To the editor

We read with great interest the research letter by Bahit et al¹ published recently in Polish Archives of Internal Medicine. The study investigated the potential association between glucagon‑like peptide‑1 receptor agonist (GLP‑1 RA) use and serious adverse ocular events, including nonarteritic anterior ischemic optic neuropathy, blindness, papilledema, and optic neuritis. Given the exponential rise in the number of GLP‑1 RA prescriptions and recent conflicting reports,^2-4 this large retrospective cohort study provides timely and valuable data.

The authors should be congratulated for recruiting a substantial Polish national cohort (over 36 000 GLP‑1 RA users matched with over 36 000 controls) and assessing a range of GLP‑1 RA agents over 2‑year follow‑up. The primary finding—that overall GLP‑1 RA use was not independently associated with an increased risk of the composite ocular outcome, as compared with matched controls—is reassuring. This conclusion was supported by very low absolute incidence rates reported for individual events and the multivariable Cox model for the composite outcome (adjusted hazard ratio [HR], 1.34; 95% CI, 0.94–1.91; P = 0.11; Supplementary material, Table S3), which notably identified diabetes mellitus itself as a significant predictor (adjusted HR, 2.51; 95% CI, 1.13–5.57; P = 0.02).

However, several findings warrant further discussion and investigation. First, the authors reported significant variability in blindness incidence among users of different GLP‑1 RAs (unadjusted P <⁠0.001), with dulaglutide emerging as a potential predictor in logistic regression.¹ While the overall class effect appears neutral, this heterogeneity raises questions about potential drug‑specific effects or residual confounding that require careful consideration and validation. Second, clarification regarding the time‑to‑event analysis would be beneficial. In the main text, it is stated that the median time to the first combined event was significantly shorter in the GLP‑1 RA group than in the controls (P = 0.03).¹ Yet, the Kaplan–Meier curve provided in Supplementary material (Figure S2) that focuses specifically on blindness and low vision shows a nonsignificant difference (P = 0.13). Understanding the apparent discrepancy in the unadjusted time‑to‑first‑event analysis for the composite outcome is essential. Third, in subgroup analyses, the adjusted HR for the composite outcome in patients with diabetes only approached statistical significance (HR, 1.62; 95% CI, 0.98–2.66; P = 0.06; Supplementary material, Table S6), suggesting potential effect modification that merits further exploration. Furthermore, while including various GLP‑1 RAs is a strength, we should also note the evolving landscape of this drug class. Older agents (eg, liraglutide) have longer market presence, whereas newer ones (eg, tirzepatide, or higher doses of semaglutide specifically approved for obesity) became available later within the study’s observation window (January 2018–June 2024).⁵ Consequently, patients initiating these newer therapies, potentially at higher doses and for indications beyond diabetes, inherently have shorter follow‑up durations in this analysis than those starting older agents at an earlier time point. This systematically truncated follow‑up for the most recent entrants could potentially mask longer‑term adverse effects or risks associated explicitly with higher cumulative exposure, newer formulations, or use in different patient populations (eg, primarily for weight management). The strengths of the study are clear; however, as the authors acknowledge, the retrospective design using diagnostic codes without individual case validation and a lack of baseline ophthalmologic examinations are important limitations, alongside the potential for observation time bias with newer agents.

In conclusion, the study by Bahit et al¹ contributes significantly to the ongoing debate, largely providing reassurance regarding the ocular safety of GLP‑1 RAs as a class in this large cohort. However, the potential signal for heterogeneity among agents, the need for clarity on time‑to‑event findings, the borderline result in the diabetes subgroup, and the inherent limitations related to truncated follow‑up for newer agents highlight the need for continued pharmacovigilance and further research. Prospective studies or analyses of large, diverse datasets incorporating detailed, validated clinical and ophthalmic information, with sufficient follow‑up across all agents and indications, are crucial to clarify these risks.