Quantifying the Effects of Sleep on Sensor-Derived Variables from Upper Limb Accelerometry in People with and without Upper Limb Impairment

doi:10.21203/rs.3.rs-3838376/v1

Background: Despite the promise of wearable sensors for both rehabilitation research and clinical care, these technologies pose significant burden on data collectors and analysts. Investigations of factors that may influence the wearable sensor data processing pipeline are needed to support continued use of these technologies in rehabilitation research and integration into clinical care settings. The purpose of this study was to investigate the effect of one such factor, sleep, on sensor-derived variables from upper limb accelerometry in people with and without upper limb impairment and across a two-day wearing period.

Methods: This was a secondary analysis of data collected during a prospective, longitudinal cohort study (n = 127 individuals, 62 with upper limb impairment and 65 without). Participants wore a wearable sensor on each wrist for 48 hours. Five upper limb sensor variables were calculated over the full wear period (sleep included) and with sleep time removed (sleep excluded): preferred time, non-preferred time, use ratio, non-preferred magnitude and its standard deviation. Linear mixed effects regression was used to quantify the effect of sleep on each sensor variable and determine if the effect differed between people with and without upper limb impairment and across a two-day wearing period.

Results: There were significant differences between sleep included and excluded for the variables preferred time (p < 0.001), non-preferred time (p < 0.001), and non-preferred magnitude standard deviation (p = 0.001). The effect of sleep was significantly different between people with and without upper limb impairment for one variable, non-preferred magnitude (p = 0.02). The effect of sleep was not substantially different across wearing days for any of the variables.

Conclusions: Overall, the effects of sleep on sensor-derived variables of upper limb accelerometry are small, similar between people with and without upper limb impairment and across a two-day wearing period, and can likely be ignored in most contexts. Ignoring the effect of sleep would simplify the data processing pipeline, facilitating the use of wearable sensors in both research and clinical practice.

Stroke

sleep

wearable sensors

accelerometry

breast cancer

multiple sclerosis

fracture

Wearable sensors have enormous potential to improve the delivery and outcomes of rehabilitation care (1, 2). This potential is being realized in both the research and clinical care realms. Research labs are exploiting wearable sensor technology to measure activity outside the clinic/laboratory throughout a course of rehabilitation care (3, 4) or an experimental intervention (5–8). Investigations have shown that improvements observed in the clinic/laboratory (i.e., activity capacity) do not necessarily translate to improvements in what a person actually does in their free-living environment (i.e., activity performance). While the use of these technologies has been primarily confined to research labs, their findings have resulted in a more recent emphasis to integrate these devices into clinical care settings (2, 9–11).

Despite the growing uptake of wearable sensors in rehabilitation, the use of these technologies pose significant burden on data collectors (i.e., patients, research participants) and analysts (1, 2, 11, 12). For example, in the research setting, support personnel are needed to carefully inspect and process the data, requiring significant time and financial resources. The challenges become particularly problematic, if not unreasonable, when attempting to integrate these technologies into clinical settings. Busy clinicians do not have time to carefully scrutinize or perform manual computations of wearable sensor data. Thus, efforts to integrate these technologies into clinical care setting necessitate a fast, efficient, and least burdensome data processing pipeline.

Expediting the data processing pipeline to minimize burden on both research and clinical staff requires a thorough understanding of factors that may influence it (11–13). One of the factors that is common among all persons and involves periodic movements is sleep (14). The unexplored impact of sleep on the data processing pipeline has important consequences for both the development of research protocols and the resources required to process wearable sensor data. For example, some research protocols require participants to remove the wearable sensor during periods of sleep if sleep is not a primary interest (15). This, however, requires participants to remember to don the device after periods of sleep, placing additional burden on the participant and increasing opportunities for data to go missing. On the contrary, protocols that permit participants to wear the device during sleep may increase burden on research support staff to remove periods of sleep if movement during sleep is not of interest. Understanding the effect of sleep on the wearable sensor data processing pipeline would determine whether these additional burdens are necessary and could streamline the use of these technologies in both research and clinical care settings.

The purpose of this investigation was to determine the effect of sleep on sensor-derived variables from upper limb accelerometry in people with and without upper limb impairment. We focused on the upper limb for two main reasons. The first is that upper limb impairment is a common sequelae of many conditions, such as stroke (16, 17), multiple sclerosis (18, 19), breast cancer (20, 21) and upper limb orthopedic conditions (22–24). The second is that a common application of wearable sensor technology in rehabilitation research is to measure upper limb movement (25–28). We therefore had two objectives for this analysis: (1) quantify the effect of sleep on sensor-derived variables from upper limb accelerometry, and (2) determine if the effect of sleep differs between people with and without upper limb impairment and across a two-day wearing period. The results of this analysis could help to minimize the personnel and computational burden to utilize wearable sensor technology in rehabilitation research labs and inform future efforts that seek to integrate this technology into clinical care settings.

