Retina Specialists, Auckland
Drug safety · semaglutide, tirzepatide and related drugs

GLP-1 Drugs and NAION

Does semaglutide or tirzepatide cause optic nerve stroke (non-arteritic anterior ischaemic optic neuropathy, NAION)? This page sets out every published estimate: cohort studies, randomised trials and meta-analyses, with the absolute rates where reported. Every ratio compares people on a GLP-1 drug with people on another treatment; above 1 means more NAION on the GLP-1 drug. Every interval is a 95% confidence interval. Numbers in brackets point to the numbered sources at the foot of the page.

Non-arteritic anterior ischaemic optic neuropathy

Sudden painless loss of vision from poor blood flow to the front of the optic nerve. The concern began with a 2024 Boston cohort and has since been tested in national registers in Scandinavia, the United Kingdom and the United States.1

NAION, observational studies in diabetes
1.85times the rate
hazard ratio against other diabetes drugs
95% CI 1.20 to 2.85 · 8 studies pooled2
Extra NAION risk, observational studies in diabetes
1 in 12,800per year
people on the drug who have an extra NAION each year
1 in 3,700 per year on drug against 1 in 5,300 on comparator2
NAION, randomised trials in diabetes
1.76times the risk
risk ratio against placebo or other treatment
95% CI 0.43 to 7.25 · 5 against 1 cases, not significant2
Regulator
Very rare
up to 1 in 10,000
European Medicines Agency, 20253

Clinical takeaways

  1. NAION is probably a real but rare effect in type 2 diabetes. Most large cohort studies and every pooled observational analysis point the same way (relative risk about 1.4 to 2.8), and the European Medicines Agency lists NAION as a very rare side effect of semaglutide, affecting up to 1 in 10,000 users.3,2,4
  2. The absolute excess is small. About 8 extra cases per 100,000 person-years in the pooled diabetes data, or a number needed to harm of about 2,700 to 3,300 over 18 months to 2 years.2,5,6
  3. The risk window is from about 6 months to 2 years of treatment, not the first weeks, with no clear excess after that, and studies weighted by design quality show smaller effects (the US Veterans study with chart-checked codes found 1.35).7,8
  4. Evidence in obesity without diabetes is thin and inconsistent; the large early relative risk (7.6) came from a small referral cohort and has not been reproduced.1,9
  5. The main susceptibility is a small, crowded optic disc (cup-to-disc ratio 0.2 or less), but no study has yet measured whether GLP-1 drugs add more risk in such discs. Stopping or withholding the drug is not recommended on current evidence; a new NAION while on treatment warrants discussion with the prescriber.10,11

Every study estimate

Filled squares are single studies; open diamonds are pooled analyses. Hover or focus a row for detail and the comparator.

Show

Absolute risk as reported

Every study that reports a yearly event rate, on the drug (dark dot) and on the comparator (open dot), in events per 100,000 people per year on a log scale. The right-hand bars show the extra rate on the drug: the paper's reported difference where given, otherwise the drug rate minus the comparator rate. Red is more events on the drug, green is fewer. Studies reporting risk over 18 months, 3 years or 5 years were divided by the follow-up. Rates differ a hundredfold between studies because populations, case definitions and follow-up differ, so compare the two dots within each row. Hover a row for the figures as 1 in so many people per year.

Who seems more susceptible

  • A small, crowded optic disc: cup-to-disc ratio 0.2 or less carried 46 times the odds of NAION in 2,169 cases; disc shape was the same whether or not people were on a GLP-1 drug.10
  • Men, people under 50, smokers, and people whose HbA1c fell by 1% or more (UK primary care).15
  • Men and people over 50 in US insurance data.6
  • Semaglutide more than liraglutide; the signal for tirzepatide is weaker, and one study found less NAION on tirzepatide than on other GLP-1 drugs.39,42,30
  • Possible mechanism: a mouse and small human study found that GLP-1 drugs lower intracranial pressure and slow retinal blood flow, reversibly.43 Case reports also describe papillitis and paracentral acute middle maculopathy.44

Why the studies disagree

  • The comparator matters. SGLT2 inhibitors may themselves protect the optic nerve, which would enlarge the apparent GLP-1 effect.22,19
  • Diagnosis codes for NAION are unreliable. Chart review confirmed 74.5% and 84.7% of coded cases in two Veterans studies.14,8
  • Studies from Denmark and Norway report larger effects than those from the global record network, possibly from method rather than population.13,28
  • Randomised trials have too few events: 8 against 4 cases across 69 trials.32
  • Rapid improvement in blood sugar is itself a suspected trigger, as with diabetic retinopathy.15

Search and methods

Sources searched, in order

  • A personal Zotero reference library ("GLP-1" and "Research Review" collections). Items held there are marked Library in the list of sources.
  • PubMed, searched on 28 October 2026: 22 queries covering NAION and optic neuropathy, glaucoma and eye pressure, AMD, diabetic retinopathy and macular oedema, vascular occlusion, uveitis, cataract, dry eye, intracranial hypertension, trials, meta-analyses and pharmacovigilance, for all GLP-1 receptor agonists and tirzepatide. 300 records were fetched, 151 abstracts were read, and full text was opened for research letters without abstracts.
  • Not found: full text of several research letters (Fung 2025, Goldenberg 2025, Abboud 2026). No study has measured NAION risk by optic disc shape on GLP-1 drugs.

How figures are shown

  • Ratios are shown as published (hazard, risk, odds or rate ratios), oriented so that above 1 means more events on the GLP-1 drug.
  • No new pooling was done: the pooled rows are the published meta-analyses. Many overlap, as they draw on the same cohorts.

Gaps noted

  • No cohort has been stratified by disc at risk, and there is no validated NAION diagnosis code.
  • Suggested New Zealand projects: a disc OCT case-control study, validation of coded NAION, and an interrupted time series of NAION incidence by ethnicity.

Sources

Numbered in order of first citation down the page. Links open PubMed or the DOI.