Two separate eye findings travel under one headline, and they are not the same claim. One is a rare optic-nerve stroke called nonarteritic anterior ischemic optic neuropathy, or NAION, which entered the semaglutide literature in 2024 and is still unsettled. The other is worsening of diabetic retinopathy, which entered it in 2016, has an agreed mechanism, and sits on the label. Different tissue, different decade, different strength of evidence. They also concern different people, which matters most for anyone buying on a cash plan rather than the diabetes indication.
Where the NAION signal came from
A retrospective matched cohort study at a single academic institution searched a neuro-ophthalmology registry covering December 2017 through November 2023 and found 16,827 patients with no history of NAION. Among the 710 with type 2 diabetes, 17 NAION events occurred in the 194 prescribed semaglutide against 6 in the 516 on non-GLP-1 diabetes drugs: a 36-month cumulative incidence of 8.9% (95% CI, 4.5% to 13.1%) against 1.8% (95% CI, 0% to 3.5%), hazard ratio 4.28 (95% CI, 1.62 to 11.29).[1]
The overweight and obese cohort produced the larger number. Among 979 patients, 20 events occurred on semaglutide against 3 on other weight-loss drugs, for cumulative incidences of 6.7% (95% CI, 3.6% to 9.7%) and 0.8% (95% CI, 0% to 1.8%) and a hazard ratio of 7.64 (95% CI, 2.21 to 26.36).[1] The authors called it an association and said causality would need further study.
Hold the incidences, not the ratios. A 6.7% three-year risk of an irreversible optic neuropathy would be an emergency. That figure comes from a population already referred to a neuro-ophthalmology service, where the reason for the visit is frequently the outcome being counted.
National registers found a smaller version of the same thing
Denmark ran the question across everyone. A five-year cohort covering all 424,152 people with type 2 diabetes in the country between 2018 and 2024 split them into 106,454 exposed to once-weekly semaglutide and 317,698 not. Over 1,915,120 person-years, 218 developed NAION. The incidence rate was 0.228 against 0.093 per 1,000 person-years, hazard ratio 2.19 (95% CI, 1.54 to 3.12), with a median 22.2 months from first prescription to event.[2]
A Danish-Norwegian study compared semaglutide initiators against initiators of SGLT-2 inhibitors — a fairer comparator, since both groups are being escalated for the same reason. Across 44,517 Danish and 16,860 Norwegian users, 32 NAION events occurred in total. The pooled hazard ratio was 2.81 (95% CI, 1.67 to 4.75), and the incidence rate difference was +1.41 per 10,000 person-years (95% CI, +0.53 to +2.29).[3] Its own conclusion pairs the two: an increased risk, and an absolute risk that remains low.
Sweden then published the most careful version. Among 107,518 GLP-1 initiators and 185,898 SGLT-2 inhibitor initiators, 62 and 64 cases of anterior ischemic optic neuropathy occurred. Risk was 0.04% against 0.02% at one year and 0.12% against 0.07% at five, a risk difference of 0.05 percentage points (95% CI, 0.00% to 0.10%).[4]
The sensitivity analysis is the part worth reading twice. Restricting to patients already on metformin at baseline — a crude way of matching on how advanced the diabetes is — pulled the risk ratio down to 1.24 at three years and 1.23 at five, with confidence intervals crossing 1 and risk differences of 0.01 to 0.02 percentage points.[4] The authors read that as a sign the excess may reflect residual confounding by diabetes severity rather than the drug.
The subgroup that reverses is the one this site sells into
A multinational cohort drew on 160 health care organizations across 21 countries and split the population three ways: 37,314 with type 2 diabetes only, 129,690 with obesity only, and 130,216 with both, each matched 1:1 against non-GLP-1 comparators.[5]
In the diabetes-only group the three-year hazard ratio was 1.51 (95% CI, 0.71 to 3.25). In the obesity-only group — the group the referral-cohort study scored at 7.64 — every point estimate fell below 1.0: 0.41 (95% CI, 0.08 to 2.09) at one year, 0.67 (95% CI, 0.20 to 2.24) at two, and 0.72 (95% CI, 0.24 to 2.16) at three.[5]
That is the shape of this whole subject. The largest reported hazard ratio and a point estimate suggesting no harm at all describe the same kind of patient, and neither is precise: the obesity-only interval spans a twelvefold range and includes 1 comfortably. The honest reading is not that semaglutide protects the optic nerve in people without diabetes. It is that nobody has enough events in that population to say.
What the randomized trials can and cannot settle
A 2025 systematic review pooled 78 randomized trials covering 73,640 participants. Semaglutide neither increased nor reduced eye disorders overall (OR, 1.01; 95% CI, 0.91 to 1.12) or diabetic retinopathy (OR, 1.04; 95% CI, 0.92 to 1.17). NAION returned an odds ratio of 3.92 on a 95% CI of 1.02 to 15.02 — statistically significant by the narrowest possible margin.[6]
Its trial sequential analysis then said the useful thing: the accumulated sample is large enough to be conclusive about diabetic retinopathy and is not large enough to be conclusive about NAION.[6] Randomized evidence has closed one of these two questions and left the other open.
