In this week’s issue

• A new TriNetX study shows NSAID use is associated with a lower risk of exudative and non-exudative AMD. 
• A large retrospective cohort study shows that anti-VEGF injections within 6 weeks of a stroke or MI have no significant risk of cardiovascular outcomes compared with holding anti-VEGF injections. 
• In a large prospective cohort study, nearly half of eyes changed ≥1.0 D over 5 years once axis was considered, suggesting caution when considering toric IOLs in children.
• A new AI fundus tool called Reti-Pioneer, while not accurate enough yet, is a start for a low-cost tool to screen for systemic diseases such as diabetes, gout, hypertension, and hyperlipidemia.

Ophthalmology

Could NSAID use reduce the risk of developing age-related macular degeneration?

Can common anti-inflammatory medications protect the macula? Age-related macular degeneration (AMD) remains a leading cause of irreversible vision loss, and inflammation is increasingly recognized as a contributor to both non-exudative and exudative disease pathways. Because nonsteroidal anti-inflammatory drugs (NSAIDs) inhibit cyclooxygenase-mediated inflammatory signaling, they have been studied as possible modifiers of AMD risk, though prior evidence has been inconsistent. In this large retrospective TriNetX cohort study, investigators compared patients prescribed NSAIDs between 2015 and 2024 with propensity score-matched controls. NSAID prescription was associated with lower subsequent AMD risk across multiple follow-up points, including the full study period, with reduced risk also seen for both non-exudative and exudative AMD. Protective signals were observed for aspirin and non-selective COX inhibitors. Taken together, these findings link NSAID exposure with lower AMD risk and further support inflammation as a potential target in AMD pathogenesis. 

JAMA Ophthalmology 

Can retina specialists stop holding their breath after a stroke or MI?

Maybe the real risk was delaying vision-saving treatment all along. Retina specialists often face a dilemma about whether to continue, delay, or switch anti-VEGF agents in patients who experience a stroke or myocardial infarction. Current guidance suggests delays or agent substitution in the peri-cardiovascular event (CVE) period, but strong prospective data is lacking. In this large retrospective cohort study, a multi-institutional US EHR network (TriNetX, 2005–2025) used propensity score–matched cohorts to compare patients who received intravitreal anti-VEGF within 14 days before or up to 6 weeks after a CVE versus matched controls who also had a history of anti-VEGF use but received no injections within 3 months before or after the CVE (n = 1,526 matched pairs for stroke). Peri-event anti-VEGF exposure was not associated with worsened cardiovascular recovery and was associated with lower all-cause mortality at 90 days (3.7% vs 9.6%; RR 0.39) and 1 year (9.3% vs 17.8%; RR 0.52). No differences in outcomes were observed among the commonly used agents aflibercept, bevacizumab, and ranibizumab. These findings suggest that delaying anti-VEGF therapy or modifying agent selection after a CVE may be unnecessary.

American Journal of Ophthalmology 

Reassessing astigmatism stability following pediatric cataract surgery

Astigmatism after pediatric cataract surgery: stable…or just looks that way? With increasing interest in toric intraocular lenses (IOLs), understanding how astigmatism behaves in the long term after pediatric cataract surgery is crucial. While early postoperative astigmatism is often reassuring, children’s eyes continue to grow and remodel for years, potentially affecting both the magnitude and axis of astigmatism, which are key considerations for refractive planning and amblyopia management. This post hoc analysis of a large, prospective registry included 213 children (266 eyes) under 13 years who underwent lensectomy with primary IOL implantation and compared refraction at an early postoperative visit (≥60 days) with that at 5 years postoperatively. Using standard clinical notation that ignores axis, the mean astigmatism change over 5 years was a modest +0.37 D. However, when axis changes were incorporated using power vector analysis, mean change increased to 1.06 D, with 47% of eyes changing by ≥1.00 D and ~16% by ≥2.00. Changes were not associated with age at surgery, though greater early postoperative astigmatism predicted larger long-term shifts. Astigmatism after pediatric cataract surgery may look stable, but clinically meaningful magnitude and axis changes are common, warranting caution when considering toric IOLs in children.

British Journal of Ophthalmology  

What does red-green color vision deficiency look like inside the retina?

