Document resource
WHAT IS ALREADY KNOWN ON THIS TOPIC Both glucagon-like peptide-1 receptor agonists (GLP-1 RAs) and metformin have demonstrated potential neuroprotective effects in type 2 diabetes mellitus (T2DM), but no head-to-head real-world comparisons have evaluated their relative efficacy in preventing dementia.WHAT THIS STUDY ADDS This large-scale, propensity score–matched cohort study shows that GLP-1 RAs significantly reduce the risk of overall dementia—particularly Alzheimer’s disease and non-vascular dementias—compared with metformin when used as first-line therapy in patients with T2DM.HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY These findings support the early use of GLP-1 RAs in patients with T2DM at risk for cognitive decline and may influence future diabetes treatment guidelines to prioritize therapies with dual glycemic and neuroprotective benefits.Introduction Type 2 diabetes mellitus (T2DM) is a growing global public health challenge, with its rising prevalence imposing substantial burdens on healthcare systems worldwide. 1 Beyond the well-documented metabolic complications, T2DM significantly increases the risk of cognitive impairment and dementia, particularly Alzheimer’s disease (AD), and other non-vascular dementias (non-VaDs).2 These conditions drastically reduce patients’ quality of life and create immense economic strain on families, healthcare systems, and societies.2 3 Epidemiological studies indicate that individuals with T2DM have a 1.7-fold higher risk of developing dementia compared with the general population.4 With the aging population further exacerbating this issue, preventing T2DM-related dementia has become an urgent priority for public health. Early interventions and effective management strategies are crucial not only for improving individual outcomes but also for alleviating societal and economic burdens.3 5 Both glucagon-like peptide-1 receptor agonists (GLP-1 RAs) and metformin, widely used in T2DM management, have been investigated for their potential to mitigate dementia risk.4–10 GLP-1 RAs have demonstrated robust neuroprotective effects, including reducing neuroinflammation, alleviating oxidative stress, and improving cerebral glucose metabolism.11–17 Several clinical trials have shown that GLP-1 RAs, such as liraglutide or dulaglutide, can enhance cognitive function and reduce cerebral amyloid burden in AD.7 8 18 19 Metformin, the first-line therapy for T2DM,20–23 has also shown promise in reducing oxidative stress and improving insulin sensitivity, potentially offering cognitive benefits.24–28 However, the evidence on metformin’s impact on cognitive function remains inconsistent.9 10 Despite these advancements, there are no direct comparative studies evaluating the effectiveness of GLP-1 RAs versus metformin in preventing T2DM-related dementia, leaving a critical gap in the literature.Given the significant clinical and societal burden of T2DM-related dementia, determining which therapy—GLP-1 RAs or metformin—provides superior protection against cognitive decline is crucial.4–10 While metformin’s status as the first-line therapy is well-established,20–23 if GLP-1 RAs demonstrate superior efficacy in dementia prevention, they could redefine treatment guidelines for T2DM. The potential to prioritize GLP-1 RAs as a first-line therapy could significantly enhance long-term cognitive outcomes for patients. Clarifying this therapeutic distinction will offer crucial guidance for clinicians, policymakers, and researchers, potentially reshaping dementia prevention strategies in T2DM management.Methods Data source This retrospective cohort study used the TriNetX global federated health research network, a robust platform comprising deidentified electronic health records (EHRs) from over 98 healthcare organizations worldwide. This database includes comprehensive longitudinal data on demographics, diagnoses, medications, laboratory results, and healthcare utilization, enabling precise cohort identification and robust outcome analyses. All data are deidentified in compliance with the Health Insurance Portability and Accountability Act (HIPAA) and the General Data Protection Regulation (GDPR), ensuring patient privacy. 