Associations between parent-reported and objectively measured sleep duration and timing in infants at age 6 months

Abstract

Study Objectives

To compare the estimates of sleep duration and timing from survey, diary, and actigraphy in infants at age 6 months, overall and by select demographics and other factors.

Methods

In total, 314 infants participating in the Rise & SHINE (Sleep Health in Infancy & Early Childhood study) cohort in Boston, MA, USA, wore an actigraph on their left ankle for 7 days. Parents concurrently completed a sleep diary and the expanded version of the Brief Infant Sleep Questionnaire. Concordance between parent-reported and objective sleep estimates was assessed using Bland–Altman plots, Spearman’s rank correlations, intraclass correlations, and linear regression models.

Results

Mean infant age was 6.4 (0.6 SD) months; 51% were female and 42% were Non-Hispanic white. Mean total sleep duration using actigraphy was 526 (67 SD) minutes per night, 143 (42 SD) minutes per day, and 460 (100 SD) minutes during the longest nighttime sleep period. Relative to actigraphy, parent-completed survey and diary overestimated total day (by 29 and 31 minutes, respectively) and night sleep duration (67 and 43 minutes, respectively) and underestimated the longest sleep (58 minutes), with the highest agreement for sleep onset and offset timing (differences < 30 minutes). There was a tendency toward greater bias among short- and long-sleeping infants. Self-reporting bias for diary-measured longest nighttime sleep and total night sleep duration was higher in infants of parents reporting a problem with their baby’s night awakenings and in low-income families, respectively.

Conclusions

Our findings underscore the need to be cautious when comparing findings across studies using different sleep assessment methods.

Statement of Significance.

We showed that parent report relative to actigraphy overestimates daytime and nighttime sleep but underestimates the longest nighttime sleep in infants. Agreement between different sleep assessment methods was best for sleep timing. Measurement imprecision tended to increase in those infants with extremely short or long sleep durations. Parents reporting a problem with their baby’s night awakenings misperceived their baby’s sleep as compared with actigraphy.

Introduction

Research sleep studies conducted in infants have relied mostly on parent-reported sleep measures obtained from parent-completed surveys, sleep diaries, and interviews [1]. A problem with parent-reported estimates, however, is that they are sensitive to reporters’ expectations around sleep and actual awareness of sleep behaviors. For example, parent-reported infant sleep can be underreported in infants who have learned to self-soothe following brief awakenings at night or overestimated when parents are not aware of short night awakenings [2]. As such, brief awakenings may not be recognized by parents because the infant was not signaling or signaling was not perceived by the parent.

Actigraphy, on the other hand, is able to detect brief wake episodes and is recommended by the American Academy of Sleep Medicine as an approach for objectively assessing sleep patterns in infants and children [3]. Previous studies support the validity of actigraphy in infants, ages 1–12 months, against the gold standard polysomnography for identifying sleep state [4, 5].

Few studies have compared parent-reported sleep data to actigraphy data in infants. Ceratto et al. [6] compared data from actigraphy with data from parental diaries in 55 hospitalized infants, aged 1–12 months, and found diary data overestimated sleep duration compared with actigraphy, with concordance between the two methods ≤0.66. In 20 healthy term infants studied over the first 12 months of life, So et al. [7] found significant differences in the amount of sleep and wake over 24 hours recorded by actigraphy and diary report, ranging from 3.8% to 9.4%, depending on the child age at the time of recording. Differences were highest at 1 month of age. Del-Ponte used actigraphy to investigate the validity of the Brief Infant Sleep Questionnaire (BISQ) [8] in 586 Brazilian children at 3, 6, 12, and 24 months [9]. At all ages, sleep durations were more similar between the two methods than night awakenings or WASO (wake after sleep onset) [9]. Also, Tikotzky and Volkovich [10] reported poor agreement between actigraphy and parental reports in 226 infants, with increasing differences across the period of observation (3, 6, 12, and 18 months).

