Proposed section structure
For this topic, a clinically appropriate structure is: Highlights; Clinical background and rationale; Study design and methods; Key results; Clinical interpretation; Strengths and limitations; Implications for practice and research; Funding, registration, and disclosures; References. This structure emphasizes the main safety question, the observational methods used to address it, and the practical meaning for diabetes and retina specialists.
Highlights
Semaglutide was not associated with an increased risk of neovascular age-related macular degeneration (NVAMD) in adults with type 2 diabetes across 12 OHDSI-network databases.
Results were directionally consistent across two analytic frameworks: active-comparator new-user cohort analyses with propensity score adjustment and self-controlled case-series analyses.
Network-wide estimates comparing semaglutide with dulaglutide, empagliflozin, sitagliptin, and glipizide showed no statistically significant difference in NVAMD risk using either a diagnosis-code definition or a more specific diagnosis-plus-procedure definition.
The study offers reassuring real-world evidence for clinicians concerned about ocular safety signals around semaglutide, while also underscoring the limits of observational data for rare ophthalmic outcomes.
Clinical background and rationale
Semaglutide is now central to type 2 diabetes management and is also widely used for obesity treatment. As use has expanded, clinicians have become increasingly attentive to possible ophthalmic adverse events. Most discussion has focused on diabetic retinopathy, particularly after findings from SUSTAIN-6 suggested more diabetic retinopathy complications among semaglutide-treated participants, likely related at least in part to rapid glycemic improvement in susceptible patients rather than a direct retinal toxic effect.
Neovascular age-related macular degeneration is a distinct retinal disorder characterized by choroidal neovascularization, exudation, hemorrhage, and potentially irreversible central vision loss. It is biologically and clinically different from diabetic retinopathy, but both conditions are common in older adults with type 2 diabetes, making drug safety questions clinically relevant. Because type 2 diabetes and older age coexist frequently, even a small increase in NVAMD risk associated with a widely prescribed medication would have substantial public health implications.
The present study by Cai and colleagues addressed this concern using the Observational Health Data Sciences and Informatics (OHDSI) network, a large federated real-world data ecosystem that allows harmonized analyses across multiple databases. The central question was straightforward: among adults with type 2 diabetes, does semaglutide exposure alter the risk of NVAMD compared with alternative glucose-lowering therapies or within-person baseline risk periods?
Study design and methods
Data source and study period
This was a retrospective observational study conducted across 12 databases in the OHDSI network from December 1, 2017, through December 31, 2024. OHDSI studies typically use a common data model and standardized analytics across participating datasets, then pool site-level estimates with meta-analytic methods. That approach can improve consistency and scalability while preserving local data governance.
Population and exposure groups
The study included adults with type 2 diabetes who were new users of semaglutide. Comparator groups included other GLP-1 receptor agonists, specifically dulaglutide and exenatide, as well as non-GLP-1 receptor agonist agents with differing mechanisms: empagliflozin, sitagliptin, and glipizide. In total, 227,971 new semaglutide users were included.
The choice of active comparators is important. Comparing semaglutide with other agents used in similar clinical contexts helps reduce confounding by indication and healthcare utilization patterns. Inclusion of both within-class and outside-class comparators broadens interpretability, although residual confounding remains possible in all nonrandomized drug safety studies.
Outcome definitions
The investigators used two NVAMD outcome definitions. NVAMD-C relied on condition codes alone. NVAMD-CP required condition codes plus procedures, presumably to improve specificity by capturing patients with coded disease who also underwent interventions associated with neovascular AMD care, such as intravitreal therapy or related ophthalmic procedures. Using both a broader and a more specific outcome definition is methodologically sensible for a rare event in claims and electronic health record data.
Analytic approaches
Two complementary methods were used.
First, an active-comparator cohort design estimated hazard ratios using propensity score-adjusted Cox proportional hazards models. This is a standard pharmacoepidemiologic approach for evaluating comparative safety in routine care settings.
Second, the study used a self-controlled case-series design analyzed with conditional Poisson regression to estimate incidence rate ratios. The self-controlled case-series method compares risk periods and nonrisk periods within the same individual, inherently controlling for fixed patient characteristics such as genetics, chronic comorbidity burden, and many socioeconomic factors. It cannot, however, fully address time-varying confounding.
Finally, random-effects meta-analysis was applied to generate network-wide hazard ratio and incidence rate ratio estimates. This is appropriate when combining multiple heterogeneous real-world data sources.
Key results
Overall finding
The study found no evidence that semaglutide increases or decreases the risk of neovascular age-related macular degeneration among adults with type 2 diabetes. This null finding was consistent across both the active-comparator cohort analyses and the self-controlled case-series analyses.
Comparative cohort results
Compared with dulaglutide, semaglutide was not associated with a statistically significant difference in NVAMD risk. For the condition-code definition, the hazard ratio was 0.57 with a 95% confidence interval of 0.21 to 1.57 (P=.28). For the condition-plus-procedure definition, the hazard ratio was 0.25 with a 95% confidence interval of 0.05 to 1.27 (P=.10). Although both point estimates were below 1.0, the confidence intervals were wide and crossed the null, indicating imprecision and no confirmed protective effect.