29 To ensure data security and compliance with HIPAA and GDPR, stringent measures were implemented. Patient data were fully deidentified, removing direct identifiers like names or social security numbers. Advanced encryption safeguarded EHR transmission and storage, while role-based access restricted data usage to authorized personnel. Analyses were conducted in a secure, ISO/IEC 27001-certified cloud environment provided by TriNetX, with regular audits ensuring regulatory adherence. Aggregated datasets further minimized re-identification risks, maintaining the highest standards of data privacy and security. Advanced tools in the TriNetX platform mitigate confounding factors and enhance data reliability, supporting high-impact real-world studies in diabetes and dementia research.29 Study population and design This study compared adults with T2DM who initiated either GLP-1 RAs or metformin as first-line therapy between November 20, 2004, and November 20, 2024. Patients were identified from a global federated health network encompassing 98 healthcare organizations. Eligibility criteria included a T2DM diagnosis (ICD-10-CM E11.x) and a prescription for GLP-1 RAs (ATC A10BJ) or metformin (ATC A10BA). The index date was defined as the earliest prescription date post-diagnosis, with a minimum of 24 months of follow-up to ensure rigorous outcome assessment and reduce potential bias. The 24-month follow-up period was selected to ensure sufficient time to observe dementia outcomes while accommodating real-world study constraints. This duration captures incident cases and reduces biases from short-term fluctuations or diagnostic misclassifications. It aligns with prior research showing significant cognitive outcomes often emerge within 2–5 years of intervention. 3 Additionally, it provides ample data to evaluate the sustained effects of GLP-1 RAs and metformin while minimizing attrition bias.Exposure definitions Patients were stratified into two treatment cohorts based on their initial antidiabetic therapy: the GLP-1 RA cohort and the metformin cohort. Cross-over between the two drug classes was prohibited to maintain treatment consistency, although the use of additional glucose-lowering medications—such as sulfonylureas, thiazolidinediones, and insulins—was allowed if initiated after the index date. This strategy ensures that differences in observed outcomes can be attributed to the primary treatment, thereby minimizing confounding.To strengthen methodological rigor, patients who had been prescribed newer agents, including sodium-glucose cotransporter-2 inhibitors (SGLT2i; ATC A10BK) or dipeptidyl peptidase-4 inhibitors (DPP-4i; ATC A10BH), were excluded from the analysis. Furthermore, only patients with continuous prescriptions for their assigned therapy over a minimum duration of 6 months were included in the study population. This inclusion criterion ensured sustained exposure during the early treatment phase, thereby reducing the likelihood of treatment misclassification and enhancing the reliability of outcome assessment.Exclusion criteria To ensure analytical validity, patients were excluded if they had prior exposure to the alternate drug class, pre-existing dementia diagnoses (ICD-10-CM F01: vascular dementia (VaD), G30: Alzheimer’s disease, F03: unspecified dementia), significant cognitive or neurological impairments, psychiatric disorders (ICD-10-CM F32–F39: depressive and mood disorders), Parkinson’s disease (ICD-10-CM G20), cerebral atherosclerosis (ICD-10-CM I67.2), or alcohol-related cognitive disorders (ICD-10-CM F10.27). Additionally, individuals with end-stage renal disease (ICD-10-CM N18.6) or undergoing dialysis within 6 months before the index date were excluded. These exclusion criteria ensured a precise and comprehensive comparison of GLP-1 RAs and metformin in reducing dementia risk, thereby providing reliable insights for clinical and real-world applications.Endpoints measures The primary outcome was the incidence of overall dementia, defined as the first occurrence of a diagnosis of VaD, AD, unspecified dementia, or dementia associated with other diseases. Secondary outcomes included the incidence of specific dementia subtypes, such as VaD, AD, and other forms of dementia, as well as all-cause mortality.To address potential biases from pre-existing or undiagnosed dementia and to allow sufficient time for the therapeutic effects of the medications to manifest, a 6-month washout period was applied. Outcomes occurring within this interval were excluded from the analysis. Follow-up began on day 181 after the index date and continued until the first occurrence of an outcome, loss to follow-up, or the conclusion of the study period.Patients with prior diagnoses of dementia or who experienced all-cause mortality before the washout period were excluded. Additionally, participants with less than 24 months of follow-up after the washout period were excluded to ensure an adequate observation period for capturing meaningful clinical outcomes.Covariates To