Wearables flag insulin resistance
Wearables are moving beyond steps — new reports claim devices and analytics can predict insulin resistance, shifting wearables toward proactive metabolic‑health management rather than passive tracking. That opens the door to earlier interventions for people at risk of type‑2 diabetes. (healthperformancenexus.substack.com)
The study, called WEAR‑ME and led by Ahmed A. Metwally and colleagues, was published in Nature on 16 March 2026. (nature.com) WEAR‑ME remotely enrolled 1,165 adults across the U.S.; cohort medians were BMI 28 kg/m2, age 45 years, and HbA1c 5.4%, and the team used HOMA‑IR as the ground‑truth measure with a cutoff of 2.9. (nature.com) A multimodal model that combined wearable time‑series data with routine blood biomarkers and demographics achieved AUROC 0.80, sensitivity 76% and specificity 84% for detecting insulin resistance. (nature.com) The authors report higher performance in high‑risk subgroups—93% sensitivity and 95% adjusted specificity in obese, sedentary participants—and reproduced results on an independent validation cohort (n=72) where WFM‑derived features raised AUROC from 0.66 (demographics only) to 0.75. ( ) To process wearable signals the team fine‑tuned a wearable foundation model (WFM) pretrained on roughly 40 million hours of sensor data, and adding WFM representations to a model with demographics, fasting glucose and a lipid panel increased AUROC from 0.76 to 0.88. (nature.com) Data were collected via the Google Health Studies app from devices including Fitbit and Google Pixel watches and routine blood testing was automated through a partnership with Quest Diagnostics; the group also prototyped an “Insulin Resistance Literacy and Understanding” agent built on the Gemini LLM family to contextualize predictions and emphasized the work is for research/informational purposes. ( )