Overall, the existing literature indicates that the association between parent-reported and objective sleep measures is modest or poor [11]. Sleep measures derived from both methods show better concordance for sleep onset and offset times and total sleep duration than for the number of night awakenings and WASO [9, 12–14]. Evidence for different magnitudes of reporting bias or measurement error by race/ethnicity and adverse sleep can be found in the adult literature [15, 16]. Research in adult samples has shown that correlations between self-reported and actigraphy-assessed sleep variables are influenced by demographics, including race/ethnicity, mood, perception of poor health, and underlying sleep disorders [16–18]. In adolescents, discrepancies between actigraphic and sleep diary measures of sleep were greater among boys than girls [19].

Very few studies have examined factors linked with discrepancies between parent-reported and objective sleep assessment in infants [2, 20]. For example, parental-reported infant sleep duration varied by infant sleep location and mode of feeding [20]. We postulate that attending a childcare center may reduce parental awareness of daytime sleep, resulting in the underreporting of daytime sleep [21].

To fill these important research gaps, our study had two aims: First, to examine the concordance among different assessment methods of infant sleep, namely diaries, the expanded version of the BISQ, and actigraphy, in a large sample of infants studied at home at age of 6 months. We hypothesized that compared with multiday actigraphy, parent report would overestimate sleep periods with the lowest agreement using the expanded version of the BISQ (administered on a single point) compared with multiday diary. The second aim was to delineate if sex/gender, race, ethnicity, income, daycare attendance, and survey-reported poor infant sleep quality are associated with discrepancies between subjective and objective sleep assessment methods. We hypothesized that the degree of concordance between these different sleep assessment methods would be lower in racial/ethnic minorities, infants attending a daycare center, and poor sleepers. For this analysis, we analyzed data from the 6-month visit of the Rise & SHINE (Sleep Health in Infancy & Early Childhood) study since actigraphy has been shown to be less reliable in newborns [22]. As a novel aspect of our study, we examined the agreement of three sleep assessment methods for both daytime and nighttime sleep as well as their associations with a broad range of infant characteristics.

Methods

Participants

The Rise & SHINE study is an ongoing prospective cohort study with the overarching goals of examining sleep patterns and growth in early life among infants recruited from the newborn unit of Massachusetts General Hospital in Boston, MA, USA, between May 2016 and June 2018. Details of the study protocol and recruitment/retention procedures are available elsewhere [23]. In brief, full-term singletons with no genetic or congenital abnormalities and their biological mothers without any chronic health conditions participated in this study. Other eligibility criteria included mother’s fluency in either English or Spanish, mothers’ age (at least 18 years), and the family living within 40 miles of Boston without plans to move from the area during the study period.

The study was approved by the institutional review board (IRB) of Massachusetts General Hospital. Signed informed consent was obtained from participating mothers during enrollment. When the infants were 1, 6, 12, and 24 months of age, two clinical research coordinators conducted home visits where they collected survey data and gave instructions to the parents on home-based actigraphy and sleep diary recording. Data were collected and organized on REDCap (Research Electronic Data Capture) electronic data capture tools hosted at Massachusetts General Hospital. REDCap is a secure web-based software platform designed to support data capture for research studies [24, 25]. Mothers completed the survey at the home visit and their infants wore an actigraph on the left ankle for seven continuous days, during which time the parents completed sleep diaries. After completing the assessment and sleep record, each family received a gift card and a study newsletter that included information on infant sleep and growth.

Sleep survey

During the home visit, mothers completed a subset of questions of the expanded version of the BISQ to describe their children’s sleep during the past week. The BISQ included questions on: the infant’s daytime and nighttime sleep patterns, sleep environment, and sleep-related behaviors. Sleep measures assessed using the expanded version of the BISQ have been validated against the corresponding sleep measures derived from actigraphy and sleep diary in a subsample of an internet-based sleep survey [8, 26]. The validation demonstrated a high test–retest reliability (r = 0.81 to 0.95) of BISQ and positive, albeit variable correlations with actigraphy (Pearson correlations of the sleep variables from BISQ with actigraphy and sleep diary r = 0.23–0.83). Since the BISQ does not include a question about morning awakening, we used a modified question of the Children’s Sleep Habits Questionnaire (CSHQ) for our assessment of parent-reported wake time [27].

