Data-Driven Markov Decision Process Approximations for Personalized Hypertension Treatment Planning.

Area of Science:

• Health economics
• Biostatistics
• Clinical decision-making

Background:

• Markov decision process (MDP) models offer optimal treatment policies but face challenges in computation time and physician interpretability.
• Physician interpretability and computational efficiency are key factors for the clinical adoption of advanced treatment planning models.

Purpose of the Study:

• To evaluate the efficacy of Poisson regression in approximating optimal hypertension treatment policies derived from MDP models.
• To assess whether a simplified Poisson model can maintain the high accuracy and health outcomes of complex MDP models.

Main Methods:

• Utilized Poisson regression to approximate optimal treatment policies for hypertension management.
• Compared the approximated policies against those generated by a standard Markov decision process (MDP) model.
• Calculated expected discounted quality-adjusted life years (QALYs) to measure patient health outcomes.

Main Results:

• The Poisson regression approximation achieved 99% accuracy in matching the optimal MDP treatment policy.
• This high fidelity resulted in nearly identical patient health outcomes compared to the MDP-derived policy.
• The Poisson approximation yielded an additional 104 QALYs per 1000 patients versus the Seventh Joint National Committee guidelines.

Conclusions:

• Poisson model approximation demonstrates high fidelity to optimal MDP treatment policies for blood pressure management.
• This approach enhances the usability and transparency of personalized treatment strategies.
• The findings suggest a viable method for improving clinical decision support in hypertension care.