GLP-1 drug side effects in FAERS: which adverse events are reported disproportionately
Between July 2016 and June 2026, FDA received 90,598 serious adverse event reports (after removing duplicates) naming one of six GLP-1 receptor agonists as a suspect drug: semaglutide, tirzepatide, dulaglutide, liraglutide, exenatide or lixisenatide. We asked ClinLit AI, which now includes FDA’s adverse event reports, labels and approvals, which reactions stand out in those reports compared with every other drug, and how that squares with the labels, the literature and the regulators. (For where these drugs are being tested next, see Beyond diabetes and obesity: where GLP-1 drugs are being tested in 2026.)
Key findings
- Delayed gastric emptying leads the class. “Impaired gastric emptying” is reported disproportionately for all six drugs, with proportional reporting ratios (PRRs) of 10 to more than 120. More than half of FAERS’s gastroparesis reports in the window name one of these six drugs as a suspect.
- Pancreatitis is a signal for every drug in the class, from a PRR of about 10 for semaglutide to about 20 for liraglutide and dulaglutide. Pancreatitis terms appear in 5% to 11% of each drug’s serious reports.
- Intestinal obstruction and ileus stand out for semaglutide, tirzepatide and dulaglutide (PRR 5.5 to 7.2), but not for exenatide or lixisenatide.
- Reports of ischaemic optic neuropathy (NAION) went from almost none to hundreds a year for semaglutide: 1 or 2 a year until 2023, then 91 in 2024, 414 in 2025 and 331 in the first half of 2026, after the first published study in 2024. The EMA concluded in June 2025 that NAION is a very rare side effect of semaglutide; FDA lists it as a signal it is evaluating, and no current US label in our data mentions it.
- Report counts track use and publicity, not risk. FAERS reports are self-reported and unchecked, so every number here is biased by who chooses to report and why. Tirzepatide, approved in 2022, already gets about as many serious reports a year as semaglutide. That says more about prescriptions and attention than about either drug.
What FAERS can and can’t tell you
FAERS collects reports from doctors, pharmacists, patients and drug makers, who must forward what they receive. For everyone but the drug makers, reporting is voluntary, and most adverse events are never reported at all. What does get reported is a biased sample: it depends on publicity, lawsuits, how new the drug is, and who reports and why. A report says a reaction happened while someone took a drug, as the reporter saw it. FDA doesn’t verify the reports or judge whether the drug caused the reaction, and nobody knows how many people took each drug, so counts are not rates and a signal is not a risk.
What FAERS can show is disproportionality: whether a reaction makes up a larger share of a drug’s reports than of all other drugs’ reports. The usual measure is the proportional reporting ratio (PRR). A PRR of 10 means the reaction is ten times as common among this drug’s reports as among the rest, not that the drug makes it ten times as likely. Signals are leads for review, and they are inflated by publicity, litigation, the first years on the market and the patients a drug is given to.
Our copy holds serious reports only (death, life-threatening, hospitalisation, disability, birth defect or another medically important event), so every comparison here is against other serious reports.
What stands out for GLP-1 drugs as a class
ClinLit AI pooled the six drugs and compared their reactions with all other serious reports. The strongest and most consistent group is slowed stomach and gut: impaired gastric emptying (a signal for all six, with PRRs of 10 to 128), gastrointestinal hypomotility and cyclic vomiting syndrome. Then come pancreatitis (signal for all six), the gallbladder (gallbladder injury, biliary colic, cholelithiasis, cholecystitis), medullary thyroid cancer (164 reports; the C-cell tumour risk is in the labels’ boxed warning), optic ischaemic neuropathy (996 reports) and ketoacidosis, including starvation ketoacidosis.
Per drug, the most disproportionate reactions are not always clinical. Exenatide’s top ten are almost all device and injection-site terms (device leakage, incorrect dose administered by device), which reflect its injection devices rather than its pharmacology. Liraglutide’s top ten (lung function tests, mite and dust allergy, pulmonary vasculitis) are, as ClinLit AI put it, a cluster that “looks like a single reporting source’s idiosyncratic coding rather than a plausible drug effect”.
Reproduce this with ClinLit AI: the class signals and each drug's top ten
We asked:
Which adverse events are reported disproportionately for GLP-1 receptor agonists in FAERS?
