2.1. Study Population
We used data from the 2013–2017 Korea National Health and Nutrition Examination Survey (KNHANES), which is a representative national survey of Korea for the analysis of this study. KNHANES is ongoing, and has been annually conducted by the Korea Centers for Disease Control and Prevention (KCDC) since 1998. KNHANES consists of a health and life style interview, health examination and nutrition survey. Of the 30,553 adults aged ≥20 years participating in the 2013–2017 KNHANES, we only included 13,686 participants based on the following exclusion criteria: Energy intake < 500 kcal/day or > 5000 kcal/day (n = 4128); pregnant or lactating (n = 274); taking medicine or undergoing treatment for diabetes, cardiovascular diseases, or thyroid disease (n = 9519); or they had missing values on the sleep duration question, their waist circumference and socio-demographic or life-style related variables (n = 2946). The survey protocol was approved by the Institutional Review Board of the KCDC, and informed consent was collected from all participants.
2.4. Other Covariates
Covariates included sex, age, smoking, drinking, physical activity, family history of chronic disease, stress level, household income, living area, education and nutrient intake, and these variables were obtained from the health and life style interview or nutrition survey conducted by trained interviewers. Sociodemographic variables included sex, age, household income, living area and education. Living area was categorized into urban and rural, and household income was categorized into ≤middle-upper income and highest income. Education level was divided into ≤middle school graduate, high school graduate and ≥college graduate.
Smoking, drinking, physical activity, family history of chronic disease and stress level were included in the health and life style variables. The participants’ answers to the question, “Do you currently smoke?”, included ‘everyday’, ‘sometimes, ‘smoked in the past but do not currently smoke’ or ‘never’, and they were categorized as follows: ‘everyday’ or ‘sometimes’ were categorized into ‘yes’; and ‘smoked in the past but do not currently smoke’ or ‘never’ were categorized into ‘no’. Drinking data were obtained from the question, “How often did you drink during the year before the interview?”, and participants were categorized into as ‘never or rarely’, ‘2–4/month’, and ‘≥2/week’.
Physical activity was expressed as metabolic equivalent of task (MET) hours/day based on the physical activity questions. The participants answered the question, “How many days, hours, and minutes have you usually spent more than 10 min of physical activity in a week?”, according to the following physical activity levels: Vigorous, moderate, and walking. The number of days and duration time were estimated as hours a day.
Family history of chronic disease was categorized into ‘yes’ and ‘no’ based on the question, “Have your family members ever been diagnosed with chronic disease, such as hypertension, hyperlipidemia, cardiovascular diseases, or diabetes, etc. from a doctor?”. The participants’ answers to the question, “How often do you feel stress in your daily life?”, included ‘rarely stressed’, ‘slightly stressed’, ‘moderately stressed’, or ‘highly stressed’. Finally, calorie, carbohydrate, protein and fat intakes were obtained from the 24-h dietary recall data of the nutrition survey conducted by well-trained dietitians, and the proportions of calorie intake from carbohydrates, proteins and fats were estimated. KNHANES conducts its food intake survey using a 24-h dietary recall method. All food intakes are recorded by energy and nutrition analysis software, and intakes of calories and all the macro- and micro-nutrients are calculated by the software using a nutrition composition database. The proportion of calorie intake from carbohydrate, protein and fat was calculated by percentage of energy derived from each nutrient composition of all food intakes out of their total energy intake. We did not control other dietary variables, such as snacking, because snacks generally are high in fat and/or carbohydrate, and it can be covered by the fat and carbohydrate intake variables.
2.5. Statistical Methods
Because the primary objective of this study was to examine the effect of fat intake and stress on the association between sleep duration and abdominal obesity, our initial approach was to set up a logistic regression model as follows: Prevalence of abdominal obesity for dependent variable; sleep duration variables for independent variable; and self-reported stress level variables and quartiles of proportion of calorie intake from fat for the third control variables.
The first test investigated if fat intake and stress variables have a mediating effect. The tests were performed by adding variables sequentially using hierarchical logistic regression analyses [39
]. If the fat intake and stress variables are mediators, then the addition of confounder in the logistic regression would mediate the significance and/or size of impact of the sleep variable. The quartiles of the proportion of calorie intake from fat and the self-reported stress level were added on top of the baseline and baseline plus dietary variable models. The hierarchical logistic regression modeling approach was employed to examine the effect of fat intake and stress on the association between sleep and abdominal obesity by sequentially adding control variables. Our baseline model, model 1, consisted of sleep duration variables, namely sex, age, health and lifestyle variables (such as smoking, drinking, physical activity and family history of chronic disease), and other socio-demographic variables (such as household income, living area and education). Model 2 included model 1 variables and the following dietary factor variables: Total calorie intake and the proportion of calorie intake from carbohydrates. Model 3 included model 2 variables plus the quartiles of the proportion of calorie intake from fat or self-reported stress levels.
The second test investigated if the fat intake and stress variables modify the effects of sleep duration on abdominal obesity. To assess the effect modifications by fat intake and stress variables, the effect of sleep duration on abdominal obesity was assessed in stratified groups of fat intake and stress levels [40
]. We stratified logistic regression analyses over the quartiles of the proportion of calorie intake from fat and self-reported stress levels, and we examined the sleep duration effects across the stratified logistic regression analyses. To test if the stratified model is statistically superior to the aggregated model, log-likelihood tests were conducted for these logit regression models [41
]. This log-likelihood test determines the better model between stratified models and one aggregated data model. If structural breaks across stratified data exist, the log-likelihood test favors the stratified model over the aggregated model.
All statistical analyses were conducted using STATA S/E 15.0 (StataCorp, College Station, TX, USA). For categorical variables and continuous variables, the Chi-square test and one-way analysis of variance (ANOVA) were performed, respectively, to estimate the differences and statistical significance in the distribution across sleep duration. To estimate the association between sleep duration and abdominal obesity, odds ratios (ORs), 95% confidence intervals (95% CIs) and p-trends were calculated using multivariable logistic regression analysis with the group of subjects sleeping less than 5 h a day as the reference group. Statistical significance was determined at a p-value < 0.05.