UC San Diego

Background Type 2 diabetes (T2D) elevates dementia risk through vascular injury, neuroinflammation, and perturbed insulin signaling. Antidiabetic classes differ in extra-glycemic actions that could modify neurodegeneration, yet head-to-head comparative data remain limited. Objective To compare time to incident dementia among older adults with T2D exposed to SGLT2 inhibitors, GLP-1 receptor agonists (GLP-1 RAs), metformin, or two-way combinations, and to quantify the contribution of comorbidities and neuroactive co-medications. Methods We performed a retrospective cohort study in the UCHDW (2012–2024). Adults aged 55–80 were assigned to the earliest qualifying exposure group and followed from index to first coded Alzheimer’s disease or unspecified dementia; vascular/multi-infarct dementias were excluded by design to reduce etiologic heterogeneity. Events recorded within 84 days of index were not considered incident outcomes (individuals were retained and censored at 84 days). Covariates included age, sex, cardiovascular disease, hypertension, chronic kidney disease, obesity, smoking, anticholinergic/overactive bladder agents, tricyclic antidepressants, proton-pump inhibitors, opioids, benzodiazepines, and ever-exposure to insulin, DPP-4 inhibitors, sulfonylureas, and thiazolidinediones. Analyses comprised Kaplan–Meier (KM) curves with log-rank testing, a pooled multivariable Cox model with a six-level exposure factor, and one-versus-rest Cox models complemented by stabilized, truncated inverse probability of treatment weighting (IPTW). Results The cohort included 22,677 SGLT2-only (150 events; 0.66%), 219,523 metformin-only (2,174; 0.99%), 35,012 GLP-1-only (149; 0.43%), 37,317 SGLT2+metformin (453; 1.21%), 7,210 SGLT2+GLP-1 (62; 0.86%), and 41,595 metformin+GLP-1 (421; 1.01%). KM curves differed significantly (log-rank χ²=38.5, p<0.001). In the pooled Cox model (reference SGLT2-only), metformin-only (HR=1.24, 95% CI 1.05–1.47; p=0.011) and SGLT2+metformin (HR=1.21, 95% CI 1.00–1.46; p=0.045) had higher hazards; GLP-1-only (HR=0.94, 95% CI 0.74–1.20; p=0.614) and SGLT2+GLP-1 (HR=0.85, 95% CI 0.64–1.15; p=0.299) did not differ significantly; metformin+GLP-1 trended higher (HR=1.18, 95% CI 0.97–1.42; p=0.092). Absolute risk differences were small (0.23–0.78 percentage points vs GLP-1-only). One-versus-rest models yielded directionally consistent estimates; IPTW with 1st–99th percentile truncation improved covariate balance and produced stable weighted estimates. Median follow-up (years) was 1.96 (SGLT2-only), 5.34 (metformin-only), 1.81 (GLP-1-only), 4.13 (SGLT2+metformin), 3.44 (SGLT2+GLP-1), and 4.84 (metformin+GLP-1). Conclusions In routine care of older adults with T2D, GLP-1 RA monotherapy demonstrates the most favorable dementia profile, whereas metformin monotherapy and SGLT2+metformin identify groups with comparatively higher risk. Given low absolute differences and absence of additional confounding covariates, cautious interpretation is warranted. Prospective studies incorporating glycemic control and exposure duration are needed to determine causal class effects and guide neuroprotective diabetes management.