Ask ClinLit AI this question →
ClinLit AI defined the class by FDA’s pharmacologic class (“GLP-1 Receptor Agonist”) and listed the signals all six share, for example:
- Impaired Gastric Emptying — 6,817 reports in total; PRR 10.3–128
- Pancreatitis — 5,077; PRR 11.7–25.3; Pancreatitis Acute — 1,241; PRR 6.9–14
It added its own warning about what drives these numbers:
A high PRR can reflect publicity and litigation rather than pharmacology: gastroparesis terms surged after media coverage of GLP‑1 weight‑loss use, and the NAION cluster followed the 2024 published case series.
(The 6,817 counts a report once per drug, so reports naming two of the drugs count twice; 4,685 distinct reports name one of them, below.)
We asked:
In FAERS, for each of these GLP-1 receptor agonists: semaglutide, liraglutide, dulaglutide, exenatide, lixisenatide, and tirzepatide (a dual GIP/GLP-1 agonist), give the number of serious reports with the drug as a suspect, and its 10 most disproportionately reported reactions with at least 20 reports, ranked by the lower 95% bound of the PRR. Show reports, PRR and its lower bound for each.
Ask ClinLit AI this question →
On the more unusual entries it added:
The same applies to exenatide’s device/injection-site cluster (a pen-device artefact, not a pharmacological finding) and to liraglutide’s allergy/lung-function cluster, which looks like a single reporting source’s idiosyncratic coding rather than a plausible drug effect.
We asked:
How many distinct primary serious FAERS reports list at least one of semaglutide, liraglutide, dulaglutide, exenatide, lixisenatide or tirzepatide as a suspect drug, and what share is that of all primary serious reports with a suspect drug?
Ask ClinLit AI this question →
90,598 distinct primary serious reports in FAERS list at least one of the six GLP-1/GIP receptor agonists as a suspect drug.
Pancreatitis: a signal for all six
Counting 12 pancreatitis terms once per report (every pancreatitis term except those naming another cause, such as alcoholic, autoimmune, hereditary, infectious, traumatic or procedural), against reports of all other drugs:
| Drug | Serious reports with pancreatitis | Share of the drug’s serious reports | PRR (95% CI) |
|---|---|---|---|
| Liraglutide | 1,094 of 10,136 | 10.8% | 20.4 (19.2–21.6) |
| Dulaglutide | 1,586 of 15,051 | 10.5% | 20.1 (19.2–21.1) |
| Exenatide | 302 of 3,961 | 7.6% | 14.1 (12.6–15.7) |
| Tirzepatide | 1,766 of 25,374 | 7.0% | 13.4 (12.8–14.0) |
| Lixisenatide | 50 of 733 | 6.8% | 12.5 (9.6–16.4) |
| Semaglutide | 2,115 of 38,878 | 5.4% | 10.5 (10.1–11.0) |
Every label in the class already warns about acute pancreatitis (below). The order of the drugs shouldn’t be read as a ranking of risk: liraglutide has been on the market longest, the newer drugs are more often used for weight loss, and publicity has differed. The spread is compatible with differences in who reports, and for which patients, as much as with differences between the drugs.
Reproduce this with ClinLit AI: pancreatitis by drug
We asked:
In FAERS, for semaglutide, liraglutide, dulaglutide, exenatide, lixisenatide and tirzepatide, count the primary serious reports with the drug as a suspect that list any of these preferred terms, counting each report once: Pancreatitis, Pancreatitis Acute, Pancreatitis Chronic, Pancreatitis Necrotising, Pancreatitis Haemorrhagic, Haemorrhagic Necrotic Pancreatitis, Oedematous Pancreatitis, Pancreatitis Relapsing, Subacute Pancreatitis, Idiopathic Pancreatitis, Obstructive Pancreatitis, Ischaemic Pancreatitis. For each drug give those reports, its total primary suspect reports, the share, and the PRR with its 95% CI, comparing with primary reports that have a suspect drug but not this one.
Ask ClinLit AI this question →
It confirmed all 12 terms exist and showed the cells behind each ratio:
the 2×2 cells behind the ratios are, for example, semaglutide a=2,115, b=36,763, c=35,221, d=6,772,942.
