GLP‐1 receptor agonists and reduced dementia risk: Real‐world evidence stacks up

Glucagon-like peptide-1 receptor agonists (GLP-1RAs), including semaglutide and the dual GLP-1/Gastric Inhibitory Polypeptide receptor (GIPR) agonist tirzepatide, are well established for their potent glycaemic control and weight-lowering effects. These agents belong to a class of highly effective antihyperglycaemic therapies that not only reduce HbA1c (up to ~1.8% with semaglutide and ~2.3% with tirzepatide), but also provide significant cardiorenometabolic benefits due to their pleiotropic effects. Their anti-obesity efficacy and safety have also been well demonstrated in multiple phase 3 trials. For semaglutide, the STEP 1 study showed an average weight loss of 14.9% over 68 weeks, with 83.5% of participants achieving ≥5% weight reduction. The STEP 5 trial confirmed sustained benefits over 2 years, while the SELECT trial demonstrated cardiovascular risk reduction (MACE). Tirzepatide achieved even greater weight loss in SURMOUNT-1 (20.9% with 15 mg weekly), with ≥5% weight loss in 90.9% of participants, and proved superior to semaglutide in SURMOUNT-5. Cardiovascular benefits of tirzepatide in patients living with obesity are anticipated based on findings from SURPASS-CVOT and SUMMIT trials; however, final conclusion will be set after completion of SURMOUNT-MMO trial. A summary of cardiovascular outcome trials investigating subcutaneous semaglutide and tirzepatide is presented in Table 1.1, 2 This summer, a wave of newly published data has drawn attention to another emerging health benefit of GLP-1RAs: their potential role in reducing the risk of dementia and other neurocognitive disorders. One of the most methodologically rigorous investigations is a retrospective cohort study recently published in JAMA Network Open.3 Drawing on data from the TriNetX global research platform, which compiles real-world data from more than 120 million patients across multiple healthcare organizations, these researchers constructed two 1:1 propensity score-matched cohorts (30 430 individuals per arm) to compare outcomes between patients using GLP-1RAs and those on other antidiabetic medications. The analysis demonstrated significantly lower risks of dementia (HR 0.63; 95% CI: 0.50–0.81), ischaemic stroke (HR 0.81; 95% CI: 0.70–0.93), and all-cause mortality (HR 0.70; 95% CI: 0.63–0.78) in the GLP-1RA group. Notably, the median follow-up duration was 41.1 months, providing sufficient time to capture long-term outcomes. Subgroup analyses further revealed that these protective effects were most pronounced in older (≥60 years) women, and individuals with class I or II obesity (BMI 30–40 kg/m2), suggesting a differential benefit among populations at elevated cardiometabolic and neurocognitive risk. The authors also performed sensitivity analyses and negative outcome controls to address potential unmeasured confounding, further strengthening the robustness of their results. However, one should bear in mind that observational findings require caution. Even with extensive adjustment, confounding by indication, unmeasured variables, differential healthcare contact, immortal time bias, and variable accuracy of EHR dementia diagnoses remain. These limitations preclude causal inference.4, 5 Lin et al.'s findings are echoed by several other real-world analyses, all published in July 2025. Cheng et al.,6 using a multi-cohort dataset comprising 109 778 individuals, found a similar association between GLP-1RA use and reduced dementia incidence (adjusted HR 0.74; 95% CI: 0.63–0.87). Inoue et al.7 conducted a target trial emulation using administrative claims data and demonstrated a lower incidence of dementia among GLP-1RA users versus comparators (HR 0.81; 95% CI: 0.71–0.93). In another retrospective cohort study comparing between GLP-1RAs and metformin (87 229 matched patients per cohort), Sun et al.8 showed that GLP-1RAs were associated with a lower risk of overall dementia (adjusted HR [AHR] 0.90; 95% CI 0.85–0.95), Alzheimer's disease (AD) (AHR 0.88; 95% CI 0.83–0.94), and non-vascular dementias (non-VaDs) (AHR 0.75; 95% CI 0.70–0.81). This new wave of retrospective cohort studies is distinguished by their statistical power and methodological rigour. Data on tens of thousands of individuals were drawn from validated electronic health record systems and national registries. Advanced analytical methods, including propensity score matching and target trial emulation, help mitigate common biases such as immortal time bias and unmeasured confounding, enhancing the credibility of causal inference. Despite these strengths, retrospective cohort studies are inherently limited by their observational nature. Unmeasured confounders, such as cognitive reserve, baseline neurocognitive function, or medication adherence, may still bias the findings. Nevertheless, these data corroborate earlier results from randomized trials. Notably, a recent meta-analysis by Seminer et al.,9 spanning 26 trials, also reported a significant reduction in dementia and cognitive impairment among GLP-1RA users (OR 0.55; 95% CI: 0.35–0.86). However, these associations require caution. Randomized evidence to date does not clearly show cognitive benefit with GLP-1RAs. A systematic review of placebo-controlled randomized controlled trials (RCTs) by Pierret10 found no significant cognitive effects (SMD 0.01; 95% CI −0.02 to 0.04). The GRADE trial11 likewise reported no differences in cognitive outcomes with liraglutide. Similarly, the randomized trial in antipsychotic-treated obese patients12 showed no cognitive benefit of exenatide. Taken together, current RCT data do not support a causal cognitive effect despite encouraging observational signals. Nevertheless, ongoing and upcoming trials with primary cognition-focused endpoints and refined patient selection criteria may provide clearer evidence in the near future. Overall, the available findings raise the possibility that GLP-1RAs may influence pathways relevant to neurodegeneration; however, current evidence remains insufficient to confirm preventive or disease-modifying effects. Interestingly, both clinical9 and real-world evidence3, 7 demonstrate that GLP-1RAs are consistently superior to dipeptidyl