Sleep, Hormones, and Metabolic Factors

All analyses were cross-sectional and performed using SAS/STAT 8.2. Leptin, ghrelin, and adiponectin were square-root transformed and insulin log transformed based on the distribution of residuals from the multivariate regression models. We evaluated the relationship of age, sex, BMI, and time of storage of blood sample on hormones using multiple regression. Partial correlations adjusted for age, sex, and BMI were calculated for hormones and QUICKI, with and without control of other potential confounders. The relationships between hormones and sleep were evaluated using multiple linear regression after control for potential confounders including age, sex, BMI, SDB, and morningness tendencies (as measured using the Horne-Ostberg questionnaire, an indirect surrogate of earlier rising time). In all analyses involving insulin, glucose, and QUICKI (but not leptin, ghrelin, and adiponectin), participants with diabetes (self-reported diagnosis, or currently taking insulin or diabetic medications, or with glucose >300 mg/dl) were excluded. Participants with SDB were not removed from the analyses shown. When controlling for AHI in models, participants who used continuous positive airway pressure or who had inadequate sleep were excluded. Because controlling for AHI did not significantly change the sleep-hormone regression coefficients, these analyses are not shown. The relationship of BMI with average nightly sleep was evaluated using a quadratic fit. This was examined using multiple visits (n = 1,828) from 1,040 participants with sleep diary data available. Mixed modeling techniques were used to account for within-subject correlation for participants with multiple visits. SAS procedure mixed was used for modeling and hypothesis testing using robust standard errors and a compound symmetric within-subject correlation structure. All reported p values are two-sided. For illustrative purposes, changes in leptin, ghrelin, and BMI for different sleep amounts were calculated at the average values and sex distribution of the relevant sample.