GLP-1 Agonists and Retinopathy

GLP-1 Safety

1.76x hazard ratio

Semaglutide was associated with a 76% higher rate of retinopathy complications in SUSTAIN-6, but 83.5% of affected patients had pre-existing retinopathy at baseline.

Marso et al., NEJM, 2016

Marso et al., NEJM, 2016

View as image

When the SUSTAIN-6 trial reported that semaglutide increased retinopathy complications by 76% compared to placebo, the finding made headlines and earned a label warning on Ozempic and Wegovy. The number was real, but the context behind it transforms the story. Nearly all affected patients had pre-existing retinopathy. The risk tracked with the magnitude and speed of HbA1c reduction, not with the drug itself. And the same phenomenon has been documented for decades with insulin therapy. The question is not whether GLP-1 receptor agonists damage eyes. The question is whether rapid glucose improvement in patients with advanced diabetic eye disease triggers a well-known complication that would occur with any effective glucose-lowering intervention. For an overview of GLP-1 safety concerns across organ systems, see our pillar article on GLP-1 drug interactions.

The SUSTAIN-6 Signal: What Actually Happened

The SUSTAIN-6 trial randomized 3,297 patients with type 2 diabetes and high cardiovascular risk to semaglutide (0.5 mg or 1.0 mg weekly) or placebo for 104 weeks. While the primary cardiovascular endpoint showed a 26% reduction in major adverse cardiovascular events, an unexpected finding emerged: retinopathy complications, defined as vitreous hemorrhage, blindness, or conditions requiring intravitreal agents or photocoagulation, occurred in 3.0% of the semaglutide group versus 1.8% of the placebo group (HR 1.76; 95% CI 1.11-2.78; p=0.02).^[1]

This finding required careful interpretation. The trial population was not representative of typical GLP-1RA users. Approximately 58% of SUSTAIN-6 participants had retinopathy at baseline, and 83.5% of those who experienced retinopathy complications had pre-existing diabetic retinopathy. Many were on insulin and had poor baseline glycemic control (mean HbA1c 8.7%). These are precisely the patients known to be vulnerable to early worsening of diabetic retinopathy when glucose control improves rapidly.

The semaglutide groups achieved HbA1c reductions of 1.1% (0.5 mg) and 1.4% (1.0 mg) by week 16, substantially faster than the placebo group. This rapid improvement in patients with advanced baseline retinopathy is the pattern most associated with early worsening.

The HbA1c Mediation: Speed of Glucose Lowering as the Culprit

Vilsboll et al. (2018) conducted a detailed post hoc analysis of the SUSTAIN-6 retinopathy data, combining it with data from the SUSTAIN 1-5 trials and the Japanese semaglutide trials. Their findings were clarifying.^[2]

First, there was no imbalance in diabetic retinopathy adverse events across SUSTAIN 1-5 and the Japanese trials. The retinopathy signal was isolated to SUSTAIN-6, the trial with the highest baseline retinopathy prevalence and the poorest baseline glycemic control.

Second, a mediation analysis using the initial change in HbA1c at week 16 as a covariate found that the majority of the retinopathy effect could be attributed to the magnitude and rapidity of HbA1c reduction. When the analysis adjusted for how quickly glucose levels dropped, the excess retinopathy risk with semaglutide largely disappeared.

Third, the at-risk subgroup was specific: patients who had pre-existing diabetic retinopathy, poor glycemic control at baseline (HbA1c above 8.5%), and concurrent insulin use. In patients without pre-existing retinopathy, no excess risk was observed. Semaglutide-treated patients with no known DR at baseline had a lower event rate (10.0%) than the placebo group (13.8%).

The authors concluded that the findings were consistent with the well-established phenomenon of early worsening of diabetic retinopathy, which has been associated with rapid glycemic improvement since the DCCT and UKPDS trials in the 1990s. They recommended that guidance similar to insulin labeling regarding DR monitoring should be applied to semaglutide.

The DCCT Precedent: This Has Happened Before

Early worsening of diabetic retinopathy with intensive glucose control was first documented in the Diabetes Control and Complications Trial (DCCT), published in 1993. In that trial, 13.1% of patients assigned to intensive insulin therapy experienced significant DR worsening at 6 or 12 months, compared to 7.6% on conventional therapy. The magnitude of HbA1c reduction was the dominant risk factor.

The mechanism is incompletely understood but appears to involve retinal vascular changes triggered by rapid metabolic shifts. One hypothesis: chronic hyperglycemia causes compensatory upregulation of VEGF (vascular endothelial growth factor) in retinal tissue. When glucose normalizes rapidly, the VEGF-driven neovascularization and vascular permeability that were metabolically "stable" under chronic hyperglycemia become destabilized, leading to hemorrhage or edema in vulnerable retinal vessels. Over the long term, improved glycemic control reduces retinopathy progression. The DCCT demonstrated this clearly: despite early worsening, patients in the intensive treatment group had dramatically lower rates of retinopathy at 10 years.

