Human-in-the-loop AI predictive digital twin to extend virtual precision diabetes care between visits

Area of Science:

• Digital health interventions
• Artificial intelligence in medicine
• Diabetes self-management

Background:

• Type 2 diabetes (T2D) management requires continuous patient engagement.
• Traditional diabetes care often lacks personalized, real-time support between clinic visits.

Purpose of the Study:

• To evaluate a digitally enabled "human-in-the-loop" care support model using a predictive AI digital twin.
• To provide personalized daily feedback via short message service (SMS) for adults with T2D.

Main Methods:

• A 6-month randomized trial with a subset of 19 T2D participants.
• An AI predictive control model with a transfer-learning artificial neural network digital twin.
• Weekly retraining of the digital twin using participant self-monitoring data (weight, diet, activity, glucose).
• Personalized behavioral recommendations generated by a particle swarm optimization controller.

Main Results:

• The AI model achieved ≥80% prediction accuracy.
• AI feedback group showed trends toward increased daily steps and better adherence to caloric/carbohydrate targets.
• Significantly greater weight loss in the AI group (5.87 lbs) compared to controls (3.57 lbs; p<0.012).
• Stable glucose levels were maintained in both groups (p=0.661).

Conclusions:

• AI-enabled predictive digital twin models offer a scalable approach for precision diabetes self-management.
• This technology can extend personalized support beyond traditional clinical settings.
• The "human-in-the-loop" model effectively integrates AI for enhanced patient care.