Wrist sensors in UK Biobank: gait differences seen up to 6.8 years before Parkinson's diagnosis

A team from the University of New South Wales in Sydney analysed data from 73,413 UK Biobank participants who wore a wrist device for seven days. As reported in the Journal of Neural Transmission, participants were followed through linked electronic health records for up to 10 years. Of 73,294 participants without Parkinson's disease at the time of the accelerometry, 314 were diagnosed during follow-up springer.comJournal of Neural Transmissionspringer.comOpen source ↗.

How gait was measured

Seventeen digital gait biomarkers were derived with the Watch Walk algorithm. The software finds the stretches of the recording that capture walking and estimates step count, walking speed and step regularity. The orientation and movement of the wrist help it distinguish a swinging arm from a hand held near the head or in front of the body.

Associations with time to diagnosis were assessed with Cox regression models adjusted for age and sex. The participants who later developed the disease were also grouped by the interval between the recording and the diagnosis.

Differences before diagnosis

People who developed Parkinson's disease took fewer steps per day: 6,370 in those who did not develop it versus 4,043–5,585 in those who did. The difference was larger the closer the recording was to the diagnosis. According to the publication of the study, those diagnosed within 40 months of wearing the sensor averaged about 4,000 steps a day, against about 6,400 in people without a recorded diagnosis springer.comJournal of Neural Transmissionspringer.comOpen source ↗.

At baseline, the future patients also walked more slowly, had altered step regularity and showed reduced arm swing (all p < 0.001). Five measures were consistently associated with a later diagnosis across the groups by time to onset: daily step count, maximum walking speed, step regularity, the proportion of long walking bouts, and time spent walking with static arm positions (all p < 0.001).

The authors report that these changes in real-world gait were observable up to 6.8 years before a clinical diagnosis and suggest that digital gait biomarkers may help characterise early motor changes.

Limits of the findings

The results describe differences between groups. They do not allow an individual to be told whether Parkinson's disease will develop, and pain, medication and other conditions can also affect how a person walks. Each person was recorded during a single monitoring period, so the study cannot show when a given person's arm swing began to change. Repeated recordings of the same people at regular intervals would be needed to follow that change over time.