High levels of sleep disturbance across early childhood increases cardiometabolic disease risk index in early adolescence: longitudinal sleep analysis using the Health Outcomes and Measures of the Environment study

Abstract

Study Objectives

This study examines the impact of sleep duration, bedtime, and sleep disturbance during early childhood on the risk of cardiometabolic disorder (CMD) in early adolescence.

Methods

Within the Health Outcomes and Measures of Environment Study, we examined sleep patterns of 330 children from ages 2 to 8 years and the relationship of these sleep patterns with cardiometabolic risk measures at age 12 (N = 220). We used a group-based semi-parametric mixture model to identify distinct trajectories in sleep duration, bedtime timing, and sleep disturbance for the entire sample. We then examined the associations between sleep trajectories and CMD risk measures using general linear models using both an unadjusted model (no covariates) and an adjusted model (adjusting for child pubertal stage, child sex, duration of breastfeeding, household income, maternal education, and maternal serum cotinine).

Results

In the unadjusted and adjusted models, we found significant differences in CMD risk scores by trajectories of sleep disturbance. Children in the “high” disturbance trajectory had higher CMD risk scores than those in the ‘low’ disturbance trajectory (p’s = 0.002 and 0.039, respectively). No significant differences in CMD risk were observed for bedtime timing or total sleep time trajectories in the unadjusted or adjusted models.

Conclusions

In this cohort, caregiver-reported sleep disturbance in early childhood was associated with more adverse cardiometabolic profiles in early adolescence. Our findings suggest that trials to reduce CMD risk via sleep interventions—which have been conducted in adolescents and adults—may be implemented too late.

Graphical Abstract

Graphical Abstract.

Statement of Significance.

Study findings unveil a crucial link between early childhood sleep disturbance and the risk of cardiometabolic disorder (CMD) in early adolescence. By analyzing the sleep patterns of 330 children aged 2 to 8 years and their subsequent CMD risk measures at age 12, this study pioneers a comprehensive understanding of the impact of sleep duration, bedtime, and disturbance on cardiometabolic health. Significantly, it identifies a high-risk trajectory of sleep disturbance in early childhood associated with adverse CMD profiles, emphasizing the need for timely interventions. This research not only challenges existing paradigms but also underscores the potential efficacy of early-life sleep interventions in preventing CMD, thereby reshaping strategies for pediatric and adolescent healthcare practitioners.

Introduction

Childhood obesity is a public health crisis [1], with obesity prevalence rising each year [2]. Obese children are at increased risk for developing cardiometabolic disorders (CMDs) [3], typically characterized by increased body fat, insulin resistance, impaired glucose tolerance, hypertension, and dyslipidemia [4]. The obesity epidemic led to the rise in childhood CMDs and portends subsequent rises into adulthood [5]. CMDs, the leading cause of morbidity and mortality in the United States [6–8], result in $351 billion dollars of annual costs within the United States [6, 9]. It is estimated that the majority of CMDs can be prevented through modifiable behaviors [6, 10], though <10% of adults adhere to the guidelines outlined by the American Heart Association [11]. The most frequently targeted health behaviors are physical activity and dietary intake, though these approaches have had disappointingly small effects [12]. Therefore, it is vital to identify additional modifiable targets, including both individual and population strategies, that may be simpler to implement in efforts to prevent CMDs.

Sleep—which is related to physical activity, sedentary activity, dietary intake, and obesity risk in children and adolescents—may be such a modifiable target [13–15]. During late childhood, inadequate sleep is linked with poor insulin sensitivity and high blood pressure [16, 17]. Indeed, sleep has been so consistently related to factors associated with CMD risk that the American Heart Association has updated its model of CMD prevention for children and adults by including recommendations on obtaining optimal sleep [18]. Unfortunately, most children get insufficient sleep nearly every night [19]. Experimental trials are underway to reduce obesity and CMD risk via sleep interventions in adults, adolescents, and older children [20–22], but poor sleep patterns and CMD risk factors often emerge early in childhood [23, 24], suggesting that such trials might occur too late in development.

Adolescence is an optimal time to examine CMD risk factors because they predict CMD outcomes in adulthood [24, 25]. Few studies have examined early childhood sleep patterns with obesity and CMD risk factors in adolescence [17], and the relevant broader pediatric literature has methodological weaknesses including: small samples, cross-sectional design, limited and piecemeal examination of CMD risk factors, and a focus on sleep duration while ignoring other aspects of sleep quality [16, 17, 26]. Later bedtime and low-quality sleep may impart unique risks for obesity in youth [27–29] and CMD in adults [30]. There is a critical gap in the literature identifying specific aspects of sleep most related to CMD risk in childhood. Such information could help inform preclinical models aimed to improve childhood sleep as a modifiable health target to reduce CMD risk.

We capitalized on data from a prospective birth cohort study that serially assessed caregiver reports of children’s sleep from 2 to 8 years of age and multiple CMD risk factors at 12 years of age. These data allowed us to assess whether the trajectory of children’s sleep duration, bedtime, and sleep disturbance across early childhood predicted CMD risk factors in early adolescence. To accomplish this, we first characterized the trajectories for the three sleep predictors (duration, timing, and disturbance) from 2 to 8 years. Next, we determined how each of these unique trajectories predicted CMD risk in a subset of adolescents at 12 years. The primary outcome was a previously established composite CMD risk score [31] calculated at 12 years. We hypothesized that children who had greater sleep disturbance, later bedtimes, and shorter sleep duration, or showed trajectories of worsening sleep in one or more of these areas across childhood, would have a higher composite CMD risk score at 12 years. Second, we aimed to examine how the longitudinal sleep trajectories in sleep duration, bedtime, and level of sleep disturbance relate to individual CMD risk factors.

