Managing diabetes is a daily challenge faced by nearly 40 million Americans. It involves tracking food intake, timing medication and engaging in physical activity. Getting it wrong can lead to serious health issues; therefore, developing better prediction tools is a vital part of effective diabetes care.
To support better diabetes management, researchers funded by multiple U.S. National Science Foundation grants are developing innovative tools that help patients predict blood sugar levels more precisely without compromising the privacy of their health data. This cutting-edge approach could transform how people with diabetes monitor and manage their condition in real-time.
At the core of this technology is a method called federated learning, which allows artificial intelligence models to be trained across many patients' devices without sending any personal data to a central server. This setup is ideal for healthcare, where data privacy is paramount and patients often use battery- and memory-limited smart devices. But early federated learning systems struggled to adapt to individual differences, like how people eat, move or react to insulin.
To address this challenge, the research team grouped patients based on their carbohydrate (e.g., sugar and starch) intake levels. The idea is that people who eat in similar ways tend to show similar glucose patterns. By training the AI on these grouped behaviors, the model became more effective at making personalized blood glucose predictions.
To test their approach, the team evaluated two machine learning models using data generated from an FDA-approved Type 1 diabetes simulator. As simulated data accumulated, model accuracy improved. Notably, even with limited input, the system could build personalized models - a key advantage for newly diagnosed patients or those just beginning to use digital tools to manage their care.
Because traditional AI systems typically require gathering large amounts of data in a central location, which can pose privacy risks, especially when dealing with sensitive health information, federated learning provides a good solution for the field. It keeps personal data on each individual's device - like a phone or wearable sensor - and only shares the model's learning, never the raw data. This protects patient privacy while still allowing the system to improve over time.
While the early results are promising, the researchers note that the models still rely on detailed food intake data - something not all patients can easily provide. They plan to expand their system to include other factors like exercise and medication, and to test it with larger patient groups. In the long term, the researchers hope to extend this personalized, privacy-preserving AI approach to other chronic conditions like heart disease or asthma, where individualized care is equally important.
With diabetes costing the U.S. economy over $300 billion yearly, innovations that enable earlier intervention and personalized care can drive down long-term costs and improve population health outcomes.
This project highlights how public investment in cutting-edge research drives innovation that benefits not just individual patients but the entire U.S. healthcare system.