address potential confounding factors, we adjusted for an extensive set of baseline covariates. These included demographic variables (age, sex, race/ethnicity), clinical characteristics (eg, hypertension, ischemic heart disease, cerebrovascular disease), and laboratory measurements such as hemoglobin A1c and body mass index (BMI). Concomitant medication use, including insulin, sulfonylureas, thiazolidinediones, alpha-glucosidase inhibitors, and statins, was also accounted for ( online supplemental table 1).SP110.1136/bmjdrc-2025-004902.supp1Supplementary data Propensity score matching (PSM) with a 1:1 nearest-neighbor algorithm and a caliper width of 0.1 was used to balance these covariates between the GLP-1 receptor agonist and metformin cohorts. The success of matching was validated by ensuring standardized mean differences (SMDs) of less than 0.1 for all covariates.Statistical analysis Time-to-event outcomes, including dementia, were analyzed using Kaplan-Meier survival curves to estimate cumulative incidence and log-rank tests to compare dementia incidence between cohorts. Cox proportional hazards models were used to calculate HRs with 95% CIs, addressing residual imbalances in covariates following PSM. Schoenfeld residuals validated the proportional hazards assumption, ensuring the robustness of the models. To account for potential time-dependent effects of treatment, time-varying covariates were incorporated, allowing for dynamic assessment of changes in patient profiles and treatment exposures during follow-up.Follow-up started on day 181 post-index, following a 6-month washout period designed to reduce reverse causation bias by excluding outcomes that might be influenced by pre-existing conditions or early treatment effects. Sensitivity analyses were performed using the pre-PSM population to validate the primary findings. A multivariable Cox proportional hazards model adjusted for all baseline covariates from online supplemental table 1, including demographics, comorbidities, and medication use. Results aligned with the primary analysis, confirming the robustness of the association between GLP-1 RAs and reduced dementia risk compared with metformin, while accounting for potential confounding factors.Subgroup analyses examined potential effect modifications based on baseline characteristics such as age, sex, race/ethnicity, concurrent medication use (eg, insulin, sulfonylureas, statins), and glycemic control categories (eg, hemoglobin A1c levels). Statistical significance was determined using a two-sided p value threshold of <0.05. All analyses were performed on the TriNetX platform, which adheres to stringent quality control standards and is validated for real-world data analyses. To enhance generalizability, key findings were replicated across independent database subsets and stratified by healthcare settings and geographic regions.Results Baseline characteristics of patients with T2DM treated with GLP-1 RAs versus metformin In this analysis, we compared the baseline characteristics of patients with T2DM receiving GLP-1 RAs or metformin as first-line antidiabetic therapy before and after PSM. Before matching, significant imbalances were noted in age, sex distribution, and comorbidity prevalence, including higher proportions of male patients and greater comorbidity burden among those treated with metformin. Additionally, metabolic measures, such as BMI and glycated hemoglobin (HbA1c), reflected more pronounced obesity and glycemic dysregulation in the GLP-1 RA group.After PSM, the two cohorts demonstrated excellent balance across key covariates, with absolute standardized mean differences (ASMDs) consistently below 0.1. The matched cohorts achieved comparability in age (mean age ~58 years), sex distribution (~36% male), comorbidity profiles (eg, similar rates of coronary artery disease and cerebrovascular disease), and metabolic parameters (eg, HbA1c and BMI categories). This matching minimized baseline differences, ensuring the validity of subsequent analyses comparing dementia incidence and other outcomes between these two treatment groups.Incidence of dementia and mortality following GLP-1 RA versus metformin use in type 2 diabetes Among 87,229 matched patients in each cohort, GLP-1 RA use was associated with a significantly lower cumulative incidence of overall dementia compared with metformin (2.4% vs 4.8%; AHR 0.90; 95% CI 0.85 to 0.95) ( table 1 and figure 1). Subtype analyses revealed reductions in AD (1.2% vs 2.6%; AHR 0.92; 95% CI 0.85 to 0.99) and other dementia subtypes (1.0% vs 2.4%; AHR 0.88; 95% CI 0.81 to 0.96). However, no significant