Methods
Subjects
The Department of Pediatrics at Texas Tech University Health Sciences Center operates both general and subspecialty pediatric clinics. Ages of the patients range from 4 weeks to 18 years; 55% of them are girls. The ethnicity distribution is black (8%), Hispanic (47.6%), white (38.1%), and other (6.3%). The payer sources include commercial insurance (27.7%), Medicaid (57.5%), self-pay (5.7%), government programs (8.6%), and miscellaneous (0.5%). In these clinics, pediatric patients whose BMIs exceed the 90th percentile for their age and sex are referred to the Healthy Kids Program in the same clinic for further evaluation and intervention by a licensed dietitian. Siblings at risk for obesity and children whose growth curve indicates that they are likely to meet the definition of obesity within the next 2 years also are referred to this program. The dietitian's initial evaluation includes an interview with the children and parents and distribution of questionnaires related to dietary habits and other behavioral patterns. These children were counseled on healthy dietary and lifestyle activities related to obesity at the initial visit and at follow-up visits every 2 to 4 weeks. We obtained the charts of 77 children enrolled in the Healthy Kids program. The primary goal of this pilot study was to identify possible relations between sleep habits and obesity in this population of obese and at-risk children preintervention; we only analyzed information from the children's initial visits to the dietitian. This study was approved by the institutional review board at Texas Tech University Health Sciences Center.
Measures
The outcome of interest was the child's BMI at time of the initial dietitian visit. Pediatric obesity usually is determined by whether a child's BMI exceeds the 95th percentile for his or her age and sex. This definition is not useful in this group of patients, and we modeled the BMI as a continuous response variable in subsequent regression analysis rather than as an indicator variable for pediatric obesity because our study population is either already obese or at-risk for obesity. The focus of the analysis was to identify factors associated with significant increases/decreases in BMI in this population and not which factors are associated with pediatric obesity.
Our primary predictor variables of interest were sleep duration (hours), quality of life, and sleep habits. Sleep duration was measured for both weekdays and weekends and calculated from the self-reported typical bedtimes and wake times. Quality of life was measured using the Pediatric Quality of Life Inventory (PQLI; PedsQL version 4.0). The inventory includes physical functioning, emotional functioning, social functioning, and school functioning. A five-point response scale from 0 to 4 is used; higher scores correspond to a poorer quality of life.
Other sleep-related instruments included the Pediatric Sleep Questionnaire (PSQ), the Pediatric Daytime Sleepiness Scale (PDSS), and a generic sleep questionnaire. The PSQ is a validated instrument to assess the presence of childhood sleep-related breathing disorders and prominent symptom complexes, including snoring, daytime sleepiness, and related behavioral disturbances in children. It consists of 22 questions in 2 parts (part 1, sleep; part 2, behavior), with a score of 0 or 1 allotted to a negative and positive response, respectively. A higher score corresponds to the presence of these symptom constructs. Sleepiness was assessed using the PDSS, which consists of 8 questions scored on a scale of 0 to 4. The 0-to-4 scale corresponds to a response of "never" to "always" to questions assessing symptoms related to sleepiness in children. A higher score is consistent with increased sleepiness and correlates with low academic achievement, shorter total sleep time, and a significantly higher level of diurnal sleepiness. An additional generic sleep questionnaire included questions about sleep habits (sleep and wake-up times on weekdays and weekends), naps (their length), whether family members share the room or bed with the child, and the child's self-assessment of how he or she feels upon waking (still tired vs rested). In addition, children were asked to rate their body figure using a well-described body figure scale for pediatric patients. The scale (from 1 to 9) consists of separate images of boys and girls ranging from emaciated to obese. Depending on the child's age, the parents also helped children to complete questionnaires.
Statistical Analyses
The control variables in this study included age and sex. Age was calculated in years and months. All of the statistical analyses were conducted using the statistical software package R version 2.8.1 (Bell Laboratories, Murray Hill, NJ). General demographics and physical characteristics were summarized with means and standard deviations, medians and ranges, or categorical counts and percentages. For the univariate analysis, t tests were used to test for significant subgroup differences in both the average BMI and average weekday sleep duration and to identify possible confounding factors in the underlying relation between sleep habits and BMI.
Given our small sample size, we reduced the number of predictor variables to achieve a more robust, stable final model. In addition, we were interested in characterizing the children by their sleep behavior given the battery of validated surveys and questionnaires. We clustered the children by all of their survey scores: PQLI Physical, PQLI Emotional, PQLI Social, PQLI School, PSQ1, PSQ2, and PDSS, using complete linkage hierarchical clustering. In hierarchical clustering, the distance between all of the scores is used to measure the similarity of the children (small distance, similar survey scores, similar children). Complete linkage simply defines the distance among groups as the maximum distance between children in each group. We chose the number of clusters/groups by looking at a tree of distances between the children (ie, a dendrogram) and by looking for obvious separation between groups of subjects (groups connected at high heights/large distances). This approach assumes no hypothesis about the number or type of groups. It determines the group structure solely using the distances between children and can be viewed as an exploratory technique.
Before the multivariate analysis, we explored the univariate relation between weekday sleep duration and BMI depending on age. Children subsequently were categorized into three subgroups: age ≤ 8 (young), 8 < age ≤ 12 (preteens), and age > 12 (teenagers); we used linear regression to model the relation in each subgroup. These results led us to build a multivariate linear regression model that allowed for different sleep duration effects for children 12 years or younger and children older than 12 years. This model also included sex, ethnicity (white vs nonwhite), our survey score groups, self-assessment of current body figure, naps on weekends, and sharing a bedroom.