Sleep diary

During the 7-day sleep actigraphy recording, parents were asked to record sleep, wake, “sleep in motion” (e.g. rocking, stroller, and swing), and off time of the actigraph in 5-minute epochs using a paper-based sleep diary (the diary is shown in the Supplementary Figure S1) [13, 28]. Sleep measures derived from parent-reported sleep diaries have been shown to be correlated with those from BISQ and actigraphy, especially in terms of schedule-related events, such as sleep onset time (r = 0.96 for diary–actigraphy and r = 0.61 for diary–BISQ) [8]. If less than five valid days were available from the sleep diary or actigraph, parents were asked to repeat the protocol for an additional 7 days (n = 11). The final analysis set included participants with a minimum of three days and nights of valid actigraphy data (n = 314).

Actigraphy protocol

We used the Philips Actiwatch 2 (Philips Healthcare, Andover, MD, USA) on the infants’ left ankles. Actigraphs collected count data in 30-second epochs at a sampling rate of 32 Hz [13]. A trained research assistant, blinded to other clinical data, manually annotated the records, denoting rest intervals using the diary, and analyzed with Respironics Actiware 6 software (Version 6.09, Philips/Respironics) using a low threshold of 80 counts per epoch for wake detection [4]. An older version of the Philips actiwatch, the AW64 model, has been deemed to be a valid tool for assessing sleep and wake in infants as compared with polysomnography [4, 29]. In-house laboratory testing with matched pairs in healthy adults by Philips Healthcare has shown no differences in the performance of the two models [30].

Sleep variables

Night sleep duration, longest nighttime sleep, daytime sleep duration, sleep onset time, and sleep offset time obtained from all three sleep assessment methods (i.e. survey, diary, and actigraphy) were analyzed. The nighttime sleep period was defined as the time period between 07:00 pm and 07:59 am, while the daytime sleep period was between 08:00 am and 06:59 pm, reflecting the cutoffs in the expanded BISQ for nighttime and daytime sleep periods.

Scorers manually edited rest and active intervals based on caregiver-completed logs and observation of a sharp decrease/increase in activity [11]. The start of nighttime rest interval was identified based on reported sleep on the sleep diary and the actogram showing reduced activity (>5 minutes) within or overlapping the night period (07:00 pm to 07:59 am). The end of a rest interval (beginning of an active interval) was based on diary report of awakening and increased activity on the actogram (>5 minutes). An active interval was defined as the time between two rest intervals. Additional active intervals were annotated if periods greater than 60 consecutive minutes of activity were observed. Within each rest interval, epochs of sleep and wake were determined based on the actogram activity using a low threshold of 80 counts for sleep [4]. Sleep onset was determined as the first 5-minute consecutive period of immobility (calculated by the algorithm) within the first rest interval. The longest nighttime sleep duration was calculated based on the sum of sleep epochs within the rest interval with the most sleep epochs during the nighttime. Table 1 summarizes all sleep variables used in the analysis.

Table 1.

Definition of sleep variables

Variable	Definition
	Survey	Sleep diary	Actigraphy
Sleep onset time	What time does your baby usually fall asleep for the night? Is this am or pm?	Start of “Time ruler” representing sleeping closest to 07:00 pm	Start time of sleep epoch corresponding to the diary-reported sleep onset time, the first minute when activity counts dropped for five consecutive minutes within the first nighttime rest interval
Sleep offset time	What time does your baby usually wake up in the morning? Is this am or pm?	End of “Time ruler” representing sleeping closest to 07:59 am	End time of last sleep epoch corresponding to the diary-reported sleep offset time, the first minute when activity counts increased for five consecutive minutes
Night sleep duration	How much total time does your baby spend sleeping during the NIGHT (between 07:00 pm in the evening and 08:00 am in the morning)?	“Time ruler” representing sleeping in 5-minute epochs between 07:00 pm and 07:59 am with subsequent split into their respective daytime and nighttime components	Sum of all sleep epochs (30 s each) within the rest intervals that at least partially overlapped the hours between 07:00 pm and 07:59 am
Number of night awakenings	How many times does your baby typically wake during the night?	Number of wakefulness episodes (diary epochs without “Time ruler” representing sleeping) between 07:00 pm and 07:59 am
Longest night sleep	On a typical night, what is the longest stretch of time that your baby is asleep during the night without waking up?	Longest continuous sleep episode between sleep onset and sleep offset time	Sum of sleep epochs within the rest interval with the most sleep epochs during the nighttime (i.e. that at least partially overlapped the hours between 07:00 pm and 07:59 am)
Daytime sleep duration	How much total time does your baby spend sleeping during the DAY (between 08:00 am in the morning and 07:00 pm in the evening)?	“Time ruler” representing sleeping in 5-minute epochs between 08:00 am and 06:59 pm with subsequent split into their respective daytime and nighttime components	Sum of sleep epochs (30 s) in the rest intervals that at least partially overlapped the hours between 08:00 am and 06:59 pm

