The Use of Actigraphy to Study Sleep Disorders in Preschoolers: Some Concerns about Detection of Nighttime Awakenings

Abstract

Study Objectives:

This study compared actigraphy with videosomnography in preschool-aged children, with special emphasis on the accuracy of detection of nighttime awakenings.

Design:

Fifty-eight participants wore an actigraph for 1 week and were videotaped for 2 nights while wearing the actigraph.

Setting:

Participants were solitary sleepers, studied in their homes.

Participants:

One group (n = 22) was diagnosed with autism, another group (n = 11) had developmental delays without autism, and a third group (n = 25) were typically developing children; age ranged from 28 to 73 months (mean age 47 months); 29 boys and 29 girls.

Interventions:

N/A.

Measurements and Results:

Nocturnal sleep and wakefulness were scored from simultaneously recorded videosomnography and actigraphy. The accuracy of actigraphy was examined in an epoch-by-epoch comparison with videosomnography. Findings were 94% overall agreement, 97% sensitivity, and 24% specificity. Statistical corrections for overall agreement and specificity resulted in an 89% weighted-agreement and 27% adjusted specificity.

Conclusions:

Actigraphy has poor agreement for detecting nocturnal awakenings, compared with video observations, in preschool-aged children.

Citation:

Sitnick SL; Goodlin-Jones BL; Anders TF. The use of actigraphy to study sleep disorders in preschoolers: some concerns about detection of nighttime awakenings. SLEEP 2008;31(3):395-401.

THE USE OF THE ACTIGRAPH FOR ACTIVITY-BASED SLEEP ASSESSMENT HAS INCREASED WIDELY IN BOTH ADULT AND CHILD CLINICAL RESEARCH STUDIES. Actigraphs are miniaturized wristwatch-like devices that detect body movements in different axes and translate the movements to digital counts in predetermined epoch intervals.¹ Actigraphic assessment of sleep is cost-effective and permits monitoring of activity for extended periods of time to infer sleep-wake patterns. Recommended guidelines have been described, although no consensus has been reached on which device or algorithm is best for any particular sample.²

The Standards Committee of the American Academy of Sleep Medicine has deemed actigraphy useful for children when characterizing and monitoring circadian rhythms and sleep disturbances.³ The nonintrusive nature of actigraphy has, in fact, led to its widespread popularity for use in studies of children's sleep. Actigraphy has been used to establish normative values for activity-based sleep assessment for preschool-aged populations 4 and to investigate activity in children with autism spectrum disorders,^5–9 attention-deficit/hyperactivity disorder,^10,11 and sleep disorders.¹² Despite the increasing popularity of actigraphy, however, whether or not actigraphy can accurately detect nocturnal awakenings is of particular concern in studies of children with fragmented sleep.

Sleep problems are common in typically developing young children, affecting 20% ¹³ to 28% ¹⁴ of toddlers and preschoolers. The most common sleep problems in young children are called “behavioral insomnias of childhood,” which consist of prolonged sleep-onset latency times at the beginning of the night and/or prolonged and frequent awakenings during the middle of the night.¹⁵ Because wakefulness characterizes a large proportion of sleep problems in young children, the accuracy of actigraphy as a technology for assessing awakenings is especially important to confirm. A decrease in accuracy of actigraph-scored wakefulness has been well documented in adult patients with insomnia in whom quiet wakefulness has been mistaken for sleep.¹⁷ However, its accuracy in active young children is less well known.²

When compared with polysomnography in epoch-by-epoch analyses, the ability of actigraphy to detect sleep is greater than its ability to detect periods of awake. In a recent review, Tryon¹⁶ described 14 separate studies that compared actigraphy with polysomnography. Although not systematically explored, these studies used various actigraph models and varying sensitivities. The overall sensitivity of actigraphy for detecting activity scored as sleep, as compared with the polysomnographic epochs of sleep, ranged from 78% to 99%. Less robust results were reported for agreements between the 2 methods for identifying wake behavior after sleep onset (WASO). In typical adult sleepers, the agreement for nocturnal awakenings for epoch-by-epoch comparisons, termed specificity, ranged from 46% to 87%¹⁶ but was as low as 34% for adults with sleep disturbances.¹⁷ Although similar results have been found in infant populations,¹⁸ the accuracy of actigraphy with preschool-aged children has yet to be explored.

