A study published this month in Ophthalmology Retina looked for a link between GLP-1 drugs and non-arteritic anterior ischemic optic neuropathy, a sudden and usually permanent loss of vision in one eye. Across 19,505 adults with type 2 diabetes it found 29 cases in total, and could not rule out either a real effect or none at all. The headline comparison was not statistically significant. If you are taking one of these drugs, this study does not establish that it will damage your sight.
What the study actually measured
The researchers used United States commercial insurance claims from 2012 to 2024 and applied target trial emulation, a design that mimics the structure of a randomised trial using observational data. They restricted the analysis to new users and compared them against active comparators rather than against nobody: 9,213 people starting a GLP-1 receptor agonist against 10,292 starting an SGLT-2 inhibitor or a DPP-4 inhibitor. Propensity score methods balanced baseline demographic and clinical characteristics across the groups.
The adjusted hazard ratio for NAION among GLP-1 users was 1.87, with a 95% confidence interval of 0.85 to 4.12. That interval spans everything from a 15% reduction in risk to slightly more than a fourfold increase, which is another way of saying the data are consistent with the drug doing nothing. The reason the interval is so wide is the event count: 29 cases across 19,505 people, roughly one in 670 over the study period. Subgroup findings pointed to higher risk among liraglutide users within 12 and 18 months of follow-up, and among men and older adults.
How this fits what came before
NAION has been the most contested safety question attached to this drug class. Earlier work produced inconsistent results, and the authors say so directly, attributing the inconsistency to differences in why the drug was prescribed, what it was compared against, how long people were followed and who was in the study population. This paper is an attempt to fix several of those problems at once, which is what the active comparator and the new-user restriction are for: both reduce the chance that the sickest or most treatment-experienced patients end up loaded into one arm.
What it changes about the advice on this site is modest. We have covered a separate cohort linking GLP-1 drugs to more recorded diabetic retinopathy without more vision loss. This is a different condition with a different mechanism, and the two should not be merged. Neither result gives a reason to stop treatment. Both give a reason to mention any sudden change in vision to a prescriber promptly rather than at the next scheduled visit.
The central limitation is arithmetic. Twenty-nine events cannot settle a question like this, and a confidence interval running from 0.85 to 4.12 is not a null result so much as an absence of information. The subgroup signals for liraglutide, for men and for older adults come from slicing an already small number of events into smaller ones, which is exactly the circumstance in which a chance finding looks like a pattern. The cohort is people with type 2 diabetes in private United States health plans, so it says nothing directly about people taking these drugs for weight loss alone, who are younger on average and carry different baseline eye risk. The authors call for confirmatory work and for a mechanism, and neither exists yet.
Sources: Lee JJ, Brothers T, Kim LA, Ward KE, Yusuf H, Wen X. GLP1-RA Use and Risk of Non-arteritic Anterior Ischemic Optic Neuropathy in Patients with Type 2 Diabetes. Ophthalmology Retina, published online 3 September 2026. doi:10.1016/j.oret.2026.08.029.