Other measurements

Mothers reported their infant’s race and ethnicity, their household annual income, and the highest level of educational attainment after delivery. Mode of child care at 6 months was assessed with the question “Who typically cares for your child during the day?” with response options including: You, Father, Family member (inside home or other), In-home nanny, Daycare/childcare outside the home, and Other.

Statistical analyses

Descriptive statistics were reported for our sample, using means and standard deviations for continuous variables and frequencies and percentages for discrete variables. We inspected the continuous variables for data anomalies (e.g. outliers) using boxplots and found no major concerns with the data.

We calculated the mean values of the sleep variables across both objective (e.g. actigraphy) and parent-reported (e.g. sleep diary and BISQ) measures and reported the degree of association between each pair of sleep outcomes using Spearman’s rank correlations. Spearman’s rank correlations were chosen over Pearson correlations due to the nonnormal distributions of each sleep outcome in our sample. Mean differences in sleep variables and their corresponding 95% confidence intervals were calculated for each method. We then calculated intraclass correlations (ICCs) and their 95% confidence intervals for each sleep variable to approximate the level of agreement between objective and parent-reported measures. ICC values were obtained via PROC MIXED in SAS (SAS Institute, Cary, NC), with the type of measure (i.e. survey, sleep diary, or actigraphy) as fixed effects and a random subject effect. We considered a less than chance agreement = <0, slight agreement = 0.01–0.20, fair agreement = 0.21–0.40, moderate agreement = 0.41–0.60, substantial agreement = 0.61–0.80, and almost perfect agreement = 0.81–0.99 [15].

We also modeled parent-reported sleep outcomes as a linear function of objective sleep outcomes. The model’s intercept (centered on eight hours), also known as the bias, captures the extent to which parent-reported measures overestimate (if the intercept is positive) or underestimate (if the intercept is negative) sleep. If there was no bias, the intercept would be zero. The model’s slope captures the degree of calibration between the two measures or the change in the parent-reported measure relative to a change in the objective measure. A slope of one indicates perfect calibration, while a slope of zero suggests that there is no relationship between the parent-reported and objective measures. Wider confidence intervals around the slope suggest more uncertainty in calibration (i.e. more variability in the relationship between parent-reported and objective sleep). To assess whether the agreement between parent-reported and objective sleep outcomes was affected by selected participant characteristics (e.g. gender), we added each characteristic as a covariate and the interaction between the objective sleep outcome and each characteristic to the unadjusted models, separately. All models used robust standard errors to account for potential heteroskedasticity in the residuals.

Bland–Altman plots were employed to visualize the agreement between the three sleep variables [31–33]. In each Bland–Altman plot, the y-axis shows the mean difference, or the estimated bias, between each pair of sleep outcomes, while the x-axis represents the average of each measure. The limits of agreement were taken to be ±1.96 standard deviations of the differences. Data points that lie outside these limits were considered to have a poor agreement. All statistical analyses were conducted using SAS 9.4.