Materials and Methods

Participants

Between March 2003 and January 2006, pregnant women were recruited to participate in the Health Outcomes and Measures of the Environment Study, a longitudinal pregnancy and birth cohort study initially created to determine whether early-life environmental toxicant exposures influence children’s health [32]. Women were identified who were living in a nine-county region centered around Cincinnati, Ohio using medical scheduling systems of nine prenatal practices. Eligibility was determined through clinical records and phone interviews with the women. Inclusion criteria included: women were <19 weeks pregnant, >18 years of age, residing in a home built in or before 1978 (not including a mobile or trailer home), HIV-negative, not taking medications for seizures or thyroid disorders, fluent in English, and planning to continue prenatal care at collaborating hospitals and continue living in the greater Cincinnati region for the next year. Exclusion criteria included maternal diagnoses of diabetes, bipolar disorder, schizophrenia, or cancer that resulted in radiation treatment or chemotherapy. Women who self-identified as black were oversampled (31%) to allow for investigation of potential health disparities.

Procedures

Following delivery, the caregiver and children completed follow-up study visits at ages 1 day, 4 weeks, and 1, 2, 3, 4, 5, 8, and 12 years of age. At these visits, anthropometric measurements were collected, and a series of questionnaires were completed. Telephone interviews were also collected frequently between in-person study visits. For present analyses, sleep data from ages 2, 2.5, 3, 4, 5, and 8 were analyzed, predicting the detailed CMD risk factor measurements conducted at age 12. Retention rates across study visits ranged from 48% (age 4) to 94% (4 weeks); retention rate at the 12-year follow-up was 58% [32, 33].

Measures

Early childhood sleep duration, bedtime timing, and sleep disturbance (longitudinal predictors).

Caregivers in this study completed an adapted version of the Children’s Sleep Habits Questionnaire (CSHQ) when their child was ages 2, 2.5, 3, 4, 5, and 8. The CSHQ was originally designed as a 45-item version for children between ages 4 and 10 [34], but was later validated for children down to 2 years of age [35]. The measure is constructed so that the caregiver is asked to reflect on a typical week’s sleep while completing the questionnaire. The CSHQ queries the frequency of sleep-related behaviors, including: difficulties in falling asleep independently or staying asleep (e.g. “child awakes more than once during the night”), speed in which the child is able to fall asleep at night (e.g. “child falls asleep within 20 minutes after going to bed”), behavioral resistance towards bedtime (e.g. “child struggles at bedtime”), bedtime fears/nightmares (e.g. “Child is afraid of sleeping alone”), or evidence of problematic nighttime behaviors, including the presence of organic sleep disorders (e.g. “child seems to stop breathing during sleep.”)

The CSHQ has adequate internal consistency (α = 0.68 for a community sample and α = 0.78 for a clinical sample [34]). As caregivers were first administered this questionnaire when their child was aged 2, nine items were not administered due to developmental inappropriateness (e.g. “Child wets the bed at night.") To maintain consistency, these items were excluded for the remainder of the study. Rather than use the cut point derived by Owens et al. to categorize “poor sleepers” [34], we derived a total sleep disturbance score by summing remaining CSHQ items; higher scores reflect greater sleep disturbance. More specifically, higher scores may reflect greater issues with falling asleep independently, greater frequency of night wakings, longer sleep onset periods, increased nighttime fears, or symptoms that may reflect organic sleep disorders (see Supplementary Table S1 for the final subset of items that were included). Within our sample, this sleep disturbance score had adequate internal consistency (α’s ranging from 0.76 to 0.80 across all waves). The CSHQ also asks the caregiver to write in the child’s typical bedtime; this was used to determine our bedtime variable. Finally, the CSHQ also assessed total sleep duration by asking for “child’s usual amount of sleep each day (combining nighttime sleep and naps).” We used this value to determine our sleep duration variable.

CMD risk factors (outcome variables at age 12).

Body mass index. Height and weight were measured in triplicate using Holtain Harpenden Stadiometer and a ScaleTronix 5002 scale (Hill-Rom Inc., Chicago, Illinois) and were averaged across all three measurements. Height and weight estimates were then used to determine body mass index, which was then converted to age- and sex-adjusted BMI z-scores.

Hip and waist circumference: Using a Gulick-II tape fiberglass, no-stretch tape measure (County Technology Inc., Gays Mills, Wisconsin), iliac crest circumference at the top of the hipbones and hip circumferences at the maximum protuberance of the buttocks were measured in triplicate and averaged across the three measurements.

Body fat: Using dual x-ray absorptiometry (Hologic Horizon densitometer; Hologic Inc., Marlborough, Massachusetts), we obtained measures of whole-body fat mass (adipose tissue, in kilograms) and visceral fat area (cross-sectional area of adipose tissue within the abdomen, measured in centimeters squared) [36]. We used the National Health and Nutrition Examination Survey body composition analysis option when deriving these estimates. Fat mass index was calculated by dividing total fat mass (kilograms) by the square of height (meters). Then converted to age- and sex-adjusted standardized z-scores using National Health and Nutrition Examination Survey estimates [31, 36].

Blood pressure: A Dinamap Pro100 automated monitor measured blood pressure on three separate occasions, 1 minute apart, occurring after a 5-minute rest period. Diastolic pressure and systolic measurements were averaged across the second and third measurements in mmHg units. We also calculated SBP percentile z-scores using the American Academy for Pediatrics Guidelines for the computation of the CMD risk score [37].