difference was observed for VaD (0.7% vs 1.3%; AHR 1.00; 95% CI 0.91 to 1.14).Table 1Comparison of outcome incidence in patients with type 2 diabetes receiving GLP-1 RAs versus metformin as first-line therapy following propensity score matchingOutcomePatients with outcome, No./Total No. (%)AHR (95% CI)GLP-1 RA groupMetformin groupPrimary endpoint Cumulative incidence of overall dementia2130/87,229 (2.4%)4215/87,229 (4.8%)0.90 (0.95 to 1.00)Secondary endpoints Incidence of vascular dementia618/87,229 (0.7%)1110/87,229 (1.3%)1.00 (0.91 to 1.14) Incidence of Alzheimer’s disease dementia1080/87,229 (1.2%)2300/87,229 (2.6%)0.92 (0.85 to 0.99) Incidence of other dementia subtypes902/87,229 (1.0%)2088/87,229 (2.4%)0.88 (0.81 to 0.96) All-cause mortality rate4201/87,229 (4.8%)7698/87,229 (8.8%)0.89 (0.81 to 0.95)AHR, adjusted HR; GLP-1 RA, glucagon-like peptide-1 receptor agonist.Figure 1Kaplan-Meier curves showing dementia incidence in patients with type 2 diabetes receiving glucagon-like peptide-1 receptor agonists (GLP-1 RAs) compared with metformin as initial therapy after propensity score matching.In terms of secondary outcomes, the all-cause mortality rate was notably reduced in the GLP-1 RA group (4.8% vs 8.8%; AHR 0.89; 95% CI 0.81 to 0.95). These findings underscore the potential advantages of GLP-1 RAs in mitigating dementia and mortality risks among patients with T2DMs.Subgroup analysis of dementia risk Across all dementia outcomes, GLP-1 RAs demonstrated significant reductions in risk compared with metformin, particularly in older adults, females, and White and Black patients. For overall dementia, the aHR was 0.87 (95% CI 0.84 to 0.91; p<0.0001), with consistent findings across subgroups. The benefits were most pronounced in patients aged ≥60 years, with significant reductions noted in those aged 60–79 years (aHR 0.85; 95% CI 0.80 to 0.90) and ≥80 years (aHR 0.80; 95% CI 0.74 to 0.88). Both sexes benefited, although females had a slightly lower aHR (0.83; 95% CI 0.79 to 0.89) compared with males (0.90; 95% CI 0.84 to 0.97).For AD, GLP-1 RAs reduced risk by 12% overall (aHR 0.88; 95% CI 0.83 to 0.94). Older age groups, particularly those ≥60 years, experienced the strongest benefits, with females showing a more substantial reduction (aHR 0.82; 95% CI 0.75 to 0.89). Racial subgroup analysis revealed significant reductions among White patients (aHR 0.85; 95% CI 0.79 to 0.92), while other racial groups showed limited effects.VaD risk showed no significant difference in the overall population (aHR 1.00; 95% CI 0.92 to 1.08), although females had a modest risk reduction (aHR 0.88; 95% CI 0.79 to 0.99). No consistent trends were observed across age or racial subgroups.For other dementia subtypes, GLP-1 RAs provided a robust protective effect (aHR 0.75; 95% CI 0.70 to 0.81), with pronounced benefits in patients aged 40–79 years. Both sexes and White and Black patients demonstrated significant reductions in risk. Concurrent medication use, such as statins (aHR 0.82; 95% CI, 0.74 to 0.91) and insulin (aHR 0.78; 95% CI 0.70 to 0.86), further enhanced the protective effects of GLP-1 RAs.These findings suggest GLP-1 RAs as a promising therapeutic option for mitigating dementia risk in type 2 diabetes, particularly in older adults, females, and specific racial groups, while highlighting the need for tailored interventions based on individual characteristics.Discussion This study represents the first and largest direct comparison of GLP-1 RAs and metformin in mitigating dementia risk among patients with T2DM. Historically, metformin has been the cornerstone of first-line therapy for T2DM due to its well-established efficacy in glycemic control and its potential neuroprotective effects. 20–28 However, whether initiating GLP-1 RAs as first-line therapy could offer superior benefits in reducing diabetes-related dementia risk has remained unaddressed. The findings demonstrated that GLP-1 RAs significantly reduced the incidence of overall dementia compared with metformin (2.4% vs 4.8%; AHR, 0.90; 95% CI 0.85 to 0.95), with pronounced benefits observed in AD and other dementia subtypes. Notably, GLP-1 RAs also provided substantial reductions in all-cause mortality. These findings address a key knowledge gap by directly comparing the neuroprotective efficacy of GLP-1 RAs and metformin in dementia prevention. Given the increasing global burden of diabetes-related cognitive decline and the lack of head-to-head comparisons between GLP-1 RAs and metformin, our findings provide actionable insights for clinical decision-making and may inform future guidelines. Additionally, the study