Application of a Novel Actigraphy Algorithm to Detect Movement and Sleep/Wake Patterns in Children with Autism Spectrum Disorder

Abstract

Objective

Actigraphy is a non-intrusive method of recording rest/activity cycles and a surrogate for sleep/wake activity. Standard actigraphy analysis is limited in ascribing discrete movement events to wake status during sleep. We applied a novel algorithm to overnight actigraphy data recorded simultaneously with video polysomnography-electroencephalography (video PSG-EEG) to determine its ability to define movement and sleep/wake patterns in children with autism spectrum disorder (ASD) and age-comparable typically developing (TD) controls.

Methods

A previously published novel algorithm uses mathematical endpoints to analyze actigraphy data without assumptions about sleep/wake status, and smooths data using moving windows of increasing length. Nighttime activity level “S” events (S1-S5) determined by this algorithm (n = 273) were identified in 15 children ages 3-10 years (nine with ASD and six TD) who wore an AW2 Spectrum Actiwatch (Philips Respironics) while undergoing simultaneous video PSG-EEG. Data were analyzed to identify the time each activity level “S” event occurred, video movement events (movements captured by video and scored based on level of severity), and sleep/wake status defined by PSG-EEG. The relationships among activity level “S” events, video movement events, and sleep/wake status were analyzed statistically.

Results

Activity level “S” events and the presence and severity of video movement events, and sleep-wake status, were significantly associated. These associations were present in both participants with ASD and those who were typically developing.

Conclusion

This actigraphy algorithm shows promise for detecting nighttime movements and sleep/wake status and warrants further study in larger datasets of neurotypical children and those with neurodevelopmental disorders.

1. Introduction

Sleep disturbance is common in children with neurodevelopmental disorders (Wiggs & Stores, 2001). Sleep disturbance is as high as 50-80% in children with autism spectrum disorder (ASD) compared to 9-50% in children of typical development, with difficulty falling asleep or night wakings the most common concerns (Couturier et al., 2005; Krakowiak, Goodlin-Jones, Hertz-Picciotto, Croen, & Hansen, 2008; Richdale & Schreck, 2009; Souders, 2009).

While subjective measures of sleep, such as parent report, are paramount to obtain in children with ASD, it is also essential to have objective measures that document improvement in sleep patterns in this population. Reasons include (a) the high prevalence of sleep problems in this population; (b) limited ability of the child to verbally report sleep onset delay or night wakings to their parents; and (c) children with ASD may not otherwise signal to their parents that they are awake at night.

Polysomnography (PSG) has long been considered the “gold standard” for sleep measurement due to its ability to objectively measure not only wake and sleep time, but also sleep architecture. There are many limitations to PSG studies, particularly in children with ASD who have poor tolerance of PSG due to sensory sensitivities, anxiety or both. Cost and burden on families in coming to a sleep facility are also concerns. Accordingly, actigraphy has been used in descriptive (Goldman et al., 2017; Baker et al., 2019) and interventional studies (Cortesi et al., 2012; Malow et al., 2014) in ASD to provide an objective measure of sleep patterns that is non-intrusive and minimally sensory aversive.

Actigraphy devices contain an accelerometer, which quantifies motion that has been used to differentiate sleep from wakefulness in various clinical conditions (Meltzer, 2008). Sleep and wake patterns can be cost-effectively measured over days, weeks, and months of accelerometer measurements (Grutsch et al., 2011).

Actigraphy scoring algorithms, however, also have limitations (Patel et al., 2015). Actigraphy scoring algorithms vary across devices and often underestimate sleep onset latency and overestimate sleep duration in the typically developing population (Ustinov & Lichstein, 2013). Another limitation documented by Lauderdale et al. (2006) is that actigraphy defined sleep duration may be significantly less than self-reported sleep duration. This limitation may reflect variability in actigraphy sleep algorithms. To overcome this variability, sleep scoring programs have default manual and automatic activity scoring settings – however, these default settings may not be “one size fits all” when applied to different populations with variable movement characteristics, and can significantly impact final sleep actigraphy statistical analyses. For example, actigraphy shows less variability for normal sleepers than in those in whom the endogenous sleep architecture is altered (Kushida et al., 2001).

