In a multinational, retrospective cohort study of 214,442 people living with type 2 diabetes without evidence of neurodegenerative disorders, the initiation of GLP-1 RAs was associated with a 19% relative risk reduction and 0.6% absolute risk reduction of new-onset neurodegenerative disorders compared to DPP4 inhibitors. Initiation of GLP-1 RAs, as opposed to the comparator, was associated with a lower risk of any dementia, Alzheimer’s disease, or vascular dementia onset. There was no evidence of a difference in the risk of developing Parkinson’s disease or any synucleinopathy with GLP-1 RAs compared with DPP4 inhibitors. Results showed similar trends in sensitivity analyses across baseline subgroups, when considering only confirmed events, and when using basal insulin as a comparator. These findings support the rationale for conducting dedicated clinical trials to test the hypothesis that GLP-1 RAs possess neuroprotective properties in people living with type 2 diabetes and without prior evidence of neurodegeneration.
Several studies have assessed the association of GLP-1 RAs therapy with neurocognition and neurodegeneration outcomes in populations of people with type 2 diabetes mostly without prior history of neurodegeneration. An analysis of the REWIND randomized controlled trial (RCT) found a lower risk for cognitive decline with dulaglutide versus placebo, after adjusting for individual baseline cognitive function, among 8828 people with type 2 diabetes and high cardiovascular risk, although the pre-specified unadjusted main analyses failed to find a significant difference between the groups [18]. An exploratory analysis of the GRADE RCT failed to find evidence of a slower deterioration in cognitive function among people with type 2 diabetes treated with liraglutide compared to sitagliptin, glimepiride, or insulin glargine (N = 3721), although the number of individuals with a new diagnosis of dementia was not reported [31]. In a meta-analysis of three RCTs (LEADER [34]SUSTAIN-6 [35]and PIONEER-6 [36]) of 15,820 people with type 2 diabetes and high cardiovascular risk, dementia onset occurred in 15 and 32 participants in the GLP-1 RAs and placebo arms, respectively (HR 0.47 [0.25–0.86]) [21]. A lower risk of dementia or Alzheimer’s disease with GLP-1 RAs compared to other glucose-lowering agents has been observed in case–control studies of people with type 2 diabetes [21, 22, 29]. A TriNetX-based cohort study found a lower risk of new-onset Alzheimer’s disease among 17,104 people with type 2 diabetes who initiated treatment with semaglutide, with 24 to 27 participants in the semaglutide group developing Alzheimer’s disease [32]. A real-world study from the United Kingdom found that 10,684 people with type 2 diabetes who initiated GLP-1 RAs had a 62% reduction in the risk of Parkinson’s disease, with 17 participants in the GLP-1 RA group developing the disease [30]. A 2025 study of the U.S. Veterans type 2 diabetes database found that initiation of GLP-1 RAs versus DPP4 inhibitors was associated with 10% (95% CI 6–13) lower risk of neurocognitive disorders without evidence for a difference in the risk of Parkinson’s disease onset [37]. Collectively, these studies support the hypothesis that GLP-1 RAs are associated with improved neurocognition and neurodegeneration outcomes in people with type 2 diabetes, although many of the studies employ case–control designs [21, 22, 29]use data from a single country [21, 29,30,31]have relatively small sample sizes or short follow-ups [21, 29,30,31,32]test only limited outcomes, or do not assess an array of GLP-1 RAs separately [20, 21, 29, 31, 32, 37].
Our study adds multiple novel aspects to this existing literature. First, we utilize a global cohort of participants, supporting the generalizability of the results. Second, the large cohorts of 214,442 matched participants with approximately 866,873 patient-years enabled the identification of 5455 new cases of neurodegeneration, reinforcing the robustness of the results. The large sample size also provided the power to separately test the onset of various neurodegenerative disorders, demonstrating distinct differences in the risk of Alzheimer’s disease, Parkinson’s disease, synucleinopathy, dementia, vascular dementia, and other dementias among people using GLP-1 RAs compared with other glucose-lowering agents. Third, by applying a new-initiator, active comparator, propensity-score-matching, cohort design, we reduced the risk of biases and confounding factors, although not entirely eliminating them. Fourth, we performed multiple sensitivity analyses, including comparisons of GLP-1 RAs with two separate glucose-lowering agents, considering the new use of neurodegeneration-related drugs in outcome assessments, subgroup analyses according to demographic and medical background, and analysis limited to confirmed cases with events recorded on two different occasions. The results of these sensitivity analyses were consistent with the main analysis, suggesting that initiation of GLP-1 RAs compared with DPP4 inhibitors or basal insulin is associated with a lower risk of neurodegeneration onset in people with type 2 diabetes, across various demographic and medical backgrounds. These results support dedicated RCTs that will assess whether GLP-1 RAs improve neurocognition, neurodegeneration, and dementia outcomes in individuals with type 2 diabetes without a history of neurodegeneration.
