Observational evidence has suggested potential neuroprotective effects of glucagon‑like peptide‑1 receptor agonists (GLP‑1 RAs). The present study aimed to determine whether measured educational attainment and Alzheimer disease genetic risk were associated with initiation of GLP‑1 RAs and with incident dementia, and whether adjusting for these measured factors materially altered the estimated association between GLP‑1 RA initiation and dementia among adults with type 2 diabetes.
The investigators analyzed linked electronic health record, survey, and genetic data from participants in the All of Us Research Program. The analytic sample included 14,364 individuals with type 2 diabetes. Reported cohort characteristics included a mean age of 60.2 years, 54.2% female, and a mean follow‑up duration of 4.3 years. The study is reported as a preprint and has not undergone peer review.
The primary exposure was initiation of a GLP‑1 receptor agonist. The primary comparator group comprised initiators of non–sodium‑glucose cotransporter 2 (non‑SGLT2) second‑line glucose‑lowering therapies. The investigators also specified a separate comparison using SGLT2 inhibitor initiators. Models compared dementia incidence following treatment initiation across these groups.
To assess Alzheimer disease genetic susceptibility the authors included APOE ε4 status and a non‑APOE genetic risk score. Educational attainment was measured from linked survey data. These measured variables were evaluated both as predictors of treatment initiation and as potential confounders of the association between GLP‑1 RA initiation and incident dementia.
The study estimated associations of educational attainment and Alzheimer disease genetic risk with treatment initiation and with incident dementia. Hazard ratios for incident dementia comparing GLP‑1 RA initiators with comparator initiators were estimated before and after additional adjustment for educational attainment, APOE ε4, and non‑APOE genetic risk. Specific modeling details and covariates beyond these measured factors are described in the full preprint and supplementary materials available through the All of Us Researcher Workbench.
Among the 14,364 participants, the hazard ratio for incident dementia comparing GLP‑1 RA initiation with non‑SGLT2 second‑line therapy initiation was 0.85 (95% CI 0.64–1.12) in models estimated prior to adjustment for education and Alzheimer disease genetic risk. After adjustment for measured educational attainment, APOE ε4, and non‑APOE genetic risk, the estimated hazard ratio was 0.84 (95% CI 0.64–1.12). Thus, incorporation of these measured social and genetic risk factors produced minimal change in the point estimate and did not meaningfully alter the confidence interval reported.
In this All of Us cohort of adults with type 2 diabetes, measured educational attainment and Alzheimer disease genetic susceptibility (APOE ε4 and non‑APOE genetic risk) were associated with treatment initiation and dementia risk assessments but, when included in models, produced little change in the estimated association between GLP‑1 receptor agonist initiation and incident dementia. These findings suggest that, within this dataset and analytic approach, adjustment for these specific measured social and genetic variables did not materially confound the observed association.
The authors emphasize that these results do not exclude confounding by education or genetic risk in other study populations. They also note the possibility of residual confounding from other socioeconomic, clinical, behavioral, and health‑care‑related factors that were not fully captured or adjusted for. The report is a preprint and therefore has not been certified by peer review; findings should not be used to guide clinical practice without further validation.
Funding sources and investigator support from multiple NIH NIA awards are declared in the preprint. One institutional disclosure notes that Brown University received grant funding on behalf of an author from an industry sponsor for unrelated projects. Data used in this analysis are available to authorized researchers through the All of Us Researcher Workbench, subject to program access requirements.