Аннотация
Arterial hypertension (AH) is a major modifiable risk factor for cardiovascular disease and remains a significant public health challenge. Early identification of hypertension severity is essential for effective risk stratification, yet traditional approaches struggle to integrate heterogeneous clinical data effectively. This study presents an integrated framework combining large language model (LLM)-based feature extraction with machine learning for multi-stage hypertension prediction using real-world hospital data from Kazakhstan. We developed an automated pipeline that transforms unstructured cardiology consultation notes into structured datasets with 83 clinical features. Four supervised machine learning models (Logistic Regression, Random Forest, Support Vector Machine, and XGBoost) were evaluated for four-class hypertension stage classification on 553 patient records that were collected from three clinics in the Karaganda region. XGBoost achieved the best performance with 85.58 % accuracy and a macro-averaged F1-score of 0.8406, outperforming the logistic regression baseline. In addition, a recommendation generation module was implemented to produce structured outputs that support individualized blood pressure management. The proposed framework demonstrates the possibility of applying machine learning to real-world clinical data for stage-aware hypertension prediction and provides a foundation for scalable clinical decision support and risk assessment on population level.