Results

Descriptive characteristics

Of the 433 consented infants at birth, 385 were eligible for the 6-month visit at the time of analysis. Of these, 352 completed the BISQ, [8] and 314 also completed actigraphy and hence were included in this analysis. Compared with the overall sample, the analytical sample tended to have slightly more white infants (41.7% vs. 38.8%), a higher proportion of caregivers with college degrees (76.1% vs. 65.8%), and greater annual household income (61.1% vs. 55.2% ≥ $80,000). Infants had a mean (SD) age of 6.4 (0.6) months and 51% were girls (Table 2). Approximately, 58% of infants were of racial/ethnic minorities. The majority (61%) of infants lived in households with an annual income greater than $80,000/year. Approximately, 40% of the infants were attending a daycare center.

Table 2.

Descriptive characteristics of the study population (N = 314)^†

Characteristics
Mother
Mean age in years (SD)	33.3 (4.97)
College graduate	239 (76.1%)
Annual household income ≥ $80,000	192 (61.1%)
Married or cohabitating	280 (89.2%)
Problem with baby’s night waking’s—Yes	83 (26.4%)
Child
Female	161 (51.3%)
Child race/ethnicity
White	131 (41.7%)
Black	22 (7.0%)
Hispanic/Latino	106 (33.8%)
Asian	52 (16.6%)
Other	3 (1.0%)
Mean infant age in months (SD)	6.4 (0.59)
Weight in kg (SD)	7.9 (1.01)
Length in cm (SD)	66.3 (2.68)
Childcare attendance—Yes	127 (40.4%)

^†Frequencies and proportions or means and standard deviations (SD).

The average nightly sleep duration (between 07:00 pm and 07:59 am) was between 526 and 594 minutes, depending on the respective assessment method. Infants slept on average 143 to 173 minutes during the day (between 08:00 am and 06:59 pm). The sleep onset time by actigraphy was on average at 08:12 (0:54) pm and 18 minutes later when assessed by survey. Infants awoke on average at 07:00 (0:48) am by actigraphy and diary and at 06:36 am (2:00) by survey. The longest mean sleep duration at night was 402–460 minutes (Table 3).

Table 3.

Mean (SD) values for survey (BISQ), sleep diary, and actigraphy for different sleep variables

Sleep variables	Survey	Sleep diary	Actigraphy
Total night sleep duration (07:00 pm to 07:59 am, min)	593.5 (77.4), N = 314	571.0 (69.7), N = 303	526.4 (67.0), N = 314
Longest continuous night sleep (min)	402.2 (150.1), N = 314	403.4 (137.1), N = 303	460.4 (100.1), N = 314
Total day sleep duration (08:00 am to 6:59 pm, min)	171.8 (86.2), N = 314	173.3 (51.9), N = 303	142.9 (42.3), N = 314
Sleep onset time (clock time)	08:30 pm (2:12), N = 278†	08:18 pm (0:54), N = 303	08:12 pm (0:54), N = 314
Sleep offset time (clock time)	06:36 am (2:00), N = 273‡	7.0 (0:48), N = 303	7.0 (0:48), N = 314

SD, standard deviation.

^†missing participants because individual indicated child does not have a usual time of falling asleep.

^‡missing participants because individual indicated child does not have a usual time of waking up.

Comparison of objective and parent-reported sleep methods

Agreement (ICCs and Spearman’s rank correlations)

The ICCs between sleep diary and actigraphy were generally stronger than when actigraphy was compared with survey. Agreement was modest between survey and sleep diary, with the highest level of agreement observed for the longest sleep duration (ICC = 0.55). There was small to modest agreement between survey (i.e. the BISQ) and actigraphy for all sleep variables (ICCs 0.08—0.34) (Table 4). We did not compare sleep timing between actigraphy and sleep diary since data were highly dependent due to the actigraphy scoring with consideration of a sleep diary to determine sleep onset and offset.

Table 4.

