In the first round of analyses, we analyzed the weight-sleep-biomarker interrelationships from a weight perspective. The median respiratory disturbance index (RDI) was 0.7 (interquartile range: 0.38–1.4), and values were significantly higher for obese children (normal weight, 0.9 ± 1.9; overweight, 0.8 ± 1.3; obese, 2.4 ± 4.5; F2,244 = 6.2; P = .003). The groups were comparable in their distribution according to gender and race/ethnicity, but the obese group was slightly older; therefore, when applicable, the subsequent analyses included age as a covariate ().
Age, Gender, and Race/Ethnicity Distributions Among Normal-Weight, Overweight, and Obese Children
TST fluctuated around 8 hours, irrespective of school day or weekend day. However, substantial variability emerged in TST (Supplemental Table 5 and ). Among normal-weight, overweight, and obese groups, comparisons of daily TST values showed no significant differences either for weekdays or for weekends (Supplemental Table 5). However, analysis of SV values within BMI z score-defined groups revealed that, for obese children, TST on weekends was more variable than that on school days (t195 = 2.4; P = .016), compared with children with normal BMI z scores (t296 = −1.30; P = .20) or overweight children (t102 = 0.68; P = .49), for whom sleep duration on school days versus weekends was rather stable (Supplemental Table 5). Daily TST comparisons across the 3 BMI z score-defined groups revealed no significant differences, but TST on weekend days was shorter for the obese children (analysis of covariance, P = .03) ( and Supplemental Table 5).
Mean TSTs for normal-weight, overweight, and obese children. *P < .05.
A moving average plot depicting the trend characteristics of sleep duration (or SV) for each of the 3 weight groups is shown in . Normal-weight children displayed rather regular sleep duration during the week, with a tendency for longer sleep on weekends. In contrast, although sleep duration seemed regular on school days for obese children, sleep duration became increasingly shorter as the week progressed, reaching maximal differences on the weekend. For overweight children, a mixed pattern emerged; these children slept longer as the week progressed and further prolonged their sleep over the weekend. A group-day effect (F14,2135 = 2.07; P = .01) was found, in which Friday night was the turning point (F2,304 = 3.7; P = .03), with the overweight and obese children being radically different from each other (). shows the mean levels of the metabolic markers for the normal-weight, overweight, and obese groups.
One-week trends of sleep durations for normal-weight, overweight, and obese children. *P < .05.
Metabolic Marker Levels Among Normal-Weight, Overweight, and Obese Children
BMI z scores were not partially correlated (age as covariate) with sleep duration for normal-weight or obese children, but a significant association occurred for overweight children (school days: r = 0.31; P < .05; weekend days: r = 0.29; P < .05) (Supplemental Table 6).
Metabolic marker levels were partially correlated (age as covariate) with sleep duration for the normal-weight, overweight, and obese groups. For obese children, the variability in sleep duration on school days was positively correlated with triglyceride levels (r = 0.31; P < .05), with no other significant associations being found (Supplemental Table 7).
In a second round of analyses, we analyzed the weight-sleep-biomarker level interrelationships from a sleep perspective. All children were regrouped on the basis of their sleep patterns after we standardized TST with adjustment for age (ie, created TST z scores for our sample), thus controlling for potential age fluctuations, to determine a cutoff value for TST at which health is at risk (eg, analogous to a BMI z score). We used 3 cutoff values that is, 1 SD, 1.5 SDs, and 2 SDs above and below the mean sleep duration for weekdays and weekends. Accordingly, 9 sleep pattern groups could be defined (Supplemental Fig 3 and Supplemental Table 9).
Gender distributions over the sleep pattern groups were equal, irrespective of the 3 cutoff approaches. Age differences were found between the NS and NN groups with the 1-SD cutoff value, with the latter being younger (F8,297 = 2.1; P = .04), and between the NS and LN groups with the 1.5-SD cutoff value, with the latter being younger (F6,299 = 2.9; P = .01) (Supplemental Table 8). Furthermore, with respect to race/ethnicity distribution, there was a trend toward significantly more black children in the SS and LS groups with the 1-SD (χ162 = 32.7; P = .008) and 1.5-SD (χ122 = 21.4; P = .04) cutoff values. RDI values were not different among sleep profile groups (Supplemental Table 8). No statistical differences among sleep pattern groups were found in BMI z scores (Supplemental Table 8), with similar variances in BMI z scores occurring for the sleep pattern groups. Finally, to illustrate further the interactions between BMI z scores, sleep duration on school days and weekend days, and the effect of SV, a surface plot that depicts the complex interrelationships between BMI z scores and sleep patterns was constructed (Fig 4).
Close inspection of the metabolic variables in the context of the 9 sleep subgroups, as defined above, suggested that optimal sleep was best represented by the LL group. With this latter group as the reference, LN, SN, and SS groups demonstrated significant associations with insulin, LDL, and CRP levels ().
Correlations Between Mean Weekly Sleep Duration and Metabolic Marker Levels for 9 Sleep Pattern Subgroups
On the basis of the observation that BMI z scores and metabolic marker levels were not associated in the LL group, this group was defined as the reference group (). BMI z scores were moderately correlated with metabolic marker levels in the NN group, irrespective of the SD cutoff values used. Similarly, BMI z scores was significantly associated with CRP levels and insulin levels in the SN and SS groups ().
Correlations Between Metabolic Marker Levels and BMI z Scores for 9 Sleep Pattern Subgroups