In the summer of 2024, a group at a Boston eye hospital published a retrospective study reporting a higher rate of NAION among patients prescribed semaglutide than among matched patients on other diabetes and weight medications.
Within about four hours, the headline had become Ozempic linked to blindness, and my inbox filled up.
What NAION is
The optic nerve head has a fragile blood supply. When it is compromised, the nerve is damaged, and the result is sudden, usually painless vision loss in one eye — often noticed on waking.
It is not glaucoma and not retinal detachment. There is no established treatment that reverses it. Damage that occurs is generally permanent, and roughly a fifth of people go on to be affected in the other eye.
It is also rare. Baseline incidence in the general population is on the order of a handful of cases per hundred thousand people per year, concentrated in adults over fifty with a particular optic disc anatomy — a small, crowded disc sometimes called a “disc at risk.”
Why this particular study is hard to interpret
The research was conducted at a tertiary neuro-ophthalmology centre. That matters enormously.
People do not attend a specialist neuro-ophthalmology clinic at random. They arrive because something has gone wrong with their optic nerve. A study drawn from that population is looking at a group in which NAION is vastly over-represented compared with the general public, which makes the rate observed there impossible to translate directly into a population risk.
Add to that the standard problems of retrospective observational work:
- Confounding by indication. People prescribed semaglutide have diabetes, obesity, hypertension, sleep apnea. Every one of those is independently associated with vascular optic neuropathy.
- Surveillance bias. Patients on a heavily publicised drug may be more likely to seek care and be worked up.
- Small numbers. When the total case count is in the dozens, a handful of events move the estimate substantially.
The biologically plausible mechanism
There is a reason ophthalmologists took it seriously rather than dismissing it.
Rapid improvement in glycaemic control is a recognised trigger for early worsening of diabetic retinopathy, documented long before GLP-1 drugs existed and observed in the semaglutide cardiovascular outcomes trial in patients with pre-existing retinopathy. The proposed mechanism involves haemodynamic and autoregulatory changes in the retinal and optic nerve circulation when glucose falls quickly.
A similar mechanism at the optic nerve head is not far-fetched. Plausibility is not proof, but it is the difference between a signal worth investigating and statistical noise.
A finding can be simultaneously worth regulatory attention and far too weak to justify stopping a medication that is treating you well.
What this actually means for you
For most people taking these drugs: very little change in behaviour, and a symptom to know about.
Worth a specific conversation with your clinician if you have:
- Existing diabetic retinopathy, particularly moderate or worse — this is a well-established consideration independent of NAION and warrants ophthalmology input before rapid glycaemic improvement
- A previous episode of NAION in either eye
- A known “disc at risk” identified during an eye examination
- Poorly controlled sleep apnea, which is independently associated with NAION
How to read the next one of these
The thyroid cancer boxed warning is the other safety signal worth reading this way — a reproducible rodent finding, faithfully carried onto a human label, and not confirmed in people over fifteen years.
There will be another. This class is now taken by tens of millions of people, which means every rare condition on earth will occur in someone taking it, and some of those coincidences will reach statistical significance.
Three questions get you most of the way:
- Association or causation? Almost always association, in a retrospective study.
- What is the absolute risk, not the relative one? A doubling of a one-in-ten-thousand event is still a one-in-five-thousand event.
- Who was studied? A specialist referral population is not the public.
There is a version of this that depends on you. Signals like this one are assembled out of individual reports, and a report is only useful if it carries dates — when you started, when the dose last changed, when the symptom appeared, and how quickly your weight and glucose were moving at the time. Almost nobody can reconstruct that afterwards. A running log in Zenday means that if something does happen to you, what you hand your ophthalmologist is a timeline rather than a guess.
Regulators reviewed this signal, and labelling in some jurisdictions has been updated to mention it. That is the system working — a signal detected, investigated, and communicated in proportion. It is not the same as a drug being found unsafe.