a pool of 26. See that article for why the pool is small and cannot currently grow.

What we checked before publishing, and did not change

Reading every source in a base before quoting it is the part of this series that keeps finding things. This time it found nothing to withdraw, and two things a reader should know.

The two paper sources both survive, one of them only because a gate caught it earlier. SURPASS-2 (PMID 34170647) states in its abstract: *"nausea, 17 to 22% and 18%; diarrhea, 13 to 16% and 12%; and vomiting, 6 to 10% and 8%, respectively"* — tirzepatide first, semaglutide second. Our store holds two rows for that paper: the tirzepatide row is withheld with the reason "range_midpoint", because 13-to-16% is a range and averaging it to 14.5% would be inventing a figure the paper never states, and the semaglutide row carries the verbatim 12%. That is the write-time rate gate doing exactly what it was built for after it caught this very paper in September.

The second paper is a narrative review, not a trial, and we have an open question about it. PMID 41445996 is a 2025 perioperative review in *Cureus* whose abstract restates SUSTAIN-3's figures — *"GI adverse events including nausea (22%), diarrhea (11%), and vomiting (7%)"*. The 11% is genuinely stated and correctly attributed to semaglutide, but the source is a secondary account of a trial that is not otherwise in our corpus, and it currently counts as one of the "23 distinct studies" above. Whether a secondary source should count as a distinct study at all is a methodology question we raised on 17 September and have not yet settled; when it is settled, the count will move and we will say so.

Nothing in the registry half of the base needed correcting. Every one of the 21 registry records was re-read from its live results section for this article, and in each case the figure our API publishes for that source is exactly the unweighted mean of the non-placebo arms the registry posts today, carried at the size of its largest arm — SURMOUNT-1's 22.2% is the mean of 20.3, 22.5 and 23.7; the Phase 1 healthy-volunteer study's 0.7% is the mean of its 21 drug arms, 19 of which report zero. No stored arm figure disagreed with the record.

If you are on a GLP-1 now

This is educational content, not medical advice. The label lists diarrhoea among the most common adverse reactions, at roughly twice the placebo rate in the trials it pools, and rarely as the reason people stop (0.7% against 0.1%). Two things the patient information says are worth knowing rather than paraphrasing: it warns that *"Diarrhea, nausea, and vomiting may cause a loss of fluids (dehydration) which may cause kidney problems"*, and it instructs patients to *"Tell your healthcare provider right away if you have nausea, vomiting, or diarrhea that does not go away."* The label separately notes post-marketing reports of acute kidney injury *"in some cases requiring hemodialysis"*, mostly in patients who had become dehydrated from gastrointestinal reactions. Diarrhoea that will not settle, or that comes with dizziness, reduced urination or an inability to keep fluids down, is a reason to contact a clinician rather than to wait it out. Nothing here should be used to start, stop or change a medication or its dose without your prescriber.

Every figure in this article was read on 18 September 2026 from the cited primary source or from our live production API, and is stated as of that date. Query everything behind it — all 15 effects, both tracks, every source with its URL — at magistra.health/en/data-api, with the full eligibility methodology at magistra.health/en/methodology.