Reply to “Associations of semaglutide with first-time diagnosis of Alzheimer's disease in patients with type 2 diabetes: Target trial emulation using nationwide real-world data in the US”

Abstract

I read with great interest Wang et al.’s excellent article on the association of semaglutide with new diagnosis of Alzheimer's disease (AD).¹ The authors utilized a rigorous target trial emulation framework to assess the risk of new AD diagnosis after the initiation of semaglutide compared to seven other anti-diabetic medication classes in patients with type 2 diabetes. The primary outcome was a new diagnosis of AD after initiation of anti-diabetic medication. Notably, eligibility criteria required that patients had no use of any anti-diabetic medications within the prior 6 months. The authors performed propensity score matching to attempt to eliminate confounding between treatment groups.

The results were impressive, with a hazard ratio of receiving a new diagnosis of AD within 3 years of 0.33 for semaglutide compared to insulin. The results were consistent across different medication classes and patient subgroups. Interestingly, examination of the Kaplan–Meier curves indicates an almost immediate split (within days) for semaglutide versus insulin, sodium–glucose cotransporter 2 inhibitors, metformin, sulfonylureas, and thiazolidinediones. The Kaplan–Meier curve for semaglutide versus other glucagon-like peptide-1 receptor agonists (GLP-1Ras) even appears to diverge by ≈ 90 days.

Despite the exciting preclinical evidence and observational studies suggesting a potential benefit of semaglutide in preventing or delaying dementia, the immediate separation of the Kaplan–Meier curves is biologically implausible given the pathogenesis of AD, and highly suggestive of unmeasured confounding,² despite the rigorous methods used by the authors. As noted in the description of the AHEAD 3-45 Study, “the AD continuum begins with a long asymptomatic or preclinical stage, during which amyloid beta (Aβ) is accumulating for more than a decade prior to widespread cortical tauopathy, neurodegeneration, and manifestation of clinical symptoms.”³ These authors note that “individuals with preclinical AD have an increased risk of cognitive decline over 3–5 years” compared to those without.³

Therefore, seeing a reduction in incident AD within days or weeks of a new medication is highly suggestive of residual confounding. Despite the appropriately rigorous matching, it is probable that residual confounding remains, as patients who receive semaglutide versus insulin or a thiazolidinedione are almost certainly inherently different (in baseline education levels, socioeconomic status, or patient engagement with the medical field, which may alter patient desire for semaglutide). Physician differences between those prescribing semaglutide versus medications such as sulfonylureas or thiazolidinediones are possible as well, given that these are generally poor choices for the initial treatment of type 2 diabetes. Patients receiving some of these medications may be receiving suboptimal care in other domains as well, which may alter the risk of dementia.

I am excited about the possibility of GLP-1RAs reducing the incidence of dementia, and suspect these results will be proven right by future randomized trials. In the interim, I would urge caution when observational data suggest risk reductions that seem biologically implausible.

I have no financial or other conflicts of interest that could influence this work. Author disclosures are available in the supporting information.