Gastroparesis, ileus and intestinal obstruction
| Drug | Gastroparesis reports | PRR (95% lower bound) | Obstruction or ileus reports | PRR (95% lower bound) |
|---|---|---|---|---|
| Semaglutide | 3,498 | 126 (121) | 1,707 | 7.2 (6.9) |
| Dulaglutide | 1,532 | 102 (96) | 512 | 5.5 (5.0) |
| Tirzepatide | 1,281 | 49 (46) | 919 | 5.9 (5.5) |
| Liraglutide | 455 | 39 (35) | 175 | 2.8 (2.4) |
| Lixisenatide | 12 | 13.4 (7.6) | 6 | 1.3 (0.6) |
| Exenatide | 51 | 10.6 (8.0) | 35 | 1.4 (1.0) |
Gastroparesis here is impaired gastric emptying, diabetic gastroparesis and postoperative gastroparesis; obstruction is 14 intestinal obstruction and ileus terms. The gastroparesis ratios are extreme because these drugs dominate the term: 4,685 of FAERS’s 8,370 gastroparesis reports in the window (56%) name one of the six, 3,498 of them semaglutide. Reports rose sharply from 2023 (semaglutide: 85 in 2023, 641 in 2024, 1,843 in 2025), the pattern of stimulated reporting described under NAION below. For obstruction and ileus, exenatide and lixisenatide show no disproportionality.
The labels describe both. Each says the drug “is not recommended in patients with severe gastroparesis” (in a severe gastrointestinal adverse reactions warning), each states that the drug delays gastric emptying (in the pulmonary aspiration warning and the pharmacology section), and each lists ileus and intestinal obstruction among its postmarketing reports.
Reproduce this with ClinLit AI: gastroparesis and obstruction
We asked:
In FAERS, for semaglutide, liraglutide, dulaglutide, exenatide, lixisenatide and tirzepatide, count the primary serious reports with the drug as a suspect, each report once, for two concepts. Gastroparesis: Impaired Gastric Emptying, Diabetic Gastroparesis, Gastroparesis Postoperative. Intestinal obstruction or ileus: Intestinal Obstruction, Small Intestinal Obstruction, Large Intestinal Obstruction, Gastrointestinal Obstruction, Duodenal Obstruction, Ileus, Ileus Paralytic, Mechanical Ileus, Postoperative Ileus, Subileus, Gastric Ileus, Intestinal Pseudo-Obstruction, Colonic Pseudo-Obstruction, Distal Intestinal Obstruction Syndrome. For each drug and concept give the reports and the PRR with its 95% lower bound, comparing with primary reports that have a suspect drug but not this one. Also say how many of all primary reports with each concept list any of the six drugs.
Ask ClinLit AI this question →
It checked that all 17 terms exist, and answered the last part:
Gastroparesis: 8,370 primary reports in the whole suspect-drug universe; 4,685 of them (56%) list at least one of the six as a suspect drug
We asked:
In FAERS, how many serious reports of Impaired Gastric Emptying list semaglutide or tirzepatide as a suspect drug, by year received from 2019 to 2026?
Ask ClinLit AI this question →
We asked:
What do the current FDA labels of Ozempic, Wegovy, Mounjaro and Zepbound say about pancreatitis, delayed gastric emptying or gastroparesis, ileus, and ischaemic optic neuropathy? Quote the label sections.
NAION: from almost no reports to hundreds a year
Non-arteritic anterior ischaemic optic neuropathy (NAION) is a sudden loss of blood flow to the optic nerve. FAERS has no NAION term; it is coded as optic ischaemic neuropathy.
| Year received | Semaglutide | Tirzepatide | Liraglutide | Dulaglutide |
|---|---|---|---|---|
| 2019–2023 | 5 | 0 | 1 | 1 |
| 2024 | 91 | 4 | 1 | 0 |
| 2025 | 414 | 85 | 16 | 18 |
| 2026 (January–June) | 331 | 55 | 1 | 9 |
| PRR (95% lower bound) | 137 (126) | 22 (18) | 7.5 (4.9) | 6.7 (4.6) |
(Liraglutide also has one report each in 2017 and 2018.)
The jump follows the literature. The first study, a matched cohort from one neuro-ophthalmology centre, was published in JAMA Ophthalmology in 2024 and reported hazard ratios of 4.3 in type 2 diabetes and 7.6 in overweight or obesity (PMID 38958939). Larger registry and claims studies followed, mostly with smaller associations: a Danish-Norwegian cohort of people with type 2 diabetes found a hazard ratio of 2.8 against SGLT-2 inhibitors, about 1.4 extra cases per 10,000 person-years (PMID 40098249). A pooled analysis of the manufacturer’s placebo-controlled trials of semaglutide and liraglutide found no increase (3 confirmed cases among 64,917 treated participants against 5 cases in 4 of 31,912 participants on placebo; PMID 42049287). The North American Neuro-Ophthalmology Society and the American Academy of Ophthalmology concluded that the evidence, mostly retrospective, suggests at most a small increased risk, with a low absolute risk (PMID 42132708). ClinLit AI found 85 PubMed articles on GLP-1 drugs and NAION; none before August 2024.