Polysomnography is the most reliable method for studying sleep, yet it presents several limitations when used with young children. Children can be especially resistant to instrumentation and sensitive to changes in their customary sleep locations. Also, polysomnography with special needs children, such as those with attention-deficit/hyperactivity disorder, autism, and cognitive delays, can be even more challenging. Yet, it is these children who are reported as most likely to have an increased rate of sleep disorders.^9,11,19–21 Although recent studies suggest that polysomnography is possible with some special needs populations,⁹ most literature continues to rely on parent report and actigraphy. The accuracy of actigraphy in comparison with polysomnography in these clinical groups has not been widely investigated.²³

Videosomnography provides another method of recording sleep and is often used adjunctively with polysomnography. Time-lapse videosomnography provides the ability to film sleep in the natural setting of the child's home without instrumentation and has provided important information on night waking in infants.^{22, 23} In toddlers and preschool-aged children, our clinical and research experience using videosomnography in the home suggests that nighttime awakenings are typically characterized by the child sitting up in bed and looking around the room or moving in ways that are not typical of the phasic activity more characteristic of rapid eye movement sleep—behaviors that are easily captured on video. Although these observations have not been confirmed by polysomnography, we believe that the use of videosomnography may provide a method of assessing the accuracy of actigraphy in populations where the use of polysomnography is limited. Moreover, if actigraphy is not accurate in recording the awakenings characteristic of behavioral insomnias, then videosomnography might provide an alternative or adjunctive method.

The aim of this study was to evaluate actigraphy compared with videosomnography in preschool-aged children, and more specifically, the accuracy of actigraphy compared with videosomnography for detecting the number and duration of nighttime awakenings.

METHODS

This study was part of a larger actigraphic study examining sleep-wake patterns in 3 groups of children: children with autism, developmentally delayed children without autism, and typically developing children. All procedures were approved by the University of California, Davis Institutional Review Board, and parents provided informed consent for their children. Parents received nominal compensation for their participation.

Participants

As outlined in the research grant that supported this study, a random subset of parents from the larger study (n = 194) was asked to participate in 2 consecutive nights of home video recordings of their children's sleep. The original intent of this component of the larger study was to examine the concurrence of actigraphy and videosomnography. Data from 58 subjects (29 boys) were included in the final analysis. Subjects ranged in age from 28 to 73 months, with a mean age of 47.8 months (SD= 12.7) months. Of the subjects who were included, 22 children were diagnosed as having autism, 11 with developmental delays of mixed etiologies without autism, and 25 were identified as typically developing. Diagnoses were substantiated through the use of the Autism Diagnostic Observation Scale (ADOS), the Autism Diagnostic Interview, Revised (ADIR) and the Mullen Scales of Early Learning (MSEL). The ethnic composition of the group was Caucasian (60%), Hispanic (10%), African American (4%), Asian American (4%), and mixed (22%). Most of the parents were married (85%), with a fairly high level of education (60% college graduates). The sociodemographic characteristics of the video subset did not differ significantly from the larger sample on proportions of diagnostic group members, ethnicity, parent education, parent occupation, developmental quotient, or adaptive functioning within each diagnostic group. The subset, however, had a significantly larger proportion of girls.

Procedures

At the time of enrollment, the placement, care, and use of the actigraph were reviewed with the parents. Sleep diaries for daily recording were also provided and explained to parents. During the week that the child wore the actigraph, 1 of the experimenters conducted a home visit to set up the videosomnography equipment for 2 consecutive nights. Parents were responsible for activating the video at the beginning of the child's bedtime routine and ending the recording after the child had risen from bed in the morning. They were also responsible for changing the cassette tape for the second night's recording. After the second night of recording, the experimenter returned to the home to collect the video equipment.

