Estimating Dynamic Treatment Regimes in Mobile Health Using V-learning

Area of Science:

• Biomedical Informatics
• Machine Learning
• Digital Health

Background:

• Precision medicine aims to personalize healthcare using individual patient data.
• Mobile technologies enable real-time health monitoring and intervention delivery.
• Current methods for dynamic treatment regimes are limited to fixed, infrequent decision points.

Purpose of the Study:

• To develop a novel reinforcement learning method for estimating optimal dynamic treatment regimes.
• To adapt these methods for mobile health data with continuous monitoring and decision-making.
• To address the limitations of existing approaches for out-patient settings.

Main Methods:

• Proposed a new reinforcement learning algorithm tailored for mobile health data.
• Accommodated an indefinite time horizon and minute-by-minute decision-making.
• Applied the method to estimate optimal dynamic treatment regimes for type 1 diabetes.

Main Results:

• The proposed estimators demonstrated consistency and asymptotic normality under mild conditions.
• Successfully applied the method to control blood glucose levels in type 1 diabetes patients.
• The approach is suitable for real-time data from mobile health applications.

Conclusions:

• The novel reinforcement learning method effectively estimates optimal dynamic treatment regimes for mobile health.
• This approach supports personalized, real-time interventions in out-patient settings.
• The method shows promise for managing chronic conditions like type 1 diabetes.