A further limitation to actigraphy analyses relates to the identification of night wakings through analysis of movement. Actigraphy algorithms detecting wake time after sleep onset based on movement may not differentiate between relatively high intensity brief movements and relatively low intensity longer duration movements, even though these movements reflect very different clinical conditions. For example, a child may have restless sleep with 30 movements lasting an average of 20 seconds each. These movements would result in a value for wake time after sleep onset that is similar to that of the same child having one prolonged movement lasting ten minutes, although these two scenarios have different implications. In the scenario with the prolonged movement, the child is more likely to wake up fully, and possibly call out for his/her parents. In the scenario with the briefer but more numerous movements, the child is more likely to remain asleep, and may actually have minimal “wake time after sleep onset.” Reporting mean actigraphy values does not differentiate these events. Small, physiologically insignificant movements can also interfere with the analysis of actigraphy data, so the relevant movements need to be distinguished from the irrelevant movements.

The Clinical Trials Transformation Initiative (CTTI, 2019), established by the US Food and Drug Administration and Duke University in 2007, conducts projects to better understand the range of current practices, assess alternative approaches, understand barriers to change, and propose recommendations for improvement. There are multiple approaches to data analysis encompassed by this initiative (Moreau et al., 2018) The CTTI (2019) novel endpoint project provides initiatives to create innovative measures that can make clinical trial data analyses and measurements more specific and timely.

In the context of CTTI (2019) and given the common sleep disturbances identified in children with ASD, their limitations in reporting symptoms, and the shortcomings of PSG in this population, the objective of our study was to explore novel actigraphy endpoints that characterize movement events during sleep. We applied a novel algorithm to overnight actigraphy data recorded simultaneously with video polysomnography-electroencephalography (video PSG-EEG). Endpoints generated by the algorithm were compared to scoring of visual movements captured by simultaneous video, and sleep/wake status captured by PSG-EEG. This previously published algorithm by Peterson et al. (2016) uses descriptive statistical endpoints to analyze actigraphy data without any assumptions about sleep/wake status. This algorithm minimizes the time needed to analyze large amounts of actigraphy data and produces descriptive endpoints to further define the sleep patterns in terms of movement. The algorithm also utilizes a smoothing process to help separate insignificant from significant movements. The resulting endpoints provide an easily interpretable measure of activity that may be used to measure the effectiveness of an intervention.

2. Methods

2.1. Participants

This study received approval from the Institutional Review Board (IRB) at Vanderbilt University Medical Center. Participants included children with autism spectrum disorder (ASD), and typical development (TD). Study participants with ASD were recruited through the Vanderbilt site for the Autism Treatment Network (ATN), a network of academic medical centers across North America that uses standardized assessments to enhance medical care for children with ASD. A clinical psychologist with expertise in ASD administered the Autism Diagnostic Observation Schedule (ADOS) (Lord et al., 2000) to confirm participants’ diagnoses. Participants who were typically developing were recruited via study flyers and email announcements sent out to the Vanderbilt community. Children with ASD and those who were typically developing were 3-10 years of age. The children were able to tolerate wearing the actigraphy device for the period of data collected. Children with ASD who had a history of epilepsy or untreated sleep apnea were excluded. Typically developing children who had a medical condition requiring daily medication other than seasonal allergies were also excluded from our study.

For ASD, we identified 31 participants meeting study criteria who agreed to be approached about the study and consented 21 participants into the study, 14 of whom were sufficiently adherent to the study procedures. For the typically developing participants, we selected those who were the same sex and within one year of age of ASD participants.

2.2. Overnight Laboratory PSG Study

Each participant completed one night of video-EEG polysomnography study data collection in the Sleep Research Core of the Clinical Research Center at Vanderbilt University Medical Center while also wearing an AW Spectrum Actiwatch® (Philips Respironics, Bend, OR). Both the parent and child were educated on PSG procedures, the parent gave informed consent and the child assented, where applicable. The PSG was performed by a registered sleep technologist and included 21 channels of EEG, along with chin electromyography and electrooculography to stage sleep. Nasal pressure transducer and thermistor, respiratory effort, electrocardiogram, anterior tibialis EMG, and pulse oximetry were also monitored.

2.3. Video Events

To define a scoring system for the intensity of movements during the night, the video from a previously recorded dataset of children with ASD were reviewed by two of the co-authors (MLA and BAM). Video events were classified into one of three visual movement event categories, mild, moderate, or severe. This scoring system is presented in Table 1 below, with examples of movements in each category that commonly occurred during the video recordings.

Table 1.

