This study evaluated whether initiation of GLP-1 receptor agonists is associated with longer clinically recorded Alzheimer’s disease (AD)-type dementia–free survival compared with sulfonylureas (SU) or SGLT2 inhibitors among older adults with type 2 diabetes. The research used a target trial emulation framework and causal machine-learning methods to estimate both average and individualized treatment effects over a 48-month horizon.
Investigators used the NIH All of Us electronic health record (EHR) Controlled Tier Dataset v8 (data cutoff October 1, 2023). Eligible individuals were adults aged 55 years or older with type 2 diabetes, a 12-month washout period, and no prior clinical record of dementia before treatment initiation. Separate comparisons were constructed for GLP-1 receptor agonist initiation versus sulfonylureas and GLP-1 initiation versus SGLT2 inhibitor initiation.
To address confounding and estimate effects comparable to a randomized target trial, the study applied propensity score weighting and doubly robust estimation. For individualized treatment effect estimation and assessment of heterogeneity, the authors used causal survival forests to estimate effects on 48-month restricted mean survival time (RMST) free from clinically recorded AD-type dementia.
The GLP-1 versus SU analysis included 6,328 individuals. The average treatment effect (ATE) favored GLP-1 initiation with a small but statistically significant increase in dementia-free survival over 48 months: ATE 0.21 months (95% CI: 0.07–0.35). The authors reported an approximate 48-month number-needed-to-treat (NNT) of 202 to prevent one additional clinically recorded AD-type dementia event during follow-up. The subgroup with the highest predicted benefit gained 0.45 months (95% CI: 0.28–0.62) of dementia-free survival.
The GLP-1 versus SGLT2 inhibitor comparison included 3,070 individuals. The average effect of GLP-1 initiation on 48-month dementia-free survival was not statistically significant: ATE 0.06 months (95% CI: –0.18 to 0.31). The estimated 48-month NNT was approximately 245. Despite the null average effect, treatment effects were heterogeneous; the highest-benefit stratum in this comparison gained 0.83 months (95% CI: 0.49–1.17) of dementia-free survival.
Using causal survival forests to probe treatment-effect heterogeneity, the study found that predicted greater benefit from GLP-1 initiation was associated with several baseline characteristics: older age, concurrent insulin use, lower glycated hemoglobin (HbA1c), and lower body mass index (BMI). These associations describe which subgroups in the observational data were more likely to show larger estimated gains in 48-month dementia-free survival with GLP-1 initiation.
Authors interpret the results as hypothesis-generating. The findings suggest that initiation of GLP-1 receptor agonists may modestly delay clinically recorded AD-type dementia compared with sulfonylureas. Comparative effectiveness versus SGLT2 inhibitors appeared variable across patient subgroups, with no statistically significant average benefit but identifiable strata with larger gains.
The investigators emphasize that diabetes treatment selection should continue to be guided primarily by established glycemic, cardiovascular, renal, and patient-centered considerations, noting that the present results do not by themselves establish a change in prescribing practice. Limitations inherent to observational target-trial emulation and reliance on clinically recorded dementia outcomes in EHR data underpin the authors’ caution: these are not randomized trial results and remain hypothesis generating.
Individual-level data were obtained from the All of Us Research Program Controlled Tier Dataset and are available only to authorized researchers through the secure Researcher Workbench, subject to institutional agreements and required training. Aggregate data supporting the manuscript’s findings are reported in the paper and appendix. The All of Us Institutional Review Board determined that activities using Controlled Tier data did not constitute human-subjects research requiring individual project review under 45 CFR 46; participants provided informed consent for the All of Us Research Program. Funding for this work came from the German Federal Ministry of Research, Technology and Space (03LWH0181B). The authors declared no competing interests.
Note: All numeric estimates, cohort sizes, statistical methods, and conclusions in this summary are taken directly from the source article. The authors framed results as exploratory and hypothesis generating; no additional external evidence or data were added here.