Additional cardiometabolic risk measures: We obtained a blood sample following an overnight fast, which was used to measure triglycerides, high-density lipoprotein (HDL), low-density lipoprotein (LDL), leptin, and adiponectin. Levels of detection for leptin were 0.8 ng/mL and < 2 µg/mL for adiponectin. All assays were performed by trained technicians at the Cincinnati Children’s Hospital Medical Center’s Clinical Translational Research Center Core Laboratory using assays from Roche Diagnostics (Indianapolis, Indiana): HDL (catalog #03039773), LDL (catalog #07005717), triglycerides (catalog #20767107), adiponectin (catalog #EZHADP-61K), and leptin (catalog #EZHL-805K). We calculated the Homeostatic Model Assessment for Insulin Resistance (HOMA-IR) using a standard formula (HOMA-IR = fasting insulin [milli-international units per liter] × fasting glucose [milligrams per deciliter]/405) [38]. We also calculated a triglycerides to HDL ratio and an adiponectin to leptin ratio.

Cardiometabolic risk score: As data suggest that cardiometabolic risk components such as HOMA-IR, visceral fat, and Triglyceride to HDL ratio may be more predictive of CMD risk than more traditional components (e.g. waist circumference and blood pressure) [39, 40], we focused our primary analyses on the previously established CMD risk score [31] for this study. This score is calculated by summing the age- and sex-standardized z-scores for HOMA-IR, triglyceride to HDL ratio, visceral fat area, and systolic blood pressure. Higher CMD risk scores indicate great CMD risk.

Covariates

Based on prior relevant research, we established a set of potential covariates: baseline maternal education, baseline household income, duration of breastfeeding, child sex, child pubertal stage at age 12 years (self-assessed, which has been shown to be positively associated with hormonal biomarkers of puberty) [41], child race, parent cardiometabolic history (i.e. whether a parent had a previous CMD diagnosis, assessed at the 12-year visit), maternal age at delivery of the infant, maternal serum cotinine during pregnancy (mean of measures at 16- and 26-week gestation), and maternal pre-pregnancy BMI. Trained research assistants collected covariate information using questionnaires and medical chart abstractions. To calculate pre-pregnancy BMI, we used maternal height (measured at 4-week postpartum visit) and pre-pregnancy weight (assessed via chart abstraction, or imputed if data were missing) [42]. Duration of breastfeeding (months) was measured utilizing standardized questions regarding breastfeeding practices. Using a direct acyclic graph approach, the minimal sufficient adjustment set for estimating the total effect of child sleep on CMD risk score included: child pubertal stage, child sex, duration of breastfeeding, income, maternal serum cotinine, and maternal education (Figure 1). Therefore, we adjusted for these covariates in our analysis of the association between child sleep and CMD risk.

Figure 1.

Directed acyclic graph for determining final set of covariates. Based on the minimal sufficient adjustment sets for estimating the total effect of child sleep on CMD risk score, we included child pubertal stage, child sex, duration of breastfeeding, income, maternal education, and tobacco smoke exposure in our final adjusted models.

Statistical Analysis

We first ran a group-based semi-parametric mixture model trajectory analysis on all available longitudinal sleep data to identify underlying distinct trajectories for each of the sleep measures (sleep duration, bedtime, and sleep disturbance, examined separately) for the full sample (N = 330). Briefly, the group-based trajectory analysis involved fitting the longitudinal sleep data in a mixture model and estimating the posterior probabilities of trajectory membership for each individual. The modeling algorithm accommodates missing data in longitudinal assessments, so individuals are able to be included in the analysis if they contributed at least one-time point [43–45] (89.1% of the 330 participants had two or more timepoints; Supplementary Table S2). The modeling was carried out in a stepwise fashion to determine the optimal number of groups and trajectory shape for each group. We used the Bayesian Information Criterion (with lower value indicating a better model fit) to guide the decision on optimal model. In addition, we considered average posterior probability (value > 0.7 for each group) and adequate sample numbers in each group (no <5% for each group). Individuals were assigned membership in the group to which their posterior membership probability was the largest.

Following the trajectory analysis, we used general linear models to examine the association between trajectories in sleep disturbance, bedtime, and sleep duration and the CMD risk score in the 220 adolescents with CMD data. Finally, we used separate general linear models to examine the association between distinct trajectories in sleep disturbance, bedtime, and sleep duration and our secondary outcomes: individual CMD risk factors. The adiponectin-leptin ratio, visceral fat area, triglycerides, ratio of triglycerides to LDL, and HOMA-IR were log-transformed due to the data distribution being markedly skewed. We ran all models first without the covariates (indicated as our “unadjusted” models) and then with covariates (indicated as our “adjusted” models). All analyses were conducted using SAS (Version 9.4), with trajectory modeling implemented using Proc TRAJ.

Results

Participant characteristics

A total of 330 caregiver/singleton child-dyads had at least one-time point in which the CSHQ was completed (average CSHQ completion rate across the six assessment periods was 68.53%, with the range of completion being 58.8%–84.8%; see Supplementary Table S3 for the breakdown of completion rates across measurement year). Of these 330 dyads, 220 completed at least one aspect of the cardiometabolic panel at the 12-year study visit. The full sample of 330 children in the sleep trajectory analysis and the sample of 220 children with cardiometabolic measures at the 12-year visit were very similar in demographic characteristics (Table 1). The 220 children were 54.5% female and 58.6% white, non-Hispanic, and 47.7% of households earned <$50 000 income yearly (Table 1). At age 12 years, their average CMD risk score was −0.05 (3.45), and the average body mass index z-score was 0.35 (1.17; Table 2).

Table 1.

