Participants and data-collection
We selected a random sample of schools (n=26) from a government database of secondary schools in Flanders, a region in Belgium consisting of around 6 million inhabitants. All included schools were state-funded, as the vast majority of schools in Belgium are state funded. Eight schools (31%) agreed to collaborate in the study. The main reason for not participating was no time to set-up the survey at the school within the time frame of the study. The study took place between November 2014 and May 2015. Within each school, classes were randomly selected. We aimed to collect data among grades 7 to 12 for each school (aged 12-18). However, not all grades from each school could be reached because of other activities (e.g. being away on a school trip, a planned mandatory school test, etc.). Data collection took place at school, during one class hour. The anonymous paper-and-pencil survey was administered by the researchers, who explained at the start of the survey that students were under no obligation to participate and could withdraw at any time. Students were assured that their responses would be confidential and that no information would be shared with teachers, parents, or fellow students. Five students declined to participate, none of the parents declined consent. The study received approval from the Ethics Committee of the Ghent University Hospital (2012/307, B670201214183). Adolescents provided written informed consent, parents provided passive informed consent. Parents were informed about the study through the school and received a telephone number and e-mail address from the researchers, via which they could notify that they did not want their child to participate. They were informed that when they did not contact the researchers, they agreed with participation of their child.
Measures
General socio-demographic information
Items were selected from the Health Behavior in School-aged Children (HBSC) 2009/10 questionnaire, a cross-national survey supported by the World Health Organization (57). Socio-demographic variables included gender, age, type of education (general academic, technical or vocational track), country of birth, family living situation, self-reported weight and height (used to calculate Body Mass Index, BMI). Age, gender and BMI were taken into account as covariates in this study, as gender, age and BMI differences have been associated with adolescents’ health behaviours and (mental) health outcomes (58). For example, girls are less likely to have breakfast every weekday and boys, in general, report early and weekly smoking more often. In addition, girls are more likely to describe lower life satisfaction in comparison to boys (58). Furthermore, a negative drop in healthy behaviors is seen with increasing age. For example, 11-year-olds are more likely to meet the physical activity guidelines of at least 60 minutes of moderate to vigorous physical activity daily than 15-year-olds in almost all countries and regions (58). BMI has also been reported as inversely associated with global self-esteem in adolescents (59).
Family affluence
This part of the HBSC questionnaire also consists of the validated adolescent self-report ‘Family Affluence Scale’ (FAS), to identify family material wealth and socio-economic status (SES) of children and adolescents (56, 60-62). The FAS is used as an indicator of SES. It has widely been used to explore and explain socioeconomic inequalities in a wide range of health indicators in the HBSC study over the last 20 years (61). FAS is validated against other measures of SES and macro-economic indicators (e.g. Gross Domestic Product (GDP)) in 35 countries (56, 61). The FAS was developed to overcome the problem of inaccurate perceptions and missing data among children and adolescents of their family’s finances, especially among lower socio-economic groups which could thus lead to an underestimation of socioeconomic inequalities (37, 61). It was proposed as a less intrusive, more comprehensible approach to identify the family’s socioeconomic status (63) than inquiring about parents’ educational, occupation or income levels (56, 64). It is indicated that in contrast to for example parental occupation, the proportion of missing data on FAS items is low (61). The FAS II consists of four items: number of cars, own bedroom, computers owned and number of holidays per year (56, 63). A composite FAS score (ranging from 0-9) is calculated for each adolescent based on his or her responses to these four items. The following, international, cut-off points were used: score of 0, 1, 2 classified as low affluence; score of 3, 4, 5 as medium affluence; and a score of 6, 7, 8, 9 classified as high affluence (56).
Healthy lifestyles
Items to assess healthy lifestyles, except for sleep duration, were also taken from the HBSC survey. Several health-related lifestyle behaviors among adolescents are interrelated. Based on Principal Component Analyses on these data reported elsewhere (13), healthy lifestyles were grouped into two factors: ‘energy-balance related behaviors’, consisting of physical activity and a healthy diet, and ‘addictive behaviors and sleep duration’, consisting of alcohol consumption, smoking and perceived sleep duration. These factors will be used to discuss the results, individual behaviors are however retained in the analyses.
Energy-balance related behaviors
Physical activity was measured by the number of days they achieved ≥60 minutes of moderate to vigorous physical activity, and was defined in the questionnaire as: “bodily movements that make your heart beat faster and make you feel out of breath at some moments”. A healthy diet was measured by assessing the number of days per week adolescents had have breakfast. Eating a regular, healthy breakfast contributes to the daily recommended intake of essential nutrients (65, 66). Moreover, a daily breakfast may be used to identify adolescents at risk for unhealthy lifestyle behaviors. For example, daily breakfast intake is associated with both daily fruit and vegetable consumption, and there is an inverse relationship between daily breakfast intake and daily soft drink consumption (51).
