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

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Informatics

Background:

• Hypertension management involves complex physician decisions balancing medication efficacy and patient factors, often relying on experience difficult to quantify.
• Existing clinical guidelines may not fully capture the nuanced decision-making processes of hypertension specialists.
• Quantifying and replicating specialist expertise is crucial for improving treatment standardization and precision.

Purpose of the Study:

• To develop and validate a deep learning model simulating hypertension specialists' prescription patterns.
• To predict subsequent physiological responses (blood pressure and heart rate) to prescribed medications.
• To assess the model's ability to replicate clinical decision-making within acceptable margins.

Main Methods:

• A dual-block deep neural network (DNN) framework was designed, with one block for prescription prediction and another for physiological response forecasting.
• The model was trained simultaneously using a multi-objective approach on clinical trial data, employing the Huber loss function for robustness.
• Performance was evaluated using mean absolute error (MAE), error variance, and mean relative error (MRE).

Main Results:

• The DNN model achieved high predictive accuracy for post-medication blood pressure, with errors consistently below 10 mmHg (MAE = 6.2 ± 1.8 mmHg).
• Drug dosage predictions demonstrated strong alignment with actual specialist prescriptions, indicated by a mean relative error of 0.12%.
• The model effectively replicated physician decision-making within clinically acceptable parameters.

Conclusions:

• The developed deep learning framework successfully simulates hypertension specialists' treatment strategies and predicts patient physiological outcomes.
• This AI-driven approach has the potential to standardize care, reduce decision variability, and enhance precision medicine in hypertension management.
• The study serves as a proof-of-concept, highlighting the feasibility of the dual-block DNN architecture for clinical decision support, with future validation on larger, multicenter datasets recommended.