Sample Characteristics of the Health Outcomes and Measures of the Environment (HOME) Study Sample at Baseline

	Full sample#; N = 330 (percent or mean ± SD)	CMD sample; N = 220 (percent or mean ± SD)
Female	53.9%	54.5%
Race/ethnicity of child
White, non-Hispanic	61.8%	58.6%
Black, non-Hispanic	31.2%	35.9%
Hispanic	2.4%	1.8%
Asian/Pacific	2.7%	2.7%
American Indian	1.8%	0.9%
Family yearly income at baseline ($)
50K or less	44.5%	47.7%
>50K and ≤100K	38.5%	36.8%
>100K and ≤150K	11.5%	10.0%
>150K	5.5%	5.5%
Maternal insurance status
Private health insurance	73.9%	71.7%
Public health insurance	24.3%	26.5%
No insurance	1.8%	1.8%
Maternal age at delivery (years, mean [SD])	29.6 (5.8)	29.4 (6)
Maternal pre-pregnancy BMI (kg/m2, mean [SD])	26.3 (6.5)	26.5 (6.8)
Infant weight at delivery (grams, mean [SD])	3378 (627)	3350 (616)
Maternal education
High school/GED or less	21.1%	22.4%
Some college/trade school	24.3%	29.3%
Bachelor’s degree	31.6%	29.7%
Graduate or professional school	21.9%	18.7%
Maternal marital status
Married	66.9%	64.0%
Not married, but living with someone	12.2%	11.0%
Not married, living alone	21.0%	25.1%
Cardiometabolic disease family history in parent
None	70.3%	57.7%
One or more	29.7%	42.3%
History of any breastfeeding
No	18.0%	19.2%
Yes	82.0%	80.8%
Duration of breastfeeding (weeks, median; IQR*)	18 (2–48)	16 (1–44)

^#Full sample consisted of 330 participants who have sleep data at one or more time points;

^*Median and interquartile ranges presented due to high skewness of the data.

Table 2.

Cardiometabolic Outcomes at Year 12

CMD variable	N	Mean ± SD
Cardiometabolic risk score	176	−0.05 ± 3.45
Visceral fat area (cm2)*	208	41.29 (31.12−54.04)
Whole-body fat mass index (z-score)	208	0.21 ± 0.82
Body mass index (z-score)	220	0.35 ± 1.17
Iliac to hip ratio	214	0.86 ± 0.06
Triglycerides (mg/dL)*	179	74.00 (59.00−97.70)
High-density lipoprotein (mg/dL)	179	53.23 ± 12.19
Low-density lipoprotein (mg/dL)	179	86.35 ± 25.84
Ratio of triglycerides to low-density lipoprotein (median, IQR*)	179	1.38 (1.05−2.01)
HOMA-IR*	179	2.75 (1.90−4.22)
Mean blood pressure—systolic (mm/Hg)	214	101.64 ± 9.09
Leptin-adiponectin ratio (median, IQR*)	178	1415.15 (578.99−4246.29)

^*Median and IQR reported due to high skewness of data.

Sleep trajectories

Using a group-based semi-parametric mixture model, we identified distinct latent trajectories across the entire sample in the areas of sleep disturbance, bedtime, and sleep duration. Specifically, when examining levels of sleep disturbance across ages 2–8 years, we observed three distinct trajectories over time: low sleep disturbance (N = 176; 53.3%), medium sleep disturbance slightly decreasing over time (N = 88; 26.7%), and high sleep disturbance (N = 66; 20.0%; see Figure 2A). When examining the timing of bedtime across ages 2–8 years, we also observed three distinct categories over time: early bedtime (N = 126; 38.2% of the sample), moderately late bedtime (N = 171; 51.8% of the sample), and late bedtime slightly shifting earlier over time (N = 33; 10.0% of the sample; Figure 2B). Finally, when examining the duration of sleep across ages 2–8 years, we observed two distinct trajectories over time: short sleep duration (N = 64; 19.4%) and long sleep duration (N = 266; 80.6%; Figure 2C).

Figure 2.

(A) Sleep disturbance trajectories score across ages 2–8. (B) Bedtime tractories across ages 2–8. (C) Sleep duration trajectories across ages 2–8.

Sleep trajectories predicting primary outcome: cardiometabolic risk scores

Using general linear modeling for the subsample with CMD risk scores at age 12, we observed that within the unadjusted model, participants in the high sleep disturbance trajectory had a significantly greater CMD risk score mean (1.51; 95% CI [0.40, 2.63]) compared to those in the low (−0.55; 95% CI [−1.21, 0.12]) and medium (−0.20; 95% CI [−1.21, 0.81]) sleep disturbance trajectories (p’s = 0.002, 0.027, respectively; Table 3). After adjusting for covariates, those in the high sleep disturbance trajectory continued to have a significantly greater CMD risk score mean (1.36; 95% CI [0.11, 2.61]) compared with those in the low sleep disturbance group (−0.17; 95% CI [−1.04, 0.71]); p = 0.039); no significant differences were observed between the low and medium sleep disturbance trajectories in CMD risk scores after adjusting for covariates (p = 0.079; Table 3).

Table 3.

Associations Between Sleep Trajectories and CMD Risk Index Score Across Sleep Domains

	Unadjusted			Adjusted #
	LS mean (95% CI)	P-value (comp group 1)	P-value (comp group 2)	LS mean (95% CI)	P-value (comp group 1)	P-value (comp group 2)
Sleep disturbance
Low (N = 176)	−0.55 (−1.21, 0.12)	—	—	−0.17 (−1.04, 0.71)	—	—
Medium (N = 88)	−0.20 (−1.21, 0.81)	0.575	—	−0.05 (−1.15, 1.05)	0.854	—
High (N = 66)	1.51 (0.40, 2.63)	0.002	0.027	1.36 (0.11, 2.61)	0.039	0.079
Bedtime
Early (N = 126)	−0.49 (−1.33, 0.35)	—	—	0.04 (−0.96, 1.05)
Moderately late (N = 171)	−0.05 (−0.77, 0.68)	0.436	—	0.16 (−0.71, 1.04)	0.840
Late (N = 33)	0.99 (−0.32, 2.31)	0.065	0.178	0.79 (−0.68, 2.26)	0.397	0.438
Sleep duration
Short (N = 64)	−0.05 (−1.21, 1.11)	—	—	−0.12 (−1.37, 1.12)	—	—
Long (N = 266)	−0.05 (−0.62, 0.52)	0.999	—	0.33 (−0.43, 1.08)	0.513	—

^#Adjusted for child pubertal stage, child sex, duration of exclusive breastfeeding, income, maternal education, and maternal serum cotinine.