Participants:

This study was a secondary analysis of data collected during the baseline time point of a two-site (Washington University in St. Louis, MO and Shirley Ryan Ability Lab, Chicago, IL) prospective, longitudinal cohort study (NIH R37HD068290). This study recruits individuals without upper limb impairment (e.g., healthy controls) and individuals with upper limb impairment, including people with stroke, multiple sclerosis, distal radius fracture, proximal humerus and/or clavicle fracture, shoulder pain, or breast cancer who were undergoing physical and/or occupational therapy to improve upper limb function. Eligibility criteria for the control cohort included: (1) ≥ 18 years of age, (2) free of neurologic, musculoskeletal, or medical conditions that affect the upper limb or significantly affect physical activity in general. Eligibility for the upper limb impairment cohort included: (1) ≥ 18 years of age, (2) upper limb impairment as judged by a referring physician or surgeon, (3) referral to rehabilitation services to address upper limb impairment, (4) therapist documented goal(s) to increase or restore upper limb function, (5) no other concurrent neurologic, musculoskeletal, or medical conditions that affect the upper limb or physical activity in general, (6) no other co-morbid conditions determined by physician or therapy documentation that indicate a minimal chance for improvement in function (e.g., end-stage cancer diagnosis), (7) not pregnant or planning to become pregnant, (8) no cognitive or communication problems that would prevent them from completing the study. For this analysis, all participants with upper limb impairment were grouped together as we had no scientific reason to suspect that the effects of sleep on the upper limb performance variables would differ based on diagnosis, nor would we have the statistical power to detect these sub-group differences. Only participants who wore the sensors for the full 48-hour wear period were included in this analysis. The study was approved by Washington University’s Institutional Review Board, and all participants signed informed consent prior to engaging in any study procedures. This manuscript was developed in accordance with the STROBE guidelines.

Measures:

All study procedures occurred remotely and study data were collected and managed using REDCap electronic data capture tools housed at Washington University in St. Louis. REDCap (Research Electronic Data Capture) is a secure, web-based software platform designed to support data capture for research studies providing 1) an intuitive interface for validated data capture; 2) audit trails for tracking data manipulation and export procedures; 3) automated export procedures for seamless data downloads to common statistical packages; and 4) procedures for data integration and interoperability with external sources (29, 30). Participants in both cohorts were sent several surveys and two wrist-worn accelerometers to wear for a 48-hour period. The surveys collected demographic information, including age, sex, race, and ethnicity, as well as clinical information (upper limb impairment cohort only). Surveys were completed online or on paper, depending on the participant’s preference, and within one week of wearing the sensors. For the upper limb impairment cohort, participants completed the baseline time point within two weeks of starting outpatient rehabilitation services.

Upper limb accelerometry variables were measured using established, reliable, and valid bilateral wrist-worn accelerometry methodology (31, 32). Participants wore tri-axial GT9X Link accelerometers (ActiGraph Inc, Pensacola, Florida) on each wrist for 48 hours. Participants were instructed to go about their normal routine while wearing the sensors and to keep the sensors on during sleep. Once the sensors were returned to the lab, the data were visually inspected to verify a 48-hour wear period using ActiLife 6 software (ActiGraph Inc, Pensacola, Florida) and exported for further processing.

Upper Limb Performance Variables:

Work is ongoing to determine which sensor-derived variables from upper limb accelerometry are most important for research and clinical practice (33). Thus, we examined 25 variables in total that reflect different aspects of motor behavior of the upper limbs in daily life under the assumption that the relevance of specific variables may vary across research questions, clinical populations, and scientific fields (2, 34-37). Here, we focus on five variables that were previously shown to generate unique clusters of individuals with and without neurologic upper limb deficits (33), and present the remaining 20 variables in the Supplement. Table 1 displays the five upper limb sensor variables and their respective calculations.

Table 1. Description of Upper Limb Sensor Variables*

Variable	Description
Preferred Time	Time (in hours) that the dominant/unaffected limb is moving.
Non-Preferred Time	Time (in hours) that the non-dominant/affected limb is moving.
Use Ratio	Ratio of hours of non-dominant/affected limb movement, relative to hours of dominant/unaffected limb movement.
Non-Preferred Magnitude	Median of the accelerations of the non-dominant/affected limb, in gravitational units (gs)
Non-Preferred Magnitude Standard Deviation	Standard deviation of the magnitude of accelerations across the non-dominant/affected limb, in gravitational units (gs)

*Preferred indicates the dominant limb (control cohort) or unaffected limb (upper limb impairment cohort). Non-preferred indicates the non-dominant limb (control cohort) or affected limb (upper limb impairment cohort).

Data were sampled at 30 Hz, band-pass filtered, and down-sampled into 1-second epochs for each axis by summing the 30samples within each second. Accelerations in each axis were combined into a single vector magnitude using the formula A vector magnitude threshold of ≥ 2 was used to determine if the upper limb was active for each 1-second epoch (38, 39). To compute the variables preferred time and non-preferred time, each second of activity was summed over the wearing period and converted to hours. The use ratio was calculated as the ratio of non-preferred limb movement relative to the preferred limb movement, expressed in hours. The non-preferred magnitude and standard deviation (Table 1) were converted from activity counts to gravitational units (gs) to present our results in a device-independent manner, facilitating comparisons across the literature.