Preclinical studies have demonstrated a potential role of GLP-1 receptor activation in mechanisms of learning and memory [12]with evidence supporting potential benefits in models of Alzheimer’s disease [14, 15, 38] and Parkinson’s disease [19]. Possible mechanisms may include inhibiting neuroinflammation [17]interfering with microglial activation [19]blocking blood–brain barrier leakage [18]and modulating synaptic plasticity [18]among other potential mechanisms [13, 17]. The extent to which different GLP-1 RAs cross the blood–brain barrier remains under investigation. Dulaglutide was shown to cross at a slow rate, whereas semaglutide and liraglutide demonstrate even lower permeability [39, 40]yet even GLP-1 RAs with limited penetration have been shown to directly influence brainstem regions [41]. Our findings suggest that the initiation of either dulaglutide, liraglutide, or semaglutide is associated with a reduced risk of neurodegeneration, indicating that the underlying mechanisms may not rely on the direct crossing of the blood–brain barrier. Instead, they may reflect a broader indirect therapeutic effects of GLP-1 RAs on glycemic control, body weight, blood pressure, albuminuria, and systemic inflammation [35, 42,43,44]. These effects could potentially reduce vascular-related brain injury and promote cerebrovascular integrity, thereby reducing atherosclerotic risk among other effects. In line with this hypothesis, and as demonstrated in previous studies [8,9,10,11]we observed a lower risk of new-onset stroke or death in individuals initiating GLP-1 RAs compared with DPP4 inhibitors. Taken together, while our data suggest an association between initiation of GLP-1 RA in people living with type 2 diabetes and reduced risk of neurodegeneration, the exact mechanisms deserve further investigation.
Several RCTs tested the efficacy of GLP-1 RAs in improving neurological outcomes in people living with Parkinson’s disease. A 2013 single-blind proof-of-concept trial of 45 people with moderate Parkinson’s disease found that 12-month exenatide therapy compared with non-use was associated with improvements in the Movement Disorders Society Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) [24]. A follow-up placebo-controlled trial found that exenatide mitigated the deterioration in the MDS-UPDRS part III off-medication score among 62 people with moderate Parkinson’s disease [23]. A 2024 placebo-controlled trial found that lixisentide, compared with placebo, was associated with mitigation of progression in the MDS-UPDRS part III [23]. The 2025, phase 3, placebo-controlled, EXENATIDE-PD3 trial was the largest and longest study that assessed neurological outcomes with a GLP-1 RA in people living with Parkinson’s disease. Following 194 patients for 96 weeks, the trial did not find a significant difference in the MDS-UPDRS part III with exenatide versus placebo [28]. In our real-world study, we did not find evidence for a difference in the risk of new incidence of Parkinson’s disease among people with type 2 diabetes initiating GLP-1 RAs compared with those initiating DPP4 inhibitors. In a sensitivity analysis, the risk of Parkinson’s disease onset was 15% (95% CI 3–25) lower in those initiating a GLP-1 RA when compared with basal insulin. These discrepancies can be explained by data that support potential biochemical and clinical benefits of DPP4 inhibitors with respect to Parkinson’s disease outcomes [45, 46]although, on the other hand, it is possible that the GLP-1 RA to basal insulin comparison is subject to residual confounding. Notably, Parkinson’s disease onset in this study was detected based on ICD-10 codes, which have been shown to have approximately 50% misclassification rates with other Parkinsonian disorders [47, 48]limiting the specificity of this outcome.
Several placebo-controlled RCTs tested liraglutide in patients with mild-to-moderate Alzheimer’s disease dementia. A 2016 study of 38 participants found that 26-week liraglutide therapy prevented the decline in cerebral glucose consumption without affecting cognitive function [26]. The phase IIb ELAD trial found that liraglutide therapy in 204 people with mild-to-moderate Alzheimer’s dementia did not mitigate the decline in brain glucose metabolism (primary endpoint), although it mitigated the decline in cognitive function (secondary endpoint) and MRI-measured temporal lobe and whole cortical volume (exploratory endpoint) [49, 50]. The ongoing EVOKE and EVOKE + RCTs (NCT04777396 and NCT04777409) assess the effect of 104–156 weeks of 14 mg oral semaglutide compared with placebo on cognitive function in 1840 people with early-stage symptomatic Alzheimer’s disease, expected to be recruited from 40 different countries. The results of this large trial will provide important, and hopefully conclusive, evidence on the potential disease-modifying effects of semaglutide in people already diagnosed with Alzheimer’s disease [43].
This study has several limitations. Its retrospective, real-world design remains vulnerable to residual confounding and information bias, including detection bias. Although propensity-score matching incorporated key covariates, such as body mass index and prior cardiovascular or kidney disease in order to mitigate indication bias, it could not eliminate residual confounding. Notably, post-matching, modest imbalances persisted in some variables in the comparison of GLP-1 RAs with basal insulin, particularly higher blood pressure in those initiating GLP-1 RAs. For the matching process, due to platform constraints, continuous variables were categorized, which may have resulted in information loss. Prescription data lacked details on dose, treatment duration, and adherence, limiting the specificity of exposure-outcome assessment. Follow-up was confined to encounters captured within participating organizations, so events occurring elsewhere may have been missed. We did not have access to participants’ cognitive test assessments. Neurodegenerative outcomes were identified solely through ICD-10 codes entered by various clinicians; we lacked confirmation by neurology specialists or other adjudications. While the confirmed events analysis had potentially higher specificity, it was limited by reduced sensitivity and by the potential of immortal time bias between the first and second events. The TriNetX platform provides only aggregated data, limiting the use of advanced modeling techniques that address time-varying confounders or mediation analyses focused on changes in body weight or plasma glucose. Further research is needed to assess the association between GLP-1 RA-mediated reduction in body weight and neurodegeneration outcomes. Some specific subgroups had a small sample size, limiting the ability to draw a definitive conclusion. The platform does not support subgroup analysis by country, limiting assessment of heterogeneity of the associations by geographic variation or by types of health-care systems.
In conclusion, in a global real-world cohort of people living with type 2 diabetes without evidence of neurodegeneration, the initiation of GLP-1 RAs, compared to DPP4 inhibitors or basal insulin, was associated with a lower risk of new-onset neurodegeneration. Dedicated clinical outcome trials are needed to assess the effects of GLP-1 RAs on neurocognition and neurodegeneration in this population.
