Remote Physical Activity Monitoring in Neurological Disease: A Systematic Review

1Research over the last decade has examined accelerometer-based remote monitoring of physical activity in health and disease. Wearable physical activity monitors have also become increasingly commonplace as consumer products, primarily marketed for fitness. When considering whether remote physical activity monitoring can inform decision-making for use in clinical populations, questions about validity, reliability, feasibility and responsiveness arise.

2Physical activity is typically defined as voluntary bodily movement using skeletal muscle that requires energy beyond resting levels. Measurement of physical activity is important because of established links between physical inactivity and various morbidities. Neurological disease can also increase the risk of physical inactivity secondary to associated disability. Physical activity monitoring using accelerometers, pedometers, and gyroscopes has gained traction in healthcare, wellness and medical research. Monitoring can focus on gait, upper or lower limb function or other patterns of body movement or behavior. Potential variables that can be used to measure physical activity include step count, activity count, activity bouts, active minutes and energy expenditure. Remote physical activity monitoring provides a convenient way of assessing movement outside of the clinic setting and may correlate with disease-specific predictors, outcomes, or interventions.

3However, remote measurement of physical activity in people with neurological disease has the potential to be complicated by neurological impairments such as gait abnormalities, weakness, spasticity or tremor that could confound remote measurement in these populations. While disease-specific examination and validation of remote physical activity is needed, systematically reviewing the literature across neurological disorders may reveal lessons about feasibility, implementation and interpretation that apply across neurological indications.

4This systematic review summarizes research on remote physical activity monitoring in neurological diseases, including multiple sclerosis (MS), stroke, Parkinson’s disease (PD), dementia, traumatic brain injury (TBI), ataxia, epilepsy and migraine. To focus primarily on physical activity outside of the immediate clinical setting, studies were included that monitored physical activity for at least 24 hours.

5Original research studies were identified from the PubMed/MEDLINE, CINAHL and SCOPUS databases. Once relevant articles were identified, they were located individually in the Web of Science database and in Google Scholar to examine citing and cited-by articles. The search strategy used a combination of MeSH (Medical Subject Headings) terms and keywords. The search terms used alone and in combination were categorized according to PICO: Population: “multiple sclerosis,” “parkinson*,” “stroke,” “cerebrovascular accident,” “brain injury,” “ataxia,” “headache,” “migraine,” and “epilepsy”. Intervention/ indicator: “acceleromet*,” “activity monitor*,” “free living physical activity,” “pedometer,” “wearable sensor*”.

6Comparator/ Control: Not using the device. Inclusion criteria did not require studies to be intervention trials. Outcome: physical activity (measured heterogeneously e.g. step or activity count, movement count, bouts of activity)

7We also examined articles that reported physical activity monitoring in samples with “heart disease” or “diabetes” to identify if sub-populations of neurological conditions were evaluated. A medical librarian (P.T.) advised on search strategy, search terms, and methodology.

8Studies were included if they 1) recorded human physical activity, defined as voluntary (skeletal) muscle movement during daily functioning requiring energy expenditure; 2) monitored subjects for ≥24 hours; 3) used remote monitoring via devices that employ accelerometers, gyroscopes and/or pedometers to measure physical activity and capture data remotely for subsequent analysis; 4) enrolled adults 18 years of age or older with a diagnosis of MS, stroke, PD, dementia, TBI, epilepsy, migraine, headache or ataxia; 5) and were published between January 2004 and December 2014. Studies were excluded that recorded involuntary motor activity such as seizures or tremor; focused on movement during sleep or examined sleep as the primary outcome; extrapolated measures for average step counts from shorter monitoring periods; measured total daily energy expenditure (such as daily calorie consumption or diet interventions) without physical activity monitoring; or measured global positioning satellite (GPS) data exclusively rather than more direct measurement or corroboration of physical activity. We also excluded case reports and case studies.

9Two authors (V.B., E.P.) searched independently. Titles and abstracts were screened for relevance and supplementary review. One author (V.B.) manually searched the reference sections of complete manuscripts for additional articles. Consensus for meeting the eligibility criteria was achieved by comparing search results (V.B., E.P.).

