ABSTRACT NUMBER: ESOC2026A1693 ARTIFICIAL INTELLIGENCE–BASED PREDICTION OF HEMORRHAGIC STROKE RISK IN PATIENTS WITH TYPE 2 DIABETES MELLITUS

Abstract Background and aims To predict the risk of hemorrhagic stroke (HS) in patients with type 2 diabetes mellitus (T2DM) by integrating the HES-BLED, ATRIA, and HEMORR₂HAGES risk scores into an artificial intelligence–based model. Methods Eighty T2DM patients without prior hemorrhagic stroke were analyzed using AI-integrated HES-BLED, ATRIA, and HEMORR₂HAGES scores, applying Spearman correlation, ROC–AUC, OR, and 95% CI. Results HES-BLED and ATRIA classified hemorrhagic stroke risk as high in 18.8% (n = 15), moderate in 52.5% (n = 42), and low in 28.7% (n = 23), while HEMORR₂HAGES identified high, moderate, and low risk in 20.0% (n = 16), 47.5% (n = 38), and 32.5% (n = 26), respectively. Strong correlation was observed between HES-BLED and HEMORR₂HAGES (r = 0.72; p < 0.001), with moderate correlations between ATRIA and HES-BLED (r = 0.61; p = 0.002) and ATRIA and HEMORR₂HAGES (r = 0.58; p = 0.004). ROC analysis showed the highest diagnostic performance for HEMORR₂HAGES (AUC = 0.87; 95% CI: 0.79–0.94), followed by HES-BLED (AUC = 0.84; 95% CI: 0.75–0.92) and ATRIA (AUC = 0.78; 95% CI: 0.68–0.87). High-risk patients had increased odds of hemorrhagic stroke: OR = 3.6 for HES-BLED, 4.2 for HEMORR₂HAGES, and 2.9 for ATRIA (all p ≤ 0.006). Conclusions Among the integrated risk scores, HEMORR₂HAGES demonstrated the highest diagnostic accuracy for predicting hemorrhagic stroke risk in patients with T2DM and may serve as an effective tool for individualized prevention strategies. Conflict of interest noting to disclose