Digital biomarkers for stroke rehabilitation
Recovery is tested under supervision, but the goal of rehabilitation is what someone actually does once they leave the therapy room.
.png)
Stroke rehabilitation has a measurement problem that digital tools address unusually well. Recovery is assessed with supervised tests of what a person can do, but the goal of rehabilitation is what they actually do once they leave the therapy room. Those two things diverge, sometimes dramatically, and the divergence is clinically important. A person can regain the capacity to walk and still not walk, which is a different rehabilitation problem from one of impairment.
The measures collected here describe real world function after stroke: walking speed in daily life, how walking is distributed through the day, how often the person stands up from sitting, how much of the day is spent sedentary, and how active the affected limb is relative to the unaffected one.
Upper limb measurement deserves particular attention. Wrist worn sensors on both arms can quantify how much a person uses the affected side during ordinary activity, which is the outcome that constraint based therapies target directly. Self report is a weak substitute here, because people are poor judges of how much they favour one side.
Digital biomarkers used in stroke rehabilitation research
How these measures are used
The dominant use is measuring transfer, meaning whether gains achieved in supervised therapy appear in daily life. A trial that shows improvement on a clinic walk test and no change in daily walking has learned something important, and only continuous measurement can reveal it.
Digital measures also support dose tracking in rehabilitation research. How much therapy someone actually performs, including home exercise between sessions, is a major source of variation in trial outcomes and has historically been recorded by self report. Sensor based counts of relevant movement give a more credible estimate.
The third use is trajectory. Recovery after stroke follows a curve that is steepest early and then flattens, and the shape of that curve varies between people. Frequent measurement describes it far better than assessments at discharge and at three months. These measures are paired with standard functional scales, which supply the clinical anchors that sensor outputs are interpreted against.
What the evidence supports today
The evidence for real world mobility measurement after stroke is reasonably solid for the walking measures, which draw on the same technical validation as the wider mobility domain, and which have been shown to differ from supervised test performance in ways that carry prognostic information. Walking speed in particular has an established relationship with community ambulation status.
Upper limb activity measurement using bilateral wrist accelerometry has a credible research literature and is well suited to the specific question of limb use asymmetry. It is less established as a regulatory endpoint than the lower limb measures.
The important caveat is that post stroke gait is atypical, and algorithms developed on normal walking patterns can underperform on hemiparetic gait, including step detection itself. Studies should not assume that accuracy figures published in healthy or older adult cohorts transfer to this population. Where a device has not been validated specifically after stroke, that gap should be treated as a real limitation rather than an administrative one.
Common questions
Why measure walking outside the clinic after a stroke?
Because capacity and performance diverge. Supervised tests show what a person can do when prompted and supported; wearable measures show what they do at home, where fatigue, confidence, environment and support all intervene. Rehabilitation aims at the second, and only continuous measurement observes it.
How is affected limb use measured?
By placing accelerometers on both wrists and comparing activity between them across ordinary days. The resulting asymmetry describes how much the person relies on the unaffected side. It is the natural outcome measure for therapies designed to increase use of the affected limb, and it is far more reliable than asking someone to estimate it.
Do standard step counting algorithms work after stroke?
Not always. Hemiparetic gait is slower and more asymmetric than the walking patterns most algorithms were developed on, and step detection can undercount at low speeds. Studies in this population should use devices and algorithms validated specifically in post stroke gait rather than relying on general population accuracy figures.
Which measures matter most in early versus later recovery?
Early on, measures of activity volume and sit to stand transitions capture the emergence of basic function. Later, walking speed, bout structure and community mobility become more informative, since they describe whether recovered capacity is being used outside the home.
Run a study on these measures
WeGuide captures wearable data and patient reported outcomes in one platform, from screening through to analysis.
