Vascular stiffness–integrated ai model improves discrimination of poor responders to azilsartan–indapamide therapy

Background Response to angiotensin receptor blocker (ARB)–diuretic therapy remains heterogeneous in arterial hypertension. Baseline systolic blood pressure (SBP) is widely used to estimate treatment response, but may not sufficiently capture vascular stiffness and metabolic/renal determinants influencing therapeutic effectiveness. Purpose To determine whether an artificial intelligence (AI) model integrating vascular stiffness and clinical parameters improves prediction of 6-month response to azilsartan–indapamide beyond baseline SBP. Methods In 300 patients with arterial hypertension treated with azilsartan–indapamide, we developed an interpretable multivariable logistic regression model incorporating baseline SBP, pulse wave velocity (PWV), estimated glomerular filtration rate (eGFR), body mass index, glucose, hypertension duration, smoking status and age. Poor response was defined as inadequate SBP reduction at 6 months. Model performance was compared with an SBP-only model using 10-fold cross-validation and independent holdout validation. Discrimination was assessed by area under the receiver operating characteristic curve (AUC). ΔAUC was evaluated using bootstrap resampling. Decision curve analysis assessed clinical utility. The final model was exported into an offline browser-based tool to enable point-of-care probability estimates. Results In holdout validation, the multivariable model demonstrated superior discrimination compared with SBP-only prediction (AUC 0.91 vs 0.83; ΔAUC 0.085; p=0.033). PWV independently predicted poor response (odds ratio 1.21, p=0.026). Reduced eGFR, longer hypertension duration and smoking were also associated with inadequate treatment response. Decision curve analysis showed improved net clinical benefit across clinically relevant probability thresholds. Conclusion An interpretable AI model integrating vascular stiffness and metabolic/renal parameters improves identification of poor responders to azilsartan–indapamide beyond baseline SBP and supports precision phenotyping with feasible offline bedside deployment.Figure1_ROC.pngFor image description, please refer to the figure legend and surrounding text.Figure2_DCA.