Sleep/Wake Detection by Behavioral Response to Haptic Stimuli

Abstract

Study Objectives:

Actigraphy, the tool of choice for assessment of sleep phase disorders, is insensitive to movement-free waking. This study aimed to determine whether the detection of waking could be performed by recording instrumental responses to haptic stimuli delivered by a low-cost device.

Methods:

Twenty adults underwent 2 nights of laboratory polysomnography (PSG) while wearing a fingerless glove under which a stimulating actigraph (“Wakemeter”) was apposed to the palm. The Wakemeter, controlled by a tablet computer, delivered gentle, haptic stimuli every 10 minutes during the sleep period. If a stimulus was detected, the participant squeezed the Wakemeter. Stimulus times, response times and movements were streamed to the tablet. Concurrent PSG data were scored blind to stimuli and responses. Self-reported sleep quality ratings were collected each morning.

Results:

The Wakemeter was acceptable to 19 of 20 participants, and effects on self-reported and objective sleep were small. The probability of a response to the stimulus during a wake epoch was high regardless of movement. In contrast, actigraphy magnitude distributions were indistinguishable across epochs scored wake without movement versus sleep, confirming a known limitation of actigraphy. A simple method for calculating sleep efficiency from responses to the stimuli yielded estimates that were highly correlated with PSG-derived estimates (rho = .69, P < .001).

Conclusions:

Behavioral responses to haptic stimuli detected epochs of movement-free wake during the sleep period and may augment actigraphy in the low-burden estimation of sleep efficiency. Acceptability of the method over longer recording periods remains to be established.

Citation:

Miller KE, Bäbler L, Maillart T, Faerman A, Woodward SH. Sleep/wake detection by behavioral response to haptic stimuli. J Clin Sleep Med. 2019;15(11):1675–1681.

BRIEF SUMMARY

Current Knowledge/Study Rationale: Conventional actigraphy is an imprecise measure of sleep efficiency because it cannot detect movement-free waking, a common feature of insomnia. This study examined the ability of a low-cost system that collects responses to haptic stimulation of the palm to detect intercurrent wake and estimate sleep efficiency in a heterogeneous sample of individuals with sleep difficulties.

Study Impact: Responses to haptic stimuli can be collected during sleep and the method tested may augment actigraphy for estimating sleep efficiency. Improved estimation of sleep efficiency could aid clinicians in the delivery of insomnia treatments, and aid sleep researchers engaged in naturalistic studies in which sleep efficiency rather than sleep phase is the variable of interest.

INTRODUCTION

Conventional actigraphy is an imprecise measure of sleep efficiency, especially as insomnia worsens and the sleep period contains more movement-free waking (MFW) that is misinterpreted as sleep.^1–4 Behavioral response monitoring (BRM) systems, which measure responses to stimuli presented during the sleep period, are one approach to overcoming insensitivity to MFW.^5,6 Studies measuring responses to stimuli in order to detect sleep onset or wake have most often used button-pushes to tones, requiring both a hand button/switch and auditory stimulation.^6,7 Loudspeakers incur uncontrolled variation in stimulus intensity associated with head repositioning; and in-ear technology can interfere with lateral sleep posture.^5,8 We describe a low-cost system that integrates accelerometry with the delivery of haptic stimuli to the palm, streaming all data to a tablet computer or smartphone over Bluetooth. This stimulating actigraph, hereafter, the Wakemeter, requires only one device on the body and is relatively insensitive to position changes during sleep. The aims of this study were: (1) to examine the acceptability of the Wakemeter, (2) to estimate its effect on self-reported and objective sleep, (3) to assess, by reference to polysomnography (PSG), whether the presence/absence of responses to haptic stimuli distinguished MFW from sleep, and (4) to compare Wakemeter- and PSG-based estimates of sleep efficiency.

METHODS

Participants

Participants were recruited at the VA Palo Alto Healthcare System (VAPAHCS). All provided written informed consent in accordance with the procedures of the Stanford/VAPAHCS Institutional Review Board. Participants were aged 18 to 75 years and free of acute somatic illness. No further exclusion/inclusion criteria were applied. Forty-five individuals contacted the laboratory expressing interest. Two declined to participate after hearing a study description. Thirteen screened eligible but were lost to contact for reasons that could not be determined. One was excluded because of onset of an illness. Twenty-nine individuals were enrolled, of whom seven withdrew after night 1, five were lost to contact; one refused to continue due to time commitment, and one found the PSG-sensor application procedure aversive. There was no difference in self-reported sleep quality on night 1 between completers and those who withdrew (t₂₃ = 0.79, P = .44). Two were withdrawn due to PSG recording errors on night 1. Twenty completed both study nights.