Sleep Detection:

Actigraphy is a reliable and valid method for detecting sleep, with previous reports demonstrating acceptable to high agreement with polysomnography (40-43). For this study, we employed a multi-step approach to detect sleep using published algorithms (40, 41). Sleep time was determined from the sensor worn on the participant’s preferred upper limb using the methodology described by Schoch and colleagues (41). Briefly, this approach applies the Sadeh sleep algorithm (40) and includes several additional adjustments to improve the algorithm’s accuracy (41). The first adjustment applies a threshold to distinguish between sleep and wake time using the mean vector magnitude for each day of data multiplied by the threshold. Here, we used a threshold of 0.35 because it was best suited to identify sleep in pilot testing. The second adjustment removes periods when the sensors were not worn, which was not applicable for this analysis as all participants wore the sensors for the full 48-hour wear period. The third adjustment utilized criteria by Webster et al (44) to address instances of incorrectly identifying periods of sleep while the participant was awake, via various smoothing routines. The final adjustment involved generating a sleep graph from the previous adjustments and displaying it to the research team member processing the data. The research team member made additional adjustments (if needed) based on any notes made by the participant on their accelerometry wearing log.

For each sensor variable of interest and wearing day, the sensor variable was computed using the full wear time (sleep included) and with sleep time removed (sleep excluded). The effect of sleep was quantified as the difference between sleep included and excluded (i.e., sleep included – sleep excluded). Custom-written R scripts (R Core Team 2021, version 4.2.1) were used for all variable computations and to implement the sleep detection algorithms described above (45, 46).

Statistical Analysis:

Linear mixed effects regression (LMER) was used to address our two study objectives and account for within-participant repeated accelerometry measurement across days (47). For each sensor variable, a LMER model was tested in which the difference between sleep included and excluded was the dependent variable. Each model included a random intercept for participant and fixed effects for day and cohort and their interaction:

At the time of this analysis, 148 individuals met the eligibility criteria to participate in the larger study. Of these 148, 16 individuals had missing data due to withdrawal from the study or loss to follow-up, and 5 individuals did not meet the 48-hour wear criteria for this analysis. This resulted in 127 individuals (65 without upper limb impairment and 62 with upper limb impairment) included in this analysis. Table 2 displays the demographic and clinical characteristics of our full sample as well as each cohort. Supplemental Table 1 displays the demographic and clinical characteristics of each upper limb condition group.

Table 2. Characteristics of Study Sample

Variable

Full Sample

(n = 127)

Controls

(n = 65)

UL Conditions

(n = 62)

Age (yrs)

53.5 ± 17.8

49.4 ± 20.1

57.8 ± 13.9

Sex

Female

Male

62.2% (79)

37.80% (48)

69.23% (45)

30.77% (20)

54.84% (34)

45.16% (28)

Race

White

Black

Asian

Native Hawaiian or Pacific Islander

71.65% (91)

21.26% (27)

5.51% (7)

0.79% (1)

80.0% (52)

10.77% (7)

9.23% (6)

0% (0)

62.9% (39)

32.26% (20)

1.61% (1)

Ethnicity

Hispanic, Latino

Non-Hispanic, Non-Latino

3.15% (4)

96.85% (123)

1.54% (1)

98.46% (64)

4.84% (3)

95.16% (59)

Hand Dominance

Right

Left

Both

90.55% (115)

8.66% (11)

0.79% (1)

90.77% (59)

9.23% (6)

0% (0)

90.32% (56)

8.06% (5)

1.61% (1)

Employment

Working ≥ 37.5 hrs/wk

Working ≥ 20 hrs/wk

Working < 20 hrs/wk

Not working

37.01% (47)

5.51% (7)

7.87% (10)

49.61% (63)

49.23% (32)

6.15% (4)

10.77% (7)

33.85% (22)

24.19% (15)

4.84% (3)

66.13% (41)

Living Situation

Living alone, assistance with BADLs

Living alone, independent with BADLs

Living with others, assistance with BADLs

Living with others, independent with BADLs

1.57% (2)

24.41% (31)

7.09% (9)

66.93% (85)

0% (0)

27.69% (18)

0% (0)

72.31% (47)

3.23% (2)

20.97% (13)

14.52% (9)

61.29% (38)

Affected Side

Right

Left

Both

48.39% (30)

45.16% (28)

6.45% (4)

Concordance

Yes

No

45.16% (28)

54.84% (34)

Abbreviations: UL- upper limb, BADLs- basic activities of daily living

Our first objective examined the effect of sleep on each sensor variable. We found a statistically significant difference between sleep included and excluded for three of the five sensor variables: preferred time, non-preferred time, and non-preferred magnitude standard deviation (Table 3, intercept values). To help contextualize these differences, the left column of Figure 1 displays overlapping density plots of sleep included (grey) and sleep excluded (blue) for each of the five sensor variables. The right column of Figure 1 displays the density plot for sleep included (grey), which is then overlain by the distribution of the differences between sleep included and excluded (white density plot, vertically scaled and centered around the mean of the variable with sleep included). For example, looking at preferred time (Figure 1, top left), one can see that the distribution for sleep excluded (blue) is slightly shifted to the left compared to sleep included (grey), suggesting that excluding sleep results in a reduction in preferred time hours compared to including sleep. Correspondingly, the top right panel of Figure 1 shows the distribution of differences is shifted down by 0.31 hours relative to the mean when sleep is included (solid line vs. direction and length of arrow), but there is uncertainty in that effect, with the standard deviation of the differences being 0.18. Thus, for any given point in the black histogram, we would expect that point to be shifted by 0.31 ± 0.18 hrs (or 19 ± 11 minutes) if sleep was to be included.

Table 3. Linear Mixed Effects Regression Model Parameters for each Sensor Variable.