The dynamics of HbA1c reduction in the intensive arm of DCCT closely matched those of semaglutide in SUSTAIN-6. Both achieved approximately 1.5-2.5 percentage-point drops in HbA1c within the first 3-4 months of treatment. The parallel is not coincidental; it is the same phenomenon mediated by the same mechanism, differing only in the glucose-lowering agent used.

The Meta-Analysis: 93 Trials, Nuanced Results

Kapoor et al. (2023) published a comprehensive meta-analysis of 93 randomized clinical trials comparing FDA-approved GLP-1RAs to placebo, insulin, or oral antidiabetic medications.^[3] The results were nuanced:

Compared to placebo, GLP-1RA use was associated with a modestly increased risk of early-stage DR (RR 1.31; 95% CI 1.01-1.68) and early-stage retinal adverse events (RR 1.29; 95% CI 1.01-1.66). This is consistent with the early worsening phenomenon seen in SUSTAIN-6.

Compared to insulin, GLP-1RA use was associated with a reduced risk of late-stage DR (RR 0.38; 95% CI 0.15-0.98). This finding is notable because it suggests that while GLP-1RAs may transiently worsen early retinopathy (likely through rapid glucose lowering), they may protect against the advanced stages of diabetic eye disease over time.

The analysis revealed that the retinopathy signal was driven largely by a single agent, albiglutide, which showed significantly higher early-stage DR risk versus placebo (RR 2.18) and lower late-stage DR risk versus insulin (RR 0.25). Other GLP-1RAs, including liraglutide, exenatide, and dulaglutide, did not individually show statistically significant increases in retinopathy events.

Demographic analysis also revealed confounding: GLP-1RA and comparator groups differed significantly in age, HbA1c, body weight, BMI, diabetes duration, sex, race, and ethnicity. These baseline imbalances complicate attribution of retinopathy events to the drug rather than to patient characteristics.

LEADER: Why Liraglutide Did Not Show the Signal

The LEADER trial enrolled 9,340 patients with type 2 diabetes and high cardiovascular risk, a population similar to SUSTAIN-6. Liraglutide 1.8 mg daily reduced MACE by 13% and cardiovascular death by 22%, but showed no increased retinopathy signal.^[4]

Why did liraglutide not trigger the same retinopathy concern? Several factors may explain the difference. Liraglutide's HbA1c reduction was somewhat smaller and slower than semaglutide's: approximately 0.4 percentage points less reduction over the first months. The LEADER population, while similarly high-risk, had a slightly different baseline retinopathy profile. And liraglutide's pharmacokinetic profile (daily dosing, shorter half-life) produces more gradual glycemic improvement compared to semaglutide's once-weekly pharmacokinetics.

The absence of a retinopathy signal with liraglutide reinforces the HbA1c mediation hypothesis: the retinopathy risk correlates with how quickly glucose drops, not with GLP-1 receptor activation per se. A drug that lowers glucose more aggressively produces more early worsening in susceptible patients.

Real-World Evidence: The Signal Fades

Cai et al. (2025) published an OHDSI (Observational Health Data Sciences and Informatics) network study examining semaglutide and diabetic retinopathy across multiple real-world databases.^[5] Unlike clinical trials, which enrich for high-risk patients, real-world data captures the full spectrum of semaglutide users, including the majority who have no pre-existing retinopathy and moderate baseline HbA1c levels.

The study found no elevated retinopathy risk with semaglutide compared to other diabetes medications. This is consistent with the hypothesis that the SUSTAIN-6 signal was specific to the trial's enriched population of patients with advanced retinopathy and poor glycemic control, rather than representing a generalizable drug effect.

Preclinical evidence adds an interesting counterpoint. Cheng et al. (2025) demonstrated that semaglutide attenuates diabetic retinopathy progression in animal models by ameliorating retinal vasculopathy and neurodegeneration.^[6] Far from being retinopathic, semaglutide appears to have direct retinal protective effects through anti-inflammatory and anti-angiogenic pathways. The disconnect between these protective preclinical effects and the SUSTAIN-6 clinical signal further supports the interpretation that the clinical finding reflects the metabolic speed of glucose lowering, not an intrinsic property of the drug.

Clinical Implications: Who Is at Risk and What to Monitor

The retinopathy evidence produces a clear risk profile. Patients most vulnerable to early worsening on GLP-1RAs are those who have all three factors simultaneously: pre-existing diabetic retinopathy (especially proliferative or severe non-proliferative), poorly controlled baseline glucose (HbA1c above 8-9%), and expectation of rapid HbA1c improvement (more than 1.5 percentage points in the first 3-4 months).

