A new Bayesian meta-analysis in Neurology ↗ put a number on the NAION safety signal that the news section flagged on May 1 as the watch signal in GLP-1 ocular outcomes. Across five observational studies encompassing 1,593,554 patients, semaglutide ↗ users had a 2.52-fold higher risk of nonarteritic anterior ischemic optic neuropathy compared with non-GLP-1 users (95% credible interval 1.56 to 4.72, posterior probability that RR exceeds 1 of 99.9%). The effect was tighter in diabetic patients (RR 2.41) and persisted across leave-one-out sensitivity analyses.
The data set. Five observational studies and randomized controlled trials, searched through April 2025, included 682,456 semaglutide users and 911,098 non-GLP-1 controls. The certainty of evidence is rated low-to-moderate by GRADE, which means the meta-analytic estimate is the best current sizing of the signal but there are residual concerns about confounding and study heterogeneity (I² = 57.8% in the primary analysis, suggesting moderate variation between studies). The methodology is Bayesian random-effects with weakly informative priors, the right tool for this kind of pooled rare-event question.
The numbers worth holding. The relative risk is 2.52, but the absolute risk is what most clinicians and patients will need. The meta-analysis estimates 85 NAION cases per 100,000 GLP-1 users overall, and 118 cases per 100,000 specifically among semaglutide users. These are absolute incidences, not relative differences. NAION as a base-rate event in the diabetic population sits around 50 per 100,000 person-years; the GLP-1-attributable excess is roughly the same order of magnitude.
The clinical translation. Per yesterday's narrative review ↗, the 2025 multidisciplinary expert consensus moved the field from "avoid GLP-1 in diabetic retinopathy" to "stratify ophthalmic monitoring based on baseline risk." The Bayesian meta-analysis is the kind of evidence that consensus needed. With RR 2.52 in semaglutide and tighter RR 2.41 in diabetics, the case for explicit NAION counseling at GLP-1 initiation is now empirically anchored. Patients with prior NAION in the contralateral eye, sleep apnea, small optic disc anatomy, or other vasculopathy risk factors should know that semaglutide raises this risk roughly two and a half times. The absolute risk remains low (about one in every thousand patients per several years), but the magnitude of relative increase is large enough to be material in informed-consent conversations.
What this does not say. The meta-analysis is observational evidence with low-to-moderate certainty. It does not establish causation, only association. The leave-one-out sensitivity analyses are reassuring, but the I² of 57.8% means the studies are heterogeneous, which can reflect real biological differences (different patient populations, different disease severity at baseline, different durations of follow-up) or methodological differences (different NAION definitions, different control groups). The 95% credible interval of 1.56 to 4.72 is wide. The lower bound is meaningfully above 1.0, which is the basis for the 99.9% posterior probability of harm; the upper bound at 4.72 means the true effect could be substantially larger than the central estimate. More work is needed, and the authors say so.
What this resolves. The NAION question has been pending since the first observational signal in 2024. The first studies disagreed. The narrative review yesterday described the signal as "small in absolute terms but with a delayed temporal pattern, worth tracking and stratifying." The meta-analysis confirms the signal direction and sizes the magnitude. The question moves from "is there a signal" to "what is the magnitude and how do we counsel patients about it." Clinicians using GLP-1 drugs in diabetic patients now have a number to work with.
The platform read. The GLP-1R ↗ target page on peptidemodel hosts the GLP-1(7-36) amide ↗ and Exendin-4(4-39) ↗ cards that are the closest peptide references for the semaglutide pharmacology. The NAION result does not change which peptide molecules are useful designs against this target. It changes how those molecules need to be thought about in the broader benefit-risk frame, which is now an organ-system-wide question in GLP-1 pharmacology rather than a glycemic-control question.
The next twelve months. Three sources of additional evidence will refine the picture. RCT-level NAION data from cardiovascular outcome trials with planned ophthalmologic surveillance, larger administrative-database studies with refined NAION case definitions, and direct comparisons across GLP-1 drugs (semaglutide vs tirzepatide ↗ vs dulaglutide ↗) to determine whether the risk is class-level or molecule-specific. The class-versus-molecule question is the most clinically actionable; if NAION risk concentrates in semaglutide and not the GLP-1/GIP duals, the prescribing landscape changes accordingly.