A wearable wrist device that employs a novel machine-learning algorithm accurately tracks upper-limb motor function after stroke by capturing how patients use their arms in everyday life, which could allow clinicians to make real-time adjustments in rehab programs that could speed recovery, a new study suggests.
The digital arm performance scale (DAPS) was developed by investigators for daily assessment of upper-limb motor performance in stroke survivors using 23,000 hours of movement data collected by 215 individuals wearing wrist accelerometers.
In an analysis, DAPS was 40%-50% more accurate than commonly used clinician-administered assessments and more closely reflected patients' true level of function, said Sunghoon Ivan Lee, PhD, associate professor at the University of Massachusetts Amherst, one of the investigators who developed the device.
"The primary advantage of DAPS is that it captures patients' real-world performance rather than their in-clinic capacity," Lee told Medscape Medical News.
The study was published online September 30 in Science Translational Medicine.
A Needed Digital Biomarker
Upper-limb impairment frequently occurs after stroke and can persist long term, limiting independence and daily activities.
Rehabilitation can improve motor recovery, but progress is typically measured during in-person visits with standardized clinical assessments that capture performance only at a specific point in time, which can be influenced by factors such as fatigue, motivation, and encouragement from the examiner.
These evaluations are often conducted at the outset and conclusion of a rehabilitation program, Lee said.
"Clinicians must guide interventions without knowing whether the trajectory is moving in the right direction, and patients have no clear idea of their progress beyond how they perceive or feel it," Lee said.
Wearable devices offer an option for ongoing performance review, but existing models usually track only the movement of the affected arm, failing to capture the complexity of real motor recovery, researchers wrote.
To fill the gap between in-clinic assessments and real-world measurements, investigators set out to develop a digital biomarker, drawing on roughly 23,000 hours of wrist-accelerometer data from 215 participants.
The cohort included 61 people with subacute stroke (median age, 67 years; 26 women), 79 with chronic stroke (median age, 60 years; 27 women), and 75 healthy adults (median age, 53 years; 40 women). Participants with stroke had persistent unilateral upper-limb deficits.
The accelerometer captured movement in naturalistic environments for approximately 24 hours at a time. The investigators divided continuous movement data into individual movement segments and identified features related to velocity, jerk, and the timing of consecutive movements. These features were combined using a machine-learning approach to generate DAPS.
The scale was trained using data from the subacute stroke group and healthy control individuals and then tested independently in the chronic stroke group. Scores generally ranged from 0 to 1, with lower scores indicating greater motor impairment.
To validate DAPS, the investigators compared device measurements with results from the Action Research Arm Test (ARAT) and the Fugl-Meyer Assessment for Upper Extremity (FMA-UE). In addition to evaluating its reliability and validity, the study also measured the device's ability to distinguish between those who survived strokes and healthy individuals, along with its responsiveness to changes over time.
The subacute group was assessed up to eight times over 24 weeks, while the chronic group underwent up to 14 assessments over 20 weeks. ARAT and FMA-UE were used in the subacute group, while the chronic group completed the ARAT.
DAPS vs Standard Measures
In the subacute stroke group, DAPS measurements were highly consistent when repeated on the same day, with an intraclass correlation coefficient of 0.95. Results were more accurate than clinician assessments, with an estimated error rate that was 52%-58% lower than that of the normalized ARAT and FMA-UE scores.
DAPS scores were higher in patients who had better upper-limb function on the ARAT and FMA-UE, with correlations of 0.79 and 0.81, respectively.
Changes in DAPS scores over time also closely tracked changes in clinical measures, with correlations of 0.98 with ARAT and 0.93 with FMA-UE.
Among the 36 subacute stroke participants with multiple assessments, DAPS also distinguished patients who improved from those who did not. The area under the receiver operating characteristic curve was 0.89 for the ARAT and 0.75 for the FMA-UE.
Study limitations included the use of data from previous studies, limited improvement in the chronic stroke group, and fewer women than men in this group, which made differences in DAPS scores between the sexes difficult to interpret.
What Comes Next
Tracking daily movement could provide additional information between routine stroke rehabilitation assessments.
"Measuring movement in a patient's everyday environment provides ecological validity," Lee said. "This gives clinicians a near-real-time trajectory of recovery, making it possible to assess whether a patient is responding to a prescribed intervention and adapt therapy dynamically rather than waiting weeks to discover a plateau."
This approach could also benefit stroke rehabilitation research by reducing the number of individuals needed for clinical trials. In a chronic stroke simulation, three participants per treatment arm were needed with DAPS compared with eight with ARAT, a 62.5% reduction. In a separate simulation involving subacute stroke, seven participants per arm were needed with DAPS compared with 26 with FMA-UE, a 73.1% reduction.
However, further validation is needed before it could be used routinely in clinical practice or as a trial endpoint, the investigators noted. The algorithm is currently open sourced to facilitate independent testing across different patient populations and clinical settings.
"Our primary unanswered question is empirical validation," Lee said. "We need to replicate a successful clinical trial to verify in practice whether the sample-size reductions and efficiency gains predicted by our simulations hold up in an active trial setting."
Disclosure information for study authors is available in the original study publication.
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.