Procedures

The first laboratory visit began with participants completing a demographic questionnaire and the Pittsburgh Sleep Quality Index (PSQI).⁹ PSG sensors were then applied, including electrodes on the scalp, face, and chest, respiratory bands around the chest and abdomen, and an oximetry sensor on a finger of the nondominant hand. A light glove with tips of the fingers removed (to avoid overheating) was placed on the participant’s dominant hand and the Wakemeter slipped under the glove into the concavity of the palm. Participants were instructed to gently squeeze the device when they felt a vibration. Participants were randomized to receive haptic stimuli from the Wakemeter on night 1 or night 2, and were informed whether they should expect to receive stimuli. Perfect responding to all buzzes was not emphasized, and participants were assured that they could remove the Wakemeter if stimuli prevented them from sleeping. Participants then went to bed as they pleased. The technician remained in an adjoining room using video and audio surveillance to ensure participant safety. In the morning, the technician assisted the participant with sensor removal, and administered a standardized question about sleep quality (scaled from 1 [very poor] to 10 [very good]). A second study night was scheduled within 1 month (mean interval 18.25 days, standard deviation [SD] 7.58 days). Participants were compensated following each study night with a $50 gift card.

Polysomnography

The PSG montage included two channels of bipolar electro-oculogram, two channels of scalp electroencephalography (EEG) (Cz and Pz referenced to linked mastoids), bipolar submental and left anterior tibialis electromyogram (EMG), abdominal respiratory effort (RESP), electrocardiogram (ECG), and blood oxygen saturation. ECG was recorded from the right subclavius area to the left ninth rib. During recording, biosignals were filtered to the following bandwidths, EEG 0.1–100 Hz, EOG 0.3–30 Hz, EMG 30–300 Hz. All PSG data were digitized at 2000 Hz. Offline postprocessing downsampled all signals to 200 Hz and applied additional filtering to facilitate manual sleep scoring (EEG 0.1–30 Hz, EOG 0.1–10 Hz, EMG rectified and integrated over a 20-msec time constant, RESP 0.1–0.5 Hz). Manual scoring was performed by K.E.M, blind to stimulus presentations, and supervised by a registered PSG technician. Scoring was performed in accordance with The AASM Manual for the Scoring of Sleep and Associated Events: Rules, Terminology and Technical Specifications, Version 2.0,¹⁰ except that MFW epochs, indicated by EEG and EMG signals consistent with wake but unaccompanied by artifacts indicative of body movement, were distinguished from movement-wake (MW) epochs, indicated by artifact across multiple channels obscuring EEG.

Mattress Actigraphy

Mattress actigraphy employed a 1-inch mattress topper in which were embedded two accelerometers in the thoracic region.¹¹ Mattress actigraphy was collected and streamed synchronously with PSG solely to enable offline synchronization of PSG with the Wakemeter accelerometry (described in the next paragraphs). This acquisition process employed two analog-to-digital convertors that enabled the synchronous acquisition of two sets of signals at different sampling rates.

Wakemeter

The Wakemeter uses the MetawareR wearable sensor integration platform (Mbientlab, San Francisco, California). The circuit integrates an accelerometer, a haptic motor, and a pushbutton. The MetawareR has been replaced by the MetamotionR, which provides the same functionality as the earlier device and is compatible with the Android application used in this study (https://mbientlab.com/metamotionr/). As of October 8, 2018, the list price of the MetamotionR was $81.99.

It communicates wirelessly over Bluetooth to a smartphone or tablet computer. In the Wakemeter implementation, the circuit is encapsulated in a small, plastic pill case (Figure 1). Squeezing the hinged case depresses a pushbutton located on the circuit board. The Wakemeter app, running on the Android OS, was programmed by one author (L.B.) to deliver haptic stimuli of adjustable intensity, duration, and rate; to detect responses; and to digitize movement at 10 Hz. The Wakemeter app can be obtained by emailing wakemeter@gmail.com. The 1-second haptic stimuli were presented every 10 minutes. Their intensity was set in pilot testing to be gentle but perceptible. Accelerometer data, stimulation times, and response times were continuously streamed to the tablet and later downloaded to a personal computer. They were analyzed using software written in Matlab (The MathWorks, Inc., Natick, Massachusetts, United States) by one author (S.H.W). This software is available upon request. A squeeze occurring within 10 seconds of the stimulus was scored as a response. To estimate sleep efficiency, the number of 10-minute nonresponse epochs was divided by the total number of 10-minute epochs in the sleep period.

