(096) GLP1 comparative effectiveness against dementia onset in a large, multi-site cohort of type 2 diabetes patients: a trial emulation approach with targeted learning

Objective: To compare impact of exposure to GLP-1RA versus sodium-glucose cotransporter 2 inhibitors (SGLT2i), dipeptidyl peptidase 4 inhibitors (DPP4i), and sulfonylureas (SU) on reducing the risk of dementia using targeted learning, a rigorous causal inference approach combined with machine learning that address both baseline and time-dependent confounding.
Research Design and

Methods: Using clinical and claims data from four diverse US healthcare systems, we emulated three two-arm trials contrasting the impact of sustained treatment with GLP-1RA versus SGLT2i, DPP4i, and SU, respectively, on dementia diagnosis. Cohort eligibility required medication initiation between 01/01/2014 and 12/31/2022, no exposure to the two medications compared in the prior 2 years, no evidence of pre-existing dementia diagnosis, and age ≥ 60 years. Patients were followed until earliest of dementia, diagnosis, death, health plan disenrollment, or 12/31/2022. We estimated cumulative risk differences at 2.5 years.

Results: In Cohort 1 (60,852 patients), there was no evidence of differential dementia risk between sustained exposure to GLP1-RA and versus SGLT2i. In Cohort 2 (33,539 patients), GLP-1RA was associated with reduced risk of dementia diagnosis compared to DPP4i: risk difference (RD) -0.013 95% confidence interval (CI):[-0.017; -0.009]; number needed to treat (NNT): 126. In Cohort 3 (110,898 patients), GLP-1RA was associated with reduced risk of dementia diagnosis compared to SU: -0.016 [-0.018, -0.015]; NNT: 62.

Conclusions: Rigorous causal inference analysis suggests that sustained exposure to GLP-1RA may modestly reduce risk of dementia diagnosis, compared to DPP4i or SU exposure—but not compared to SGLT2i—in patients with T2DM aged > 60 years.