Rarely, and almost all of the measured risk sits in people who have diabetes rather than in people taking the drug for weight alone. The largest single count comes from one academic eye center, where 20 cases of sudden optic-nerve vision loss appeared among 979 overweight or obese patients prescribed semaglutide, against 3 in the matched comparison group [1]. When eight observational cohorts covering 5,916,559 people were pooled two years later, the association held among patients with diabetes at 1.84, on a confidence interval of 1.21 to 2.80 [2]. In the overweight and obese group it came to 1.68, on an interval running from 0.72 to 3.91, which is wide enough to hold both no association at all and a near-quadrupling of one.
Sudden painless vision loss in one eye is the symptom behind the cohort above, and the finding is real enough to have changed labeling conversations worldwide. A retrospective matched study at one academic center searched 16,827 patients with no history of the condition and compared those prescribed semaglutide against those prescribed other treatments. The hazard ratios are large, and the population they came from is the part that decides what they mean for anyone reading a seller’s page.
What one eye center counted
Among 710 patients with type 2 diabetes, 17 events of nonarteritic anterior ischemic optic neuropathy occurred among those prescribed semaglutide against 6 in the comparator group. Cumulative incidence over 36 months was 8.9 per cent (95% CI 4.5 to 13.1) against 1.8 per cent (95% CI 0 to 3.5), a hazard ratio of 4.28 (95% CI 1.62 to 11.29, P < 0.001).
Among 979 patients who were overweight or obese, 20 events occurred on semaglutide against 3 on the comparator. Cumulative incidence was 6.7 per cent (95% CI 3.6 to 9.7) against 0.8 per cent (95% CI 0 to 1.8), a hazard ratio of 7.64 (95% CI 2.21 to 26.36).
Where those people came from
Every patient in this study was evaluated by neuro-ophthalmologists at a single academic institution, which is to say a referral population of people already sent to specialists for eye problems. A cumulative incidence of 6.7 per cent in that group says nothing like 6.7 per cent for a general population, because the denominator has been filtered by the exact thing being measured.
The interval widths say the same thing more quietly. A hazard ratio of 7.64 with a confidence interval running from 2.21 to 26.36 is built on 23 events in total. The direction is consistent and the magnitude is unsettled, which is what a large interval on small counts looks like.
What happened when eight cohorts were pooled
A later systematic review put eight observational studies together, 5,916,559 people of whom 1,870,346 were on semaglutide, and reported a pooled relative effect of 1.93 with an interval of 1.22 to 3.08 [2]. Splitting that by indication is where it stops being one number. Among people with diabetes it was 1.84 (95% CI 1.21 to 2.80), and among overweight or obese people 1.68 (95% CI 0.72 to 3.91), which is the row almost everybody buying these drugs online belongs to.
Heterogeneity across those eight studies was 91 per cent, close enough to the ceiling that they are not estimating one common number at all. The full reading of that pooled figure, and of the absolute rate the abstract never publishes, is in the optic nerve evidence by indication.
The one eye finding that came from a randomized trial
Everything above is observational. The only randomized eye signal in this literature is diabetic retinopathy, and it came out of SUSTAIN 6. That was a two-year pre-approval cardiovascular trial in type 2 diabetes, where semaglutide was associated with a significant increase in retinopathy complications against placebo [3]. Across SUSTAIN 1 through 5 and the Japanese trials there was no imbalance, so the signal sits in the longest trial of the sickest patients.
A post hoc mediation analysis attributed most of the effect to how far and how fast HbA1c fell in the first sixteen weeks, among patients who already had retinopathy, had poor control at baseline, and were on insulin. Early worsening after a rapid improvement in glucose control is a documented phenomenon with insulin too, which makes it a screening instruction rather than a reassurance. The mechanism and its limits are set out in the retinopathy evidence.
Glaucoma, and a comparison with nobody untreated
A retrospective cohort matched 51,540 adults aged 60 and over with type 2 diabetes across 21 countries. Over two years it found less incident primary open-angle glaucoma on tirzepatide than on other GLP-1 drugs, at a hazard ratio of 0.65 (95% CI 0.44 to 0.96) [4]. Both arms were taking a drug from this class, which makes the result a ranking rather than a risk assessment.
An upper bound of 0.96 clears 1 by four hundredths, and two years is a short window for a disease that develops over decades. What the design cannot reach is whether either drug leaves an eye better or worse off than taking nothing, which is the gap described in the glaucoma comparison.
Where the eye numbers run the other way
Idiopathic intracranial hypertension raises pressure inside the skull and swells the optic nerve, and weight loss is one of the few things established to help it. A meta-analysis of nine studies and 13,257 participants reported less papilledema on these drugs at a risk ratio of 0.38 (95% CI 0.25 to 0.56) and less visual worsening at 0.51 (95% CI 0.37 to 0.68) [5].
The same analysis reports an 80 per cent reduction in mortality, in a condition that threatens sight rather than life. That is not a drug effect but a measurement of how different the two groups were before anyone was treated. That figure calibrates the eye results beside it: the direction is probably right and the size is almost certainly inflated. The full argument is in the intracranial pressure analysis.
What to do with all of it
NAION is uncommon and it is not reversible, so a small absolute risk is still worth knowing. The practical response is not to avoid the drug on the strength of one referral cohort, and it is not to dismiss the finding either. It is to treat sudden painless vision loss in one eye as an emergency, and to tell a clinician about it the same day rather than at the next refill.
That is harder than it sounds when the prescription arrived through a website. Most of the sellers tracked here publish nothing about who to contact or how quickly, which is the gap counted in what sellers will not tell you and listed in the questions they leave open. Knowing the route to a human before you need one is the part of this a buyer can actually control, and the disclosure census shows who publishes it.