underscores the importance of tailoring therapeutic strategies to optimize both metabolic and neuroprotective outcomes in high-risk T2DM populations. Further randomized controlled trials (RCTs) are warranted to confirm these findings and evaluate the long-term cognitive benefits of GLP-1 RAs.GLP-1 RAs demonstrate superior efficacy compared with metformin in reducing the risk of AD and other non-VaDs in patients with T2DM, as evidenced by our study and corroborated by other research. In our cohort, GLP-1 RA use was associated with a significant reduction in the risk of AD (AHR 0.88; 95% CI 0.83 to 0.94) and other non-VaDs (AHR 0.75; 95% CI 0.70 to 0.81) compared with metformin, underscoring their potential neuroprotective advantage. Similar findings have been reported in clinical trials and population-based studies. For instance, an RCT of liraglutide in patients with early AD demonstrated improvements in cognitive function and reductions in cerebral amyloid burden,30 aligning with the mechanisms identified in our study.31–33 Another large-scale observational study of Danish registries showed a lower incidence of dementia among GLP-1 RA users compared with other antidiabetic therapies, consistent with our results.34 Mechanistically, GLP-1 RAs reduce neuroinflammation, enhance synaptic plasticity, and modulate amyloid-beta and tau pathologies—features that are central to AD’s neurodegenerative process.15 31–33 In contrast to metformin’s predominantly systemic metabolic effects, GLP-1 RAs exert direct neuroprotective actions, enhancing blood-brain barrier integrity and promoting neurotrophic signaling.35–37 These findings collectively highlight the promise of GLP-1 RAs in mitigating the burden of neurodegenerative dementias in T2DM, emphasizing their potential as a preferential therapeutic option over metformin for this high-risk population.The differential effects of GLP-1 RAs and metformin on overall dementia and VaD highlight the complexity of their mechanisms and the pathophysiology of these conditions. Both medications demonstrate neuroprotective properties, such as reducing neuroinflammation and oxidative stress, improving insulin sensitivity, and enhancing cerebrovascular health, which likely contribute to their benefits in overall dementia.11–17 24–28 However, the multifactorial nature of VaD,38 driven by cerebrovascular damage such as small vessel disease and white matter lesions, poses significant challenges for pharmacological interventions targeting metabolic or neurodegenerative pathways.3 5 6 Interestingly, a modest but significant reduction in VaD risk was observed exclusively in female patients treated with GLP-1 RAs (aHR 0.88; 95% CI 0.79 to 0.99), whereas no significant differences were noted in the overall population (figure 2B). This finding may be influenced by estrogen’s vascular and neuroprotective effects,39 which could enhance the benefits of GLP-1 RAs or metformin in postmenopausal women.8 Sex-specific metabolic and inflammatory differences in diabetes further underscore the potential role of biological variables in modulating treatment efficacy.40 The limited impact of GLP-1 RAs and metformin on VaD may also reflect study design limitations, including potential misclassification, underdiagnosis of VaD, and insufficient statistical power due to its relatively low incidence. These findings emphasize the need for personalized therapeutic strategies and future research to explore sex-specific factors, longer follow-up durations, and synergistic effects of combined therapies to optimize dementia prevention in T2DM.Figure 2Subgroup analysis of dementia risk in patients with type 2 diabetes receiving glucagon-like peptide-1 receptor agonists (GLP-1 RAs) compared with metformin as initial therapy. aHR, adjusted HR.GLP-1 RAs demonstrate substantial promise in reducing the risk of T2DM-related dementia, particularly AD and other non-VaDs, through their broad neuroprotective mechanisms (figure 2 A-D). Unlike metformin, whose benefits primarily derive from systemic metabolic effects such as improved insulin sensitivity and reduced oxidative stress,24–28 GLP-1 RAs exert direct central nervous system effects by crossing the blood-brain barrier.35–37 These mechanisms include attenuating neuroinflammation, reducing oxidative stress, promoting the release of neurotrophic factors critical for neuronal survival and synaptic plasticity, and enhancing cerebral glucose metabolism—a key function often impaired in neurodegenerative diseases like Alzheimer’s.11–17 35–37 In addition, GLP-1 RAs uniquely target the pathological hallmarks of AD, including reducing