Real-World Walking Patterns From Wrist-Worn Wearables Predict Incident Type 2 Diabetes: A UK Biobank WatchWalk Study

AIMS: Digital gait biomarkers derived from wearable devices may capture early functional changes relevant to type 2 diabetes (T2D) risk. We investigated the associations between real-world gait characteristics and incident T2D and their contribution to risk discrimination beyond established risk factors. METHODS: We analysed data from 71 618 UK Biobank participants without diabetes at baseline who wore a wrist accelerometer for seven days. Multiple digital gait biomarkers capturing walking speed, quality and distribution patterns were derived. Incident T2D was ascertained from linked health records. Cox proportional hazards models were fitted, adjusting for traditional T2D risk factors. Model discrimination was assessed using Harrell's C-index. RESULTS: During follow-up, 2577 participants (3.6%) developed T2D. Several digital gait biomarkers were significantly associated with T2D in minimally adjusted analyses, and of these, slower walking speed, shorter maximum walking bouts and lower daily running duration remained independently associated with higher T2D risk after multivariable adjustment. A streamlined model including digital gait biomarkers and established risk factors showed good discrimination (C-index 0.80), outperforming models based on traditional risk factors and step counts alone. CONCLUSIONS: In this large population-based cohort, real-world walking patterns measured using wrist-worn devices were independently associated with incident T2D and enabled more accurate identification of at-risk individuals. Digital gait biomarkers may reflect early metabolic and functional vulnerability preceding the clinical diagnosis of T2D and may facilitate risk stratification in clinical practice and prevention initiatives.