Compared with empagliflozin, semaglutide again showed no significant difference. The hazard ratio for NVAMD-C was 0.98 (95% CI 0.54 to 1.79, P=.94), and for NVAMD-CP it was 0.79 (95% CI 0.38 to 1.64, P=.52). These estimates cluster around the null and are broadly reassuring.
Compared with sitagliptin, the point estimates were numerically above 1.0 but not statistically significant: NVAMD-C hazard ratio 2.08 (95% CI 0.90 to 4.83, P=.09) and NVAMD-CP hazard ratio 1.80 (95% CI 0.55 to 5.86, P=.33). These results illustrate an important principle in rare-event safety studies: a numerically elevated point estimate, especially with a wide confidence interval, does not establish a harmful association. The intervals are compatible with no effect and reflect limited precision.
Compared with glipizide, semaglutide also showed no significant difference: NVAMD-C hazard ratio 0.83 (95% CI 0.35 to 2.02, P=.69) and NVAMD-CP hazard ratio 0.50 (95% CI 0.21 to 1.19, P=.12).
The abstract states that no evidence of increased or decreased risk was found for semaglutide exposure, nor for the other GLP-1 receptor agonist or non-GLP-1 receptor agonist comparators. That overall consistency is more informative than any single point estimate.
Self-controlled case-series results
The self-controlled case-series analysis also showed no association between semaglutide exposure and NVAMD. For NVAMD-C, the incidence rate ratio was 0.92 (95% CI 0.67 to 1.26, P=.60). For NVAMD-CP, the incidence rate ratio was 1.02 (95% CI 0.76 to 1.36, P=.92). These values are close to 1.0 and strongly support the absence of a measurable signal in this dataset.
Because the self-controlled case-series framework controls for stable patient-level confounders, the concordance between this design and the active-comparator cohort analyses adds credibility to the primary conclusion.
How to interpret the effect sizes
For clinicians, the take-home message is not simply that the P values were nonsignificant. More importantly, the totality of the estimates does not reveal a coherent pattern of excess NVAMD risk attributable to semaglutide. There are some wide intervals, which is expected for an uncommon ophthalmic event, but the network-wide evidence does not indicate a reproducible harmful signal.
Clinical interpretation
This study is timely because semaglutide has been under intense clinical scrutiny for eye-related outcomes. It helps separate one specific concern, neovascular age-related macular degeneration, from broader discussions about retinal events in diabetes care.
Several points are clinically relevant. First, NVAMD is not diabetic retinopathy. Mechanisms thought to explain early worsening of diabetic retinopathy with rapid glucose improvement do not automatically imply a similar effect on AMD pathobiology. Second, patients with type 2 diabetes are older and have multiple vascular risk factors, so coincident occurrence of NVAMD after starting a new medication may create concern even without causation. Large comparative studies are therefore essential. Third, the absence of a detectable signal across both between-person and within-person analyses should reassure prescribers who have hesitated to use semaglutide because of concern about wet AMD.
At present, these findings do not support changing semaglutide prescribing on the basis of NVAMD risk. For patients who already require regular retinal care, standard ophthalmic surveillance should continue according to age, visual symptoms, and existing eye disease, but no additional NVAMD-specific monitoring strategy emerges from this study.
Strengths and limitations
Strengths
The major strengths are scale, design triangulation, and real-world relevance. More than 227,000 new semaglutide users were studied across 12 databases, which is notable for a retina safety outcome. The use of both active-comparator cohort analyses and self-controlled case-series methods reduces reliance on a single analytic assumption set. The inclusion of both within-class and non-class comparators is also valuable. Finally, the use of two outcome definitions addresses the trade-off between sensitivity and specificity.
Limitations
As with all observational studies using routinely collected data, outcome misclassification is possible. Administrative codes may not perfectly identify neovascular AMD, and even the condition-plus-procedure algorithm may miss or misclassify cases. The authors tried to mitigate this by using two definitions, but chart-level adjudication was not described in the abstract.
Residual confounding remains possible in the cohort analyses despite propensity score adjustment. Important AMD determinants such as smoking exposure, family history, dietary factors, genetic susceptibility, and detailed ophthalmic imaging findings are often incompletely captured in claims or structured electronic health record data.
The self-controlled case-series method controls for fixed confounders but is still susceptible to time-varying confounding, including changes in healthcare contact, co-medications, disease severity, or surveillance intensity after treatment initiation.
Event rarity likely limited precision. This is reflected in several wide confidence intervals, especially for some comparator analyses. Therefore, the study rules out a strong and consistent association more confidently than it excludes very small effects.
