Feed

Chronic Obstructive Pulmonary Disease (COPD) is a leading cause of disability and death worldwide. Early identification remains challenging, as existing prediction models largely rely on clinic-based assessments and self-report measures that are resource-intensive and prone to bias. This study aimed to determine whether real-world mobility metrics could predict incident COPD.

This prospective cohort study included 28,251 UK Biobank participants aged 60-78 years who wore wrist-worn accelerometers. Digital gait biomarkers were derived using signal-processing and machine-learning algorithms. Incident COPD was identified via linked electronic health records. Associations between digital gait biomarkers and incident COPD were examined using Cox proportional hazards models with internal validation adjusted for age, sex, body mass index, smoking pack-years, air pollution exposure, and asthma history. Model discrimination was evaluated using Harrell's concordance index.

Among 28,251 participants, 639 (2.26%) developed COPD over a mean follow-up period of 8.5 (SD=1.3) years. Lower running duration, slower maximal walking speed, shorter walking bout duration, and a lower proportion of walks longer than 8 seconds were independently associated with incident COPD. A model incorporating these four digital gait biomarkers, and four easily collectable self-report measures, age, sex, smoking pack-years, and asthma history, achieved a Harrell's concordance index of 0.80; comparable to existing models that require clinic-based tests and extensive self-report items.

Real-world mobility metrics are early indicators of incident COPD, providing an accessible and automatic approach for early risk identification in older people and enabling early intervention to delay disease progression and preserve quality of life.