Semaglutide and NAION: 2.33 Relative, 0.16 Points Absolute
In March 2026, JAMA Ophthalmology published a target trial emulation reporting that veterans who started semaglutide had 2.33 times the risk of non-arteritic anterior ischaemic optic neuropathy compared with veterans who started an SGLT2 inhibitor.
The same paper reports the absolute difference: 0.16 percentage points.
Both numbers describe the same 173 events among 102,361 people, and which one a reader remembers determines whether they walk away alarmed or reassured.
A framing note. Semaglutide is an approved medicine, this is a study of patients, and semaglutide is not in the Peptra Labs catalogue. Nothing here is guidance about human use.
What the semaglutide cohort did
The work comes from the Veterans Health Administration, with authors from Stanford and the University of Utah, using nationwide data from March 2018 to March 2025.
The design is the same family we have covered repeatedly this month: an active-comparator, new-user target trial emulation. Eligible veterans had type 2 diabetes, were currently on metformin, and had never used a GLP-1 receptor agonist or an SGLT2 inhibitor. The exposure was starting semaglutide. The comparator was starting any SGLT2 inhibitor as second-line therapy.
Confounding was handled with overlap weighting, and the authors report the groups were well balanced afterwards: mean age 60.1 years, body mass index 37.8, HbA1c 7.0 percent, 85.5 percent male, 20.7 percent Black, 8.1 percent Hispanic and 61.9 percent non-Hispanic White.
The outcome was incident NAION identified from diagnostic codes. That last detail matters and we will come back to it.
The two ways to state the result
| Framing | Number |
|---|---|
| Hazard ratio | 2.33, CI 1.54 to 3.54, P below .001 |
| Incidence rate, semaglutide | 123 per 100,000 person-years |
| Incidence rate, SGLT2 inhibitor | 67 per 100,000 person-years |
| Weighted incidence, semaglutide | 0.29 percent |
| Weighted incidence, SGLT2 inhibitor | 0.13 percent |
| Average treatment effect | 0.16 percentage points |
11,478 veterans started semaglutide and 90,883 started an SGLT2 inhibitor. Median follow-up was 2.1 years, with a maximum of 7.5. Across the whole cohort, 173 NAION events occurred.
The authors’ own conclusion holds both halves together: semaglutide initiators had a two-fold NAION risk, while the absolute risk was low, and clinicians and patients should be counselled on a rare but evident increase.
That is a well-constructed sentence. A doubling of a rare event is still a rare event, and it is also still a doubling. Coverage that keeps only the ratio and coverage that keeps only the 0.16 both misrepresent the paper.
The detail that will decide how this ages
NAION here was identified using diagnostic codes, not adjudicated by an ophthalmologist reviewing each case.
That is the standard constraint of database work and it cuts in a specific direction with this outcome. NAION is uncommon, it is diagnosed by examination, and a patient already under closer follow-up is more likely to receive the code at all. If starting a newer, more discussed medication brings a patient into contact with the health system more often, some of a signal like this can be surveillance rather than disease.
With 173 events split across two arms, a modest amount of differential coding is enough to move a hazard ratio.
A paper that has been revised and is now being argued about
The publication record around this study is itself worth reporting, because it shows how a safety signal actually settles.
The original appeared on 1 March 2026. An erratum was published in the same journal on 1 June 2026. On 27 August 2026, JAMA Ophthalmology published a piece of correspondence titled “Data and Temporal Risk Concerns of Semaglutide and NAION”, together with a reply from the original authors.
We can confirm the erratum and the correspondence exist and are indexed. We have not been able to read either letter, both are behind the publisher’s paywall with no abstract released, and so we are not characterising what they argue. The titles indicate the dispute concerns the data and the timing of risk, and that is as far as anyone should go without the text.
The point stands without the content. Five months after publication, this result has been corrected once and formally questioned once. That is science working normally, and it is also a reason to treat a single hazard ratio as a stage in an argument rather than a conclusion.
Where it sits among everything else this month
The eye is the one organ system where this compound now has convergent evidence from three independent methods.
We reported six years of European reports where eye disorders drew a signal in EudraVigilance, with a reporting odds ratio of 2.67 against liraglutide and reasonably tight intervals. A systematic review and meta-analysis of ocular adverse events with semaglutide appeared in JAMA Ophthalmology in September 2025. And this cohort provides the event-based estimate that a reporting database cannot.
Spontaneous reports, a meta-analysis and a target trial emulation are three different instruments, and they agree on direction. That is a stronger position than any of the psychiatric or cognitive signals we have covered, where the instruments point different ways.
It also contrasts with a FAERS analysis of cardiac reporting, where the authors attributed a disproportionate signal to demographics and indication rather than toxicity, and with 764,115 matched pairs in serious mental illness, where the same class was associated with lower mortality. Different organs, different directions, same broad literature.
What this has to do with the catalogue
Nothing in this study is a compound we supply for human use, and the transferable lesson is about reading, not about products.
A hazard ratio without a base rate is uninterpretable. Tirzepatide and retatrutide have no equivalent ophthalmic dataset, and the reason is the same one we noted this week about post-marketing databases generally: these methods need years of prescribing before they produce anything, so the newest compounds are invisible to them. An absence of NAION data for the triple agonist is an absence of exposure, not evidence of safety.
For laboratories, our European research buyer guide covers procurement and documentation, everything is supplied on a research use only basis, and our ranking of the most-studied research peptides tracks the published record rather than the headlines.
References
- Heberer K, Bress AP, Cogill S, et al. New-Onset Nonarteritic Anterior Ischemic Optic Neuropathy and Initiators of Semaglutide in US Veterans With Type 2 Diabetes. JAMA Ophthalmology, 1 March 2026. doi 10.1001/jamaophthalmol.2025.6262. Erratum: JAMA Ophthalmology, 1 June 2026, 144(6):571
- Yu CH. Data and Temporal Risk Concerns of Semaglutide and NAION. JAMA Ophthalmology, 27 August 2026, and the authors’ reply, same issue
- Natividade GR, Spiazzi BF, Baumgarten MW, et al. Ocular Adverse Events With Semaglutide: A Systematic Review and Meta-Analysis. JAMA Ophthalmology, 1 September 2025
- Ammendolia I, Mondello C, Esposito E, et al. Psychiatric and eye disorders associated with use of glucagon-like peptide 1 receptor agonists (GLP-1 RAs). Naunyn-Schmiedeberg’s Archives of Pharmacology, 22 August 2026
The products referenced on this site are supplied for laboratory research use only. They are not medicines and are not intended for human or veterinary use. This article summarises published research for informational purposes and is not medical advice. Statements about third-party studies belong to their authors.