Mean differences and intra-class correlation coefficients between survey, sleep diary, and actigraphy (95% CI)

	Mean difference (95% CI)	ICC (95% CI)	Mean difference (95% CI)	ICC (95% CI)	Mean difference (95% CI)	ICC (95% CI)
	Actigraphy—survey		Actigraphy—sleep diary		Survey—sleep diary
Total night sleep duration (07:00 pm to 07:59 am, min)	−67.1† (−76.3, −57.9)	0.34† (0.25, 0.44)	−42.9‡ (−50.6, −35.2)	0.51‡ (0.42, 0.59)	22.9‡ (14.0, 31.7)	0.44‡ (0.35, 0.53)
Longest night sleep (min)	58.2† (39.9, 76.4)	0.17† (0.09, 0.30)	58.2‡ (43.4, 73.0)	0.40‡ (0.32, 0.50)	−1.8‡ (−17.2, 13.6)	0.55‡ (0.47, 0.63)
Total day sleep duration (08:00 am to 06:59 pm, min)	−28.9† (−38.1, −19.7)	0.25† (0.16, 0.37)	−30.8‡ (−35.9, −25.6)	0.53‡ (0.45, 0.61)	−1.0‡ (−10.3, 8.3)	0.33‡ (0.24, 0.44)
Sleep onset time (decimal hours from midnight)	−0.3§ (−0.6, −0.1)	0.23§ (0.14, 0.35)			0.2¶ (0.0, 0.5)	0.31¶ (0.22, 0.42)
Sleep offset time (decimal hours from midnight)	0.4|| (0.1, 0.6)	0.08|| (0.02, 0.28)			−0.4# (−0.6, −0.1)	0.10# (0.03, 0.28)

^†  N = 314; ^‡N = 303; ^§N = 278; ^||N = 273; ^¶N = 270; ^#N = 263.

Spearman correlations reflect these findings: correlations were moderate for comparisons between sleep diary and actigraphy (rs: 0.37—0.60), with comparable agreement between day and night sleep duration. We observed slight-to-fair correlations between sleep variables estimated by actigraphy and survey (r ≤ 0.4), except for sleep onset time (r: 0.67), for which there was substantial agreement between these methods (Table 5).

Table 5.

Spearman correlations between self-reported (survey and sleep diary) and examined (actigraphy) sleep variables

	Actigraphy
	Total night sleep duration (07:00 pm to 07:59 am, min)	Longest night sleep (min)	Total day sleep duration (08:00 am to 06:59 pm, min)
Survey
Total night sleep duration (07:00 pm to 07:59 am, min), N = 314	0.40**	0.34**	−0.16**
Longest night sleep (min), N = 314	0.10	0.16**	−0.01
Total day sleep duration (08:00 am to 06:59 pm, min), N = 314	−0.13*	−0.14*	0.32**
Sleep onset time (decimal hours from midnight), N = 278	−0.47**	−0.33**	0.26**
Sleep offset time (decimal hours from midnight), N = 273	−0.08	−0.05	0.14*
Sleep diary
Total night sleep duration (07:00 pm to 07:59 am, min), N = 303	0.57**	0.50**	−0.38**
Longest night sleep (min), N = 303	0.24**	0.37**	−0.14*
Total day sleep duration (08:00 am to 06:59 pm, min), N = 303	−0.40**	−0.39**	0.60**
Sleep onset time (decimal. hours from midnight), N = 303	−0.40**	−0.27**	0.29**
Sleep offset time (decimal hours from midnight), N = 303	−0.03	−0.04	0.11

*p < 0.05; **p < 0.01.

Mean differences and bias

Compared with parent-reports by survey and diary, actigraphy underestimated total night sleep duration by −67.1 and −42.9 minutes, respectively (Table 4). In contrast, both parent-report methods underestimated the longest night sleep period. We detected the smallest differences for sleep offset and onset time (less than 30 minutes) across the three assessment methods.

Supplementary Table S1 further presents the bias and calibration between the various methods. Similar to the prior analyses, night sleep duration was significantly underestimated by both parent reports from survey or diary compared with actigraphy, and the calibration across the methods was modest.

We constructed Bland–Altman plots to visualize the inter-method level of agreement for total day and night sleep duration. The survey data generally overestimated total day and night sleep duration for infants who are long sleepers and underestimated total day and night sleep duration for infants who were short sleepers compared with either actigraphy or sleep diary. Sleep diary overestimated both daytime and nighttime sleep independent of the overall sleep duration (Figures 1 and 2).

Figure 1.

Bland–Altman Plots for total night sleep duration. (A) Survey versus actigraphy, (B) actigraphy versus sleep diary, and (C) survey versus sleep diary. The solid line indicates the mean of the differences or bias, and the dashed lines indicate the lower and upper 95% limits of differences.