Nighttime visual movement event severity descriptions: epoch by epoch, all participants.

Movement Event Severity	Movement Event Description
1 – Mild	• Minimal movement or barely visible movements • Moving arms, legs, and wrist • Wiggling, itching, and stretching • Rotating positions (supine, prone, right, and left) • Mother tucking child into bed
2 – Moderate	• Moving arms in a constant motion around body • Scratching face and rubbing eyes • Moving watch around • Kicking and rolling around • Gets up on elbows • Rolling back and forth • Mother rocking child to sleep
3 – Severe	• Child is awake and looking around • Child is messing with the covers • Crying, screaming, kicking, and talking • Child sits up and lays back down • Child gets out of bed • Child playing in bed

For each participant, one reviewer (MLA), who was blinded to the child’s diagnosis and results of the actigraphy analysis, reviewed the video in its entirety and documented the clinical characteristics of each movement event using the previously developed scoring system. MLA classified each visual movement event as mild (level 1), moderate (level 2), or severe (level 3) depending on the intensity of the movement (Table 1). Another co-author (BAM) independently reviewed all events in a selection of 20% of the records in this study using the scoring system. BAM agreed with the intensity classification of the visual movement event (mild, moderate, or severe) by MLA 89% of the time. The classification of the first reviewer (MLA) was used for the statistical analysis.

Each video event was scored as occurring during wake or sleep by reference to PSG using standard criteria, defined by the American Academy of Sleep Medicine (Berry et al., 2017).

2.4. Actigraphy Procedures

The participant wore the actigraphy watch on the non-dominant wrist during the overnight PSG. The actigraphy device was programmed by the Philips Respironics Actiware software for an active mode, a medium threshold, and for 1-minute epochs. Parents of participants completed a sleep diary corresponding with when their children wore the actigraphy watch and pressed the event marker on the watch at bedtime and waketime. The event marker was pressed to indicate sleep and wake times, enhancing sleep diary and parent report accuracy and validity. The sleep diary consisted of sleep and wake times, along with comments related to the child’s movement and behavior during the night and around bedtime. Research staff with expertise in working with actigraphy data downloaded and reviewed all data.

2.5. Actigraphy Endpoints

Actigraphy counts were downloaded in minute-by-minute epochs (each epoch was one minute in duration). The endpoints of this algorithm by Peterson et al. (2016) included activity level “S” values (S1-S5) corresponding to the number of times the smoothed actigraphy data crossed a specific threshold. The algorithm that identifies the onset/end of rest periods includes multiple smoothing thresholds of the actigraphy data with moving Gaussian windows of 20-100 minutes, and identification of an optimal threshold value. Lowering the threshold value increases the sensitivity of the activity level “S” values to detect movement, while raising the threshold value decreases the sensitivity of the activity level “S” events used to detect movement. It was anticipated that the activity level “S1” events would correspond to mild disturbance, activity level “S2-S4” would correspond to moderate disturbance, and activity level “S5” would correspond to severe disturbance. These activity level “S” events provide a more discrete characterization and an easily interpretable measure of activity that minimizes movement artifacts. They provide descriptive information that can be used to describe a sample and to measure the effectiveness of an intervention.

The raw data endpoints were smoothed using time windows of 5, 25, 45, 65, and 85 minutes corresponding to S1-S5 respectively. The threshold that the peak of each smoothed curve had to exceed in order to be counted as an “Sx” event was set at 25 counts per minute (CPM). We optimized these window and threshold parameters based on actigraphy data in a previously collected dataset. In addition to activity level “S” events we examined maximum activity level, sum of activity level, and duration of movement.

2.6. Alignment Process for Video and Activity Level “S” Events

As our algorithm is based on moving Gaussian windows, activity level “S” events did not consistently occur simultaneously with video events within the one-minute epoch used for comparing video events and activity level “S” events (“S1”-“S5”). The “S” events usually preceded video events due to the nature of the smoothing algorithm. Therefore, we created two analytic datasets based on the width of the time interval (window) for combining neighboring activity level “S” and video events. These time intervals consisted of one and two-minute alignment windows. We allowed the video event to immediately precede the activity level “S” event. For example, in the one-minute window, an activity level “S” event occurring at 22:00 and a video event occurring between 21:59 and 22:00 would be analyzed as occurring in the same window (e.g., correlated). Similarly, for the two-minute window, an activity level “S” event occurring at 22:00 and a video event occurring between 21:58 and 22:00 would be analyzed as occurring in the same window. See Table 2 for additional clarification of the alignment process.