Sensor Variable	Model Parameter	Estimate	95% Confidence Interval	T-Value	FDR Adj. P-Value
Preferred Time	Intercept	0.31	0.28 – 0.35	19.81	<0.001
	Day	-0.02	-0.07 – 0.02	-1.04	0.61
	Cohort	0.01	-0.05 – 0.07	0.18	0.94
	Day x Cohort	0.02	-0.07 – 0.11	0.49	0.78
Non-Preferred Time	Intercept	0.29	0.26 – 0.32	21.42	<0.001
	Day	-0.03	-0.07 – 0.01	-1.49	0.44
	Cohort	-0.03	-0.09 – 0.02	-1.24	0.58
	Day x Cohort	0.01	-0.08 – 0.11	0.22	0.92
Use Ratio	Intercept	0.004	0.00007 – 0.01	2.23	0.13
	Day	-0.001	-0.01 – 0.004	-0.39	0.82
	Cohort	0.0004	-0.01 – 0.01	0.12	0.94
	Day x Cohort	0.01	-0.001 – 0.02	1.59	0.40
Non-Preferred Magnitude	Intercept	0.00007	-0.0001 – 0.0002	0.82	0.71
	Day	0.00006	-0.0002 – 0.0003	0.49	0.78
	Cohort	-0.0005	-0.001 – -0.0002	-3.13	0.02
	Day x Cohort	.00002	-0.0005 – 0.001	0.10	0.94
Non-Preferred Magnitude SD	Intercept	0.0003	0.0002 – 0.001	4.17	0.001
	Day	0.0002	-0.00003 – 0.0003	1.65	0.38
	Cohort	-0.0001	-0.0004 – 0.0003	-0.64	0.74
	Day x Cohort	-0.0001	-0.0005 – 0.0002	-0.76	0.72

INSERT FIGURE 1

Figure 1. Differences between Sleep Included and Excluded for Each Sensor Variable. Left panel: Overlapping density plots displaying the distribution of each sensor variable with sleep included (grey) and sleep excluded (blue). The mean for each distribution is shown as a thin vertical line (solid- sleep included, dashed- sleep excluded). Right panel: Overlapping density plots displaying the distribution of each sensor variable with sleep included (grey) and the differences between sleep included and excluded (white) scaled and centered around the variable mean with sleep included. The thin vertical line displays the mean of the distribution with sleep included. The arrow depicts the direction the variable would shift by excluding sleep. The number (s = ) in each plot represents the standard deviation of the differences between sleep included and excluded for that variable, demonstrating uncertainty in the difference. Preferred indicates the dominant upper limb (Controls) or the unaffected upper limb (Upper Limb Impairment cohort).

Our second objective examined if the effect of sleep differed between individuals with and without upper limb impairment and between wearing days. Non-preferred magnitude was the only variable in which we observed a significant difference between people with and without upper limb impairment (Table 3). Compared to people without upper limb impairment (i.e., controls), people with upper limb impairment had a slightly smaller difference between sleep included and excluded for this variable. Despite being statistically significant, this difference between cohorts appears to be trivially small with values at the low end of the milli-scale, i.e., 95% CI = [-0.001, -0.0002], when individual differences in non-preferred magnitude exist at the centi- to deci-scale, see left panel of Figure 2. Figure 2 displays overlapping density plots of sleep included (grey) and the distribution of differences between sleep included and excluded (white density plot, scaled and centered around the variable mean with sleep included for interpretability) for each cohort. Overall, the effect of sleep in people with and without upper limb impairment was relatively similar, as evidenced by the Cohort estimates in Table 3 and in examining the distributions of Figure 2. The effect of day was not significant for any of the LMER models (Table 3), suggesting that the effect of sleep was not substantially different across wearing days. There was also no significant interaction between day and cohort for any of the upper limb performance variables, suggesting that the effects of sleep were relatively consistent between cohorts and wearing days (Figure 3).

INSERT FIGURE 2

Figure 2. Differences between Sleep Included and Excluded for Each Sensor Variable by Cohort. Overlapping density plots displaying the distribution of each sensor variable with sleep included (grey) and the differences between sleep included and excluded (white, scaled and centered around the variable mean with sleep included) for Control participants without UL impairment (left) and people with UL impairment (right). The thin vertical line displays the mean of the distribution with sleep included. The arrow depicts the direction the variable would shift by excluding sleep. The number (s = ) in each plot represents the standard deviation of the differences between sleep included and excluded for that variable, demonstrating uncertainty in the difference. Preferred indicates the dominant upper limb (Controls) or the unaffected upper limb (Upper Limb Impairment cohort). Abbreviations: UL- upper limb.

INSERT FIGURE 3

Figure 3. Differences between Sleep Included and Excluded for Each Sensor Variable by Cohort and by Day. Each coupled panel represents a sensor variable, with the leftward panel representing Control participants and the rightward panel representing participants with UL impairment. Each panel displays box plots of the differences between sleep included and excluded for Day 1 and Day 2. Preferred indicates the dominant upper limb (Controls) or the unaffected upper limb (Upper Limb Impairment cohort).

Supplemental Tables 2 – 21 and Supplemental Figures 1-20 display the LMER model results and visual representations of the findings for each supplemental variable, respectively.

