Type 2 diabetes mellitus (T2DM) is associated with increased risk of cognitive decline, including mild cognitive disorder, Alzheimer disease, and vascular dementia. Pathophysiologic mechanisms implicated include chronic hyperglycemia, insulin resistance, vascular injury, and neuroinflammation. Observational and mechanistic studies have suggested that some antidiabetic drug classes may exert neuroprotective effects, whereas others—especially when used in advanced disease—may associate with worse cognitive outcomes. Prior studies have often been limited by small samples, short follow-up, or focus on a single drug class, leaving uncertainty about comparative associations across commonly used regimens.
This retrospective cohort study used the TriNetX US Collaborative Network of deidentified electronic health records covering 2010–2024. The network contains records from multiple health-care organizations and includes diagnoses, prescriptions, laboratory values, and demographics coded using ICD-9/ICD-10. Eligible patients were adults aged 40–69 years with a diagnosis of T2DM and at least 1 year of follow-up after the index prescription. The reported analytic sample comprised 1,528,885 adults (mean age 58 years; 49.2% women).
Patients were classified into five mutually exclusive treatment groups according to their first recorded antidiabetic regimen and continued use during follow-up:
If patients initiated additional agents later, they were assigned to the combination group at the time the combination was first recorded.
Primary outcomes were incident diagnoses identified by ICD‑10 codes: mild cognitive disorder (G31.84), Alzheimer disease (G30), and vascular dementia (F01). Outcome ascertainment began 1 day after the index date and continued until loss to follow-up or end of the study period. Analyses were stratified by follow-up duration (<5 years versus ≥5 years) using landmark analyses.
Baseline characteristics were summarized using means with standard deviations for continuous variables and proportions for categorical variables. Propensity score matching (1:1) with a greedy nearest‑neighbor algorithm and caliper of 0.1 pooled standard deviations was used to balance measured covariates between groups. Covariates included demographic factors and clinical comorbidities; glycated hemoglobin was among the included laboratory covariates. Cox proportional hazards models produced hazard ratios (HRs) with 95% confidence intervals (CIs) for incident cognitive outcomes.
Reported key findings from the matched analyses include:
Metformin plus GLP‑1 receptor agonists was associated with lower hazards for all three outcomes: vascular dementia (HR 0.46; 95% CI 0.37–0.57), mild cognitive disorder (HR 0.79; 95% CI 0.66–0.95), and Alzheimer disease (HR 0.46; 95% CI 0.31–0.70).
Metformin plus SGLT‑2 inhibitors was associated with reduced hazard of vascular dementia (HR 0.68; 95% CI 0.54–0.87) but was not reported as significantly associated with Alzheimer disease or mild cognitive disorder in the summary estimates provided.
Metformin plus DPP‑4 inhibitors did not have a highlighted protective association in the summarized results (details were not reported in the provided source text).
Insulin monotherapy was associated with substantially higher hazards across outcomes: vascular dementia (HR 3.27; 95% CI 3.03–3.52), mild cognitive disorder (HR 1.71; 95% CI 1.56–1.87), and Alzheimer disease (HR 1.56; 95% CI 1.32–1.84).
All reported hazard ratios are cause‑specific estimates from Cox models.
The associations of metformin combined with GLP‑1 receptor agonists and SGLT‑2 inhibitors with lower vascular dementia risk are consistent with biologic mechanisms proposed for these classes (GLP‑1 effects on brain insulin signaling and neuroinflammation; SGLT‑2 effects on vascular and cardiorenal outcomes). However, the authors emphasize that the markedly higher hazards observed with insulin monotherapy are likely influenced by confounding by indication and unmeasured markers of diabetes severity (for example, longer diabetes duration, greater burden of cardiovascular or renal disease, and microvascular or macrovascular complications) rather than a direct causal drug effect.
The investigators also note an analytical limitation: standard Cox models estimate cause‑specific hazards and do not account for death as a competing event. In groups with higher mortality and censoring (notably the insulin group), reported hazard ratios may not directly translate to differences in cumulative incidence.
Strengths:
Limitations:
In this large retrospective cohort of adults with T2DM, combinations of metformin with GLP‑1 receptor agonists and with SGLT‑2 inhibitors were associated with lower hazards of vascular dementia, with GLP‑1 combinations also associated with lower hazards of Alzheimer disease and mild cognitive disorder in the reported analyses. Insulin monotherapy was associated with higher hazards for all three cognitive outcomes, a finding the authors caution likely reflects confounding by indication and higher mortality/censoring rather than a direct adverse effect. The findings suggest that antidiabetic medication choice may influence long‑term cognitive outcomes and should be considered alongside other risks and benefits when managing patients with T2DM. Details not explicitly reported in the provided source text include subgroup-specific event rates, absolute risk differences, and analyses for metformin plus DPP‑4 inhibitors; these were not available in the excerpt provided.