Background: Sodium-glucose cotransporter-2 inhibitors (SGLT2i) are a newer class of glucose-lowering drugs for type 2 diabetes (T2D) with demonstrated cardiovascular and renal benefits. However, their association with acute kidney injury (AKI) remains controversial.
Objectives: This study aimed to employ advanced causal learning methods to assess the heterogeneous effects of SGLT2i on AKI in patients with T2D.
Methods: We analyzed data from the 2015–2023 OneFlorida+ network, a statewide electronic health record (EHR) dataset encompassing over 21 million patients. The study included adults with T2D who initiated either an SGLT2i or a dipeptidyl peptidase-4 inhibitor (DPP4i) during the study period. The primary outcome was AKI, with the index date defined as drug initiation. Follow-up continued until AKI occurrence, death, last observation, or study end. Propensity score matching (PSM) controlled for confounding, and Cox regression estimated hazard ratios (HR) with 95% confidence intervals (CIs). Double robust learning was employed to identify heterogeneous treatment effect (HTE) subgroups, and causal structural learning was used to identify key causal factors associated with AKI risk post-SGLT2i initiation.
Results: The matched cohort included 34,480 patients. Cox regression indicated that SGLT2i use was associated with a 15% lower risk of AKI compared to DPP4i (HR=0.85, 95% CI: 0.77–0.94). Double robust learning confirmed an overall reduced AKI risk with SGLT2i (risk difference [RD]: -0.40% [-0.65%, -0.15%]); however, a distinct subgroup—patients with anemia and atrial fibrillation—experienced an increased AKI risk (RD: 0.8% [0.1–1.5%]). Causal discovery learning identified atrial fibrillation, anemia, cardiovascular disease, and heart failure as direct causal factors influencing AKI risk.
Conclusions: Our findings highlight heterogeneous treatment effects in the relationship between SGLT2i use and AKI risk. While SGLT2i generally reduce AKI risk, certain patient subgroups may be vulnerable to adverse outcomes. These results underscore the need for personalized therapeutic strategies to optimize treatment decisions for T2D patients at varying risk levels.
