Measurements

At 6 months and yearly from 1 to 7 years, mothers reported their children’s sleep duration in a usual 24-hour period.⁵ The main exposure was insufficient sleep at three age periods, 6 months to 2 years, 3 to 4 years, and 5 to 7 years. We first averaged sleep hours/day during each of these three age periods. Based on age-specific sleep recommendations from the National Sleep Foundation,¹³ we then categorized sleep in each period and defined insufficient sleep duration at each time period as follows: from 6 months to 2 years, sleep < 11 hours/day or 11-<12 vs. ≥ 12; from 3 to 4 years, sleep < 10 hours/day or 10-<11 v. ≥11; and from 5 to 7 years, sleep < 9 hours/day or 9-<10 v. ≥ 10.

The main outcomes were mother- and teacher-reports of child executive function, behavior, and social-emotional functioning in mid-childhood (median 7.7 years). To assess executive function, mothers and teachers were mailed the self-administered Behavioral Rating Inventory of Executive Function (BRIEF),¹⁴ a validated 86-item questionnaire designed to assess executive function behaviors in home and school environments. The BRIEF includes the following sub-scales: inhibit, shift, emotional control, initiate, working memory, plan/organize, organization of materials, and monitor. The sub-scales form 2 broadband indexes: (1) the behavioral regulation index, which indicates the ability of the child “to shift cognitive set and modulate emotions and behavior via appropriate inhibitory control” and (2) the metacognition index, which reflects the child’s ability to “initiate, plan, organize, and sustain future-oriented problem-solving in working memory.” The BRIEF indices are each scaled to a mean of 50 and standard deviation of 10. The global executive composite is the average of the 2 indices, representing a summary measure of executive function. Higher BRIEF scores represent poorer executive function.

To assess child behavior and social-emotional functioning also in mid-childhood, mothers and teachers were mailed the self-administered Strengths and Difficulties Questionnaire (SDQ), a validated 25-item questionnaire designed to assess children’s social, emotional, and behavioral functioning.¹⁵ The SDQ is used widely in research and clinical settings,¹⁶ and has five subscales: prosocial behavior, hyperactivity/inattention, emotional symptoms, conduct problems, and peer relationship problems. Possible scores range from 0–40 points. Higher total difficulties scores (with the exclusion of the prosocial scale) indicate greater difficulties. Normative data for the SDQ derive from a representative sample of US children.¹⁷

At enrollment, we collected information about maternal age, education, parity, and household income. We collected child’s race and ethnicity in early childhood. In mid-childhood (median 7.7 years), we administered the Home Observation Measurement of the Environment short form (HOME-SF),¹⁸ which assesses cognitive stimulation and emotional support in the environment. Possible scores range from 0 to 22. Higher scores indicate environments more supportive of development. In mid-childhood, we also asked parents to report the number of hours their children watched TV/videos on an average weekday and weekend day in the past month. Response categories included, “none, < 1 hour a day, 1–3 hours a day, 4–6 hours a day, 7–9 hours a day, and ≥10 hours a day”. We did not ask specifically about the content of the programming viewed.

We first examined bivariate relationships of children’s sleep duration in each age period with each covariate and with our neurobehavioral outcomes. We also examined the correlation of sleep in infancy (6 months to 2 years) with sleep at 3–4 years and 5–7 years using Pearson correlation. We then used multivariable linear regression models to examine the associations of insufficient sleep in each age period with the neurobehavioral outcomes with and without the inclusion of potential confounders. Our first model, Model 1, was adjusted for child age and sex only. We then additionally adjusted the multivariable models for potential confounders including sociodemographic factors (maternal age, parity; parental education, household income, and HOME-SF score; and child race/ethnicity) and child television viewing at mid-childhood (Model 2). The multivariate models from 3 to 4 years were adjusted for sleep from 6 months to 2 years; models from 5 to 7 years were adjusted for sleep from 6 months to 2 years and 3 to 4 years.

The confounding variables in our analyses were not available for all subjects. We therefore used multiple imputation to generate plausible values for each missing value.^19,20 We used a chained equations approach with predictive mean matching based on linear regressions for approximately continuous variables and logistic or generalized logistic regression for dichotomous or more generally categorical variables. The “completed” data set comprises the observed data and one imputed value for each missing value. We replicated this analysis across completed data sets and then combined them in a structured fashion that accurately reflects the true amount of information in the observed data, i.e., without erroneously presuming that the imputed values are known true values, but recovering the information in partially observed subjects. We generated 50 complete data sets²¹ and combined multivariable modeling results (Proc MI ANALYZE) in SAS version 9.3 (SAS Institute, Cary NC). From these multiple imputation results, we report adjusted effect estimates from regressions and 95% confidence intervals for each sleep category with the lowest risk sleep category as the reference group.

Given differences in participant characteristics by sleep duration, we also considered whether a lack of covariate overlap between exposed (sleep duration 3 to 4 years <10 hours/day) and unexposed (sleep duration >=10 hours/day) drove our results. We used propensity scores to define overlapping covariate values, or “common support.” We ran common-support regression after excluding 16 participants where one or the other exposure group provided few data; results were similar so we do not report them.²²