This research study introduces a scalable method for detecting insulin resistance (IR) by combining data from consumer wearables with standard blood biomarkers. Using a large cohort, the researchers developed deep neural networks and a specialized wearable foundation model to predict metabolic risk more accurately than traditional demographic assessments. The results demonstrate that integrating heart rate and activity patterns with glucose and lipid panels significantly improves the identification of individuals at risk for type 2 diabetes. To make these findings actionable, the study incorporates a large language model agent designed to provide users with personalized health insights and lifestyle recommendations. Ultimately, this framework offers a cost-effective, accessible tool for early intervention and the prevention of chronic metabolic diseases.

References:

Metwally A A, Heydari A A, McDuff D, et al. Insulin resistance prediction from wearables and routine blood biomarkers[J]. Nature, 2026: 1-11.