Table 2.

Alignment Process for Visual Movement Events and Activity Level “S” Events

(A) Alignment – One-Minute Window
Minute	Visual Movement	Activity Level “S”
1		B
2	A
3	A
(B) Alignment – Two-Minute Window
Minute	Visual Movement	Activity Level “S”
1		B
2
3	A
4	A

(A) Example of Alignment Process in Analytic Dataset 1 – One-Minute Window. No Activity Level “S” event is captured in Minutes 2 and 3 but a Visual Movement (Video) Event is recorded. If the previous time window (Minute 1) has a missing value for duration but has an Activity Level “S” event recorded, this Activity Level “S” event will be aligned to Minutes 2 and 3. In this scenario, the Activity Level “S” event for Minutes 2 and 3 is A.

(B) Example of Alignment Process in Analytical Dataset 2 – Two-Minute Window. No Activity Level “S” event is captured in Minutes 3 and 4 but a Visual Movement event is recorded. If the previous time window (Minutes 1 and 2) has a missing value for the Visual Movement Event but has an Activity Level “S” event recorded, this Activity Level “S” event will be aligned to Minutes 3 and 4. If both Minute 1 and Minute 2 captured an Activity Level “S” event, the closer one (Minute 2) will be used. In this scenario the Activity Level “S” event for minute 3 and 4 is B.

Alignment was also performed for activity level “S” events in relation to the other activity parameters – maximum activity, sum of activity, and duration of activity counts. In other words, each of these variables falling within a one- minute or two-minute window would be analyzed as occurring within the same window as the video event.

2.7. Statistical Analysis

Actigraphy-derived endpoints were analyzed in relation to video events and sleep/wake status (based on PSG-EEG). Agreement between activity level “S” events and sleep/wake status, as well as agreement between video events and sleep/wake status, were tested using Pearson’s Chi-square test. Because each patient had a large number of repeated measures, mixed-effect models were used to account for the correlation structure in the data. Generalized linear mixed-effects models with logit link were used to fit binary video events and sleep/wake status. Linear mixed-effects models were used to fit continuous outcome variables including maximum activity level, sum of activity level, and duration of movement. The activity levels were log-transformed to make the data less skewed. In addition to the main fixed effect of activity level “S” events or video events, we also included a nonlinear term of the time from sleep onset as a fixed effect by fitting a restricted cubic spline function. Individual participant identification was fitted as a random effect. ASD and typically developing groups were combined, except when comparing differences across groups in Section 3.5 of the results section. All analyses and model developments were carried out in R 3.5.1, separately for the one-minute window dataset and the two-minute window dataset. Significance was set at a two-sided 0.05 alpha level.

3. Results

3.1. Participants

Of the 14 participants with ASD, 9 had scorable video, PSG, and actigraphy signals. Six TD children were appropriate sex and age matches for these 9 participants with ASD. The age of the participants ranged from 3-10 years, four females and 11 males, with 14 white participants and one Asian participant. The average age (standard deviation: SD) for ASD was 6.6 (2.6) years and for TD was 7.8 (1.3) years; the average age of all participants was 7.1 (2.2). Psychotropic medications were taken by six participants with ASD and included melatonin, clonidine, diphenhydramine, quetiapine, fluoxetine, aripiprazole and methylphenidate. None of the TD participants took psychotropic medications. No participants had seizures, sleep apnea, or parasomnias recorded on video-EEG-PSG.

3.2. Activity Levels

The primary analysis results reported below will focus on the results from the one-minute window dataset. We chose to report this dataset given that the raw actigraphy data was derived in minute-by-minute epochs. The two-minute window dataset showed very similar findings to the one-minute window dataset in the statistical analysis.

Of the activity level “S” events, none occurred during sleep, 46% occurred during arousal, and 54% occurred during wake, with wake/arousal/sleep status based on PSG-EEG.

For the video events, 66% were scored as Level 1 (mild), 22% as Level 2 (moderate), and 12% as Level 3 (severe). There was a significant agreement between binary activity level “S” events (yes = S1-S5; no = no S event) and visual movement event severity (p = .0016).