The purpose of this study was to investigate the effect of sleep on the wearable sensor data processing pipeline to facilitate the use of wearable sensor technology in rehabilitation research and clinical care settings. Our first objective was to quantify the effect of sleep on sensor-derived variables from upper limb accelerometry. We observed a small, significant difference between sleep included and excluded for three of the five sensor variables: preferred time, non-preferred time, and non-preferred magnitude standard deviation. Our second objective was to determine if the effect of sleep differed between people with and without upper limb impairment and across a two-day wearing period. Here, we found that the effect of sleep differed slightly between people with and without upper limb impairment for one of the five variables, non-preferred magnitude, and that the effect of sleep was not different across a two-day wearing period. These findings have important implications for the use of wearable sensors in both research and clinical care settings.

Despite being statistically significant, the differences we observed were small. For example, excluding sleep reduces non-preferred time by 0.29 ± 0.15 (or 17.4 ± 9 minutes) relative to the mean when sleep is included. A difference of 17 minutes of non-preferred upper limb use is likely trivial in many contexts, particularly as wear time increases. The differences in non-preferred magnitude and its standard deviation were even smaller (Table 3, Figures 1 and 2). Collectively, these data suggest that the effect of sleep on sensor-derived variables from upper limb accelerometry is small and similar between people and without upper limb impairment and across a two-day wearing period and can likely be ignored in most cases.

From a research perspective, these findings may help minimize burden on both research staff and participants (2). For research staff, committing the time and resources to remove periods of sleep when processing accelerometry data from the upper limb(s) is likely not necessary in most cases. For research participants, removing a sensor opens the possibility that a participant will forget to put the sensor back on, creating opportunities for data to go missing (52). These results suggest that research protocols need not instruct participants to remove the sensors during sleep, which may improve participant adherence to wearing the devices (15). Some research protocols also instruct participants to record their activities while wearing the sensor(s) (53). If sleep is not a behavior of interest in a research study using accelerometry to measure upper limb movement, there is likely no need to burden participants with recording this activity. These reduced burdens will streamline the use of wearable sensor technology in research labs already accustomed to it, and also bring this technology within reach of labs unfamiliar to it, through a less computationally complex data processing pipeline.

From the clinical perspective, the success of deploying new technologies in clinical care settings hinges upon how well the technology is integrated into clinical workflows (54-58). This includes making the data directly available in the electronic health record (EHR) (1, 54, 59, 60). Busy clinicians do not have time to process raw data or generate summary variables. It is therefore imperative that the data processing pipeline be as simple and efficient as possible. Findings from this study suggest that the data processing pipeline can be simplified by ignoring periods of sleep when deploying wearable sensors to compute variables that reflect motor behavior of the upper limbs in people with and without upper limb impairment. This will expedite data processing, minimize burden on the clinician, and allow summary data to be in the hands of busy clinicians sooner. We also note that, in scenarios in which data are collected over multiple time points, the decision to remove periods of sleep or not should remain consistent, regardless of whether the data processing pipeline is built for research or clinical care initiatives.

There are several limitations to consider when interpreting the results of this study. The first is that wrist-worn actigraphy is not without measurement error (52, 61). Using wrist-worn actigraphy to detect sleep relies on the use of upper limb movements to determine whether periods of sleep have occurred (62). Any missed or incorrectly identified upper limb movements would have a downstream effect on a sleep algorithm’s ability to quantify sleep patterns. Actigraphy-based sleep detection algorithms themselves also suffer from accuracy challenges (63, 64). For example, if a person is awake but not moving their upper limbs, the algorithm is likely to incorrectly identify this period as sleep. For these reasons, actigraphy-based sleep detection algorithms tend to perform better at detecting sleep (i.e., higher sensitivity), rather than periods of wakefulness (i.e., lower specificity) (42, 64, 65). This means we may have over-estimated the amount of sleep in this study and potentially over-estimated the effect of sleep on the upper limb sensor variables. For this study, the Sadeh sleep algorithm was selected due to its high agreement with polysomnography and widespread use in the literature (40, 64). While other algorithms exist, they tend to perform similarly in terms of overall accuracy, sensitivity, and specificity (64). It is therefore plausible that using a different algorithm would have yielded a similar result; however, this was not formally tested in this study.

The effect of sleep on sensor-derived variables from upper limb accelerometry is small and similar between people with and without upper limb impairment and across a two-day wearing period. This small effect can likely be ignored in the upper limb wearable sensor data processing pipeline in many situations. To continue to support the uptake of wearable sensor technology in rehabilitation research and clinical care, future work should investigate the role of other factors, such as upper limb movement during walking, on the data processing pipeline. Efforts should also be directed towards understanding which sensor-derived variables from upper limb accelerometry are most important and should be carried forward in future research and clinical care initiatives.

REDCap - Research Electronic Data Capture, LMER - Linear mixed effects regression, UL- upper limb, BADLs- basic activities of daily living, EHR – electronic health record

Ethics Approval and Consent to Participate: The study was approved by Washington University’s Institutional Review Board, and all participants signed informed consent prior to engaging in any study procedures.

Consent for Publication: Not applicable.

Availability of Data and Materials: The datasets analyzed during the current study are available from the corresponding author upon request.

Competing Interests: The authors declare that they have no competing interests.

Funding: This work is funded by NIH R37 HD068290

Authors’ Contributions: AEM and CEL conceived the idea for this project. AEM and KRL analyzed the data. MDB organized, managed, and performed quality inspections of the data used for this analysis. AEM wrote the first draft of this manuscript. CEL, KRL, and MDB provided edits to the manuscript. All authors read and approved the final manuscript.

Acknowledgements: We are grateful to the individuals who participated in the study from which these data were generated.