10Data were extracted (V.B.) and checked (E.P., D.D.A., J.M.G), with final adjudication by consensus from two senior authors (D.D.A., J.M.G.). Variables included population studied; disease-specific severity levels; device name, placement and intent (i.e. patient behavior change or healthcare monitoring); intervention (if any); setting; demographic data; and study details, including design, funding sources and motivational factors (i.e. subject imbursement, visual display of data). Studies were graded for risk of bias based on methodology proposed by the Cochrane Collaborations (see S1a and S1e Table). Conclusions and lessons learned across studies were summarized.

11The systematic review identified 745 studies through the databases and an additional 25 articles through recursive and manual reference searches. Once eligibility criteria were applied, 137 studies remained (Fig 1 and S1 Fig). Individual studies are summarized in Tables 1–5. Table 6 (sections a-e) documents the sample characteristics. The risk of bias with level of evidence for interventional studies is reported in S1 Table. A description of the most common devices used in the included studies appears in S2 Table.

12The majority of the 61 studies (60/61, 98.4%) that remotely monitored activity in MS (Tables 1 and 6 section a) measured physical activity by walking; one study focused on upper extremity movement. The length of continuous monitoring ranged from 3 to 7 days for each discrete measurement period with 7 days being the measurement paradigm for the majority (41/61, 67.2%) of studies. Most of the studies (44/61, 72.2%) included both relapsing and progressive MS phenotypes; >78% of participants had relapsing MS. Although MS disease duration varied, studies primarily included persons with disease duration of less than 20 years. Fifty-two studies focused on people having mild to moderate disability (able to walk without a cane or support), and only two studies reported inclusion of people with greater levels of disability (requiring a walker or wheelchair for mobility). One research group (Department of Kinesiology and Community Health, University of Illinois at Urbana-Champaign, Urbana, Illinois) authored 49/61 studies (81.7%); results from studies conducted by other groups generally corroborated this group’s results. No studies reported direct research funding by monitoring device manufacturers.

13In two studies focused on people with MS, average daily activity and step count measured via wearable accelerometers correlated with performance-based and self-reported walking mobility and physical activity. A third study observed that accelerometers correlate only with performance-based measures of walking (6-minute walk; [6MW,] and the Timed-Up and Go, test; [TUG]) and not self-reported walking activity.

14People with MS record lower levels of physical activity than the general population and unaffected controls. People with MS also frequently fail to reach daily levels of intensity and duration recommended for the general population. Lower physical activity levels in MS are associated with higher levels of disability and lower scores in a range of clinical and self-reported outcomes such as walking speed and endurance (Timed 25-Foot Walk, 2-minute walk and 6MW), fatigue (i.e. Fatigue Severity Scale), depression (i.e. Hospital Anxiety and Depression Scale), self-efficacy, and balance (Berg Balance Scale, TUG). Higher levels of physical activity correlate significantly with better performance on mobility measures in the clinic, self-reported disability questionnaires and cognitive processing speed. Lower physical activity in MS correlates with age, disease duration, progressive forms of MS, spasticity, and unemployment, but not race. However, rate of disability accumulation over 6 months was similar in an active versus a more sedentary group in one study. Of the 4 studies in MS that tested interventions, internet-based interventions appear to be beneficial in promoting objective and self-reported physical activity and are associated with decreased disability.

15More than half of the studies (24/41 studies, 58.5%) that reported on activity monitoring post-stroke (Tables 2 and 6 section b) measured walking or gait; 14 studies (34.1%) assessed upper extremity or arm movement; and 3 (7.3%) measured both arm movement and walking. One study included participants with either a diagnosis of stroke (n = 30) or TBI (n = 20). This study is listed under both diagnostic headings and results are analyzed by diagnosis group. Monitoring duration was usually between 2 and 6 days (28 studies, 68.3%), although one study monitored step count for 4 weeks, reporting change in daily average steps between the 5 days prior and post intervention. Monitoring usually commenced between 3 and 6 months after the stroke (28, 68.3%). Fewer than 40% of studies reported details about the type of stroke (i.e. ischemic or hemorrhagic and/or neuroanatomical localization). The presence and side of paralysis or paresis was reported in 92.7% (38/41) of the articles; one article reported on the presence or absence of tremor as a potential confounder. During monitoring, participants were in the “home/ community” or “hospital—acute care” settings; none of these studies specifically monitored patient activity in acute rehabilitation or at skilled nursing facilities.