Figure 2.

Bland–Altman plots for total day sleep duration. (A) Survey versus actigraphy, (B) actigraphy versus sleep diary, and (C) survey versus sleep diary. The solid line indicates the mean of the differences or bias, and the dashed lines indicate the lower and upper 95% limits of differences.

Reporting bias varied across various demographic and other characteristics, but only for the actigraphy–diary comparisons. Here, self-reporting bias was significantly higher in infants whose parents reported problems with nighttime awakenings than in those infants without any sleep problems (−137.6 vs. −41.1 minutes). In addition, infants of low-income families had a slight, but significant higher self-reporting bias for total night sleep duration than higher-income families (62.8 vs. 60.0 minutes). We did not observe any differences in discrimination for any other characteristics (Supplementary Tables S2 and S3).

Discussion

In this study of 314 six-month-old infants studied with both parent-reported and objectively measured sleep assessments, we found that compared with actigraphy, parent reports (from survey and diary) of nighttime and daytime total sleep duration are moderately correlated but provide biased estimates. In particular, parent reports overestimate total day and night sleep duration and underestimate the longest sleep period. Of all commonly measured sleep variables, the highest level of agreement was observed for sleep onset timing and the weakest correlations are for the longest duration of nighttime sleep. Moreover, biases were larger for infants with more extreme sleep duration: mothers tended to underestimate total day and night sleep duration in short-sleeping infants and overestimate sleep in long-sleeping infants as compared with actigraphy. In particular, our data indicate that on average, total nighttime sleep duration may be overestimated by parent report by an average of 70 minutes relative to actigraphy, with duration of longest sleep period underestimated by 60 minutes.

Reporting bias by diary was significantly higher for infants of parents reporting a problem with their baby’s night awakenings. Overall, these findings underscore differences in the accuracy of alternative measurements of sleep, with potential for systematic bias, which is exaggerated in groups in whom there may be particular clinical interest in quantifying sleep (those with extreme sleep durations and sleep problems).

Our findings are consistent with previous smaller studies in infants, children, and adolescents showing that sleep timing is usually better correlated across parent-reported and objective sleep assessment methods compared with indices of duration [12, 14, 34]. For example, Werner et al. [34] found satisfactory agreement rates between actigraphy and diary for sleep timing and nocturnal sleep period with differences of less than 30 minutes, whereas agreement was poor for nocturnal wake and actual sleep time (±106 and ± 55 minutes, respectively) in 50 children, ages 4–7 years. Sex, age, and socioeconomic status did not associate with differences between assessment methods for all sleep variables.

When compared with actigraphy, the BISQ overestimated nighttime sleep duration by a mean of 67 minutes, while a 7-day sleep diary overestimated sleep duration by 43 minutes. This result is comparable to estimates from a longitudinal study in 24 infants using the BISQ and actigraphy across the first 6 months of life, where mothers overestimated nighttime sleep by 1–2 hours per day [35]. In part, the high level of agreement with sleep timing and diary assessment may reflect that nonindependence of data from these instruments since the diary was used to annotate the beginning of the actigraph record.

In contrast, parents underestimated the longest sleep duration by almost 1 hour. A behavioral sleep intervention trial of 194 infants found that the longest sleep duration was significantly higher according to diary than actigraphy at baseline, 6, and 24 weeks post-intervention [36]. The longest sleep duration is closely related to “sleeping through the night,” which has been suggested as “any 8 hours of continuous sleep at night.” [37]. This aspect of sleep is of great practical interest to parents since it is closely related to their own sleep quality. Potentially, mothers gave estimations of their infants’ longest sleep that were more reflective of their own nighttime sleep. Research has shown that all sleep dimensions are independently important when making sleep recommendations. For example, lower sleep quality (more night awakenings) in 43 infants predicted compromised attention regulation and behavior problems at 3–4 years of age, which underscores the importance of assessing sleep parameters of both duration and continuity [38].

