GLP-1 and Pancreatitis: Count Every Term the Registry Posts and the Placebo Arm Reports It More Often Than the Drug in Both Controlled Trials We Hold

Pancreatitis is the rarest effect we publish a number for — 0.1%, against 1.7% for the next-rarest — and the only one carrying an instruction: the label tells you to stop the drug and call someone. That makes it the effect where a number matters most and where we have the least right to state one casually.

Four numbers are in circulation, all of them real, none of them comparable:

The fourth is the one nobody prints, and this article is mostly about it.

This is the sixteenth 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, headache, abdominal pain, nausea, diarrhoea, fatigue, dizziness, acid reflux and what a pooled rate is made of. Every figure below was read on 19 September 2026 from the current US prescribing information, from each trial's own results record on ClinicalTrials.gov, from the openFDA endpoint, or from our live production API.

What the label says, and where it actually says it

The current US prescribing information for Wegovy (semaglutide), revised June 2026, handles pancreatitis in three separate places, and only one of them contains a number.

Section 5.2, *Acute Pancreatitis*, is a warning with no figure. It states that acute pancreatitis, *"including fatal and non-fatal hemorrhagic or necrotizing pancreatitis, has been observed in patients treated with GLP-1 receptor agonists, including WEGOVY"*, and instructs prescribers to observe patients for persistent or severe abdominal pain radiating to the back and to discontinue if pancreatitis is suspected. No rate, no count, no n.

Section 6.1, under *Other Adverse Reactions in Adults and/or Pediatric Patients*, is where the number lives: *"In WEGOVY clinical trials in adults for weight reduction, acute pancreatitis was confirmed by adjudication in 4 WEGOVY-treated patients (0.2 cases per 100 patient years) and 1 in placebo-treated patients (less than 0.1 cases per 100 patient years). One additional case of acute pancreatitis was confirmed in a patient treated with WEGOVY in another clinical trial."*

Two properties of that sentence decide everything downstream. It is adjudicated — a committee reviewed candidate cases and confirmed four — which is a stricter filter than an investigator typing a MedDRA term into a case report form. And its denominator is patient-years of exposure, not patients: 0.2 per 100 patient-years is a rate per year of treatment, so a person on the drug for two years contributes twice what a person on it for one year does. You cannot convert it into a percentage of patients without knowing total exposure, and the label does not print that.

Pancreatitis appears in none of the label's adverse-reaction tables. Table 3 lists adverse reactions at *“≥2% and Greater Than Placebo”* in adults with obesity or overweight, Table 4 at *“2% and Greater Than WEGOVY 2.4 mg and Placebo”* in the 7.2 mg trial, and Table 5 at *“≥3% and Greater than Placebo”* in patients aged 12 and older. Pancreatitis clears none of those bars, in any population, which is itself a published fact about how rare it is.

Section 6.2, *Postmarketing Experience*, lists *"acute pancreatitis and necrotizing pancreatitis, sometimes resulting in death"* among gastrointestinal disorders reported after approval — under the standard preamble that *"because these reactions are reported voluntarily from a population of uncertain size, it is not always possible to reliably estimate their frequency or establish a causal relationship to drug exposure."* Hold that sentence; it is the regulator pre-emptively explaining why the fourth number in our opening table is not a rate.

*A correction of our own, made while writing this:* until today every surface of ours that carried the 0.2-per-100-patient-years figure cited it to "section 5". The figure is right and unchanged; the pointer was wrong, because in September we quoted the sentence correctly and attributed it to the section heading it shares rather than to the section we found it in. It is recorded as item 41 of our published methodology, now at v5.30.

What the registry says when you count every term

Our pancreatitis base holds three trials. Two of them randomised people to placebo, and both post their pancreatitis rows in the serious adverse-event table, which is where an event this severe belongs.

The first thing you meet there is that "pancreatitis" is not one row. MedDRA splits it, and SELECT (NCT03574597, 17,604 participants, adverse events reported from randomisation up to 240 weeks) posts six separate pancreatitis terms:

SURMOUNT-1 (NCT04184622, 2,539 participants, baseline to week 193) posts three:

The third trial, ACHIEVE-J (NCT06010004, 401 participants with type 2 diabetes, 54 weeks), has no placebo arm — three active orforglipron doses only — and posts one pancreatitis term, in its non-serious table rather than its serious one: 0 of 132 on 3 mg, 0 of 135 on 12 mg, 1 of 134 on 36 mg.

So in both of the placebo-controlled trials we hold, counting every pancreatitis term the registry posts, the placebo arm reports the event slightly more often than the drug arms — 0.31% against 0.23% in SELECT, 0.31% against 0.21% in SURMOUNT-1. Two caveats stated rather than buried: summing across terms can count one participant twice if they were coded under two of them, so these sums are upper bounds on affected people; and at four, twenty or twenty-seven events these differences are well inside what chance produces, in either direction. That is the point. Neither trial, individually, is powered to resolve a difference at this event rate, and neither of them says the drug is safe.