16Post-stroke, people tend to have a lower frequency of moderate to vigorous bouts of physical activity and are less likely to reach generally recommended minimum levels of physical activity than healthy controls. However, one study found that the “time participants spent on their feet” was similar to healthy controls. Lower physical activity level post-stroke is associated with poor balance and greater depression scores. Four intervention studies were identified: 3 aimed at improving arm function, and 1 successfully increased daily step counts using a goal-directed step activity-monitoring program. An observational study showed little change in daily limb use with accelerometer results, despite significant improvements in clinical measures. Measuring both upper extremities post-stroke facilitated differentiation of uni- vs. bi-manual tasks, distribution of arm usage, and comparison of impaired vs. unimpaired arm function. Spontaneous early arm movement activity was associated with greater neurological recovery post stroke, although results varied regarding prediction of upper extremity recovery.

17All 20 studies that reported on activity monitoring in PD (Tables 3 and 6 section c) measured physical activity through walking. Durations of monitoring were mostly for 2–6 days (6, 35.0%) or 7 days (8, 40.0%). Thirty-five per cent of studies (7/20) reported on the presence or absence of tremor as a potential confounder.

18One activity-monitoring device (DynaPort Hybrid) was able to differentiate between ON/OFF phases and detect “missteps/ near falls” in people with PD in the clinic and home environments. Participants wore the device in the clinic while missteps were induced, an algorithm was developed to detect deviations from their gait patterns, and the algorithms were validated during an additional three days of device wear-time outside the clinic. Abnormal gait patterns, such as lower amplitude and greater step-to-step variability, were associated with fall risk in people with PD whereas total walking amount was not.

19People with PD tend to take fewer steps and do shorter bouts of physical activity than the general population. A reduction in total number of steps per day correlates with PD progression, and milder severity of PD is associated with higher physical activity levels. People with PD tend to have a smaller number of longer sedentary periods than healthy controls, although total sedentary time is similar. An intervention study aimed at increasing physical activity in people with PD resulted in increased muscle strength and flexibility, self-directed exercise frequency and duration, reduced fear of falls, but no overall change in the total amount of physical activity.

20Nine of the 11 studies (81.8%) that reported on activity monitoring in dementia (Tables 4 and 6 section d) measured physical activity as walking. Two studies focused on upper extremity or arm movement in addition to walking or gait. Monitoring typically lasted 2–6 days (6/11 studies, 54.5%). Most studies involved people with a presumed Alzheimer's dementia or a combination of Alzheimer’s dementia and frontotemporal or Lewy Body dementias (8/11 studies, 72.7%). Severity of cognitive dysfunction was usually mild to moderate (9/11 studies, 81.8%). Only 2 studies involved people with severe cognitive disability.

21Physical activity level in people with dementia depended on stage of disease. People with mild Alzheimer’s dementia have lower mean physical activity (associated with apathy and more daytime napping) and lower step count per day compared to people with mild cognitive impairment (MCI) or healthy controls. Monitoring was feasible in people with cognitive impairment and accelerometry was able to distinguish partners with and without early Alzheimer’s disease even before deficits were clinically visible. Monitoring in people with dementia distinguished “intensive wandering behavior,” which, when assessed along with estimations of energy expenditure, facilitated accurate calculation of nutritional requirement.

22The single study in TBI concluded that 7 days of accelerometry was feasible in 30 people more than 3 months post-TBI (adherence >86%). Physical activity was below recommended levels. Data were more reliable than a self-reported physical activity questionnaire to determine amount, but not type of, moderate to vigorous physical activity.

23In a single study of physical activity monitoring in ataxia, 19 participants with spinocerebellar ataxia wore a step activity monitor for 7 days; greater physical activity was associated with shorter disease duration and lower disability scores.

24The remaining studies that reported physical activity monitoring in mixed populations measured walking activity or gait (Tables 5 and 6 section e). One study observed 50 people with either TBI or stroke over the age of 50 and greater than 3 months post injury assessing various activity monitoring systems. Another study evaluated a tri-axial accelerometer (TriTrac RT3) over 7 days in a study sample of patients with stroke (> 6 months in duration) (20), PD (7), or MS (11), and sedentary healthy controls (9). Mobility was more accurately assessed using 7-day activity monitoring than with a patient reported measure. A third study measured step count in participants with PD (10), MS (10), primary muscle disorder (10) and healthy controls (30) over 7 days in free-living conditions. Neurological patients were observed to have a lower level of physical activity than healthy controls.

