A research team has introduced a conceptual framework called GlucoSense that combines smartphone motion data and blood glucose predictions to assess the risk of type 2 diabetes early on. The work by researchers from Westcliff University, SINTEF Digital, International Islamic University Chittagong, the University of Oslo, and Southeast University was published in "Frontiers in Digital Health" (DOI: 10.3389/fdgth.2026.1932366).
GlucoSense consists of two modules based on Graph Attention Networks (GATv2). One module recognizes human activities from smartphone accelerometer data across 23 activity classes; the other predicts blood glucose levels 30 minutes in advance and was trained using the OhioT1DM dataset. Unlike sequential models, GlucoSense maps relationships between movement patterns, heart activity, diet, and glucose behavior in a graph structure.
In a 90-day longitudinal study with six participants, mean glucose levels, time above hyperglycemic thresholds, and glycemic variability coefficient were calculated. The system correctly identified metabolically healthy individuals and those with prediabetic patterns, the team reported. The results were compared with HbA1c laboratory tests; the clinical outcomes matched the system's classification. The glucose module reduced RMSE from 31.82 mg/dl with an LSTM baseline to 25.60 mg/dl, and the activity module achieved 98.50 percent accuracy. All differences were confirmed by paired significance tests.
The authors emphasize that this is a proof of concept with six participants. The results did not demonstrate efficacy in the general population; larger, multicenter studies are needed. offset of 0.1 per score point.