In the unadjusted model, participants in the late bedtime trajectory approached significance in having a greater CMD risk score mean (0.99; 95% CI [−0.32, 2.31]) compared to those in the early bedtime trajectory group (−0.49; 95% CI [−1.33, 0.35]; p = 0.065), but this relationship was attenuated after adjusting for covariates (p = 0.397; Table 3). No other comparisons between trajectory groups within bedtime or sleep duration emerged as significant when predicting CMD risk scores (p’s > 0.178; Table 3).

Sleep trajectories predicting secondary outcomes: individual cardiometabolic disease risk markers

Those in the high sleep disturbance trajectory had higher estimates in several CMD risk markers. Specifically, compared with participants in the low sleep disturbance trajectory, those in the high sleep disturbance trajectory had greater whole-body fat mass index z-scores (unadjusted p = 0.007; adjusted p = 0.043) and BMI z-scores (unadjusted p < 0.001; adjusted p = 0.028; Table 4). In the unadjusted models only, those in the high sleep disturbance trajectory had greater HOMA-IR indexes (unadjusted p = 0.001; adjusted p = 0.090), visceral fat (unadjusted p = 0.048; adjusted p = 0.073), and leptin-adiponectin ratio (unadjusted p = 0.039; adjusted p = 0.356) compared with those in the low sleep disturbance trajectory. Additionally, we observed similar trends when comparing those in the medium sleep disturbance trajectory with those in the low sleep disturbance trajectory, specifically in the BMI z-scores (unadjusted p = 0.026; adjusted p = 0.309), HDLs (unadjusted p = 0.047; adjusted p = 0.089), and HOMA-IR (unadjusted p = 0.029; adjusted p = 0.208) in the unadjusted models only (Table 4). We did not observe a significant impact of sleep disturbance trajectories and the CMD markers of iliac to hip ratio, LDL, systolic blood pressure, triglycerides, and triglycerides to HDL ratio (unadjusted p’s > 0.087, adjusted p’s > 0.107).

Table 4.

CSHQ Total Sleep Disturbance Trajectories and Cardiometabolic Markers

	Unadjusted			Adjusted#
Dependent variable	Low disturbance LS mean (95% CI)	Med disturbance LS mean (95% CI)	High disturbance LS mean (95% CI)	Low disturbance LS mean (95% CI)	Med disturbance LS mean (95% CI)	High disturbance LS mean (95% CI)
Visceral fat## (cm2)	39.27 (36.22, 42.57)	42.39 (37.75, 47.59)	45.84 (40.26, 52.19)*	41.08 (37.30, 45.24)	43.03 (38.20, 48.48)	47.49 (41.48, 54.37)
Whole-body fat mass index (z-score)	0.09 (−0.06, 0.24)	0.24 (0.03, 0.46)	0.49 (0.25, 0.73)**	0.17 (−0.03, 0.36)	0.29 (0.05, 0.53)	0.49 (0.22, 0.76)*
Body mass index (z-score)	0.14 (−0.06, 0.35)	0.36 (0.06, 0.66)	0.86 (0.54, 1.19)***,˚	0.29 (0.03, 0.54)	0.47 (0.15, 0.79)	0.75 (0.40, 1.10)*
Iliac-to-hip ratio	0.86 (0.84, 0.87)	0.86 (0.84, 0.88)	0.85 (0.83, 0.86)	0.85 (0.84, 0.87)	0.86 (0.84, 0.88)	0.85 (0.83, 0.87)
LDL(mg/dL)	87.95 (82.85, 93.05)	82.83 (75.26, 90.39)	86.37 (77.79, 94.95)	85.52 (78.86, 92.19)	81.29 (73.13, 89.46)	86.09 (76.72, 95.47)
HDL(mg/dL)	53.79 (51.41, 56.18)	54.93 (51.39, 58.47)	49.48 (45.47, 53.49)˚	53.65 (50.46, 56.83)	55.22 (51.32, 59.12)	50.36 (45.89, 54.86)
HOMA_IR##	2.51 (2.23, 2.82)	2.75 (2.31, 3.27)	3.70 (3.03, 4.51)***,˚	2.74 (2.37, 3.17)	2.86 (2.39, 3.42)	3.37 (2.75, 4.13)
Systolic blood pressure (mm/Hg)	101.21 (99.53, 102.88)	100.97 (98.57, 103.37)	103.51 (100.89, 106.14)	101.43 (99.34, 103.52)	100.89 (98.30, 103.48)	102.65 (99.79, 105.51)
Triglycerides## (mg/dL)	78.66 (72.90, 85.82)	72.47 (63.68, 82.48)	82.90 (71.59, 95.99)	75.15 (66.96, 84.37)	70.78 (61.43, 81.55)	83.39 (70.87, 98.12)
TG:HDL ratio##	1.50 (1.34, 1.68)	1.35 (1.15, 1.60)	1.72 (1.42, 2.07)	1.44 (1.24, 1.68)	1.32 (1.10, 1.58)	1.70 (1.38, 2.10)
Leptin-adiponectin ratio##	1746.65 (1346.80, 2265.21)	1505.92 (1019.64, 2224.1)	1019.47 (658.4, 1578.54)*	1429.56 (1030.22, 1983.70)	1335.96 (890.69, 2003.84)	1111.57 (701.33, 1761.79)

^#Adjusted for child pubertal stage, child sex, duration of exclusive breastfeeding, income, maternal education, and maternal serum cotinine;

^##Geometric LS mean and 95% CI reported; *p < 0.05 compared to group “Low Disturbance”; **p < 0.01 compared to group “Low Disturbance”; ***p < 0.001 compared to group “Low Disturbance”; ˚p < 0.05 compared to group “Med Disturbance”; ˚˚p < 0.01 compared to group “Med Disturbance”; ˚˚ ˚˚p < 0.01 compared to group “Mid Disturbance.”