25Many studies provided evidence of the reliability of various devices. For the StepWatch Activity Monitor post-stroke, the test-retest interclass correlation coefficient (ICC) values were 0.93–0.99 over a minimum of 3 days. Other studies documented similar ICC values for Actical accelerometer activity counts (ICC >0.94; 95% CI 0.91–0.97) in people post-stroke with no differences between workdays and weekend days. In MS, test-retest ICC values were 0.91 and 0.88 for steps per day and activity counts per day (ActiGraph GT3X), respectively, over 6 months, although the ICC was smaller for people with greater disability (ICC = 0.672 for activity counts/day and ICC = 0.774 for steps/day). In a direct comparison in MS, seven days of monitoring (ActiGraph 7164) produced an ICC of 0.93 whereas three days yielded an ICC of 0.80, with no difference noted between days of the week (weekdays or weekend days) when measuring walking activity or gait. A 7-day period (using a TriTrac RT3 accelerometer) was most reliable in patients with stroke, MS or PD. In PD, 24 hours of monitoring was found to be reliable to record a participants’ functional activity (average step count, inactive vs. active minutes using an activity monitor). In spinocerebellar ataxia, internal consistency was highest with 7-days of monitoring, but 3 days of monitoring using a step activity monitor still correlated strongly with 7-day measures.

26Evidence of validity primarily comes from comparison of activity data collected remotely with established performance-based and self-report measures. In MS, number of steps per day correlates with the Expanded Disability Status Scale (EDSS), the Patient Determined Disease Steps (PDDS) scale, performance-based ambulatory measures in the clinic and patient-reported outcomes. Post-stroke, the ICC was high when comparing activity counts for the paretic and non-paretic hip (0.96), but correlation was moderate when comparing activity with patient-reported activity questionnaires. Post-stroke, activity counts for the upper extremity had high predictive value for good arm recovery; both arms are used less than by healthy controls, and less arm activity correlates with increased impairment and reduced muscle activity measured by EMG. In TBI, activity counts were more accurate than questionnaires in characterizing levels of moderate to vigorous physical activity. In spinocerebellar ataxia, average step count across 7 days correlated strongly with disability scores and moderately with walking speed.

27This systematic review examines a decade of literature on remote monitoring of physical activity in people with neurological diseases. Physical activity monitoring is feasible in these populations, including in those with impaired cognition. Some of the evidence was sparse: very few of the eligible studies used remote activity monitoring as an outcome for an intervention (9/134), indicating that use of these tools in neurological populations is still primarily in an observational or validation phase. Nevertheless, the data in some diagnostic groups indicate that remote monitoring of physical activity can be a clinically useful way to assess activity status over time.

28A wide array of variables can be used to measure physical activity. The most common are permutations of activity count or step count. However, other activity variables may provide better prognostic value in disease-specific situations. For example, length and number of moderate to vigorous activity bouts reflected differences better than total step count in some studies following stroke, whereas total step count, highest step rate in 1 minute, highest step rate in 5 minutes, and peak activity index appeared most reliable in others. Detection of upper limb recovery via accelerometer measures of arm/upper extremity movement was also favored post-stroke, and may prove helpful in other populations, such as upper limb function in MS. In PD, average number of steps per day correlated with activity level and disease progression in many studies. However, in a minority there was no correlation between activity count and patient-reported assessments of symptom severity. Physical activity monitoring using specialized devices may also be used to predict fall-risk and measure missteps in PD, functionality that, if replicated and validated, could be very be useful in other neurological populations, including MS and stroke.

29Across diagnoses, physical activity is consistently lower in neurological populations than in those without neurological disease. The total amount of activity or step counts measured via accelerometers is lower in MS (e.g.), dementia (e.g.), and stroke than in controls. In people with moderate to severe PD, pattern of activity was different (sedentary bouts were longer) but total volume of sedentary time was similar to controls. In those with mild to moderate PD, speed of turns was slower than in healthy controls, and reductions in daily ambulatory activity (volume of moderate to vigorous physical activity) were detected over a year, even without evident changes in clinical measures of gait or disease severity.