Smuck M, Odonkor CA, Wilt JK, Schmidt N, Swiernik MA. The emerging clinical role of wearables: factors for successful implementation in healthcare. npj Digital Medicine. 2021;4(1):45.
Lang CE, Barth J, Holleran CL, Konrad JD, Bland MD. Implementation of Wearable Sensing Technology for Movement: Pushing Forward into the Routine Physical Rehabilitation Care Field. Sensors (Basel). 2020;20(20).
Lang CE, Holleran CL, Strube MJ, Ellis TD, Newman CA, Fahey M, et al. Improvement in the Capacity for Activity Versus Improvement in Performance of Activity in Daily Life During Outpatient Rehabilitation. Journal of Neurologic Physical Therapy. 2022:10.1097/NPT.0000000000000413.
Rand D, Eng JJ. Disparity between functional recovery and daily use of the upper and lower extremities during subacute stroke rehabilitation. Neurorehabil Neural Repair. 2012;26(1):76-84.
Danks KA, Roos MA, McCoy D, Reisman DS. A step activity monitoring program improves real world walking activity post stroke. Disability and rehabilitation. 2014;36(26):2233-6.
Moore JL, Roth EJ, Killian C, Hornby TG. Locomotor training improves daily stepping activity and gait efficiency in individuals poststroke who have reached a "plateau" in recovery. Stroke. 2010;41(1):129-35.
Hornby TG, Holleran CL, Hennessy PW, Leddy AL, Connolly M, Camardo J, et al. Variable Intensive Early Walking Poststroke (VIEWS): A Randomized Controlled Trial. Neurorehabil Neural Repair. 2016;30(5):440-50.
Waddell KJ, Strube MJ, Bailey RR, Klaesner JW, Birkenmeier RL, Dromerick AW, et al. Does Task-Specific Training Improve Upper Limb Performance in Daily Life Poststroke? Neurorehabil Neural Repair. 2017;31(3):290-300.
Fowler King B, MacDonald J, Stoff L, Nettnin E, Jayaraman A, Goldman JG, et al. Activity Monitoring in Parkinson Disease: A Qualitative Study of Implementation Determinants. J Neurol Phys Ther. 2023;47(4):189-99.
Braakhuis HEM, Bussmann JBJ, Ribbers GM, Berger MAM. Wearable Activity Monitoring in Day-to-Day Stroke Care: A Promising Tool but Not Widely Used. Sensors (Basel). 2021;21(12).
Lobelo F, Kelli HM, Tejedor SC, Pratt M, McConnell MV, Martin SS, et al. The Wild Wild West: A Framework to Integrate mHealth Software Applications and Wearables to Support Physical Activity Assessment, Counseling and Interventions for Cardiovascular Disease Risk Reduction. Prog Cardiovasc Dis. 2016;58(6):584-94.
Dinh-Le C, Chuang R, Chokshi S, Mann D. Wearable Health Technology and Electronic Health Record Integration: Scoping Review and Future Directions. JMIR Mhealth Uhealth. 2019;7(9):e12861.
Torriani-Pasin C, Demers M, Polese JC, Bishop L, Wade E, Hempel S, et al. mHealth technologies used to capture walking and arm use behavior in adult stroke survivors: a scoping review beyond measurement properties. Disability and rehabilitation. 2022;44(20):6094-106.
Gabelia D, Mitterling T, Högl B, Wenning GK, Frauscher B. Do periodic arm movements during sleep exist in healthy subjects? A polysomnographic study. Sleep Medicine. 2014;15(9):1150-4.
Wright H, Wright T, Pohlig RT, Kasner SE, Raser-Schramm J, Reisman D. Protocol for promoting recovery optimization of walking activity in stroke (PROWALKS): a randomized controlled trial. BMC Neurol. 2018;18(1):39.
Lawrence ES, Coshall C, Dundas R, Stewart J, Rudd AG, Howard R, et al. Estimates of the Prevalence of Acute Stroke Impairments and Disability in a Multiethnic Population. Stroke. 2001;32(6):1279-84.
Virani SS, Alonso A, Benjamin EJ, Bittencourt MS, Callaway CW, Carson AP, et al. Heart Disease and Stroke Statistics-2020 Update: A Report From the American Heart Association. Circulation. 2020;141(9):e139-e596.
Cattaneo D, Lamers I, Bertoni R, Feys P, Jonsdottir J. Participation Restriction in People With Multiple Sclerosis: Prevalence and Correlations With Cognitive, Walking, Balance, and Upper Limb Impairments. Archives of physical medicine and rehabilitation. 2017;98(7):1308-15.
Lamers I, Feys P. Patient reported outcome measures of upper limb function in multiple sclerosis: A critical overview. Mult Scler. 2018;24(14):1792-4.
Hayes SC, Johansson K, Stout NL, Prosnitz R, Armer JM, Gabram S, et al. Upper-body morbidity after breast cancer: incidence and evidence for evaluation, prevention, and management within a prospective surveillance model of care. Cancer. 2012;118(8 Suppl):2237-49.
Pinto M, Gimigliano F, Tatangelo F, Megna M, Izzo F, Gimigliano R, et al. Upper limb function and quality of life in breast cancer related lymphedema: a cross-sectional study. European journal of physical and rehabilitation medicine. 2013;49(5):665-73.