beta-amyloid deposition and tau protein hyperphosphorylation,15 31–33 while restoring blood-brain barrier integrity to prevent neurotoxic insults. These findings are supported by comparative studies, which have demonstrated cognitive improvements and reduced neurodegenerative burden in patients treated with GLP-1 RAs such as liraglutide and semaglutide.7 8 18 19 Our study corroborates these results, showing significant reductions in AD risk associated with GLP-1 RAs, highlighting their potential superiority over metformin in addressing the multifaceted pathophysiology of neurodegenerative dementias. Given the severe societal, familial, and economic burden of diabetes-related dementia,2 3 these findings raise important considerations about the role of GLP-1 RAs as first-line therapies in T2DM management. While further long-term studies are warranted to validate these results, integrating GLP-1 RAs as primary therapeutic agents may represent a paradigm shift in preventing the cognitive complications of diabetes.This study represents the largest and most comprehensive evaluation of GLP-1 RAs compared with metformin as first-line antidiabetic therapies in mitigating the risk of T2DM-related dementia, with a specific emphasis on AD and other non-VaDs. Leveraging the global federated health research network, encompassing deidentified data from over 98 healthcare organizations worldwide, this analysis ensures broad generalizability and robust methodological rigor. Advanced techniques, including PSM and sensitivity analyses, minimized confounding and validated the robustness of the findings. GLP-1 RAs demonstrated significant reductions in AD risk (AHR 0.88; 95% CI 0.83 to 0.94) and other non-VaDs (AHR 0.75; 95% CI 0.70 to 0.81), consistent with prior research highlighting their neuroprotective mechanisms, such as reducing beta-amyloid accumulation, alleviating oxidative stress, and enhancing cerebral glucose metabolism.7 8 18 19 By providing direct comparative evidence, this study addresses a critical gap in the literature and underscores the potential of GLP-1 RAs as a superior therapeutic option for preventing T2DM-related cognitive decline. These findings offer actionable insights for clinicians and policymakers, emphasizing the importance of integrating neuroprotective strategies into diabetes management to address the growing burden of dementia.This study has several limitations that should be acknowledged. First, as a retrospective cohort study using real-world data, the potential for residual confounding remains, despite rigorous PSM and sensitivity analyses. Second, the reliance on deidentified EHRs introduces variability in diagnostic coding, which may lead to misclassification of dementia subtypes or underreporting of outcomes, particularly VaD. Third, the exclusion of patients with prior exposure to both GLP-1 RAs and metformin may limit the generalizability to individuals with mixed treatment histories. Fourth, the follow-up period, while sufficient for observing dementia outcomes, may not fully capture long-term cognitive effects, especially given the progressive nature of neurodegenerative diseases like Alzheimer’s. Additionally, the absence of direct cognitive performance measurements or biomarkers (eg, amyloid or tau) limits mechanistic insights into the observed benefits of GLP-1 RAs. Lastly, although the study population was diverse, it may not represent certain high-risk groups, such as those with advanced diabetes complications or significant comorbidities. Despite these limitations, the findings remain robust, as the lower mortality in the GLP-1 RA group—paradoxically reducing dementia risk due to longer survival—suggests that the study may have underestimated the true neuroprotective effects of GLP-1 RAs compared with metformin. This underscores the potential superiority of GLP-1 RAs and highlights the need for further randomized controlled trials with extended follow-up, standardized diagnostics, and neuroimaging to validate these findings.Conclusions Our study provides the largest comparison of GLP-1 RAs versus metformin in reducing T2DM-related dementia risk, particularly AD and non-VaDs. GLP-1 RAs exhibited superior efficacy over metformin in mitigating dementia risk, particularly for AD (AHR 0.88) and non-VaDs (AHR 0.75), likely through distinct neuroprotective mechanisms such as beta-amyloid clearance and enhanced cerebral glucose metabolism. Despite limitations such as residual confounding, the findings remain robust and suggest GLP-1 RAs could redefine T2DM-related dementia prevention strategies, warranting further RCTs to confirm these results.