Actigraphy

A Mini-Mitter® Actiwatch Actigraph (AW64, Mini-Mitter, Inc., Bend, Ore) was worn for 7 consecutive days and nights on the child's nondominant ankle. Two children had difficulty wearing the Actiwatch on their ankle, so they wore the Actiwatch on their nondominant wrist. The Actiwatch weighs approximately 2 ounces and was embedded in a foam pad secured by a Velcro strap around the child's ankle. As with other studies,^{4, 24} the ankle was chosen primarily because of the smaller size of upper limbs in young children. In addition, in our pilot studies, we noted that children with neurodevelopmental disorders were more likely to resist placement and/or remove the actigraph from their upper extremity. Working with our occupational therapist, we determined that ankle placement of an actigraph that was embedded in sensory-reducing foam was associated with better compliance. Although it is possible that awakenings characterized by the child's sitting up might not be reflected as an awakening on an actigraph placed on the ankle, our pilot data, again, demonstrated that there is significant body movement and position change when the child shifts from lying to sitting.

The actigraph contains a microsensor that converts movements into data and summarizes the movements in 1-minute epochs. The data consist of activity counts (frequency) stored in 1-minute epochs (bins) for later uploading to a computer. After uploading to the computer, the actigraph data are automatically scored using the manufacturer's algorithm set at medium sensitivity (Mini-Mitter, Inc.). One-minute video and actigraphy epochs were used for all comparisons.

Because the number of awakenings was excessive, compared with both the parent diary and the video recordings, the data were recoded by means of our own “smoothing” algorithm, designed to reduce the number of single-minute movement epochs that had been scored as awakenings. Secondary smoothing serves a practical purpose only and has thus far not been validated. The factory-set Actiwatch algorithm regularly produced, on average, 8 to 10 awakenings per night. The diary and video awakenings averaged 2 to 5 per night. Readjusting the factory calibration to high or low sensitivities did not significantly change these numbers.

The laboratory smoothing algorithm required that the minimum length of an awakening following sleep onset (WASO) be 2 minutes. More specifically, the laboratory smoothing algorithm coded an awakening as beginning at the first of 2 or more consecutive minutes with activity counts greater than 100. When 2 or more consecutive minutes with activity counts greater than 100 were immediately preceded by any activity count above 0, that epoch was considered the start of the awakening. An awakening was scored as ending, signifying a return to sleep, at the first of 3 consecutive 0s (no activity). Thus, this secondary filter reduced the average number of awakenings per night to a range more consistent with the parent diaries and the video recordings. The smoothing process was automated via formulae in an Excel (Microsoft, Redmond, WA) spreadsheet.

The actigraph variables used in the comparison with video were sleep-onset time, sleep-latency time, sleep end time, total sleep time, number of awakenings, and WASO. An operational definition of each variable is given in Table 1.

Table 1.

Operational Definitions of Actigraph and Video Variables

Variable	Operational definitions
	Actigraphy	Videosomnography
Sleep-onset time	The first of at least 3 consecutive minutes with an activity frequency count of 0	The first minute of the first 5 minutes when the child is observed as physically still with eyes closed
Sleep end time	The final activity frequency count of 0 before waking in the morning	The final observed awakening of the night in which the child does not return to sleep
Sleep latency	The difference between the sleep-onset time of the actigraph and the bedtime indicated in the parent diary	The difference between the time of lights out and sleep-onset time
Total sleep time	The time from sleep-onset time to sleep end time minus any WASO	The time from sleep-onset time to sleep end time minus any WASO
Number of awakenings	The number of awakenings determined by the actigraph after filtering by a smoothing algorithm	The number of awakenings determined by the child sitting up in bed and looking around the room with eyes open
WASO	Sum of the durations of all awakenings as indicated from the smoothing algorithm	The sum of the durations of all awakenings throughout the night

WASO refers to wake after sleep onset.