Participants in the late bedtime trajectory had higher HOMA-IR index scores in the unadjusted models only (unadjusted p = 0.004; adjusted p = 0.080; Table 5) compared with those in the early bedtime trajectory; all other relationships between bedtime trajectories and CMD markers were nonsignificant (unadjusted p’s < .061; adjusted p’s < 0.245). Those in the long sleep duration trajectory had higher blood pressure estimates than those in the short sleep duration trajectory in the adjusted model only (unadjusted p = 0.084, adjusted p = 0.041). No other significant associations were noted between sleep duration trajectories and CMD risk estimates (unadjusted p’s > 0.084, adjusted p’s > 0.248; Table 6).

Table 5.

CSHQ Bedtime Trajectories and Cardiometabolic Markers

	Unadjusted			Adjusted#
Dependent variable	Early bedtime LS mean (95% CI)	Middle bedtime LS mean (95% CI)	Late bedtime LS mean (95% CI)	Early bedtime LS mean (95% CI)	Middle bedtime LS mean (95% CI)	Late bedtime LS mean (95% CI)
Visceral fat##(cm2)	39.93 (36.19, 44.05)	41.76 (38.38, 45.44)	43.83 (37.38, 51.39)	41.80 (37.53, 46.55)	43.37 (39.43, 47.69)	45.49 (38.66, 53.54)
Whole-body fat mass index (z-score)	0.13 (−0.05, 0.31)	0.22 (0.06, 0.38)	0.40 (0.10, 0.70)	0.22 (0.00, 0.44)	0.27 (0.08, 0.46)	0.45 (0.12, 0.77)
Body mass index (z-score)	0.14 (−0.11, 0.39)	0.44 (0.22, 0.66)	0.61 (0.19, 1.03)	0.34 (0.06, 0.62)	0.48 (0.23, 0.73)	0.66 (0.22, 1.10)
Iliac-to-hip ratio	0.85 (0.84, 0.87)	0.85 (0.84, 0.86)	0.87 (0.85, 0.89)	0.85 (0.83, 0.86)	0.85 (0.84, 0.87)	0.87 (0.84, 0.89)
LDL(mg/dL)	87.89 (81.65, 94.13)	86.97 (81.50, 92.44)	80.64 (70.88, 90.39)	86.16 (78.77, 93.54)	85.90 (79.34, 92.46)	77.21 (66.60, 87.84)
HDL(mg/dL)	53.90 (50.14, 56.03)	53.98 (51.4, 56.56)	51.22 (46.61, 55.83)	52.49 (48.93, 56.05)	54.39 (51.23, 57.55)	52.06 (46.94, 57.18)
HOMA-IR##	2.35 (2.04, 2.72)	2.91 (2.56, 3.30)*	3.52 (2.81, 4.42)**	2.64 (2.23, 3.08)	3.00 (2.60, 3.46)	3.35 (2.66, 4.23)
Systolic blood pressure (mm/Hg)	101.31 (99.29, 102.34)	102.13 (100.40, 103.85)	100.74 (97.42, 104.06)	101.96 (99.64, 104.27)	101.72 (99.71, 103.74)	100.20 (96.66, 103.74)
Triglycerides## (mg/dL)	79.48 (71.39, 88.48)	76.35 (69.51, 83.88)	78.72 (66.56, 93.1)	76.62 (67.31, 87.23)	75.12 (66.96, 84.28)	75.75 (62.87, 91.26)
TG:HDL ratio##	1.54 (1.34, 1.77)	1.45 (1.28, 1.64)	1.58 (1.27, 1.96)	1.50 (1.27,1.78)	1.42 (1.22, 1.65)	1.50 (1.18, 1.91)
Leptin-adiponectin ratio##	1609.49 (1169.61, 2214.79)	1650.42 (1247.77, 2183.00)	977.26 (587.64, 1625.20)	1298.38 (902.34, 1868.25)	1468.95 (1063.98, 2028.06)	1021.34 (597.54, 1745.71)

^#Adjusted for child pubertal stage, child sex, duration of exclusive breastfeeding, income, maternal education, and maternal serum cotinine;

^##Geometric LS mean and 95% CI reported; *p < 0.05 compared to group “Early Bedtime”; **p < 0.01 compared to group “Early Bedtime”; ***p < 0.001 compared to group “Early Bedtime”; ˚p < 0.05 compared to group “Middle Bedtime”; ˚˚p < 0.01 compared to group “Middle Bedtime”; ˚˚ ˚˚p < 0.01 compared to group “Middle Bedtime.”

Table 6.

