News · Heart & Metabolic
Sudden vision loss hit 3 more GLP-1 users per 10,000 over 18 months than two rival diabetes drugs
An Annals of Internal Medicine target trial emulation compared GLP-1 users against two other diabetes drug classes. Ischemic optic neuropathy was more common on GLP-1s, concentrated in men, over-50s and those with eye or heart disease.
- 18-month risk: 8.5 vs 5.5 per 10,000 against SGLT2 inhibitors (risk difference 3.0).
- Against DPP4 inhibitors: 7.8 vs 4.2 per 10,000 (risk difference 3.6).
- Numbers needed to harm were 3,333 and 2,778 respectively.
- Of 81 events among GLP-1 users, 85% were in people over 50 and 70% in men.
- Risk differences were minimal in women and in people under 50.
Almost everything written about GLP-1 drugs in the last three years has been about what they do to weight. This is about an eye.
Ischemic optic neuropathy is sudden loss of vision in one eye, caused by blood supply failing to the optic nerve. It usually arrives overnight and is noticed on waking, it is usually painless, and the vision usually does not return. It is also rare, which is why establishing whether a drug raises the risk takes a very large dataset.
Writing in Annals of Internal Medicine, researchers ran a target trial emulation and found the 18-month risk for ION was 8.5 versus 5.5 per 10,000 among GLP-1RA users versus SGLT2i users, and higher again against a second comparison class.
Reading three per ten thousand properly
The relative increase sounds alarming and the absolute one is small. Both are true and the second is the one that governs a decision.
Corresponding numbers needed to harm were 3333 and 2778. That is how many people would need to take a GLP-1 for eighteen months for one extra case of this vision loss to occur. Put the other way: 3,332 of every 3,333 users would not experience it.
Rare harms matter anyway, because these drugs are now taken by millions of people, and a risk of one in three thousand across that population is a real number of blinded eyes. But it is not a reason for an individual to panic, and it is a long way from the risk profile of a drug that should be withdrawn.
Why the comparison group is the clever part
The obvious study would compare GLP-1 users against people not on a GLP-1. That study would be worthless.
People prescribed a new diabetes drug are sicker, better monitored, and different in a dozen ways from people prescribed nothing, and every one of those differences also predicts vascular events in the eye. This analysis instead compared GLP-1 users against users of two other diabetes drug classes, SGLT2 inhibitors and DPP4 inhibitors, which puts similarly diagnosed and similarly treated people on both sides.
A target trial emulation goes further, specifying in advance who is eligible, when treatment starts and how long follow-up runs, so the analysis of routine records is structured like the randomized trial nobody will run. It removes some biases. It cannot remove the possibility that doctors chose GLP-1s for patients who differed in ways the records do not capture.
The risk is not spread evenly
The subgroup pattern is unusually clear, and it is where the practical value sits.
Among GLP-1RA users, 69 of the 81 events occurred in persons older than 50 years and 57 occurred in men. Risk differences were higher in men, patients aged 50 years or older, and those with cardiovascular disease or ophthalmic conditions, with minimal differences in women and those younger than 50 years.
That concentration matches what is already known about this condition. Ischemic optic neuropathy of this kind clusters in older men with crowded optic discs and vascular disease, and those are exactly the people in whom the excess appeared. A drug effect that lands where the baseline risk already is looks more like a real interaction than a statistical accident.
There is a second gradient worth noting. Risk differences were attenuated among metformin monotherapy users compared with users of 2 or more diabetes medications, meaning the excess was smaller in people with less advanced disease.
What this does not settle
The authors flag a measurement problem in their own limitations: diagnostic codes specifically for the non-arteritic form of this condition are imperfect, so some of what was counted may be other optic nerve disease.
The mechanism is also unknown. Rapid improvement in blood sugar has been linked to transient worsening of diabetic eye disease for decades, and GLP-1s lower blood sugar and weight fast, so there is a plausible story. This study does not test it.
And the counterfactual is not “no risk”. Type 2 diabetes is a disease in which your blood glucose, or blood sugar, levels are too high, and high blood sugar damages the small vessels of the eye. A drug that controls it is preventing eye disease at the same time as this analysis finds it associated with a rare form of it.
What to actually do about this
Nothing abrupt. The finding is a rare harm identified in an observational analysis, in a drug class with substantial demonstrated benefit, and the absolute numbers are small.
What it usefully changes is a conversation for one group: an older man with existing heart or eye disease starting a GLP-1 has a specific, small, now-quantified risk worth naming out loud. For a woman under 50, on this evidence, there is essentially nothing to discuss.
The one thing that should change immediately for everyone is what to do about symptoms. Sudden painless loss of vision in one eye is an emergency regardless of cause, and the window in which anything can be done is short. That advice predates this study and it is the part of it that matters most.
People also ask
What did the study find?
The 18-month risk for ION was 8.5 versus 5.5 per 10,000 among GLP-1RA users versus SGLT2i users (RD, 3.0 [95% CI, 0.4 to 5.7]) and 7.8 versus 4.2 per 10,000 among GLP-1RA users versus DPP4i users (RD, 3.6 [CI, 1.1 to 6.1]). Corresponding numbers needed to harm were 3,333 and 2,778, respectively.
What is ischemic optic neuropathy?
Sudden, usually painless vision loss in one eye caused by interrupted blood supply to the optic nerve. It typically happens overnight, is noticed on waking, and the lost vision does not usually come back. It is rare.
What is a number needed to harm?
How many people would have to take the drug for one extra case to occur. Here, roughly 2,800 to 3,300 people on a GLP-1 for 18 months for one additional case of this specific vision loss. The condition is rare, so a meaningful relative increase is still a small absolute one.
Who was most affected?
Among GLP-1RA users, 69 (85.2%) of the 81 ION events occurred in persons older than 50 years and 57 (70.3%) occurred in men. Risk differences were higher in men, patients aged 50 years or older, and those with cardiovascular disease or ophthalmic conditions, with minimal differences in women and those younger than 50 years.
What is a target trial emulation?
A method that uses routine health records but designs the analysis as though it were the randomized trial nobody ran, specifying eligibility, treatment start and follow-up in advance. It reduces some biases of ordinary observational comparison and cannot remove confounding by indication entirely.
Why compare against other diabetes drugs rather than no drug?
Because people prescribed a diabetes drug differ from people prescribed nothing. Comparing GLP-1 users against SGLT2 and DPP4 inhibitor users puts similarly ill people on both sides, which is a fairer test than an untreated comparison.
Should someone stop their GLP-1?
No, and not on the basis of a news article. The absolute risk is small, these drugs treat conditions that themselves damage eyes and blood vessels, and stopping abruptly has its own consequences. Anyone with sudden vision change in one eye should seek urgent medical attention. This is general information rather than medical advice.