Parent Report

Following actigraphy guidelines,^3,24 parents completed a sleep diary to log daytime napping and nighttime sleep for every day of actigraphy wear. Parents were requested to complete a written record each morning for the previous 24 hours by indicating bedtime, sleep-onset time, awakenings from sleep, subsequent returns to sleep, sleep location, morning rise time, and times of daytime naps, if they occurred. The diary took less than 5 minutes to complete each morning. Parents submitted their diary records each day online or via phone to maximize compliance. The diary bedtime was used to set the actigraph bedtime. If no bedtime was given by the parent, or if the actigraph data indicated that the child had fallen asleep prior to the reported diary bedtime, then the actigraph's sleep latency was scored as missing. If the parents' reported bedtime was the same time as the sleep onset time for the actigraph, then the sleep latency for the actigraph was scored as 0:00 minutes.

Videosomnography

While wearing the actigraph, all subjects were recorded on 2 consecutive nights using a Panasonic AG6749P time-lapse video recording system (Osaka, Japan ) with a Sanyo VDC 9212 low-level illumination camera (Osaka, Japan ). The camera was mounted on a tripod near the child's bed or sleep location (e.g., couch). A microphone placed near the child's bed recorded vocalization and crying. Time-lapse video recording reduced real time by a ratio of 12:1, so that an entire night of sleep was filmed on a single 2-hour cassette. The parents started the video recorder just prior to bedtime and turned it off in the morning once the child was awake and out of bed. Recordings were possible in darkness with the use of a low-level illumination camera. Prior to recording, camera and actigraphy clocks were synchronized to ensure accurate analysis.

All videotapes were scored by 1 of 2 raters. To check reliability, 20% of the tapes were scored by both raters; 85% interrater reliability was obtained. The following variables were coded from the videotapes: lights-out time, sleep-onset time, sleep-latency time, sleep end time, number of awakenings, and the total awake duration (WASO). The operational definitions for all variables are provided in Table 1.

Data Analysis

For comparison of categorical variables (diagnostic group, sex, etc.), the Kruskal-Wallis χ² test of significance was used. For comparison of continuous variables, the paired t-test was used. The Spearman rank order test was used for correlational analyses. To establish accuracy, Tryon's methods, as reported by Ancoli-Israel et al.,²⁶ for computing overall agreement, sensitivity, and specificity were used and modified as described in more detail below.

When comparing 2 methods to establish accuracy, 1 of the 2 methods must be chosen as the standard. In laboratory sleep studies, this standard is typically polysomnography. In the current study, because awakenings were observed reliably on the videotapes by 2 raters as sitting up and looking around, videosomnography was considered the standard. To have selected the actigraph as the standard would have used nonverifiable data (i.e., absence of interrater reliability) and been contrary to methods used in previous actigraph comparisons.

Furthermore, when comparing 2 methods, distinct analyses need to examine both bout onset times and bout durations. That is, how accurate are the 2 methods in the concurrence of onset times, and how concordant are they for episode lengths? The major focus for these comparisons was on start times and durations of awakenings after sleep onset (WASO). All awakenings after an initial period of 5 minutes of continuous sleep were scored as WASO. Each video and actigraph awakening episode was compared separately. Agreements (hits) and disagreements (misses) on start times were tallied within a hit “zone.” That is, when an actigraph awakening occurred within 5 minutes of a video awakening, either prior to or subsequent to the video awakening, the actigraph-video agreement was considered a hit. If there was no actigraph awakening within 5 minutes of the onset of a video awakening, the awakening was scored as a miss. The total number of hits and misses was computed separately for each night of recording. The percentage of hits for each night was obtained by dividing the number of hits by the total number of video awakenings for the night. This variable is called concurrence of awakening start times. A mean percentage was calculated for each subject over both nights.

