Stroke is a leading cause of long-term disability worldwide.1
Although major advances have been made in the treatment of acute
ischemic stroke, innovations for recovery and rehabilitation following
stroke remain more limited.2 Notably, recent clinical trials in stroke rehabilitation yielded neutral results3–5 and longitudinal studies show tremendous heterogeneity in both patients’ trajectories and end points for recovery.6,7
To improve the efficacy of stroke rehabilitation, we need a better
understanding of the biological mechanisms underlying recovery, either
to develop new interventions or to better match existing interventions
to the most responsive patients.8,9
Omics-based approaches offer a fruitful avenue for gaining this
understanding but require data collection on a scale generally not seen
in stroke rehabilitation research; for example, large trials in stroke
rehabilitation collect data from N<400 people,4 compared with the (10s of) 1000s needed for genome-wide association studies.10,11
Although major advances have been made in the treatment of acute
ischemic stroke, innovations for recovery and rehabilitation following
stroke remain more limited.2 Notably, recent clinical trials in stroke rehabilitation yielded neutral results3–5 and longitudinal studies show tremendous heterogeneity in both patients’ trajectories and end points for recovery.6,7
To improve the efficacy of stroke rehabilitation, we need a better
understanding of the biological mechanisms underlying recovery, either
to develop new interventions or to better match existing interventions
to the most responsive patients.8,9
Omics-based approaches offer a fruitful avenue for gaining this
understanding but require data collection on a scale generally not seen
in stroke rehabilitation research; for example, large trials in stroke
rehabilitation collect data from N<400 people,4 compared with the (10s of) 1000s needed for genome-wide association studies.10,11
It
is not sufficient to collect large numbers of biospecimens, however, if
we do not also have good behavioral phenotypes to define recovery.12
Detailed clinical phenotypes are thus generally preferable to proxy
measures because they are more likely to capture biological mechanisms.
As an analogy, we can look at genome-wide association studies of alcohol
use disorder versus more easily collected measures of alcohol
consumption. Although there is some overlap in these phenotypes, they
also show distinct associations across the genome.13 Detailed clinical assessments thus provide critical information but are often more costly and difficult to deploy at scale.
is not sufficient to collect large numbers of biospecimens, however, if
we do not also have good behavioral phenotypes to define recovery.12
Detailed clinical phenotypes are thus generally preferable to proxy
measures because they are more likely to capture biological mechanisms.
As an analogy, we can look at genome-wide association studies of alcohol
use disorder versus more easily collected measures of alcohol
consumption. Although there is some overlap in these phenotypes, they
also show distinct associations across the genome.13 Detailed clinical assessments thus provide critical information but are often more costly and difficult to deploy at scale.
For instance, with respect to motor impairment, the Fugl-Meyer Assessment14
of motor recovery (a common tool used in motor recovery studies) takes
about 30 minutes to administer. This time, plus travel time for the
patient into the clinic, multiplied by the number of patients multiplied
by the number of assessments per patient, makes these kinds of
assessments costly and time-consuming. Other clinical measures, like the
modified Rankin Scale, are arguably more scalable but are cruder and
can have undesirable measurement properties, such as strong
floor/ceiling effects.15
Patient self-report measures are valuable and scalable, but
fundamentally further from biology, as self-report is inherently
filtered through a patient’s own perceptions and may track more closely
with measures of participation than with underlying impairments in body
structure/function.16,17
Thus, there is a need for a scalable metric that can capture
biologically meaningful phenotypes (ie, body structure/function
variance) relatively independent of personal or environmental
moderators.
of motor recovery (a common tool used in motor recovery studies) takes
about 30 minutes to administer. This time, plus travel time for the
patient into the clinic, multiplied by the number of patients multiplied
by the number of assessments per patient, makes these kinds of
assessments costly and time-consuming. Other clinical measures, like the
modified Rankin Scale, are arguably more scalable but are cruder and
can have undesirable measurement properties, such as strong
floor/ceiling effects.15
Patient self-report measures are valuable and scalable, but
fundamentally further from biology, as self-report is inherently
filtered through a patient’s own perceptions and may track more closely
with measures of participation than with underlying impairments in body
structure/function.16,17
Thus, there is a need for a scalable metric that can capture
biologically meaningful phenotypes (ie, body structure/function
variance) relatively independent of personal or environmental
moderators.
To that end, the goal of the
present study was to understand how wearable sensor data correlate with
less scalable, but well-validated in-clinic assessments, and how these
correlations may change over time following stroke. Specifically, we
focused on the use ratio (UR) for the paretic arm relative to the
nonparetic arm collected through bilateral, wrist-worn accelerometers.
The UR is an excellent candidate measure given that (1) it has an analog
in basic research where fore-paw asymmetry is used as an outcome in
rodent models of stroke recovery18,19; (2) human research shows the UR is feasibly collected in adults at all stages poststroke,20
(3) in neurologically intact human adults, the UR is narrowly
distributed around 1.0 but does not have a hard ceiling like many
in-clinic assessments21;
and (4) the UR is an objective real-world measure collected passively
during daily life, reducing the burden on patients and clinicians.22–24
Our in-clinic measures were the upper extremity subscale of the
Fugl-Meyer Assessment, which is considered a measure of body
function/structure within the International Classification of
Functioning, Disability, and Health (ICF) framework,25,26 and the Action Research Arm Test (ARAT), which is considered a measure of activity capacity.25,27
Given that the UR is considered a measure of activity performance
within the ICF framework, we anticipated strong positive correlations,
indicating concurrent validity, both cross-sectionally (ie, agreement
between people) and longitudinally (ie, agreement in sensitivity to
change) with our in-clinic measures.
present study was to understand how wearable sensor data correlate with
less scalable, but well-validated in-clinic assessments, and how these
correlations may change over time following stroke. Specifically, we
focused on the use ratio (UR) for the paretic arm relative to the
nonparetic arm collected through bilateral, wrist-worn accelerometers.
The UR is an excellent candidate measure given that (1) it has an analog
in basic research where fore-paw asymmetry is used as an outcome in
rodent models of stroke recovery18,19; (2) human research shows the UR is feasibly collected in adults at all stages poststroke,20
(3) in neurologically intact human adults, the UR is narrowly
distributed around 1.0 but does not have a hard ceiling like many
in-clinic assessments21;
and (4) the UR is an objective real-world measure collected passively
during daily life, reducing the burden on patients and clinicians.22–24
Our in-clinic measures were the upper extremity subscale of the
Fugl-Meyer Assessment, which is considered a measure of body
function/structure within the International Classification of
Functioning, Disability, and Health (ICF) framework,25,26 and the Action Research Arm Test (ARAT), which is considered a measure of activity capacity.25,27
Given that the UR is considered a measure of activity performance
within the ICF framework, we anticipated strong positive correlations,
indicating concurrent validity, both cross-sectionally (ie, agreement
between people) and longitudinally (ie, agreement in sensitivity to
change) with our in-clinic measures.
More at link.
