Data and Timing-Related Concerns About Semaglutide and NAION Risk

By
Chun-Hsien Yu
October 1, 2026

Jama Ophthalmology

Overview

A letter to the editor raised questions about risk-set accounting and the timing of nonarteritic anterior ischemic optic neuropathy (NAION) events in a veterans study comparing semaglutide with sodium-glucose cotransporter-2 inhibitors (SGLT2i). The authors’ reply attributed the apparent discrepancy in the at-risk table to different counting conventions and described the study’s case-ascertainment approach. The supplied material does not provide the reply’s full response to the question about time-varying hazard ratios.

Background

NAION is an optic nerve disorder examined in recent studies of semaglutide and other glucagon-like peptide-1 receptor agonists. The Brief Report discussed in the letter used a target trial emulation with overlap weighting to compare US veterans initiating semaglutide or SGLT2i, balancing 40 baseline covariates. The letter noted that the Kaplan-Meier curves appeared to separate at about 1 year and questioned whether a single hazard ratio adequately represented risk over time. It also cited an estimated positive predictive value of approximately 74.5% for administrative-code identification of NAION.

Data Highlights

Measure or issue	Reported information
SGLT2i patients at risk	90,883 on day 0 and 87,135 on day 182.5; a decrease of 3,748
Events and censoring cited in the letter	35 NAION events and 510 censored patients; together, 545 individuals
Unaccounted difference calculated by the letter	3,203 patients, approximately 3.5% of the cohort
Reported hazard ratio questioned	2.33
Timing noted by the letter	Kaplan-Meier curves begin to separate at approximately 1 year after treatment initiation
Case ascertainment	Administrative-code algorithm positive predictive value reported as approximately 74.5%
Authors’ reply on counting convention	At-risk counts refer to each time point immediately beforehand; cumulative event and censoring totals include occurrences through that time point, including day 0. The reply states that 510 SGLT2i patients were censored on day 0.

Key Findings

The letter reported a 3,748-person decline in the SGLT2i risk set between day 0 and day 182.5, compared with 545 events and censorings cited for that interval.
The letter calculated that 3,203 patients, approximately 3.5% of the cohort, were not accounted for under its interpretation of the table.
The authors’ reply said the apparent discrepancy reflected counting conventions: risk-set counts were taken immediately before each time point, whereas cumulative event and censoring totals included occurrences through that time point, including day 0.
The reply specifically reported 510 SGLT2i patients censored on day 0.
The letter observed that the Kaplan-Meier curves began to separate at approximately 1 year and asked whether the hazard ratio differed between the first treatment year and later follow-up.
The letter cited an administrative-code algorithm positive predictive value of approximately 74.5% for NAION; the supplied reply excerpt does not include a response to the time-varying hazard-ratio question.

Clinical Implications

The correspondence identifies risk-set counting conventions and the timing of curve separation as relevant when interpreting the reported hazard ratio of 2.33. The supplied material does not establish whether the hazard ratio varied by follow-up period, so that question remains unresolved in the excerpt provided.

Conclusion

The authors’ reply explains the apparent risk-set discrepancy as a difference in counting conventions, including day-0 censoring. The temporal pattern of risk and whether the hazard ratio differs across follow-up periods are not resolved by the supplied reply excerpt.