Figure 1. The Wakemeter.

The Wakemeter circuit is encapsulated in a small, plastic, hinged pill case, which is positioned against the palm of the hand underneath a light glove. The fingertips were cut off to avoid overheating.

As the PSG and Wakemeter data were collected using systems with independent clocks, synchronization was accomplished by cross-correlating all-night Wakemeter-based accelerometer data with all-night mattress actigraphic data (downsampled to 10 Hz). The cross-correlation magnitude unambiguously identified the relative delay of the two signals, which was used to synchronize Wakemeter and PSG data with approximately 1-second precision.

RESULTS

The 20 study completers were primarily male (85%), aged 25 to 74 years (mean 55.80 years, SD 14.77 years). Participants self-identified primarily as white (45%) or African American (25%). Eighty percent were veterans. The average PSQI global score (last month) was 9.45 (SD 5.40), with most of the sample (80%) endorsing poor sleep quality. One participant used continuous positive airway pressure, and six took hypnotic drugs nightly. The mean objectively defined sleep duration on night 1 for the 20 participants was 7.51 hours (SD 0.96 hours, range 6.21–9.73 hours), and the average self-reported sleep quality was 6.84 (SD 1.98). The mean sleep duration on night 2 was 7.26 hours (SD 0.72 hours, range = 5.93–8.75 hours), and the average sleep quality was 6.76 (SD 2.20).

Aim 1: Acceptability

Study completers were more likely to have experienced haptic stimulation on night 1 compared to those who withdrew (92% versus 57%). Among completers, 12 were randomly assigned to receive stimuli on study night 1 and eight on study night 2. Wakemeter data failed to record for one participant. The Wakemeter was acceptable to all but one participant who removed it after 2 hours. Therefore, data from 18 participants are included in analyses in the following paragraphs. After participating, 15 of the 18 participants indicated that they would be willing to wear the Wakemeter for up to 6 weeks during a hypothetical course of sleep treatment.

Aim 2: Effects on Self-Reported and Objective Sleep

Self-reported sleep quality exhibited no effects of night (6.9 versus 6.7, respectively; F_1,16 = 0.15, not significant [n.s.]), stimulus night order (F_1,16 = 0.05, n.s.), or their interaction (F_1,16 = 0.84, n.s.). Although not statistically significant, the delivery of haptic stimuli was associated with an approximate 0.5-point reduction in self-reported sleep quality.

The duration of the sleep period did not differ over nights 1 and 2 (457 versus 435 minutes, respectively; F_1,16 = 2.9, n.s.) or over groups receiving stimuli on night 1 versus night 2 (F_1,16 = 0.8, n.s.). The interaction of night and stimulus night order approached significance (F_1,16 = 4.1, P = .059), as the group receiving stimuli on night 2 exhibited a shorter mean sleep period on that night versus night 1 (432 versus 480 minutes, respectively). Sleep efficiency exhibited no effects of night (82.9% versus 82.5%; F_1,16 = 0.01, n.s.), stimulus night order (79.3% versus 86.1%; F_1,16 = 1.0, n.s.), or their interaction (F_1,16 = 1.2, n.s.).

In line with reduced sleep efficiency on nights when stimuli were presented, epochs following missed stimuli exhibited a slight increase of being scored wake (t₁₆ = −2.49, P = .02). This effect was accompanied by insignificant percentage reductions in the other stage percentages, presumably in favor of lighter sleep (multivariate F_2,33 = 15.28, P < .001). In Figure 2, it can be seen that the likelihood of a sleep stage persisting into the epoch following a missed stimulus was ordered as follows: rapid eye movement (REM), N2, N3, N1. Although REM sleep may be more stable following haptic stimuli, a formal analysis of transitional probabilities in a larger sample would be necessary to confirm this result.