30Remote physical activity monitoring for durations of >24 hours was feasible in the neurological populations studied; however, adherence was a potential concern. Post-stroke, the placement of sensors in pockets (confounding clothing movement with activity and increasing risk of leaving the device behind when changing clothes), impaired mental status, depression, and device discomfort (leading to withdrawal of 25% of patients from one study) all reduced adherence. In PD, patients concerned with appearance also had reduced adherence (affecting over a quarter of participants in one study). Physical activity monitoring for extended periods of time was well tolerated in people with Alzheimer's Disease, although adherence was lower (83%) compared to healthy controls (100%). Tolerability was not recorded as a significant problem in studies involving people with MS, although adherence and loss of data from attrition was noted in several studies (S1a Table).

31Intervention studies in stroke are heterogeneous with regards to adherence and walking performance. A circuit-based rehabilitation study aimed at increasing stroke patients’ amount and rate of walking in their home environment, found high adherence rates to the program. Specific mention of device adherence was not recorded. A separate intervention study recording steps per day during 4 weeks, reported ~25% attrition due to non-compliance.

32Interventional studies testing physical activity monitoring in stroke patients observed changes in clinical and patient-reported measures, but, perhaps in part due to inadequate adherence, failed to demonstrate changes in physical activity (average steps per day) in the home environment. Likewise, home intervention for increasing activity in people with PD observed improvements in strength, flexibility and a reduction in fear of falling, without noting changes is overall daily physical activity levels. Studies in MS, however, indicated that Internet-based exercise interventions can help to increase physical activity (activity/ steps per day), and improve self-reported disease symptoms and self-efficacy over 6 months.

33The few reviewed intervention studies using remote monitoring affirm that measuring activity levels of patients with minimal invasiveness in their natural environment has potential advantages over traditional self-reported and clinic-based measures. Self-reported measures are easy to obtain through questionnaires but are prone to recall bias. Performance-based measures in clinic can provide a useful snapshot of physical activity and may have prognostic value but are primarily measures of physical activity patients are capable of rather than how active patients actually are in their natural environment. Future intervention studies should continue measuring outcomes in multi-faceted ways as researchers gather more evidence of the relationship between the different categories of measures.

34The accelerometer-based activity monitors used in many of the included studies are not primarily designed or marketed for consumer use, with current prices ranging from ~$200 to $600, which do not include software (~$2000) necessary for data analysis (S2 Table). Many commercially available monitors have not yet been evaluated in neurological populations. One recent study in healthy individuals showed no systematic bias when comparing step counts recorded via commercially available activity monitors (i.e. Fitbit) versus research grade accelerometers (ActiGraph). However, the accuracy of non-research grade activity monitors remains an active source of debate, as does the failure of activity monitors to efficiently track many non-walking-based physical activities such as swimming, cycling, strength training and yoga.

35Lessons learned from this systematic review lead to several recommendations for translation of remote physical activity monitoring in neurological indications. 1) Remote physical activity monitoring research would benefit from standardization in reporting. We provide a checklist that might aid researchers and clinicians in future research and clinical use (Fig 2). 2) While remote monitoring devices and measurement protocols should be tested and validated in specific neurological conditions, solutions are likely to translate across neurological conditions that share patterns of functional impairment. 3) Activity monitors have the potential to be retooled with suites of variables specific to particular diagnostic indications. For example, a disease-specific remote monitoring suite for MS might include step and activity count, fall detection, upper extremity function and temperature sensors to correlate with possible heat-induced demyelination-related disability. Additional functionality could include reminders to exercise, take medication or keep to a schedule for bowel and bladder maintenance. For all diagnostic groups, monitors could be tailored to track adherence to home exercise programs. If worn for longer periods of time, they could detect continuation of or changes in activity after specific punctate interventions (pharmacologic, medical, telehealth, or exercise-based) aimed to increase activity levels. Further studies are needed for longer periods of time (continuously for months/years) to determine the feasibility and responsiveness of activity monitoring devices for these purposes.

36Limitations of this review include the focus on adults with neurological disease; lessons learned do not necessarily extend to pediatric populations with these conditions. This review also focuses specifically on physical activity monitoring and by necessity does not analyze advances in non-voluntary activities that can also be assessed via remote monitoring, such as seizure detection and sleep. Because only 9 of the 134 studies were interventional, our review does not include a meta-analysis.

37In conclusion, this review records emerging evidence to support the use of remote physical activity monitoring in neurological care and neurorehabilitation. Because some patients already regularly perform such monitoring on themselves using commercial wearable devices or through their smartphones, providers also need to become familiar with these technologies and strategies for interpretation and to consider this knowledge translation when planning future studies.