Regulators: on 6 June 2025 the EMA’s safety committee concluded that NAION is a very rare side effect of semaglutide (up to 1 in 10,000 people) and recommended adding it to the product information (EMA). FDA lists NAION for the whole GLP-1 class among its potential signals, with the status “FDA is evaluating the need for regulatory action” as of 6 March 2025 (FDA). None of the current Ozempic, Wegovy, Mounjaro or Zepbound labels mentions it (as of October 2026).
The FAERS surge is a textbook case of stimulated reporting: an event that was rarely reported becomes common once it is in the news and the courts. The reports confirm that the question is being asked; they can’t answer it.
Reproduce this with ClinLit AI: NAION reports, studies and regulators
We asked:
In FAERS, how many serious reports of non-arteritic anterior ischaemic optic neuropathy (NAION) or ischaemic optic neuropathy list semaglutide, tirzepatide, liraglutide or dulaglutide as a suspect drug, by year received, and is it reported disproportionately for each? List the terms you used.
Ask ClinLit AI this question →
The step change is in 2024 for semaglutide and 2025 for the others — a pattern typical of publicity-driven reporting (the 2024 NAION papers and associated litigation), not of a change in underlying risk.
We asked:
Which PubMed articles study semaglutide or other GLP-1 receptor agonists and non-arteritic anterior ischaemic optic neuropathy (NAION)? List the earliest ones and the main study types, with PMIDs.
Ask ClinLit AI this question →
We asked:
Have FDA or the European Medicines Agency taken regulatory action on semaglutide and non-arteritic anterior ischaemic optic neuropathy (NAION), and when? Cite the official sources.
Report volumes follow the market
| Year received | Semaglutide (approved December 2017) | Tirzepatide (approved May 2022) |
|---|---|---|
| 2018 | 209 | 1 |
| 2020 | 1,162 | 4 |
| 2022 | 2,380 | 303 |
| 2023 | 4,166 | 1,735 |
| 2024 | 7,837 | 4,567 |
| 2025 | 12,734 | 11,601 |
| 2026 (January–June) | 7,869 | 7,154 |
Serious reports for tirzepatide reached semaglutide’s annual level within three years of approval. Fast uptake for weight loss and close media attention explain that better than any difference in safety. Tirzepatide’s few reports before its approval most likely come from clinical trials.
Reproduce this with ClinLit AI: reports by year
We asked:
In FAERS, how many serious reports per year list semaglutide and tirzepatide as suspect drugs, from 2018 to 2026, and how does that compare with the year each was first approved by the FDA?
How we counted
- Data: openFDA’s export of FAERS (21 October 2026), loaded into ClinLit AI with serious reports only, received 1 July 2016 to 30 June 2026: 7.78 million reports. Each report is in its latest version.
- Duplicates: the same case often reaches FDA from several senders. Reports with the same sex, age, country, event date, suspect drugs and reactions are grouped, and only one per group is counted (6.85 million reports remain with a suspect drug). The grouping is approximate, so some double counting remains.
- Drugs: each reported product is mapped to its active ingredient through FDA’s substance identifiers (UNII), using openFDA’s harmonised identifiers or, where a report lacks them, the reported active substance. Only reports naming the drug as a suspect (not a concomitant) drug count.
- Reactions: FAERS codes reactions as MedDRA preferred terms. Concepts such as pancreatitis or gastroparesis span several terms; ClinLit AI chose them with its reaction-term search and listed them in each answer.
- Statistics: the PRR compares the share of a drug’s reports that mention a reaction with the share among other drugs’ reports, with a 95% confidence interval from the usual log formula. A signal means PRR of at least 2, chi-squared of at least 4 and at least 3 reports.
- Define terms explicitly. When we left a concept open (“gastroparesis”), ClinLit AI chose different term lists in different runs; in one it left out impaired gastric emptying, the main term. The tables here come from questions that list every term, which you can paste as they are.
- Our cross-check: we recomputed the pancreatitis table with our own SQL. The report counts matched exactly and the PRRs to the second decimal (liraglutide 20.35 against ClinLit AI’s 20.36).
- The literature and regulators come from ClinLit AI’s PubMed search and its web search of official sites; we opened every cited article and page.
None of this is medical advice, and none of it shows that a drug caused a reaction. For what is known about a drug’s risks, read its label and the trials behind it.
ClinLit AI can make mistakes: check its answers against the sources it cites. Nothing here is medical, investment or regulatory advice.