CSHQ Sleep Duration and Cardiometabolic Markers

	Unadjusted		Adjusted#
Dependent variable	Short sleep duration LS mean (95% CI)	Long sleep duration LS mean (95% CI)	Short sleep duration LS mean (95% CI)	Long sleep duration LS mean (95% CI)
Visceral fat##(cm2)	40.35 (35.24, 46.21)	41.60 (38.94, 44.40)	41.66 (36.32, 47.79)	43.49 (40.07, 47.21)
Whole-body fat max index (z-score)	0.26 (0.01, 0.51)	0.20 (0.08, 0.33)	0.27 (−0.01, 0.54)	0.28 (0.12, 0.44)
Body mass index (z-score)	0.61 (0.27, 0.96)	0.29 (0.12, 0.46)	0.62 (0.26, 0.98)	0.40 (0.19, 0.62)
Iliac-to-hip ratio	0.85 (0.84, 0.87)	0.85 (0.85, 0.86)	0.85 (0.83, 0.87)	0.85 (0.84, 0.87)
LDL(mg/dL)	81.61 (73.06, 90.16)	87.51 (83.29, 91.72)	81.74 (72.56, 90.92)	85.34 (79.64, 91.05)
HDL(mg/dL)	53.70 (49.65, 57.75)	53.12 (51.12, 55.12)	54.46 (50.05, 58.88)	52.94 (50.20, 55.69)
HOMA_IR##	3.05 (2.49, 3.74)	2.70 (2.45, 2.99)	2.92 (2.38, 3.57)	2.91 (2.57, 3.30)
Systolic blood pressure (mm/Hg)	99.50 (96.80, 102.20)	102.18 (100.83, 103.54)	99.09 (96.22, 101.96)	102.36 (100.62, 104.10)*
Triglycerides## (mg/dL)	73.44 (63.41, 85.04)	78.96 (73.45, 84.89)	72.28 (61.59, 84.83)	76.88 (69.61, 84.92)
TG:HDL ratio##	1.40 (1.16, 1.69)	1.53 (1.39, 1.67)	1.36 (1.11, 1.68)	1.50 (1.32, 1.70)
Leptin-adiponectin ratio##	1104.38 (712.52, 1711.74)	1636.39 (1317.44, 2032.55)	1134.42 (722.78, 1780.51)	1386.15 (1047.41, 1834.43)

^#Adjusted for child pubertal stage, child sex, duration of exclusive breastfeeding, income, maternal education, and maternal serum cotinine;

^##Geometric LS mean and 95% CI reported; *p < 0.05 compared to group “Short Sleep Duration”; **p < 0.01 compared to group “Short Sleep Duration”; ***p < 0.001 compared to group “Short Sleep Duration.”

Discussion

We found that children who had consistently higher levels of sleep disturbance during childhood had an increased CMD risk score in early adolescence compared with those who had consistently low levels of sleep disturbance. This increased risk persisted after adjusting for adolescent pubertal stage, sex, duration of breastfeeding, maternal education, and maternal serum cotinine. Interestingly, we did not observe altered CMD risk scores in either adjusted or unadjusted models based on bedtime or sleep duration obtained over time, suggesting that early childhood sleep disturbance may play a unique role in increasing adolescent CMD risk, above and beyond how much and when children sleep.

We observed distinct trajectories in sleep disturbance, bedtime, and sleep duration. Most children had low levels of sleep disturbance (as reflected by a lower CSHQ sleep disturbance score), a moderate bedtime (~9:00 pm), and a longer sleep duration that decreased over time (as one would expect with a child aging, beginning with ~13 hours per night at age 2 and decreasing to ~10 hours per night by age 8). This suggests that most children within our sample obtained adequate sleep across this 6-year developmental period, meeting the recommendations set by the National Sleep Foundation [46]. Only 10%–20% of our sample fell within the high sleep disturbance trajectory, the late bedtime trajectory (with bedtime ranging from ~9:30 pm to ~11:00 pm across time), or the short sleep duration trajectory (beginning with ~10.5 hours at age 2 per night and decreasing to ~8.75 hours per night by age 8). The patterns and percentages of our cohort with disrupted sleep were comparable with those observed in other longitudinal cohorts [47, 48].

Children in the highest trajectory of sleep disturbance had higher adolescent CMD risk scores (compared to children with the lowest trajectory of sleep disturbance), as well as increased CMD risk in several individual cardiometabolic markers, including increased body mass and body fat (in both the adjusted and unadjusted models). This suggests that children who demonstrate consistent, high amounts of sleep disturbance (e.g. difficulties in falling or staying asleep independently, quickly, or consistently night to night, evidence of behavioral resistance towards bedtime, bedtime fears/nightmares, or evidence of parasomnias or organic sleep disorders) are at increased risk for CMD development. This finding expands upon published cross-sectional and longitudinal research showing that sleep disturbances are associated with greater waist circumference in children 8–17 years old [49–51]. Furthermore, sleep disturbance appears to play a unique role in development of overweight/obesity (even independent of sleep duration) [52], though the majority of relevant research has been cross-sectional. Our findings suggest that future longitudinal research should focus on the unique impact of child sleep disturbance on CMD risk.

We also observed higher levels of insulin resistance, visceral fat, and leptin-adiponectin ratio in the higher sleep disturbance trajectory (compared with the lowest sleep disturbance trajectory) but only in the unadjusted models (though we noted a trend in a similar direction in the adjusted models). The literature is inconsistent about the relationship between sleep disturbance and insulin resistance in children and adolescents, but most studies were cross-sectional [53]. In one longitudinal study of 8–11-year-old children, Hjorth et al. [51] observed that poorer sleep quality predicted greater insulin resistance. Minimal research has been conducted on sleep quality and CMD risk factors of visceral fat and leptin-adiponectin ratio (most have examined the relationship on sleep duration), but generally speaking, poorer sleep is frequently associated with increased visceral fat in children [17], (likely due to increased sugar consumption in the late evening hours [15, 54, 55]), while the relationship with sleep and leptin and adiponectin is less clear [56].

Most research examining the role of sleep and CMD risk has focused on the duration of children’s sleep [16]. Interestingly, we found minimal differences in CMD risk for children with short sleep duration over time, compared to those with long sleep duration over time. One exception is that we did observe a small increase in systolic blood pressure for children who obtained longer sleep over time. Our largely null findings are in line with the current research literature; four recent systematic and meta-analytic reviews have examined the association between children’s sleep duration and CMD risk, and the evidence for short sleep increasing CMD risk in preschool and school-aged children was either mixed or nonsignificant [26, 57–59]. Thus simply increasing the duration of sleep in a population where most young children are sleeping the recommended amount may be insufficient to improve CMD-related health outcomes; rather, there should be a greater focus on assessing for and treating the presence of sleep disturbance.