Figure 2. The mean raw numbers of epochs of wake, N1, N2, N3, and rapid eye movement (REM) sleep when stimuli were delivered (and missed) contrasted with the sleep stage assigned to the following epoch.

Note the significant increase in wake (* P < .05) and the corresponding trends to lower numbers of epochs assigned to stages N1, N2, N3.

Aim 3: Distinguishing MFW From Sleep

A total of 754 haptic stimuli were delivered, with a mean of 42 per participant (SD 4.80, range 33–53). In line with the distribution of sleep stages, most stimuli were presented during stage N2 and N3 sleep (47.74%, combined), followed by MFW (18.56%), stage N1 (14.72%), REM sleep (13.40%), and MW (5.34%). Participants responded to 213 stimuli (28.25%), and failed to respond to 541 (71.75%). Figure 3 shows the proportion of responses and nonresponses to stimuli presented within the levels of behavioral state used (see also Figure S1 in the supplemental material). A logistic mixed-effects regression was used to model the probability of a response to the stimulus, with fixed effects being behavioral state (levels = sleep, MFW, MW); analyses were conducted in R (The R Foundation for Statistical Computing, Vienna, Austria). As presented in Table 1, responses to stimuli were far more likely in epochs scored wake versus sleep. Importantly, the odds of responding were comparable across MFW and MW levels (P = .83). A full model including both behavioral state and time of night (early versus late) did not differ significantly from the model that omitted time (χ²₃ = 2.20, P = .53). Time was binned according to a median split to ensure that this factor was decoupled from the NREM-REM cycle allowing its use as an independent factor. There was no interaction between time and behavioral state. The supplemental material includes analyses that confirm the inability of the actigraphy component of the Wakemeter to differentiate MFW from sleep, a finding not attributable to a lack of accelerometer sensitivity (Table S1, Figure S2, and Figure S3).

Figure 3. Proportions of response and nonresponse to stimuli.

Top: raw proportions of response and nonresponses to all stimuli as a function of movement-wake, movement-free wake and sleep. Bottom: corresponding within-stage proportions adding to 100%.

Table 1.

Predictors of response to stimulus.

Aim 4: Estimating Sleep Efficiency

The correlation between the Wakemeter- and PSG-based estimates of sleep efficiency in this sample was rho = .69 (P = .001). The mean difference between the Wakemeter and PSG sleep efficiency was −0.10 ± 0.16, with 95% of the differences lying between the limits of agreement of −0.41 to 0.22 (Figure 4). This difference was not statistically significant (t_32.43 = 1.47, P = .15), but may represent a true underestimate. There is some indication that Wakemeter-based sleep efficiency estimates less than 50% may be artifactual. When two outliers were removed, the correlation between the mean sleep efficiencies increased to rho = .74 (P = .001).

Figure 4. Graphical comparisons of PSG- and Wakemeter-based estimates of sleep efficiency.

Top: scatterplot including the line of identity. Bottom: Bland-Altman plot. Both plots suggest that Wakemeter-based sleep efficiency estimates below 50% may be suspect. BRM = behavioral response monitoring, PSG = polysomnography, SD = standard deviation.

DISCUSSION

This study replicates previous findings that behavioral response systems can differentiate MFW from sleep,⁵ and perhaps augment actigraphic assessment of sleep. It extends this finding to a low-cost, stimulating actigraph connected to a mobile phone or tablet. As most intercurrent wake associated with insomnia is movement free,¹² here accounting for 75% of all wake, the ability to differentiate MFW from sleep, and so to provide more accurate sleep efficiency estimates, is a desirable feature of a device aimed at improving the delivery of insomnia treatments with objective data. The correlation of PSG-based and Wakemeter-based sleep efficiency estimates at a level of rho = .70 compares very favorably to published figures for limb actigraphy (.43–.48^13,14).

Very few trials of behavioral sleep interventions have included objective measurement of sleep efficiency even though sleep consolidation is a key process variable determining the sleep prescription. Currently, these important estimates are based solely on self-reports of questionable validity, especially when insomnia is comorbid with psychopathology or memory loss. Having the addition of objective information about a patient’s total sleep time and sleep fragmentation, a clinician could assign a more accurate time in bed prescription that would lead to less significant sleep loss. The augmentation of behavioral sleep treatment by objective data is one area in which devices combining actigraphy with haptic stimulus delivery and response acquisition could advance the field. These conditions motivate our belief that a system, such as the Wakemeter, deserves consideration at this time.