Addictive behaviors and perceived sleep duration
Alcohol use was assessed by summing the frequency of six different types of alcohol consumption: beer, wine, spirits/liquor, alcopops and any other drink that contains alcohol (0-never; 4-daily. Range of summed score 0-24). Three questions were asked on tobacco use, based on the Flemish version of the HBSC 2009/10 questionnaire: 1) have you ever smoked tobacco?; 2) how often do you smoke currently?; 3) how many cigarettes per day have you smoked on average over the last 30 days?. These questions were combined and recoded to form one indicator of tobacco use frequency, namely: 0 'never smoked'; 1 'I have smoked but do not smoke now'; 2 'I smoke now, but I am not a daily smoker'; 3 'I smoke daily, but I am low dose smoker'; and 4 'I smoke daily and am a high dose smoker'. To decide on low dose and high dose among daily smokers, the median among the group of daily smokers was used as a cut-off for tobacco use frequency. The median showed around half of daily smokers smoked fewer than 11 cigarettes per day (=low dose), and half smoked 11 cigarettes or more per day (=high dose). Self-reported smoking is a reliable indicator of smoking status (67). All questions from smoking were derived from the HBSC study where items have been decided by an international expert team and have been used in numerous studies (68-70). Specifically, for the three questions on smoking ICC-values of 0.75, 0.50 and 0.85 have been reported (71). To calculate sleep duration (number of hours slept per night), adolescents were asked to report at what time they usually go to bed and get up. Self-reported duration of sleep is strongly correlated with sleep duration measured by accelerometers for weeknights and moderately correlated with sleep duration for weekend nights (72).
Mental health
Mental health was measured through feelings of depression, anxiety and stress and self-esteem. Feelings of depression, anxiety and stress were measured with the Depression Anxiety Stress Scales (DASS-21) which has good psychometric properties to measure adolescent mental health outcomes (73, 74). It consists of seven items per subscale (74). Total scores per subscale were used as dependent variables, with high reliability for each of the subscales (αdepression=0.90; αanxiety=0.84; αstress=0.87). Focusing on self-esteem is considered a core element of mental health promotion and a fruitful basis for a broad-spectrum approach (75). Positive global self-esteem was measured by a single item from the Rosenberg Self-Esteem Scale (RES), namely ‘I take a positive attitude toward myself’. Global self-esteem can be measured by a single item (76) and this specific item is a main contributor to global positive self-esteem (77, 78).
Analysis
Variables were checked on normal distribution with the values for skewness and kurtosis. The values for skewness and kurtosis between -2 and +2 are considered acceptable in order to prove normal univariate distribution (79). To test the significance of difference of the health behaviors and mental health indicators between low-medium and high FAS (Table 1), independent samples t-tests were used for variables with a normal distribution (i.e., physical activity, healthy diet, sleep duration, alcohol consumption and self-esteem) and the non-parametric variant, Mann-Whitney U tests were used for the variables that did not have a normal distribution (i.e., symptoms of depression, anxiety and stress). For smoking a χ²-test was conducted. Gamma regression models were used to account for the positively skewed distribution of the dependent variables: depression, anxiety and stress (80). The dependent variable ‘self-esteem’ showed a normal distribution. For this variable, multiple linear regression analysis was used. Regression analyses assessed the association between healthy lifestyles and mental health outcomes (RQ1); and the moderating role of family affluence in the relation between healthy lifestyles and mental health outcomes (RQ3). Analyses were controlled for individual background factors that significantly influenced mental health outcomes (namely, gender, age and BMI). Analyses were conducted stepwise, by first examining the influence of family affluence and background variables, next the healthy lifestyle variables, and finally the interaction effects between healthy lifestyle variables and family affluence. Because no interaction effects were significant, the parsimonious model was constructed based on the full model of the direct effects (see Table 2 and 3). In the first regression analyses (Table 2), BMI was not a significant predictor for any mental health outcome and therefore not included in further regression analyses (Table 3). Collinearity diagnostics were conducted examining Variance Inflation Factor (VIF) (≤10) and tolerance (≥0.1). VIF showed no multicollinearity among independent variables. Cross-tabulations were checked for empty combinations of cells or low expected frequencies (81). Continuous independent variables were mean centered. Moderator variables were created by multiplication of interaction variables. All analyses were conducted in SPSS 25.0.