Abstract Aims Evidence is limited as to who benefit the most from sodium-glucose cotransporter-2 inhibitors (SGLT2i), especially among people without elevated cardiovascular disease (CVD) risk. To address this knowledge gap, we investigated the heterogeneity in the effect of SGLT2i across CVD risk profiles. Methods Using a target trial emulation framework, we compared SGLT2i versus dipeptidyl peptidase 4 inhibitors (DPP4i) in a nationwide insurer-based database of working-age Japanese citizens in 2015–2023. The primary outcome was a composite of all-cause death, myocardial infarction, stroke, or heart failure over three years. Machine-learning causal forest was applied to assess heterogeneity by predicting individual-level risk reduction in primary outcomes by SGLT2i, and its correlation with CVD risk score. Results Overall, among 150,830 individuals included in this study (mean age, 54 years; female, 13.3%), SGLT2i was associated with decreased risk of primary outcomes (3-year risk difference, +0.38 95%CI, 0.16–0.61 percentage points). The causal forest model revealed heterogeneity in the effectiveness of SGLT2i, with estimated benefit correlating weakly with CVD risk score (r=0.287, p0.001). In particular, among 107,425 individuals with low CVD risk, 97,757 (91.0%) were predicted to benefit from SGLT2i. This subpopulation was characterized as individuals with higher blood pressure, body mass index, and fasting plasma glucose levels even with low CVD risk score. Conclusion The cardioprotective effect of SGLT2i was heterogeneous and more strongly predicted by individual patient characteristics than by overall CVD risk score, highlighting the importance of considering its benefit beyond the conventional risk stratification approach.
Mori et al. (2025) studied this question.