Nelson GN, Stepan JG, Osei DA, Calfee RP. The Impact of Patient Activity Level on Wrist Disability After Distal Radius Malunion in Older Adults. Journal of Orthopaedic Trauma. 2015;29(4):195-200.
McKee RC, Whelan DB, Schemitsch EH, McKee MD. Operative versus nonoperative care of displaced midshaft clavicular fractures: a meta-analysis of randomized clinical trials. J Bone Joint Surg Am. 2012;94(8):675-84.
Patel R, Urits I, Wolf J, Murthy A, Cornett EM, Jones MR, et al. A Comprehensive Update of Adhesive Capsulitis and Minimally Invasive Treatment Options. Psychopharmacol Bull. 2020;50(4 Suppl 1):91-107.
Bailey RR, Klaesner JW, Lang CE. Quantifying Real-World Upper-Limb Activity in Nondisabled Adults and Adults With Chronic Stroke. Neurorehabil Neural Repair. 2015;29(10):969-78.
Schwerz de Lucena D, Rowe JB, Okita S, Chan V, Cramer SC, Reinkensmeyer DJ. Providing Real-Time Wearable Feedback to Increase Hand Use after Stroke: A Randomized, Controlled Trial. Sensors. 2022;22(18).
Simpson LA, Menon C, Hodgson AJ, Ben Mortenson W, Eng JJ. Clinicians' perceptions of a potential wearable device for capturing upper limb activity post-stroke: a qualitative focus group study. J Neuroeng Rehabil. 2021;18(1):135.
Porciuncula F, Roto AV, Kumar D, Davis I, Roy S, Walsh CJ, et al. Wearable Movement Sensors for Rehabilitation: A Focused Review of Technological and Clinical Advances. Pm r. 2018;10(9 Suppl 2):S220-s32.
Harris PA, Taylor R, Thielke R, Payne J, Gonzalez N, Conde JG. Research electronic data capture (REDCap)--a metadata-driven methodology and workflow process for providing translational research informatics support. J Biomed Inform. 2009;42(2):377-81.
Harris PA, Taylor R, Minor BL, Elliott V, Fernandez M, O'Neal L, et al. The REDCap consortium: Building an international community of software platform partners. J Biomed Inform. 2019;95:103208.
Bailey RR, Klaesner JW, Lang CE. An accelerometry-based methodology for assessment of real-world bilateral upper extremity activity. PloS one. 2014;9(7):e103135.
Lang CE, Waddell KJ, Klaesner JW, Bland MD. A Method for Quantifying Upper Limb Performance in Daily Life Using Accelerometers. J Vis Exp. 2017(122).
Barth J, Lohse KR, Konrad JD, Bland MD, Lang CE. Sensor-Based Categorization of Upper Limb Performance in Daily Life of Persons With and Without Neurological Upper Limb Deficits. Frontiers in Rehabilitation Sciences. 2021;2(59).
Smith BA, Lang CE. Sensor Measures of Symmetry Quantify Upper Limb Movement in the Natural Environment Across the Lifespan. Archives of physical medicine and rehabilitation. 2019;100(6):1176-83.
Konrad J, Marrus N, Lang CE. A Feasibility Study of Bilateral Wrist Sensors for Measuring Motor Traits in Children With Autism. Percept Mot Skills. 2022;129(6):1709-35.
Hoyt CR, Van AN, Ortega M, Koller JM, Everett EA, Nguyen AL, et al. Detection of Pediatric Upper Extremity Motor Activity and Deficits With Accelerometry. JAMA Netw Open. 2019;2(4):e192970.
Porta M, Porceddu S, Leban B, Casu G, Mura GM, Campagna M, et al. Characterization of upper limb use in health care workers during regular shifts: A quantitative approach based on wrist-worn accelerometers. Appl Ergon. 2023;112:104046.
Uswatte G, Miltner WH, Foo B, Varma M, Moran S, Taub E. Objective measurement of functional upper-extremity movement using accelerometer recordings transformed with a threshold filter. Stroke. 2000;31(3):662-7.
Uswatte G, Foo WL, Olmstead H, Lopez K, Holand A, Simms LB. Ambulatory monitoring of arm movement using accelerometry: an objective measure of upper-extremity rehabilitation in persons with chronic stroke. Archives of physical medicine and rehabilitation. 2005;86(7):1498-501.
Sadeh A, Sharkey KM, Carskadon MA. Activity-based sleep-wake identification: an empirical test of methodological issues. Sleep. 1994;17(3):201-7.
Schoch SF, Jenni OG, Kohler M, Kurth S. Actimetry in infant sleep research: an approach to facilitate comparability. Sleep. 2019;42(7).
Quante M, Kaplan ER, Cailler M, Rueschman M, Wang R, Weng J, et al. Actigraphy-based sleep estimation in adolescents and adults: a comparison with polysomnography using two scoring algorithms. Nat Sci Sleep. 2018;10:13-20.
Smith MT, McCrae CS, Cheung J, Martin JL, Harrod CG, Heald JL, et al. Use of Actigraphy for the Evaluation of Sleep Disorders and Circadian Rhythm Sleep-Wake Disorders: An American Academy of Sleep Medicine Systematic Review, Meta-Analysis, and GRADE Assessment. J Clin Sleep Med. 2018;14(7):1209-30.
Webster JB, Kripke DF, Messin S, Mullaney DJ, Wyborney G. An activity-based sleep monitor system for ambulatory use. Sleep. 1982;5(4):389-99.
