Simulating a Specialist's Treatment Experience for Hypertension Using Deep Neural Networks.

Hypertension management requires precise treatment decisions that balance medication efficacy with patient-specific factors. While clinical guidelines exist, physician decision-making often incorporates nuanced experience that remains challenging to quantify. This study aimed to develop and validate a deep learning model capable of simulating hypertension specialists' prescription patterns and predicting subsequent physiological responses using clinical trial data. We designed a dual-block deep neural network (DNN) framework, where one block predicts optimal medication prescriptions and the other forecasts next-day blood pressure (BP) and heart rate (HR). The model was trained simultaneously using a multi-objective approach that captures the relationship between drug selection and physiological outcomes. Training employed the Huber loss function for robustness, and performance was evaluated using mean absolute error (MAE), error variance, and mean relative error (MRE). The model demonstrated high predictive accuracy, with post-medication BP prediction errors consistently below 10 mmHg (MAE = 6.2 ± 1.8 mmHg). Drug dosage predictions showed strong alignment with actual prescriptions (MRE = 0.12%). These results indicate that the DNN framework effectively replicates physician decision-making within clinically acceptable margins. Our findings suggest that deep learning models trained on structured clinical data can accurately simulate hypertension specialists' treatment strategies. This approach may assist in standardizing care, reducing decision variability, and enhancing precision medicine in hypertension management. This study serves as a proof-of-concept investigation, demonstrating the feasibility of our dual-block DNN architecture. While performance on our single-center dataset is encouraging, future multicenter collaborations with larger datasets are essential to validate this approach for clinical decision support.