peptidase-4 inhibitors (DPP4i) in reducing the risk of dementia. DPP4i are agents with limited the blood–brain barrier (BBB) penetration that prolong endogenous GLP-1 turnover by suppressing incretin-degrading enzymes. This implies that the neuroprotective effect of GLP-1RAs might depend on the direct activation of GLP-1 receptors within the brain, rather than the systemic increase of endogenous incretin levels. Emerging data have indicated diverse functions of GLP-1 signalling in the brain (reviewed by Chen et al.13). GLP-1 receptors are highly expressed in brain regions critical for cognition, including the hippocampus and cortex. Preclinical studies have shown that GLP-1RAs promote anti-inflammatory M2 microglial polarization, enhance regulatory T cell responses, and suppress proinflammatory cytokine release, hence mitigating chronic neuroinflammation. GLP-1RAs also reduce oxidative stress, prevent neuronal apoptosis, and restore central insulin signalling. Key signalling cascades such as PI3K/Akt, MAPK/ERK, and inhibition of GSK-3β are upregulated by GLP-1R activation, promoting synaptic plasticity and neuronal survival. These multifaceted actions likely reflect the evolutionary role of GLP-1 as a gut-brain signal that links postprandial nutrient sensing to central regulation of energy balance, cardiovascular function, and cognition, positioning it as a central regulator of metabolic and neurological homeostasis. Future investigations will elucidate how these functions are related to the aetiology of Alzheimer's disease and other dementias. Notably, in models of Alzheimer's and Parkinson's disease, GLP-1RAs have been shown to reduce amyloid-β deposition, limit tau hyperphosphorylation, and preserve cortical architecture. Additionally, GLP-1RAs improve cerebral perfusion, maintain BBB integrity, and enhance neurogenesis in the hippocampus, all of which contribute to cognitive resilience.13 Beyond their glycaemic effects, GLP-1RAs exert systemic anti-inflammatory actions, promote clinically meaningful weight loss, and beneficially modulate gut microbiome composition factors that may provide additional organ protection in metabolically vulnerable populations.14 However, despite promising population-level findings, the degree of neuroprotection conferred by GLP-1RAs may vary among individuals. Current retrospective studies adjust for age, sex, BMI, and major comorbidities, but do not account for intrinsic biological factors that may influence drug response. This becomes particularly relevant in the context of incretin resistance, a concept akin to insulin resistance, in which different individuals may exhibit heterogeneous responsiveness to GLP-1 signalling.15 Contributing factors could include polymorphisms in the GLP-1 receptor gene, variability in receptor density, and alterations in downstream intracellular signalling cascades. These considerations highlight the need for biomarker-driven stratification strategies to identify patients most likely to benefit from GLP-1RA-based interventions for dementia prevention or treatment. However, incretin resistance still remains a theoretical construct without validated biomarkers in humans. Translating these observational signals into clinical utility demands validation in rigorously designed RCTs. Ongoing and planned programmes, including GLP-1RA/GIPR's cognitive substudies embedded within cardiovascular outcome trials, will be pivotal in determining whether the observed dementia risk reduction reflects a direct neuropharmacological action or an indirect consequence of improved metabolic control. Future RCTs must adopt harmonized cognitive endpoints, integrate longitudinal neuroimaging, and incorporate high-specificity biomarkers such as plasma p-tau217 and neurofilament light chain to capture early neuroprotective effects.16 Before repurposing can be meaningfully considered, large RCTs are needed with clinical cognitive endpoints, biomarker-defined intermediate outcomes, multimodal neuroimaging, APOE-stratified and vascular-risk–stratified analyses, and comprehensive safety evaluation in older and frailer adults. Equally unresolved is the optimal therapeutic window, whether initiation in midlife confers primary prevention or later use in mild cognitive impairment achieves secondary prevention. Integrating biomarker-driven stratification, precision trial design, and real-world implementation research can shift the field from epidemiological association to actionable prevention. If realized, GLP-1RAs could redefine dementia prevention, reframing it as a metabolic–inflammatory target amenable to early pharmacological intervention. Taken together, the findings by Lin et al.3 and others are hypothesis-generating but remain insufficient to support disease-modifying claims without dedicated cognitive outcome trials. While their safety profile and metabolic benefits make GLP-1RAs attractive candidates for future repurposing, such use remains premature in the absence of definitive randomized evidence. However, side effects, including gastrointestinal intolerance, weight loss, and potential gallbladder complications, may limit tolerability in certain vulnerable groups. At the same time, expanding knowledge of the GLP-1 signalling axis offers avenues for drug discovery innovation: dual or triple agonists targeting GLP-1, GIP, and glucagon receptors may yield enhanced neurotrophic effects; small-molecule modulators and central delivery systems could optimize brain penetration and receptor binding. Ultimately, the repurposing of GLP-1RAs represents not only a promising translational opportunity but also a compelling example of how the nexus of metabolism and inflammation may shape the pathogenesis and treatment of neurological diseases. Project administration: Andrej Belančić. Investigation: All authors. Writing—original draft: All authors. Writing—review and editing: All authors. Conceptualization: Andrej Belančić. All authors have read and agreed to the published version of the manuscript. The authors have nothing to report. All authors declare that they do not have any conflict of interest concerning content of this manuscript. No new data were generated. All information is available upon reasonable request from the corresponding author.