Initiation of glucagon-like peptide-1 receptor agonists was associated with a higher adjusted 18-month risk of ischemic optic neuropathy than initiation of sodium-glucose cotransporter-2 inhibitors or dipeptidyl peptidase-4 inhibitors in a large observational analysis of commercially insured US adults with type 2 diabetes. The absolute difference was about 3 to 4 cases per 10,000 patients, and the researchers cautioned that residual confounding could explain some or all of the association.
Researchers used the Merative MarketScan Commercial Claims and Encounters database to emulate a hypothetical randomized trial from January 2017 through December 2022. The active-comparator, new-user analysis included adults aged 18 to 65 years with type 2 diabetes who initiated a glucagon-like peptide-1 receptor agonist (GLP-1 RA), an SGLT2 inhibitor, or a DPP-4 inhibitor.
Patients with prior use of any of the three study drug classes, prior ischemic optic neuropathy, other diabetes subtypes, or major contraindications to the study medications were excluded. The final cohorts included 161,489 GLP-1 RA initiators, 122,114 SGLT2 inhibitor initiators, and 86,047 DPP-4 inhibitor initiators.
The primary outcome was incident ischemic optic neuropathy (ION), which served as a proxy for nonarteritic anterior ischemic optic neuropathy (NAION) because claims data lack a diagnosis code specific to NAION. The single-code definition used in the primary analysis had an estimated 75% positive predictive value for NAION.
The researchers adjusted for more than 80 measured baseline characteristics using inverse probability of treatment weighting. Measured characteristics were balanced following weighting, although the observational method could not account for unmeasured or inadequately captured factors.
At 18 months, the adjusted cumulative risk of ION was approximately 9 cases per 10,000 GLP-1 RA initiators and 6 per 10,000 SGLT2 inhibitor initiators, a difference of 3 cases per 10,000. In the DPP-4 inhibitor comparison, the corresponding risks were approximately 8 and 4 cases per 10,000, a difference of about 4 cases per 10,000.
The GLP-1 RA cohort had 81 ION events. There were 48 events among SGLT2 inhibitor initiators and 33 among DPP-4 inhibitor initiators. Median follow-up ranged from 424 to 477 days, with weighted methods used to estimate cumulative risk through 18 months.
Results were attenuated when the researchers applied more specific outcome definitions. Requiring two ION diagnosis codes reduced the adjusted risk differences to about 2 cases per 10,000 in both comparisons. When three codes were required, the differences were approximately 1 case per 10,000 and did not reach statistical significance.
Of the 81 ION events among GLP-1 RA initiators, 85% occurred among patients older than 50 years and 70% occurred among men. Risk differences were also numerically larger among patients with preexisting ocular conditions. However, the small number of events limited precision, and the study was not powered to establish differences across subgroups or clinical conditions.
Baseline treatment-intensity findings differed by comparator. In the SGLT2 inhibitor comparison, the risk difference was larger among patients receiving two or more glucose-lowering medications than among those receiving metformin alone. The two groups had nearly identical risk differences in the DPP-4 inhibitor comparison. The researchers noted that early separation of the curves in the multidrug SGLT2 inhibitor analysis could indicate residual confounding by diabetes duration or severity.
In exploratory analyses of individual GLP-1 RAs, the highest point estimates were observed for liraglutide, followed by semaglutide and dulaglutide. The agent-specific analyses were too imprecise to establish differences among the three drugs.
In separate cohorts of patients with overweight or obesity but without diabetes, 14 ION events occurred among GLP-1 RA initiators and none occurred in either active comparator group. The researchers did not calculate risk differences because the comparator groups had no events.
The researchers described the prior evidence as heterogeneous. They cited studies that reported no association as well as others that reported higher estimates among patients receiving semaglutide, while noting limitations including referral-based populations, exposure definitions, limited methodological detail, and possible immortal time bias.
Because treatment assignment was not randomized, unmeasured differences in diabetes duration, severity, body mass index, hemoglobin A1c, and smoking may have influenced the estimates. Outcome ascertainment was also limited by the absence of a NAION-specific claims code. The rarity of ION reduced the precision of secondary and subgroup analyses, and the study population was limited to commercially insured adults aged 18 to 65 years.
“This study adds to the limited evidence on a possible link between GLP-1 RAs and ION,” wrote lead study author Kamika R. Reynolds, PhD, of Rutgers University, and colleagues. The researchers added that the findings should be interpreted in light of residual confounding related to type 2 diabetes duration and severity and that any potential safety signal should be weighed against the established cardiometabolic benefits of GLP-1 RAs.
Disclosure forms were available with the article online.
Source: Annals of Internal Medicine