Limitations and Future Directions

Although the current results are promising, acceptability of the Wakemeter remains to be firmly established. Though only one study completer (5%) found the Wakemeter unacceptable, multinight acceptability remains to be assessed. Similarly, although only small reductions were observed in self-reported sleep quality and objective sleep efficiency, these too must be examined over multiple nights in larger samples, including those with comorbid sleep disorders and hypnotic drug use. Current acceptability results and the effect on arousal relate only to the delivery frequency tested here (every 10 minutes). It is unknown how these variables would be affected by more frequent stimuli presentations or by individual adjustment of stimulus intensity. We suspect that presenting stimuli at higher rates would further compromise sleep, imposing a limitation on epoch-by-epoch validation against PSG as conventionally scored. Although all-night sleep efficiency scores are frequently reported and utilized in the delivery of insomnia treatments, finer parsing of the night is often desirable. The utility of a 10-minute sample period remains to be established.

A substantial minority of missed stimuli were presented during PSG-defined wake periods. Though the finding that sleepers can respond to stimuli during PSG-defined sleep and can fail to respond to stimuli during PSG-defined wake is well established,¹⁵ and though perfect responding was not emphasized in instructions to participants, such phenomena deserve further consideration. Examination of the EEG immediately preceding stimuli missed during a “wake” epoch or responded to during a “sleep” epoch could be revealing.

Finally, it seems likely that the fusion of behavioral response with movement data will yield additional precision in the estimation of sleep efficiency; however, movement amplitudes may not be informative in this effort. Movement amplitudes during PSG-scored MFW and sleep were highly overlapping; and we can report that movements proximal to responses were not larger than movements proximal to nonresponses. The higher sampling rates available on newer actigraphs may provide an alternate avenue for distinguishing wake-associated movements from those not associated with wake by reference to the patterning of movements in time.

DISCLOSURE STATEMENT

All authors have seen and approve this manuscript. Work for this study was performed at National Center for PTSD, Dissemination and training Division, VA Palo Alto Healthcare System, 3801 Miranda Avenue, Palo Alto, CA 94304. This study was supported in part by The National Center for PTSD, Dissemination and Training Division, of the VA Palo Alto Healthcare System and by a postdoctoral fellowship provided to K.E.M. by the Department of Veteran’s Affairs, Office of Academic Affiliations, Advanced Fellowship Program in Mental Illness Research and Treatment. Dr. Miller is now affiliated with the Mental Illness Research, Education, and Clinical Center at the Cpl. Michael J. Crescenz VA Medical Center. The SensorBed technology used in this work was invented by one of the authors (S.H.W.); however, the patent rights are the property of Stanford University and the Department of Veterans Affairs. The other authors report no conflicts of interest. The authors have no financial relationships with Mbientlab, the manufacturer of the circuit constituting the Wakemeter. The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs or the United States government.

EDITOR'S NOTE

The Emerging Technologies section focuses on new tools and techniques of potential utility in the diagnosis and management of any and all sleep disorders. The technologies may not yet be marketed, and indeed may only exist in prototype form. Some preliminary evidence of efficacy must be available, which can consist of small pilot studies or even data from animal studies, but definitive evidence of efficacy will not be required, and the submissions will be reviewed according to this standard. The intent is to alert readers of Journal of Clinical Sleep Medicine of promising technology that is in early stages of development. With this information, the reader may wish to (1) contact the author(s) in order to offer assistance in more definitive studies of the technology; (2) use the ideas underlying the technology to develop novel approaches of their own (with due respect for any patent issues); and (3) focus on subsequent publications involving the technology in order to determine when and if it is suitable for application to their own clinical practice. The Journal of Clinical Sleep Medicine and the American Academy of Sleep Medicine expressly do not endorse or represent that any of the technology described in the Emerging Technologies section has proven efficacy or effectiveness in the treatment of human disease, nor that any required regulatory approval has been obtained.

ABBREVIATIONS

BRM

behavioral response monitoring

ECG

electrocardiogram

EEG

electroencephalography

EMG

electromyogram

EOG

electro-oculogram

MFW

movement-free wake

MW

movement-wake

PSG

polysomnography

PSQI

Pittsburgh Sleep Quality Index

REM

rapid eye movement

RESP

respiratory effort

VAPAHCS

VA Palo Alto Healthcare System