What our own number is, and the three things it leaves out

Our live API publishes, for pancreatitis: a pooled clinical estimate of 0.1%, 95% CI 0–1%, from 11 stated rates across 3 distinct studies (5 source entries after collapsing), source diversity graded low, with the disclosure that 2 of the 3 sources are serious adverse-event rates and therefore a floor on all-cause incidence. That is an honest figure and it is also a narrow one. Three things it does not include, each of which this article is the first place we have set out:

It counts two of the six terms. Our collector tracks a fixed list of MedDRA terms, and for this effect that list is *Pancreatitis* and *Pancreatitis acute*. In SELECT that captures 14 of the 20 affected-participant entries on semaglutide; obstructive, oedematous, traumatic and relapsing pancreatitis are in the corpus's blind spot. Our number is therefore low by construction, and we would rather say so than widen the term list on a hunch in the same week we publish it.

It has no placebo comparator, by design. Placebo-arm rows are withheld from every estimate we publish — a drug's rate should not be pooled with a sugar pill's — so the 0.1% is composed of drug arms only. For nausea, where trial arms run 18% to 44% against placebo in the teens, that convention costs a reader little. For an event at two or three per thousand it is the difference between a number that means something and a number that means nothing, which is why the placebo columns above are printed in full.

Its weight sits almost entirely on one trial. The composition field on the same endpoint says tirzepatide supplies 6 of the 11 rows and orforglipron 3, but semaglutide's 2 rows carry 92.6% of the pooled figure — because weight follows stated sample size and SELECT's arm is 8,803 people against SURMOUNT-1's 630. The 0.1% is, to a first approximation, SELECT's two tracked terms and nothing else. (We published that composition as an API field yesterday precisely so a reader does not have to take our word for which trial is moving a number.)

For completeness: the same endpoint reports 63 corpus data points filed under pancreatitis, of which 12 clinical rows state no rate at all, 3 are April-2026 seed rows the eligibility rules exclude, and 2 are FAERS shares — which brings us to the last number.

Why 4.5% is not a rate, in the regulator's own words

Our corpus holds two FDA FAERS rows under pancreatitis: PANCREATITIS reported 2,343 times for liraglutide (5.4%) and 2,011 times for exenatide (2.1%), captured on 7 and 12 April 2026. They are excluded from every estimate and published in a separate block, labelled as shares of spontaneous reports rather than incidence.

That labelling was not specific enough, and we fixed it today. The denominator behind those percentages is not the number of reports for the drug. The collector reads openFDA's term-ranking endpoint, keeps a drug's 30 most-reported reaction terms, and divides by the sum of those 30 counts — so a single report naming nausea, vomiting and pancreatitis is counted under each of the three, and every term outside the top 30 is in neither numerator nor denominator. Re-queried live on 19 September 2026: liraglutide's top-30 counts sum to 52,864, of which PANCREATITIS is 2,356 — a 4.5% share — while openFDA returns 50,705 reports for liraglutide in total. Two quantities of similar size that are not the same thing. From today the API carries a `spontaneousReportSharesNote` saying exactly that beside the figures.

Even computed correctly, the share is not a rate, for the reason the Wegovy label itself gives about its own postmarketing list: the reports come *"from a population of uncertain size"*, so there is no denominator of exposed patients anywhere in the calculation. A drug taken by millions with an actively litigated adverse event will generate a high share. What FAERS legitimately tells you is that pancreatitis ranks fifth among all reaction terms ever reported for liraglutide — a statement about reporting, which is worth knowing and is not worth confusing with risk.

What we will not claim

Not this: that GLP-1 drugs do not cause pancreatitis. Nothing above supports it. Two trials each counting fewer than thirty events cannot rule out a real effect of the size regulators are worried about; the FDA's adjudicated analysis of the weight-reduction programme points the other way (4 cases against 1); pancreatitis carries a warning section in the label and a postmarketing entry that includes fatal cases; and adjudication and investigator-coded MedDRA terms are different instruments that will not agree.

What the evidence in our own base does support is narrower and, we think, more useful: our published 0.1% should never be read as "the risk of pancreatitis on a GLP-1", because the two controlled trials it mostly consists of report the same event at least as often in people taking nothing. Anyone quoting our pancreatitis figure without the placebo columns beside it is quoting a number we can tell them is not distinguishable from background. That disclosure is the product.

If you are on a GLP-1 now

This is educational content, not medical advice, and none of it describes what any individual should expect or do. Pancreatitis is rare in every source above — single-digit counts in trials of thousands — and it is also the one effect on our list where the label's instruction is unambiguous: severe abdominal pain that persists, that may radiate to the back, and that may or may not come with nausea or vomiting is a reason to stop the medication and contact a clinician or emergency service straight away, not to wait and see. The label's advice does not depend on which of the four numbers above you find most convincing, and neither should yours. Nothing here should be used to start, stop or change a medication or its dose without your prescriber.

Reproduce it

Every figure was read on 19 September 2026 and is stated as of that date; the pooled figure moves with each collection run, so verify at the endpoint before citing it. The full dataset — 15 effects, both tracks, every source with its URL and stated sample size, CC BY 4.0 — is at magistra.health/en/data-api, with the eligibility methodology at magistra.health/en/methodology.