Looking beyond color perception to cone structure. Red-green color vision deficiency (CVD) has traditionally been evaluated by observing functional color discrimination, yet the extent of associated structural retinal abnormalities has been more difficult to elucidate. In this cross-sectional multimodal imaging study, 36 patients with red-green CVD and 38 age-matched controls underwent adaptive optics scanning laser ophthalmoscopy (AO-SLO), optical coherence tomography (OCT), OCT-angiography, and microperimetry to evaluate cone morphology, retinal layers, and retinal sensitivity. Patients with CVD demonstrated significantly reduced cone density and regularity, increased cone spacing and dispersion, and thinning of the myoid and ellipsoid zone (25.2 ± 2.1 vs 28.4 ± 2.5 µm) and photoreceptor outer segments (29.8 ± 2.1 vs 31.5 ± 2.5 µm) compared with controls. Retinal sensitivity was also reduced, with temporal cone density correlating with retinal sensitivity specifically in CVD eyes (r=0.410). These findings suggest that red-green CVD is associated with measurable abnormalities in cone architecture and outer retinal integrity in addition to altered functional color discrimination, supporting quantitative cone imaging as a potential structural biomarker for visual dysfunction in CVD.

Eye

Are the eyes truly the windows to the rest of the body?

Can artificial intelligence analysis of retinal fundus photographs accurately detect systemic chronic disease? Diagnosing conditions such as osteoporosis, thyroid disease, and diabetes often requires specialist evaluation, laboratory testing, or imaging that may be inaccessible in low-resource settings. To address this, investigators developed Reti-Pioneer, a multitask AI algorithm designed to screen for six diseases simultaneously: hypertension, hyperlipidemia, type 2 diabetes, thyroid disease, osteoporosis, and gout. This prospective study analyzed more than 107,000 color fundus photographs from over 53,000 participants across China, Singapore, and the UK, in a variety of geographic settings, in clinics and hospitals. The model was trained on real-world retinal photographs, including poor-quality images affected by cataracts or blur, enabling the algorithm to account for the variable imaging quality commonly encountered in primary care settings. Reti-Pioneer demonstrated the strongest performance for type 2 diabetes and gout (AUROC 0.833 and 0.832), moderate performance for osteoporosis, hypertension, and hyperlipidemia, and weaker performance for thyroid disease (AUROC 0.699). External validation showed lower performance across most diseases, highlighting persistent concerns regarding generalizability. While not accurate enough to independently diagnose disease, retinal AI screening may eventually provide a rapid, low-cost tool for identifying at-risk patients in community settings.

Nature Biomedical Engineering

Synthetic retinas may offer real diagnostic gains

Can AI generate reliable and accurate training data for rare disease AI models? Diagnostic models have far fewer examples to learn from for rare ocular diseases, such as Stargardt Disease, compared to models for more common conditions, like diabetic retinopathy. EyeDiff uses natural language prompts to generate lesion-preserving ophthalmic images for common and rare diseases. The model was trained on 42,048 images from eight datasets, spanning 14 imaging modalities and 84 ocular diseases. In a medical realism test, ophthalmologists mistook 62-67% of EyeDiff’s generated images for real ones, affirming the authenticity of the generated images. The authors then tested whether augmenting real images from validation datasets with EyeDiff-generated images would improve rare disease diagnosis. Rare-disease classification AUROC (area under the receiver operating characteristic curve) using the Rare Diseases dataset and the RETFound model rose from 0.871 to 0.919, which outperformed the oversampling method (0.893) traditionally used to make up for the lack of rare disease examples available for AI learning. EyeDiff’s generated images may help augment existing data, which will be particularly helpful for rare disease AI models.

Wisconsin Epidemiologic Study of Diabetic Retinopathy - 1984

To develop or not to develop… a study on diabetic retinopathy. Diabetic retinopathy is the second leading cause of blindness in the United States. This population-based epidemiological study over 25 years aimed to determine the incidence, prevalence, and severity of diabetic retinopathy and visual impairment and associated risk factors in diabetic patients. The original study was 1980-1982 and there have been six patient follow-ups 1984-86, 1990-92, 1995-96, 2000-01, 2006-07, and 2012-14. There were 996 type 1 diabetic patients and 1370 type 2 diabetic patients who were analyzed for the incidence rate and severity of diabetic retinopathy using stereoscopic color fundus photographs. 

Key Points:  

• Diabetic retinopathy was found to be related to longer duration with the disease, high levels of glycosylated hemoglobin, proteinuria, increased diastolic BP, and male sex.
• Incidence and progression of diabetic retinopathy was greatest in those who were younger at the time of diagnosis
• After 20 years of diabetes mellitus, nearly 99% of patients with type 1 and 60% with type 2 disease demonstrated some degree of diabetic retinopathy. 

Overall, after 25 years there have been over 230 reports from the WESDR study to show glycemic control is associated with a decreased risk of diabetic retinopathy and other diabetic complications. It furthermore identified additional risk factors for diabetic retinopathy in those with diabetes.