Interestingly, while bedtime trajectory was not predictive of our composite adolescent CMD risk score, children in the late bedtime trajectory had higher levels of insulin resistance than those with an earlier bedtime trajectory, though not after adjusting for covariates. While prior research has not examined the role of a young child’s bedtime on their later cardiometabolic risk, one recent adolescent study found that adolescents with later sleep timing had increased insulin resistance, compared to those with earlier sleep timing [60]. Furthermore, adolescents with later weekend bedtimes had higher BMI z-scores [61]. These results have led to calls to focus on sleep timing when promoting health, rather than simply sleep duration [62, 63].

These findings indicate that assessment and treatment of sleep disturbances (and perhaps late bedtime) early in childhood may be useful in lowering later adolescent CMD risk. Primary care providers can build assessment for sleep disturbances routinely into child well-checks and provide evidence-based intervention, when necessary. Most pediatricians agree that patients should be counseled on sleep, but only 18% reported ever receiving formal training on pediatric sleep disorders and, when quizzed about sleep health knowledge, most pediatricians answered the questions incorrectly [64]. We encourage pediatricians to ask questions about whether the child (1) has difficulty falling asleep, whether due to an inability to initiate sleep independently due to maladaptive sleep onset associations or due to behavioral limit setting issues, (2) whether the child has difficulty staying asleep, whether due to seeking out a parent in the night or because of organic sleep concerns (e.g. snoring, pauses in breathing, and restlessness), (3) whether the child has significant bedtime fears or frequent nightmares that interfere with sleep, and (4) whether there are any concerns regarding the child’s timing of sleep (i.e. falling asleep at the same time each night, falling asleep quickly, falling asleep at a developmentally appropriate bedtime). To facilitate comfort in screening, a number of brief questionnaires can be built into medical practice to routinely and systematically screen for sleep disturbances, including the CSHQ that was used in this study [34] (45 items long, with norms available for children ages 2–10), or the pediatric insomnia severity scale (six items long, available for children ages 2–10 or 11–18) [65]. Additionally, public health measures to help decrease sleep disruption should also be encouraged, including later school start times [66], particularly for minority populations who disproportionately demonstrate disrupted sleep in childhood [67].

This study had several strengths. First, our prospective and longitudinal design allowed for examination of sleep across multiple years across childhood and the examination of how these trajectories predict later CMD risk. Second, our study examined how multiple dimensions of sleep over time relate to CMD risk; previous studies examining this topic have relied on cross-sectional methodologies or have focused on a singular domain of sleep [16]^. Third, this study uncovered novel indexes of early childhood sleep trajectories that may be used in future research to reduce multiple comparisons. Fourth, we used fasting blood samples to derive our CMD outcomes, a gold standard practice that is often overlooked in younger populations but essential for accurate assessment of glucose and insulin levels. Finally, our inclusion of a CMD risk score [31] that includes standardized variables that have been strongly associated with CMD (i.e. HOMA-IR, triglyceride to HDL ratio) [40, 68] strengthens our conclusions about the relationship between sleep and CMD risk.

Our study had some limitations. First, as is the case with all longitudinal studies, we experienced loss in follow-up at the year 12 assessment. While the moderate sample size may have reduced power to detect effects of sleep dimensions on CMD risk, our sample benefitted from increased family participation at year 12 than previous years (4, 5, and 8). Additionally, we did not detect significant differences between the full sample and those with cardiometabolic measurements, though acknowledge that bias may still be present without statistically significant differences in our primary outcomes. Second, our exclusion of several questions in the CSHQ limits our ability to draw clinical conclusions regarding sleep disturbance and serves as a significant limitation in this study. Specifically, the CSHQ was initially designed with a clinical cut-point. Because we removed several questions that were developmentally inappropriate during very early childhood, we were unable to determine the clinical threshold of sleep disturbance, and instead, we relied on the sleep disturbance score as a continuous variable where higher scores reflect more disturbance; future research should include all items of the CSHQ to better utilize the clinical threshold in understanding sleep disturbance. Third, we relied on caregiver reports of sleep across all measurement timepoints; caregivers may become less reliable reporters of their child’s sleep as the child ages. Additional sleep assessment methods, such as actigraphy or child self-report of sleep disturbances [69, 70], may be warranted in future studies. Fourth, we did rely on unadjusted model estimates in our interpretation of some study findings; it is possible that after adjusting for relevant covariates, individual CMD components may not be significantly related to sleep disturbance unless added together into a more comprehensive CMD risk score. Finally, by examining individual CMD components, we increased the likelihood of type 1 error. This potential of false–positive results, paired with the novelty of this study design and findings, necessitates replication of our study findings. We encourage future research to use the novel CMD risk score used in this article to address the fourth and final limitations.

Despite these limitations, this study is amongst the first to examine sleep across early childhood to predict adolescent CMD risk using gold-standard CMD risk assessment methodologies. We found that children who show a high level of sleep disturbance across childhood have increased CMD risk in early adolescence. Additionally, we found that the timing of bedtime and sleep duration had less of an influence on adolescent CMD risk. Though further replication is needed, these findings suggest that assessment and treatment of sleep disturbances in early childhood may be beneficial in lowering adolescent CMD risk.

Funding

This study was funded by the National Institute of Environmental Health Sciences grants PO1 ES11261, R01 ES014575, R01 ES020349, R01 ES027224, and R01 ES025214.

Disclosure Statement

Financial disclosure: None. Nonfinancial disclosure: None.