Table 3 shows the distribution of activity level “S” events across all of the recorded epochs in all participants. Most of the “S” events were classified as either S1 or S2 (mild). There was a significant association between the level of “S” event (S1-S5) with visual movement events (p <0.001) and PSG status (p < 0.001). The higher the level of “S” event, the more severe was the visual movement event and the likelihood of awake PSG status.

Table 3.

Distribution of Activity Level “S” Events in Relation to Visual Movement Event Severity and sleep/arousal/wake status (as defined by PSG). A higher “S” level was associated with more severe visual movement events. and with wake status.

	S1	S2	S3	S4	S5	Pearson	Df	p-value
N (percent) Visual Movement Events						Chi-Square
Mild	129 (85)	37 (58)	6 (24)	6 (29)	2 (18)	150	12	<0.001
Moderate	15 (10)	17 (27)	18 (72)	7 (33)	4 (36)
Severe	8 (5)	10 (16)	1 (4)	8 (38)	5 (45)
Total Epochs	152	64	25	21	11
PSG Status	S1	S2	S3	S4	S5	Pearson Chi-Square	Df	p-value
Sleep	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)	37	12	<0.001
Arousal	77 (51)	24 (38)	13 (52)	7 (33)	4 (36)
Wake	75 (49)	40 (62)	12 (48)	14 (67)	7 (64)
Total Epochs	152	64	25	21	11

On 147 occasions, a video event occurred without an “S” being coded. These were mainly due to minor physical movement events with a short duration (due to the moving Gaussian window of 5 minutes, events less than 5 minutes were not detected by the algorithm); N = 86 (59%). The other less common reasons were the physical movement event was contained within a movement event that was coded as an “S” event; N = 53 (36%) or the physical movement events occurred during sleep onset period or morning awakenings N = 8 (5%). These video events were usually associated with arousals from sleep and had a duration of fewer than 5 minutes.

Figure 1 shows illustrative examples of an ASD participant and a TD participant. For each participant, the activity levels in counts per minute (y-axis) are displaced in relation to visual movement events, sleep/wake status, and activity level “S” events across the night. Additional ASD and TD participant examples are included in the Supplemental Figure.

Figure 1.

Examples of ASD (participant 9; first graph) and TD (participant 15; second graph) participants are presented. The graph shows activity (counts/minute) derived from actigraphy in relation to time (minutes after lights out). Below the activity graph and the X axis are three rows of indicators, which coincide with activity levels. The top row depicts visual movement events 1-3 (1 is mild, 2 is moderate, and 3 is severe), the middle row depicts wake status based on polysomnography (PSG) with 1 = arousal and 2 = wake, and the bottom row depicts the level of “S” event (1-5, with 5 being the most severe activity level disturbance).

3.3. Agreement of Activity Level “S” Events and Other Endpoints

The linear mixed-effect analyses of maximum activity level “S” events and sum of activity level revealed that both activity level “S” events (both P<0. 001 for maximum activity and for sum of activity) and video events (P=0. 042 for maximum activity; P=0. 002 for sum of activity) were highly correlated after adjusting for time from sleep onset. Time from sleep onset was significantly associated with sum of activity when using activity level “S” events as a metric (P=0.03). The fitted restricted cubic splines of the time variable indicate on average, sum of activity first decreases and then increases as the night progresses. Both activity level “S” events and video events were also significantly associated with duration of movement (both P<0. 001). Results from fitting generalized linear mixed-effect models indicate that both sum of activity (P<0. 001) and maximum activity (P=0. 002) were significant metrics of binary PSG event (sleep/arousal vs. wake), adjusted for time from sleep onset. In addition, duration and severity of movement were also significantly associated with night wakings based on PSG-EEG (both P<0. 001).

3.4. Association of Activity Levels and Duration of Movement with Night Wakings

Generalized linear mixed-effects models were fitted to the binary night waking outcome based on PSG-EEG (wake vs. sleep status). Sum of activity, maximum activity, and duration of movement were significantly associated with night wakings (all p<0.001). For instance, each minute increase in duration is associated with 2.18 times (95% CI: 1.71, 2.88) higher a chance of having a night waking.

The severity of video events was associated with a higher sum of activity (p<0. 001), higher maximum activity (p<0. 001), and longer duration of movement (p<0. 001). A one unit increase in the log-scale sum of activity was associated with a 7.62 times (95% CI: 5. 35, 11. 3) higher a chance in having a PSG wake event. A one unit increase in the log-scale maximum activity was associated with an 8.17 (95% CI: 5.62, 12.4) times higher a chance in having a PSG wake event. A one-minute increase in the duration of movement was associated with a two times higher likelihood of a PSG wake event (OR=1.96; 95% CI: 1.67, 2.34).

