Insulin resistance prediction from wearables and routine blood biomarkers - PubMed
3 hours ago
- #wearable technology
- #insulin resistance
- #machine learning
- Study presents WEAR-ME, a large remote study on insulin resistance (IR) prediction using wearable devices and routine blood biomarkers.
- Multimodal model achieved robust performance (AUROC = 0.80) with wearable data, demographics, and blood biomarkers against HOMA-IR.
- Fine-tuned wearable foundation model (WFM) improved IR prediction in an independent validation cohort (AUROC = 0.75 vs. 0.66).
- Adding WFM-derived representations to models with demographics and blood biomarkers significantly enhanced performance (AUROC = 0.88 vs. 0.76).
- IR prediction integrated into a large language model for personalized recommendations and early metabolic risk detection.
- Study funded by Google, with several authors being Alphabet employees.