A large propensity-matched cohort study published in Diabetes, Obesity and Metabolism on 27 August 2026 found that people who started a GLP-1 receptor agonist were modestly more likely to be diagnosed with diabetic retinopathy, and more likely to undergo a retinopathy procedure, than matched people who started a DPP-4 inhibitor instead. The same study found no significant difference in non-arteritic anterior ischaemic optic neuropathy (NAION), and none in vision impairment or blindness. If you have diabetes and are starting one of these drugs, this is an argument for keeping up eye exams, not for avoiding treatment.
The numbers, and what the intervals mean
The authors built two new-user cohorts from people who initiated either drug class between 2006 and 2021, matched on propensity score. The main cohort contained 98,082 matched pairs. Non-vision-threatening diabetic retinopathy occurred in 11.05% of GLP-1 starters versus 9.98% of DPP-4 starters — an odds ratio of 1.12 (95% CI 1.09 to 1.15), or roughly one extra diagnosis per hundred people. Vision-threatening retinopathy or diabetic macular oedema occurred in 2.17% versus 1.98%, an odds ratio of 1.09 (95% CI 1.03 to 1.16), about one extra case per five hundred people. Retinopathy procedures ran 4.16% versus 3.68%, odds ratio 1.14 (95% CI 1.09 to 1.19).
The two outcomes that matter most to a patient did not separate. NAION gave an odds ratio of 1.11 with a 95% confidence interval of 0.96 to 1.29, and vision impairment or blindness gave 1.04 (95% CI 0.97 to 1.11). Both intervals contain 1.00, meaning the data are compatible with no difference — though the NAION interval also stretches up to a 29% increase, so it is not evidence of safety either, just an absence of a detected signal. A second cohort restricted to 15,440 matched pairs with chronic kidney disease, chosen because they are more prone to microvascular complications, reproduced the smaller findings: non-vision-threatening retinopathy odds ratio 1.10 (95% CI 1.03 to 1.18) and procedures 1.14 (95% CI 1.03 to 1.26), with the other outcomes not significant.
How this fits what came before
Eye safety has trailed this drug class for years. Early semaglutide trial data flagged worsening of existing retinopathy, with the leading explanation being that rapid improvement in blood glucose can transiently worsen retinal disease rather than the drug damaging the eye directly. Separately, reports of NAION prompted regulators in several markets to look at the question. This study speaks to both, and it does so with an active comparator rather than untreated controls, which removes some of the confounding you get when you compare people on a new drug with people on nothing.
The practical upshot is narrow and worth stating precisely: a small excess of retinopathy coding and retinal procedures, no detected excess of blindness. For readers with diabetes, that supports a baseline eye examination before starting and follow-up during the first year — a conversation to have with a prescriber rather than a reason to stop. It is also a reminder to bring existing eye disease up when you are being assessed, which is one of the things worth asking about when choosing a telehealth provider, since a rushed intake may never ask. Our GLP-1 research hub covers the wider safety picture for the class.
What is not yet known
This is observational. Propensity matching balances measured characteristics only, and people prescribed a newer injectable may simply be seen more often and coded more often — detection bias would produce exactly this pattern of more diagnoses without more blindness. The exposure window ends in 2021, so most of the GLP-1 use captured predates the high-dose semaglutide and tirzepatide era, and the results should not be assumed to transfer to those regimens. The study reports odds ratios rather than hazard ratios, and cannot separate a drug effect from the effect of rapid glucose lowering. The authors themselves note that administrative codes have poor validity for blindness, which weakens the reassuring finding as much as it weakens an alarming one.