3.5. Relationship of Autism to our Findings

Among all 15 participants, nine were diagnosed with ASD and six were typically developing. Using linear mixed-effect models similar to those described above, ASD status was not found to be significantly associated with maximum activity (p = 0.201), sum of activity (p = 0.441), or night wakings (p = 0.16).

4. Discussion

In this work, we apply a previously published algorithm for actigraphy-derived data (Peterson et al., 2016) to define nighttime movement events, including severity and relation to sleep/wake status, in children with ASD and those of typical development. This algorithm uses a moving Gaussian window to define events across the night into five categories (ranging from mild to severe, S1-S5). Our findings support a strong association between actigraphy events and the presence and intensity of nighttime movement activity (based on video scoring), and between actigraphy events and night wakings (based on PSG-EEG scoring).

The “S” categories, along with sum of activity, maximum activity, and duration of movement events, were found to be highly indicative of a nighttime movement event that resulted in a night waking. Furthermore, these actigraphy parameters were highly correlated with movement event severity.

This algorithm yields promising evidence for the measurement of sleep in children with ASD. The novel endpoints may predict severe night wakings, a common sleep disturbance in children with ASD. The novel algorithm can also be used to measure movement intensity during sleep, thereby allowing sleep intervention assessment to be made possible. Further, the endpoints will increase our knowledge and understanding of the types and severity of sleep disturbances in children with ASD.

In the context of the CTTI (2019) Initiative, the novel endpoints presented in this article may have clinical significance and can be used to assess pre- and post-treatment analyses. These endpoints provide a more concise description of the types of movement events that occur throughout the night. Previous algorithms provide a mean CPM score for movements that occurred during the night. Providing only a mean score limits the description of the variability and timing of movement and does not take into consideration the severity of movements. In contrast, the actigraphy algorithm described in this manuscript provides information related to timing, amount, severity, and frequency in order to better delineate movements during sleep across the night. Furthermore, this algorithm provides a cost-effective and minimally-intrusive way to measure sleep-related movement patterns across multiple nights. These movement patterns, and how they change with interventions, can be used to examine treatment effectiveness. As an example, before initiation of a medication or behavioral treatment for sleep, there may be five S5 events, six S4 events, and seven S3 events, with a decrease to two S5 events, two S4 events, and three S3 events after initiation of the treatment. Moreover, this algorithm provides a solution to underreported night wakings that frequently occur in the pediatric population. Parent report is used in the pediatric population to provide sleep quality and quantity data. Descriptions of what happens during a night waking are frequently not reported due to the limitations of parent report. Children often sleep in separate rooms, which decreases and limits the night waking information provided through parent report. Measurement of visual movement event severity can be used for specified sleep treatments and therapies in this population, thus serving as a promising measurement tool that could be used in the clinical setting.

The findings of this study are exploratory due to the small sample size. Future work will require larger sample sizes, along with more heterogeneous patient populations in which insomnia is a concern. For example, future studies should include children with ASD who are taking psychotropic medications as well as those who are free of these medications. This heterogeneity, combined with a larger sample size, will allow us to perform a rigorous comparison between ASD and TD groups. The narrow age range is also a limitation and a wider age range should be examined in future studies. Future work would also measure the relationship between actigraphy parameters and daytime behavioral parameters, including attention, hyperactivity, aggression, and mood. Finally, future work may include an intervention to determine if the algorithm can detect change.

In conclusion, the novel activity level “S” endpoints, activity sum and duration are significant metrics of visual movement severity and PSG arousals. The activity level “S” endpoints from our study provide a novel and characterization of nighttime movement severity. Furthermore, our study findings suggest that the endpoints generated from this novel actigraphy algorithm increase the sensitivity of nighttime movement event severity measured by standard actigraphy measure (such as wake time after sleep onset).

Highlights.

1. Standard actigraphy measurements (e.g., wake time after sleep onset) are poor metrics of nighttime arousals and visual movement severity

2. The novel activity level “S” endpoint algorithm provides a sensitive way of characterizing nighttime movement events.

3. There was a significant association between the level of “S” event (S1-S5) with visual movement events and sleep status by polysomnography.