Background: Whether lower dementia risk with SGLT2 inhibitors (SGLT2i) versus DPP-4 inhibitors (DPP4i) reflects reverse causation, residual confounding, or a class effect is unclear.
Methods: We conducted a federated new-user study across 11 Korean hospital EHR databases standardized to the OMOP Common Data Model. Adults aged ≥50 years with type 2 diabetes without dementia initiating SGLT2i, sulfonylureas (SU), thiazolidinediones (TZD), or DPP4i were included. Large-scale propensity score matching (1:2), Cox models with a 180-day lag, random-effects meta-analysis of databases meeting diagnostics, and calibration using 121 negative controls were applied. Alternative lags, on-treatment follow-up, and leave-one-out analyses were assessed.
Results: After matching, 9,370 SGLT2i, 16,536 SU, 6,770 TZD, and 14,784 DPP4i initiators were analyzed. All-cause dementia occurred in 219 SGLT2i and 460 DPP4i initiators (6.36 vs 8.48 per 1,000 person-years). SGLT2i was associated with lower dementia risk (calibrated HR, 0.69; 95% CI, 0.56–0.85; I²=0%), Alzheimer’s disease (0.72, 0.57–0.92), and dementia with anti-dementia medication (0.68, 0.52–0.88). SU (1.01, 0.90–1.14) and TZD (0.97, 0.82–1.16) did not differ from DPP4i. The SGLT2i HR was 0.74 with no lag, 0.60 with 365-day lag, 0.54 (0.35–0.82) on treatment, and 0.64–0.72 in leave-one-out analyses.
Conclusions: Lower dementia risk with SGLT2i was consistent and class-specific, arguing against reverse causation and confounding.