Wickham H AM, Bryan J, Chang W, McGowan LD, François R, Grolemund G, Hayes A, Henry L, Hester J, Kuhn M, Pedersen TL, Miller E, Bache SM, Müller K, Ooms J, Robinson D, Seidel DP, Spinu V, Takahashi K, Vaughan D, Wilke C, Woo K, Yutani H. Welcome to the tidyverse. Journal of Open Source Software. 2019;4(43):1686.
cowplot: Streamlined Plot Theme and Plot Annotations for 'ggplot2'. In: Wilke CO, editor. 2020. p. R package version 1..
Long JD. Longitudinal Data Analysis for the Behavioral Sciences Using R: SAGE Publications; 2012.
Benjamini Y, Hochberg Y. Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society Series B (Methodological). 1995;57(1):289-300.
Bates D, Maechler M, Bolker B, Walker S. Fitting Linear Mixed-Effects Models Using lme4. Journal of Statistical Software. 2015;67(1):1-48.
AICcmodavg: Model selection and multimodel inference based on (Q)AIC(c). In: Mazerolle MJ, editor. 2020. p. R package version 2.3-1.
Kuznetsova A, Brockhoff PB, Christensen RHB. {lmerTest} Package: Tests in Linear Mixed Effects Models. Journal of Statistical Software. 2017;82(13):1-26.
Cho S, Ensari I, Weng C, Kahn MG, Natarajan K. Factors Affecting the Quality of Person-Generated Wearable Device Data and Associated Challenges: Rapid Systematic Review. JMIR Mhealth Uhealth. 2021;9(3):e20738.
Zemlin C, Stuhlert C, Schleicher JT, Wörmann C, Altmayer L, Lang M, et al. Longitudinal Assessment of Physical Activity, Fitness, Body Composition, Immunological Biomarkers, and Psychological Parameters During the First Year After Diagnosis in Women With Non-Metastatic Breast Cancer: The BEGYN Study Protocol. Front Oncol. 2021;11:762709.
Lavallee DC, Lee JR, Austin E, Bloch R, Lawrence SO, McCall D, et al. mHealth and patient generated health data: stakeholder perspectives on opportunities and barriers for transforming healthcare. mHealth. 2019;6.
Langan J, Subryan H, Nwogu I, Cavuoto L. Reported use of technology in stroke rehabilitation by physical and occupational therapists. Disabil Rehabil Assist Technol. 2018;13(7):641-7.
Liu L, Miguel Cruz A, Rios Rincon A, Buttar V, Ranson Q, Goertzen D. What factors determine therapists’ acceptance of new technologies for rehabilitation – a study using the Unified Theory of Acceptance and Use of Technology (UTAUT). Disability and rehabilitation. 2015;37(5):447-55.
Morris J, Thompson N, Wallace T, Jones M, DeRuyter F, editors. Survey of Rehabilitation Clinicians in the United States: Barriers and Critical Use-Cases for mRehab Adoption. Computers Helping People with Special Needs; 2020 2020//; Cham: Springer International Publishing.
Jim HSL, Hoogland AI, Brownstein NC, Barata A, Dicker AP, Knoop H, et al. Innovations in research and clinical care using patient-generated health data. CA Cancer J Clin. 2020;70(3):182-99.
Tiase VL, Hull W, McFarland MM, Sward KA, Del Fiol G, Staes C, et al. Patient-generated health data and electronic health record integration: a scoping review. JAMIA Open. 2020;3(4):619-27.
Abdolkhani R, Gray K, Borda A, DeSouza R. Patient-generated health data management and quality challenges in remote patient monitoring. JAMIA Open. 2019;2(4):471-8.
Cederberg KLJ, Jeng B, Sasaki JE, Lai B, Bamman M, Motl RW. Accuracy and precision of wrist-worn actigraphy for measuring steps taken during over-ground and treadmill walking in adults with Parkinson's disease. Parkinsonism & Related Disorders. 2021;88:102-7.
Sadeh A. Iii. Sleep assessment methods. Monogr Soc Res Child Dev. 2015;80(1):33-48.
Blackwell T, Ancoli-Israel S, Redline S, Stone KL. Factors that May Influence the Classification of Sleep-Wake by Wrist Actigraphy: The MrOS Sleep Study. Journal of Clinical Sleep Medicine. 2011;07(04):357-67.
Patterson MR, Nunes AAS, Gerstel D, Pilkar R, Guthrie T, Neishabouri A, et al. 40 years of actigraphy in sleep medicine and current state of the art algorithms. npj Digital Medicine. 2023;6(1):51.
Marino M, Li Y, Rueschman MN, Winkelman JW, Ellenbogen JM, Solet JM, et al. Measuring sleep: accuracy, sensitivity, and specificity of wrist actigraphy compared to polysomnography. Sleep. 2013;36(11):1747-55.

No competing interests reported.

Supplement20231212.docx
Description: Supplemental Table 1 provides demographic and clinical characteristics by diagnosis. Supplemental Figures 1 – 20 and Supplemental Tables 2 – 21 visualize and present the results of the linear mixed effects regression models for the supplemental upper limb sensor variables, respectively.

Quantifying the Effects of Sleep on Sensor-Derived Variables from Upper Limb Accelerometry in People with and without Upper Limb Impairment

Status:

Version 1

Abstract

Figures

INTRODUCTION

METHODS

RESULTS

DISCUSSION

CONCLUSIONS

Abbreviations

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1