Generalizability is strongest for adults with type 2 diabetes receiving semaglutide in health systems represented in OHDSI data. The findings should not automatically be extrapolated to patients without diabetes, younger obesity populations, or people with unusual retinal comorbidity profiles.
Implications for practice and research
For endocrinologists, primary care clinicians, and ophthalmologists, the central practical implication is reassurance. Current evidence from this large network study does not indicate that semaglutide meaningfully changes the risk of neovascular AMD in adults with type 2 diabetes. Medication choice should therefore continue to be driven by established benefits and risks, including glycemic control, weight effects, cardiovascular and kidney outcomes, tolerability, and known retinal considerations such as pre-existing diabetic retinopathy status when glycemic improvement is expected to be rapid.
For researchers, this study highlights the value of federated real-world evidence networks for evaluating rare ophthalmic safety outcomes that would be difficult to study in conventional randomized trials. Future work could strengthen the evidence base by linking claims and electronic health record data to imaging-confirmed retinal diagnoses, anti-VEGF treatment records, and longitudinal visual acuity outcomes. Stratified analyses by age, baseline AMD, smoking, glycemic trajectory, and concomitant therapies would also be informative.
It would be particularly useful to distinguish incident NVAMD from prevalent but newly coded disease, and to examine whether risk differs by semaglutide formulation, dose escalation, treatment duration, or indication. Similar analyses in obesity populations without diabetes may also become increasingly relevant as semaglutide use expands.
Funding, registration, and disclosures
The abstract provided does not report funding details or a ClinicalTrials.gov registration number. Because this was a retrospective observational database study rather than an interventional trial, ClinicalTrials.gov registration may not apply. Readers should consult the full Ophthalmology article for complete disclosure statements, funding sources, database-specific governance details, and any protocol registration information available through OHDSI channels.
Conclusion
This OHDSI network study provides substantial real-world evidence that semaglutide is not associated with an altered risk of neovascular age-related macular degeneration among adults with type 2 diabetes. The conclusion is strengthened by concordant findings across active-comparator cohort analyses and self-controlled case-series analyses and by the use of two outcome definitions. While observational limitations remain, the data do not support a clinically meaningful NVAMD safety signal for semaglutide at this time. For current practice, that is a reassuring message.
References
1. Cai CX, Toy B, Martin B, Fan R, Westlund E, Tran D, Nishimura A, Lee H, Leng T, Nagy P, Mathioudakis N, Zhang L, Hribar M, Chen A, Armbrust K, Goetz K, Baxter S, Boland MV, Brown EN, Tsui E, Barkmeier AJ, Wang S, Mehta N, Stocking JC, O’Keefe G, Lee CS, Payne PRO, O’Brien WJ, DuVall S, Alshammari T, Falconer T, Dorr DA, Humes I, McCoy D, Adibuzzaman M, Mahmood R, Morgan-Cooper H, Desai P, Kothari SY, Sena A, Blacketer C, Ostropolets A, Shoaibi A, Rao G, Hripcsak G, Ryan P, Suchard MA. Semaglutide and Neovascular Age-Related Macular Degeneration Among Adults with Type 2 Diabetes: An OHDSI Network Study. Ophthalmology. 2026-06-01. PMID: 42229813.
2. Marso SP, Bain SC, Consoli A, Eliaschewitz FG, Jódar E, Leiter LA, Lingvay I, Rosenstock J, Seufert J, Warren ML, Woo V, Hansen O, Holst AG, Pettersson J, Vilsbøll T. Semaglutide and cardiovascular outcomes in patients with type 2 diabetes. N Engl J Med. 2016;375(19):1834-1844.
3. Bain SC, Klufas MA, Ho A, Matthews DR. Worsening of diabetic retinopathy with rapid improvement in systemic glucose control: A review. Diabetes Obes Metab. 2019;21(3):454-466.
4. Heerspink HJL, Sattar N, Pavo I, Weerakkody GJ, Van de Velde F, Bain SC. Effects of semaglutide on diabetic retinopathy and related outcomes in SUSTAIN 6. Diabetes Care. 2018;41(9):e122-e123.
5. Hripcsak G, Ryan PB, Duke JD, Shah NH, Park RW, Huser V, Suchard MA, Schuemie MJ, DeFalco FJ, Perotte A, Banda JM, Reich CG, Schilling LM, Matheny ME, Meeker D, Pratt N, Madigan D. Characterizing treatment pathways at scale using the OHDSI network. Proc Natl Acad Sci U S A. 2016;113(27):7329-7336.
6. Wong WL, Su X, Li X, Cheung CMG, Klein R, Cheng CY, Wong TY. Global prevalence of age-related macular degeneration and disease burden projection for 2020 and 2040: A systematic review and meta-analysis. Lancet Glob Health. 2014;2(2):e106-e116.
7. American Diabetes Association Professional Practice Committee. 12. Retinopathy, neuropathy, and foot care: Standards of Care in Diabetes—2025. Diabetes Care. 2025;48(Suppl 1). Please consult the current Standards of Care version for the exact pagination and wording.