The duration analyses compared minute-by-minute agreements between videosomnography and actigraphic scoring when episodes coincided. Sensitivity (sleep agreement), specificity (awake agreement), and overall agreement were derived according to the method described by Tryon in Ancoli-Israel, et al.²⁶ Each minute was scored as true wake (TW) when the actigraph and video both agreed on an awake minute; false wake (FW) when the actigraph scored awake and the video was scored as asleep; true sleep (TS) when both methods agreed on the minute as being asleep; and, false sleep (FS) when the actigraph scored sleep and the video was scored as awake. Sensitivity (sleep agreement) was calculated by dividing the TS epochs by the total video sleep epochs. Specificity (awake agreement) was calculated by dividing the TW epochs by the total wake epochs in the video. Finally, overall agreement was computed by dividing the TS + TW epochs by the total number of epochs in the video. Mean values for each index were then calculated for the sample. Sensitivity, specificity, and overall agreement were computed using the following equations:

• Sensitivity: TS/(Total Video Sleep Epochs)

• Specificity: TW/(Total Video Wake Epochs)

• Overall agreement: (TS + TW)/(Total Video Epochs)

Two statistical issues, not regularly associated with comparison studies but relevant for sleep studies because of the organization of sleep-wake data, deserve discussion. First, sleep and waking epochs across a night are highly skewed. The data are biased toward sleep because only a small percentage of the night is spent awake. Thus, if actigraphy is better at detecting sleep, (i.e., higher sensitivity) and the majority of the night is spent asleep, the total percentage of agreement using the (TW+TS)/Total Video Epochs equation will be biased because of the greater number of sleep epochs that agree. A weighted-agreement analysis corrects for this bias. To provide equal weights to sleep and waking epochs, a prevalence-adjusted and bias-adjusted Kappa (PABAK) was computed.²⁵ This adjustment sums all epochs for each night and divides this sum by the sum of all epochs over all subjects, multiplied by a weighting factor, 2 x (TS+TW)/(N−1).

Second, an adjustment to the specificity analysis is also required. Since the formula to compute specificity is based on the percentage of TW epochs (TW epochs/Video Awake epochs), subjects who have no awakenings observed on the video have a 0 in the denominator of the formula. That is, they have no awakenings to compare to the actigraph record, which results in a 0% specificity. In order to correct for this bias, the data from the 10% of subjects who had no video awakenings were discarded from the PABAK analysis. Similarly, subjects who have video awakenings but no TW epochs, also result in 0% specificity, regardless of the number of epochs that the video shows the subject awake. Thus, for example, a subject who has no TW epochs but 5 video awake episodes would result in 0% specificity, just as a second subject who had no TW epochs but 10 video awake episodes would result in 0% specificity. Although these 2 subjects both have 0% specificity, the accuracy of the actigraph for the 2 subjects differs. In order to correct for this error in the PABAK analysis, 0.1 was added to both the numerator and denominator for all subjects, resulting in an adjusted-specificity equation of (TW + 0.1)/(Video Wake + 0.1).

RESULTS

A Kruskal-Wallis χ² test revealed no significant differences among diagnostic groups, sex, or age on any of the actigraph-video comparisons. Therefore, the 3 diagnostic groups and 2 sexes were collapsed. The Kruskal-Wallis χ² test also revealed no significant differences between nights 1 and 2. Therefore, nights 1 and 2 were combined, and means were computed for each child. Finally, no significant differences between wrist and ankle placement were noted for the 2 children who wore the actigraph on their wrist, so all actigraph data were combined.

Sleep Descriptors

The overall means and standard deviations for each of the sleep variables—sleep-onset time, sleep latency, sleep end time, total sleep time, number of nocturnal awakenings, and WASO—are presented in Table 2. A paired t-test revealed that when compared with videosomnography, actigraphy showed significantly longer sleep-onset latency times, more nocturnal awakenings, and longer mean WASO durations; furthermore, sleep end times for actigraphy were significantly later and sleep-onset times significantly earlier.

Table 2.

Sleep Measures

Variable	Video	Actigraphy
Sleep-onset time, clock timea	21:42 (1:19)	21:33 (1:15)
Latency, minb	0:15 (0:14)	0:25 (0:19)
Total sleep time, hour: minutes	8:35 (1:51)	8:19 (1:08)
Sleep end time, clock timea	6:36 (1:38)	7:00 (1:00)
Number of awakeningsa	2.3 (1.7)	3.1 (2.4)
WASO,minutesa	12.6 (19.6)	19.1 (19.8)

Data are expressed as mean (SD) of videosomnography and actigraphy variables. WASO refers to wake after sleep onset.