For patients without pre-existing retinopathy, the data shows no increased risk. For patients with mild non-proliferative retinopathy and moderate baseline HbA1c, the risk appears minimal. For patients on semaglutide who have advanced retinopathy and very high HbA1c, the SUSTAIN-6 data and the DCCT precedent suggest that ophthalmologic monitoring during the first 6-12 months of treatment is appropriate, the same approach recommended when initiating intensive insulin therapy in similar patients.

The timing of the risk is also informative. Early worsening events cluster in the first 6-12 months after treatment initiation, corresponding to the period of most rapid glycemic change. After this window, the risk subsides, and long-term glycemic control becomes protective. In the DCCT, over 50% of patients who experienced early worsening recovered within 6 months, with few requiring ophthalmic intervention. At 10 years, intensive glycemic control reduced retinopathy risk by 76% compared to conventional therapy. The short-term cost of early worsening was decisively outweighed by the long-term benefit.

This parallel with insulin is important for calibrating risk. Insulin has been associated with early DR worsening for three decades, yet no one considers insulin retinopathic. The mechanism is metabolic, not pharmacological. The same framing applies to GLP-1RAs. For related GLP-1 safety discussions, see do GLP-1 agonists cause pancreatitis? and gastroparesis and GLP-1 drugs. For other potential adverse effects, see GLP-1 agonists and thyroid cancer and GLP-1s and gallbladder problems.

The Bottom Line

The retinopathy signal from SUSTAIN-6 reflects a metabolic phenomenon, not a pharmacological one. Rapid HbA1c reduction in patients with pre-existing diabetic retinopathy triggers early worsening that has been documented with insulin since the 1990s. Post hoc analyses, meta-analyses, and real-world data all converge on the same conclusion: GLP-1 receptor agonists are not intrinsically retinopathic. Preclinical evidence suggests they may actually protect retinal tissue. The clinical implication is straightforward: patients with advanced retinopathy starting any potent glucose-lowering therapy, whether insulin, semaglutide, or another agent, warrant ophthalmologic monitoring during the initial months of treatment.

Frequently Asked Questions

Does semaglutide cause retinopathy?

Semaglutide does not appear to directly cause retinopathy. The SUSTAIN-6 trial found a higher rate of retinopathy complications with semaglutide, but post hoc analysis by Vilsboll et al. (2018) showed the effect was attributable to rapid HbA1c reduction in patients with pre-existing eye disease. In patients without baseline retinopathy, no increased risk was observed. A 2025 real-world OHDSI network study found no elevated retinopathy risk with semaglutide.

Why did SUSTAIN-6 show retinopathy complications with semaglutide?

SUSTAIN-6 enrolled high-risk diabetic patients, 58% of whom had retinopathy at baseline. Semaglutide rapidly lowered HbA1c by 1.1-1.4 percentage points in 16 weeks. This rapid glucose improvement triggered early worsening of existing retinopathy, a phenomenon documented since the DCCT trial with insulin in the 1990s. The 83.5% of affected patients who had pre-existing retinopathy confirms this was a metabolic effect, not a drug-specific toxicity.

Is liraglutide safer than semaglutide for retinopathy?

The LEADER trial showed no retinopathy signal with liraglutide. This likely reflects liraglutide's somewhat smaller and slower HbA1c reduction compared to semaglutide, rather than a fundamental pharmacological difference. Both drugs activate the same GLP-1 receptor. A meta-analysis by Kapoor et al. (2023) found that the retinopathy signal across GLP-1RAs was modest and largely driven by albiglutide, not semaglutide or liraglutide specifically.

Should I get an eye exam before starting semaglutide?

Current guidelines recommend regular diabetic eye examinations for all patients with diabetes regardless of medication. The SUSTAIN-6 data suggests additional monitoring may be appropriate for patients starting semaglutide who have pre-existing proliferative or severe non-proliferative diabetic retinopathy combined with poorly controlled baseline glucose (HbA1c above 8-9%). For patients without pre-existing retinopathy, the data does not support increased concern.

Do GLP-1 agonists protect or damage the retina long-term?

Long-term data suggests protection. The Kapoor 2023 meta-analysis found that GLP-1RAs reduced late-stage retinopathy risk by 62% compared to insulin. Preclinical evidence from Cheng et al. (2025) shows semaglutide has direct retinal protective effects through anti-inflammatory and anti-angiogenic pathways. The initial worsening seen in SUSTAIN-6 parallels the DCCT insulin data, where early worsening was followed by dramatically lower retinopathy rates at 10 years with intensive glycemic control.

What is early worsening of diabetic retinopathy?

Early worsening is a transient increase in retinopathy severity that occurs when blood sugar improves rapidly, especially in patients with pre-existing eye disease and poor baseline glucose control. It was first documented in the DCCT trial with intensive insulin therapy in the 1990s, where 13.1% of intensive-treatment patients experienced worsening at 6-12 months. Over 50% recovered within 6 months, and long-term retinopathy outcomes were better with intensive control. The same phenomenon explains the SUSTAIN-6 semaglutide findings.