Misperception of sleep by diary was particularly high in infants of parents reporting a problem with their baby’s night awakenings. Potentially, mothers who are distressed by their infants signaling at night may have particular challenges in accurately estimating their baby’s sleep. On the other hand, it is also possible that actigraphy is less accurate in infants with frequent awakenings, thus reducing the level of agreement [39].

Direct comparison to other studies may be difficult due to the small numbers of children at this age studied before. Generally, the magnitude of bias in the infants in this study seemed larger than what has been reported for older children [10, 22, 34, 40, 41]. Among infants of 6 months or younger, a larger magnitude of bias may be due to the high amount of daytime sleep in infants, the less strong circadian rhythms in this population, the developmental differences in movement during sleep, and use of an alternative actigraphy placement (ankle instead of wrist) [42, 43]. On the other hand, sleep variables from the ankle and wrist were not statistically different in a study of 12 children ages 2–5 years [44].

Several limitations of this study merit discussion. We did not compare sleep measures to polysomnography or videosomnography, which are gold standards for sleep assessment. Despite being objective, actigraphy has a low specificity for identifying wakefulness [39, 45]. Of note, sleep measures can be highly disrupted during in-laboratory polysomnography, when compared with measures obtained in home settings [46]. Diary and actigraphy-assessed sleep parameters were not independent since we used the diary to annotate actigraphy data, thus inflating levels of agreement for sleep onset and offset. In addition, there is debate on how to best assess sleep in infants via actigraphy regarding device model, epoch length, and scoring algorithms [47]. Determining transitions between sleep and wake can be challenging with actigraphy and we used both diaries and the actogram to identify rest and activity intervals, with sleep epochs calculated only from rest intervals. This approach may underestimate the total sleep duration if short sleep bouts occurred within active intervals. Also, our estimate of longest sleep duration was based on the sum of sleep epochs within a given rest interval, and thus does not necessarily reflect continuous sleep, but may map well to parent report of “longest stretch” of sleep. Therefore, the differences between caregiver report of longest continuous sleep and actigraphy longest sleep actually may be greater than estimated. External motion (e.g. stroller and rocking) can result in inaccuracy of actigraphy data although we asked parents to indicate periods of external motion in the diary [48]. In addition, it should be noted that the BISQ was completed prior to the start of actigraphy tracking. Consequently, if an infant had an abnormal week or was undergoing changes to their sleep, this might at least partially explain discordance between the survey and actigraphy-assessed sleep patterns. While all recordings were planned to occur while infants were in their typical state of health, it is possible that some differences in health or other contextual issues influenced the representativeness of data collected. Of note, 80% of the analyzed days were marked as a typical day by the parents. Daytime and nighttime sleep need to be interpreted cautiously, understanding that we used the expanded version of the BISQ cutoffs of day versus night sleep and these cutoffs (07:00 pm and 6:59 am) do not necessarily mark a child’s behavioral pattern.

Finally, the large number of exploratory tests may increase the chances of a false-positive discovery in our analyses. Nevertheless, despite these concerns, important study strengths exist. A major strength is the relatively large sample size of 314 infants with concurrent actigraphy, diary, and survey data. Other strengths are the assessment of sleep over 24 hours across multiple days and the collection of many covariates along with the consideration of demographics as potential moderators.

Conclusion

Overall, the degree of concordance between parent-reported and objectively measured sleep in infants differs by the particular sleep variable being investigated, the infant’s individual sleep patterns (e.g. good or poor sleepers; extreme vs. average sleep duration), and parent-report instrument (survey or diary). Our findings underscore the need to be cautious when comparing findings across studies using different sleep assessment methods. Although objective monitoring is more burdensome for both participants and researchers than self- or parent-reported measures, and usually not feasible in large epidemiological studies, it may be particularly important when evaluating infants with suspected sleep problems.

Funding

This research was funded by the National Institute of Diabetes and Digestive and Kidney Diseases (Grant # (R01 DK 107972-01). S.R. was partly funded by NHLBI R35 HL 135818. C.L.J. was funded by the Intramural Program at the National Institutes of Health (NIH), National Institute of Environmental Health Sciences (NIEHS, Z1AES103325-01).

Conflict of interest statement. Financial disclosure: none. Non-financial disclosure: none.