^aP ≤ 0.05

^bP ≤ 0.01

Correlation Between Methods

Spearman correlations between videosomnography and actigraph were conducted for sleep-onset time, sleep latency, sleep end time, total sleep time, number of awakenings, and total sleep duration. All variables, except the number of nocturnal awakenings, showed a statistically significant correlation (P ≤ 0.01). The number of nocturnal awakenings also was significantly correlated (P ≤ 0.05). Correlation coefficients are listed in Table 3.

Table 3.

Correlations Between Videosomnography and Actigraphy for all Sleep Variables

Sleep onset time	Sleep latency	Total sleep time	Sleep end time	Number of awakenings	WASO
0.962b	0.502b	0.670b	0.559b	0.255a	0.435b

^aP ≤ 0.05 (2-tailed)

^bP ≤ 0.01 (2-tailed)

WASO refers to wake after sleep onset.

Concurrence of Awakening Start Times and Minute-by-Minute Analyses

Concurrence of awakening start times for each WASO episode was 32.4%. An example of a miss is shown in Figure 1.

Figure 1.

An awakening detected by videosomnography but not actigraphy. The subject was observed to have an awakening and leave his bed at 02:46. His mother returned him to his bed at 02:47 and stayed with him until he fell asleep at 03:08. The actigraph showed movement but not enough to constitute an awakening. The parent diary showed an awakening from 03:00 to 03:08, even though the parent was observed with the child for 20 minutes on the video.

The sensitivity, specificity, overall agreement, weighted-agreement, and adjusted specificity of the epoch-by-epoch comparisons are reported in Table 4. Sensitivity for sleep resulted in the highest percentage of agreement, followed by overall agreement. Specificity for awake agreements was poorest. After adjusting for weighted-agreement and adjusted-specificity, the weighted agreement dropped slightly in comparison with the overall agreement, and the adjusted specificity improved slightly in comparison with the original specificity value. No significant differences were found with sensitivity, specificity, overall agreement, weighted-agreement, and adjusted-specificity when comparing problem and no-problem sleepers.

Table 4.

Comparison of Actigraphy and Videosomnography

Concurrence of awakening start times	32.4% (34.3)
Overall agreement	94.6% (5.3)
Sensitivity	97.6% (3.1)
Specificity	24.3% (23.0)
Weighted-agreement	89.4% (8.2)
Adjusted-specificity	27.3% (20.9)

Data are expressed as mean (SD).

DISCUSSION

The objective of this study was to examine the accuracy of actigraphy compared with videosomnography in 3 groups of preschool-aged children, 2 of which are presumed to have a high rate of sleep-onset and night-waking problems. Even in typically developing children of this age, however, night awakenings are reported to have a high rate of occurrence.

Videosomnography was used as the standard in this study because of its observable, objective coding criteria for awakenings and its ease of use in children with special needs. Using actigraphy, periods of awake can be falsely labeled as sleep when the child is sedentary (e.g., television watching or riding in the car), just as periods of active rapid eye movement sleep with a large amount of motor activity may be falsely labeled as awake. The direct observations of sleep behavior on video minimize these sources of potential error. Video awakenings were coded only when the child sat up, with eyes open, looking around the room. The use of such strict coding criteria fostered a conservative approach to scoring, supporting the use of video as the standard for video-actigraph comparisons.

Although the correlational analysis between actigraphy and videosomnography revealed that all measures were significantly correlated, statistical significance is not equivalent to co-occurrence.²⁶ Positive correlations indicate the relative, rather than the absolute, concordance of actigraphy. Even after using a secondary laboratory smoothing algorithm, actigraphy still expressed significantly more awakenings per subject than videosomnography (mean = 3.1 vs 2.4). One assumption might be that actigraphy was detecting the same awakenings as videosomnography, with extraneous movement during sleep accounting for the additional actigraph “awakenings.” However, using a 5-minute window to allow for possible asynchrony of the clocks on the 2 instruments, the concurrence of actigraphic and video awakening hits was only 32.4%. That is, only one third of the observed video awakenings were detected by the actigraph.

