Data-Driven Mobile Health: System Identification and Hybrid Model Predictive Control to Deliver Personalized Physical...

Area of Science:

• Behavioral medicine
• Control systems engineering
• Computational modeling

Background:

• Behavioral medicine seeks personalized interventions for healthy habits like physical activity.
• Control systems offer a framework for designing adaptive and individualized interventions.

Purpose of the Study:

• To apply system identification and hybrid model predictive control within the control optimization trial (COT) framework.
• To design individualized behavioral interventions for promoting physical activity.
• To demonstrate the efficacy of the COT approach using participant data.

Main Methods:

• System identification to estimate ARX models from participant data.
• Development of a three-degree-of-freedom Kalman filter-based Hybrid Model Predictive Controller (3DoF-KF HMPC).
• Closed-loop simulations (nominal and Monte Carlo) to evaluate controller performance and robustness.

Main Results:

• Selected ARX models balanced predictive ability and data fit for individual participants.
• The 3DoF-KF HMPC provided "ambitious but doable" goals for physical activity.
• Simulations confirmed the controller's performance and robustness under model mismatch.

Conclusions:

• The control optimization trial (COT) approach provides a proof of concept for personalized behavioral interventions.
• This method shows promise for fostering consistent engagement in physical activity.
• The approach is currently being evaluated in the clinical trial YourMove.