Regulators reached two different answers
Europe's pharmacovigilance committee reviewed non-clinical data, trial data, post-marketing surveillance and the literature, and concluded in June 2025 that NAION is a very rare side effect of semaglutide — the labeling band meaning up to 1 in 10,000 people. The product information for Ozempic, Rybelsus and Wegovy was updated accordingly, with instructions to contact a doctor on sudden or rapidly worsening vision loss and to stop semaglutide if NAION is confirmed.[7]
The United States label says something else by saying nothing. The Wegovy prescribing information, at the version revised in June 2026, carries no entry for NAION, nonarteritic optic neuropathy or ischemic optic neuropathy anywhere in its warnings. Its ocular warning, section 5.8, is diabetic retinopathy complications.[8] Two regulators, the same published evidence, two different conclusions about whether it belongs on a label — which is a fair description of how settled this is.
Retinopathy is the older question, and it has an answer
SUSTAIN-6 randomized 3,297 patients with type 2 diabetes to semaglutide or placebo for 104 weeks. Retinopathy complications — vitreous hemorrhage, blindness, or conditions needing an intravitreal agent or photocoagulation — occurred at a hazard ratio of 1.76 (95% CI, 1.11 to 2.78; P = .02) against placebo.[9] That result sits inside a trial otherwise remembered for its cardiovascular benefit.
The label prints the split that the hazard ratio hides. Complications occurred in 3% on semaglutide against 1.8% on placebo overall. Among patients with a history of diabetic retinopathy at baseline the figures were 8.2% against 5.2%. Among patients with no known history they were 0.7% against 0.4%.[8] The absolute increase is three percentage points in one group and three tenths of a point in the other.
A post hoc mediation analysis across the SUSTAIN program explained why. Most of the effect was attributable to the magnitude and rapidity of the HbA1c drop during the first 16 weeks, in patients who had pre-existing retinopathy, poor glycemic control at baseline, and insulin treatment. No imbalance appeared in SUSTAIN 1 through 5 or the Japanese trials.[10] Early worsening of retinopathy is a documented consequence of correcting long-standing hyperglycemia quickly, including with insulin, and the SUSTAIN-6 pattern matches it. That makes the pace of the first months the variable, which is the logic behind a slow dose escalation.
The trial built to settle it has not reported
FOCUS was designed for exactly this question: a long-term trial of semaglutide against placebo in people with type 2 diabetes, with a primary endpoint of at least three-step progression on the Early Treatment Diabetic Retinopathy Study scale at year 5. It began in May 2019 and targets roughly 1,500 participants. It is active and no longer recruiting, its estimated primary completion is November 2027, and no results have been posted.[11]
So the answer to what FOCUS found is: nothing yet, and nothing before 2027 at the earliest. What exists in the meantime is observational. A target trial emulation in United States claims data compared 30,911 semaglutide initiators against 32,844 dulaglutide initiators with type 2 diabetes and no advanced retinopathy at baseline, and found no difference in treatment for diabetic macular edema or proliferative retinopathy (HR, 0.88; 95% CI, 0.70 to 1.11).[12]
How much of this reaches a cash-pay buyer
Most sellers on this site dispense compounded semaglutide to people who do not have diabetes. Compounded drugs are not FDA-approved, and are not reviewed by the FDA for safety, efficacy or quality before dispensing — a distinction set out in the compounded-versus-brand article. On the eye evidence, that population sits outside almost every number above, and it cuts both ways.
The retinopathy finding probably does not transfer. Early worsening requires retinopathy to worsen and a rapid glycemic correction to drive it. A person with no diabetes and a normal HbA1c has neither. Even the 0.7% against 0.4% figure describes people who had type 2 diabetes and no diagnosed retinopathy, which is a step closer to the trial than a cash-pay buyer is.
The NAION evidence does not transfer either, and that is not reassurance. The Swedish cohort names limited generalizability to users without diabetes as one of its stated limitations, and diabetes is itself a risk factor for NAION, so neither the exposed nor the comparator group maps onto someone taking a weekly injection for weight alone. The only study that looked hard at an obesity-only population found 129,690 patients and still could not narrow its confidence interval past 1.[5] An untested population is not a safe one.
What changes, and what does not
On the numbers as they stand, the excess NAION risk attributable to semaglutide in people with type 2 diabetes is on the order of one to five additional cases per 10,000 people over several years, and shrinks further when the comparison is adjusted for how sick the diabetes is.[3][4] That is a real signal and a very small absolute risk, and both halves of that sentence are load-bearing.
What follows is modest. A diagnosed history of diabetic retinopathy is the one condition on this page that belongs in front of a prescriber before the first dose, alongside the rest of the contraindication list. Sudden painless vision loss in one eye is an urgent in-person examination, not a message left in a portal. And an intake form that asks nothing about eye disease before shipping a vial has told you what kind of assessment it is, which is one of the patterns in the telehealth article.
Nothing on this page is a reason to start or stop a prescription. The sellers and their prices are set out in the compounded semaglutide board, and how a figure gets established before it is published is in the methodology.