Regardless of a low concurrence rate for WASO start times, it was still possible that the duration of awakenings might overlap. That is, awakenings could still be taking place during the same time period but not beginning at the same time. Although, the overall agreement of actigraphy and video for all epochs throughout the night was high, as was the agreement for sleep (sensitivity = 97%), the actigraph's agreement for WASO durations (specificity = 24%) was low.

Even after adjusting agreements for prevalence and bias (PABAK) and removing subjects with no video awakenings, for the reasons described previously, the overall weighted-agreement and adjusted specificity remained relatively unchanged. The actigraph recorded more “awakenings” than did videosomnography, but actigraphy was only able to detect awakenings characterized by the child sitting up and looking around the room 27% of the time. When the video recorded subjects who sat up and looked around the room, the actigraph most often miscoded the activity as sleep.

There are many possible reasons for the low specificity of actigraphy, as compared with videosomnography. Perhaps the placement of the actigraph on the ankle rather than the wrist might account for the low specificity. That is, the child might sit up without moving his or her legs. Observing the videotape, however, makes this explanation unlikely. The video awakenings were associated with gross body movements and general position changes. Such an explanation also seems unlikely because the overall results of this study are similar to those of previous polysomnography studies with subjects wearing the actigraph on their wrists. Although ankle placement on children with periodic limb movement disorder or restless leg syndrome might confound actigraph recordings, none of the subjects in this study were suspected of having periodic limb movement disorder or restless legs syndrome after coding the videotapes. Although a previous study has reported that the sensitivity level set for the actigraph effects accuracy,¹⁸ in the current study, the laboratory smoothing algorithm reduced the number and duration of actigraph-scored awakenings beyond that determined by the internal sensitivity settings of the actigraph.

These results do not differ from those reported by others for infants, older children, and adults.^12,18,27 Actigraphy is a poor predictor of nocturnal awakenings. The results of this study support previous research that has cautioned against using actigraphy as a diagnostic tool for sleep disorders because the actigraph infers sleep and waking from activity patterns alone.² Because no consensus has been reached on which device or what sensitivity setting and smoothing algorithm are most accurate, more research is needed to determine the parameters that are most appropriate with different populations.

Much like polysomnography and actigraphy, videosomnography is not without limitations. Although it is less intrusive than polysomnography, some children may be uncomfortable or preoccupied with the presence of the camera. The child may intentionally move to a section of the bed that is off camera or seek to hide under pillows or blankets when falling asleep. Once asleep, however, sleep activity usually brings the child back on camera. Videotaping also depends on the parents' activating the equipment prior to sleep onset. Some parents fail to start the equipment prior to the child falling asleep, making bedtime routines and sleep-onset behaviors difficult to observe. Finally, because video equipment is stationary, children who leave the bed in the middle of the night to sleep elsewhere preclude further observations of sleep that night.

In summary, although the results of this study are suggestive, they may not be definitive. It possible that the middle-of-the-night awakenings characteristic of toddlers and preschool-aged children (sitting up and looking around) are not characteristic of older children and adults, making video recording at these ages less useful. Moreover, other actigraph brands, better “smoothing” algorithms for raw actigraph data, or both may provide more accurate data. Certainly the results of this study suggest that more comparison studies, controlling for type of sleep disorder and age, that compare actigraph, polysomnography, and video are indicated. There is no single perfect way to study sleep, especially in young children. Each method has its strengths and limitations. Before selecting a single method, investigators and clinicians need to be clear about the question being addressed. The actigraph may be fine as a screening tool or for large-scale epidemiologic studies. Polysomnography may be best for definitive diagnoses of serious disorders in individual patients. And video recordings may be best for home recording of insomnias. In some situations, it may be prudent to use multiple methods. However, for those interested in studying nocturnal awakenings in preschool-aged children, caution should be exercised in using actigraphy as the sole method of assessment.