A Hybrid Ensemble Model of XGBoost and LSTM for Personalized Hypertension Medication Recommendation from Electronic Health Records

Hypertension care is impeded by high inter-patient variability in medication response, typically leading to a prolonged trial-and-error process for establishing blood pressure control. Conventional therapeutic practices, based on broad population standards, do not fully use the rich, multimodal data available in modern Electronic Health Records (EHRs). This study proposes a unique hybrid ensemble model for customized antihypertensive drug recommendation that cohesively integrates both static and dynamic patient data to predict treatment efficacy. The suggested architecture leverages a dual-pathway system designed to process disparate data types in parallel. A Recurrent Neural Network (RNN) route ingests longitudinal clinical data, such as time-series of blood pressure measurements and important laboratory findings, to successfully simulate the patient's physiological trajectory and temporal health patterns. Simultaneously, a gradient-boosted decision tree model (XGBoost) examines static, tabular data, including patient demographics, baseline comorbidities, and genetic markers, to capture complex, non-linear correlations within the patient's basic profile. High-level feature representations taken from both the temporal and static channels are carefully combined in a fusion layer. This generates a comprehensive, unified patient vector that encompasses both their stable traits and their dynamic health state. This fused representation is subsequently given to a final prediction layer to forecast the efficacy of prospective drugs. By synergistically combining complementary data sources, this methodology is designed to give more precise and tailored treatment recommendations, potentially shortening the time to blood pressure control and improving long-term cardiovascular outcomes. This strategy marks a big step towards data-driven precision medicine in hypertension management.