A 98,000-pair study finds a small retinopathy excess with GLP-1 drugs

The retinopathy question has followed GLP-1 receptor agonists since 2016. SUSTAIN-6, the cardiovascular outcomes trial that randomised 3,297 people with type 2 diabetes to once-weekly semaglutide or placebo for 104 weeks, found a lower rate of the primary cardiovascular composite in the semaglutide arm — 6.6% versus 8.9%, hazard ratio 0.74 (95% CI, 0.58–0.95) — and, alongside it, a significantly higher rate of retinopathy complications: vitreous haemorrhage, blindness, or conditions requiring intravitreal treatment or photocoagulation, at a hazard ratio of 1.76 (95% CI, 1.11–2.78; P = 0.02) [s2].

That signal has never been fully resolved. The dominant explanation has been early worsening — the long-observed phenomenon in which rapid improvement in glycaemic control transiently worsens existing retinopathy — rather than a direct drug effect. A study published in Diabetes, Obesity and Metabolism on 28 August adds a large observational data point that is consistent with a small real excess and does not settle the mechanism [s1].

The design

The authors built a retrospective propensity-score-matched cohort of adults initiating either a GLP-1 receptor agonist or a DPP-4 inhibitor between 2006 and 2021, using a new-user design with DPP-4 inhibitors as the active comparator [s1]. Two matched cohorts were created: an overall cohort of patients with diabetes, and a chronic kidney disease cohort representing patients at higher likelihood of microvascular complications [s1].

Primary outcomes were non-vision-threatening diabetic retinopathy, vision-threatening retinopathy or diabetic macular oedema, and non-arteritic anterior ischaemic optic neuropathy — NAION, the optic-nerve event that has generated its own separate safety discussion around semaglutide [s1]. Secondary outcomes were vision impairment or blindness, and the occurrence of retinopathy procedures [s1].

What it found

The matched overall cohort contained 98,082 pairs [s1]. GLP-1 users had a higher risk of non-vision-threatening retinopathy (11.05% vs 9.98%; odds ratio 1.12; 95% CI, 1.09–1.15), of vision-threatening retinopathy or macular oedema (2.17% vs 1.98%; OR 1.09; 95% CI, 1.03–1.16), and of retinopathy procedures (4.16% vs 3.68%; OR 1.14; 95% CI, 1.09–1.19) [s1].

Two outcomes did not differ. There was no significant difference in NAION (OR 1.11; 95% CI, 0.96–1.29) or in vision impairment or blindness (OR 1.04; 95% CI, 0.97–1.11) [s1].

In the chronic kidney disease cohort — 15,440 matched pairs — GLP-1 use was associated with higher risk of non-vision-threatening retinopathy (OR 1.10; 95% CI, 1.03–1.18) and of retinopathy procedures (OR 1.14; 95% CI, 1.03–1.26), but not the other outcomes [s1].

What size of risk this describes

The absolute differences are small. Non-vision-threatening retinopathy differed by roughly one percentage point across 98,082 pairs; vision-threatening disease by about two-tenths of a percentage point [s1]. Effects of that magnitude are exactly the range in which residual confounding, differential surveillance and coding practice can generate a signal that is real in the data and absent in the patient.

Differential surveillance deserves particular weight here. People started on a GLP-1 agonist are frequently people whose diabetes has been escalating, and escalation prompts referral. If GLP-1 initiators get more eye examinations than DPP-4 initiators, they will accumulate more coded retinopathy diagnoses and more procedures regardless of whether their retinas differ. The procedure finding — a 14% relative increase [s1] — is as consistent with more looking as with more disease.

The authors flag a specific limitation on the reassuring end of the result: administrative codes have notably low validity for blindness diagnosis [s1]. The null on vision impairment and blindness is therefore weaker evidence than the positive findings, not stronger. It is a null measured with a blunt instrument.

Their own conclusion is that GLP-1 use was associated with a modest increase in the risk of diabetic retinopathy and retinopathy-related procedures, independent of baseline retinopathy status, but was not associated with vision impairment or blindness [s1].

What it changes

Not the balance of the drugs, on this evidence. A one-percentage-point difference in coded retinopathy sits against a class with established cardiovascular and renal benefit in trial populations [s2]. What the study does is keep an open question open at a moment when the prescribing base is expanding rapidly beyond the populations that were randomised.

The clinically substantive gap remains the one SUSTAIN-6 opened: whether the excess is a transient consequence of rapid glycaemic improvement in eyes that already have disease, or something the drug does independently. An observational cohort matched on baseline characteristics cannot separate those, because the rate of glycaemic change after initiation is precisely what differs between the arms. That question needs a trial designed around retinal imaging, not a database.

Sources

• [s1] "Association of GLP1-Receptor Agonists With Diabetic Retinopathy," Diabetes, Obesity and Metabolism, 28 August 2026. https://doi.org/10.1111/dom.71267
• [s2] "Semaglutide and Cardiovascular Outcomes in Patients with Type 2 Diabetes," New England Journal of Medicine, 15 September 2016. https://doi.org/10.1056/NEJMoa1607141