Erythrocytosis after SGLT2 inhibitor initiation and anticoagulated outcomes in atrial fibrillation: a real-world analysis with counterfactual modeling

Background: Erythrocytosis has emerged as a hematologic response to sodium–glucose cotransporter 2 (SGLT2) inhibitor therapy, yet its occurrence and potential impact on bleeding and thromboembolic outcomes among patients with atrial fibrillation receiving anticoagulant therapy remain unknown. Methods: We conducted a population-based cohort study using the Tianjin Health and Medical Data Platform, including adults with atrial fibrillation (AF) receiving oral anticoagulants from 2011 to 2024. In Part 1, employing a new-user design with a stratified pseudo-time strategy, SGLT2 inhibitor initiation was compared with non-use for incident erythrocytosis, estimated by Poisson regression to calculate incidence rate ratios. In Part 2, restricted to SGLT2 inhibitor users, erythrocytosis served as the exposure for thromboembolic and bleeding outcomes, analyzed with Fine–Gray and Cox models. Active comparator analyses, propensity score matching, subgroup and sensitivity analyses, and counterfactual mediation and decision modeling tested robustness and individualized treatment thresholds. Results: Among 47,401 eligible patients with AF receiving anticoagulant therapy, a stratified pseudo-time strategy was applied to minimize immortal time bias, identifying 39,190 patients for analysis (mean age 72.5 ± 9.7 years; 48.1% female). Of these, 6,697 were SGLT2 inhibitor users. Erythrocytosis occurred in 11.3% of users compared with 3.1% of non-users within 6 ± 2 months, corresponding to an adjusted incidence rate ratio of 3.85 (95% CI 2.78–5.31). In the full follow-up, erythrocytosis was independently associated with a higher risk of thromboembolic events (Model 5 aHR 2.34, 95% CI 1.39–3.94; p = 0.001) but this was not statistically significant for bleeding (aHR 1.42, 95% CI 0.81–2.49; p = 0.22). Subgroup and sensitivity analyses yielded consistent results across age, sex, kidney function, and anticoagulant type. Mediation and counterfactual analyses indicated that the hemoglobin rise acted primarily as a risk marker rather than a causal mediator. Conclusions: SGLT2 inhibitor-associated erythrocytosis is a frequent hematological response in patients with AF receiving anticoagulant therapy and marks a phenotype predisposed to thromboembolism rather than bleeding. Based on counterfactual mediation analysis, the rise in hemoglobin levels likely represents a risk marker rather than a causal mediator, warranting early monitoring for personalized risk management.