Participants and procedures
This study utilizes baseline data collected from adolescent participants of an eight-month e-health obesity intervention, which included anthropometric measurements, questionnaires (Additional file 1), and three 24-h dietary recall assessments. In addition, one of their parents completed a baseline questionnaire on the home food environment. Participants were recruited from newspaper advertisements (62%), invitations sent to previous patients of a Children’s Hospital Endocrinology & Diabetes Unit (13%) and healthy weights clinic (15%), and other sources (e.g., word of mouth) (10%). Eligible adolescents were 11–16 years old and had BMI z-scores greater than one standard deviation from the mean, according to WHO age-and-gender matched growth charts [18]. Participants had to be residents of the greater Vancouver area with no plans to move within the study period, read at the grade 6 level and speak English. Exclusion criteria included comorbidities that required immediate medical attention, medical reasons that made physical activity too difficult, use of medication affecting body weight, diagnosis of Type 1 diabetes, or participation in another weight-loss program. Of the 183 parent-child pairs who completed the baseline assessment, seven did not meet eligibility requirements (e.g. BMI, reading level), three did not complete any 24-h dietary recalls, and six parents did not complete the home environment questionnaire yielding a sample of 167 parent-adolescent pairs for the present analyses. Written consent was obtained from all participants and this study was approved by the University of British Columbia and the University of Waterloo ethics boards.
Measures
Outcome variable
Dietary Intake was assessed using a previously validated [19], computer-based 24-h dietary recall program employing a three-pass technique where participants were asked to report all foods/beverages that they consumed the previous day at breakfast, morning snack, lunch, afternoon snack, dinner, and evening snack. Over 900 brand or generic food items were available and participants were instructed to substitute foods not found (20% of recalls had at least one food item substituted). Photographs depicting measured portion sizes helped to estimate portion sizes and prompts allowed for the selection of toppings commonly eaten with certain foods (e.g. spreads on toast). A summary screen allowed participants to confirm or delete their selections. Dietary data were downloaded from the web survey and processed with The Food Processor software package (version 8.0, ESHA Research, Salem, OR, 2002) that uses the 2007 Canadian Nutrient File data (http://www.hc-sc.gc.ca/fn-an/nutrition/fiche-nutri-data/index-eng.php) to calculate nutrient and Canadian food group estimates.
Of the 167 adolescents examined in the present study, 76 provided all three days of dietary recalls, while 46 provided two days and 45 provided only one day. No differences by number of dietary recalls completed were found except for consumption of desserts/treats, which was significantly greater among those who completed more days of dietary recall (data not shown). Because few differences were found, dietary intakes were averaged across all available recalls to obtain daily estimates of: 1) servings of FV, 2) percentage of energy from total fat (Fat), 3) servings of sugar-sweetened beverages (SSB), 4) servings of desserts or treats (Desserts/treats), and 5) percentage of energy from snacking occasions (Snacks). Desserts/treats included food items commonly consumed for dessert or as a treat (e.g. cookies, cake, candy, chocolate, ice cream and chips), which are typically energy dense yet nutrient poor. Servings of SSB and desserts/treats were dichotomized (any vs. none) because they had a highly left-skewed distribution.
Independent variables
Parent Modeling was assessed with five items from the adolescent questionnaire: 1) My parents eat vegetables when I am with them; 2) My parents eat fruits when I am with them; 3) My parents eat salad at a restaurant when I am with them; 4) My parents eat low-fat snacks when I am with them; 5) My parents eat low-fat dressings with salads when I am with them. Responses to each item were coded on a 4-point scale (Never, Sometimes, Frequently, Always). These items were adapted from Cullen’s 15-item parent modeling scale [20], which also included additional items specific to particular meal times. Similar items have also been used to predict diet outcomes in adolescent samples [21].
Parenting Style was assessed with eleven items from the parent questionnaire such as wanting to hear about my child’s problems, knowing where my child is after school, and telling my child that I like him/her just the way he/she is. Responses to each item were coded on a 4-point scale (Never, Sometimes, Often, Always). These items were derived from Cullen’s 11-item authoritative parenting scale [22].
Family Meal Practices was assessed with seven items drawn from the Family Nutrition and Physical Activity Screening Tool [23], which was completed by parents: 1) eating breakfast together, 2) eating at fast food restaurants, 3) eating while watching television, 4) eating fruits and vegetables with meals or as snacks, 5) using pre-packaged foods for meals, 6) eating dessert regularly after dinner, and 7) eating dessert regularly in the evening. Responses were coded on a 4-point scale so that a higher score indicated more healthful meal practices.
Home Food Availability was assessed with eight items from the parent questionnaire. Participants were asked if the following seven food types were available in the past week (yes/no) and if they were low-fat (yes/no): 1) cookies, pies, cakes or snack cakes; 2) chips (e.g. potato, corn, tortilla or Doritos chips); 3) ice cream or frozen yogurt; 4) granola bars; 5) bacon/sausage; 6) hot dogs; and 7) frozen dinners. Similar to previous studies that summed food items into the total number of core foods versus non-core foods available in the home or the number of energy-dense snack foods [24, 25], availability items were split into two indices and summed to generate: 1) Availability of high-fat foods (bacon/sausage, hot dogs, frozen dinners; range = 0–3), and 2) Availability of high-fat treats (cookies/pies/cakes/snack cakes, chips, ice cream/frozen yogurt, and granola bars; range = 0–4). Items identified as low-fat versions were omitted. Availability of soft drinks was assessed with the following item: “Did you have regular sodas or soft drinks in your home in the past week?” These items were derived from a list of 15 items used in the Girls Health Enrichment Multisite Study [26, 27]. Similar items have been used to predict dietary intake in adolescent samples [21].
Covariates
Adolescent Age and Gender, Parent Ethnicity, Maternal Education and Household Income were based on parent self-report. Highest degree, certificate, or diploma of mother was obtained and responses were grouped into three categories: 1) Less than or equal to high school education; 2) Trade certificate, diploma, non-university certificate, or university certificate below a bachelor level; and 3) University degree or greater. Total income, before taxes and deductions, of all household members from all sources in the past 12 months was obtained and responses were collapsed into four categories: 1) ≤ $40,000; 2) $40,001–$80,000; 3) $80,001–$120,000; and 4) ≥ $120,000. Body Mass Index z-scores, based on sex and age, were computed from measured height and weight using the WHO method for children and adolescents (5–19 years old) [18].
Analysis
Confirmatory Factor Analysis (CFA) was performed to determine if scale factor structures were supported in this sample. Availability of high-fat foods was conceptualized as an index and availability of soft drinks was assessed by only one item; therefore, they were not examined using CFA. Model fit was assessed using commonly accepted fit indices: Chi-square goodness of fit test (p-value ≥.15), Comparative Fit Index (CFI > .95), Root Mean Square Error of Approximation (RMSEA<.06 with an upper CI ≤ .08 and a p-value > .05), and the Standardized Root Mean Square Residual (SRMR<.08) [28]. Since the chi-square test is highly influenced by model complexity and sample size, and CFI and SRMR are highly influenced by the inclusion of non-significant paths, the RMSEA was the main index used to determine model fit [28]. A single model was built with all three latent constructs and the Maximum Likelihood Estimator was used. Internal consistency of items in each scale was determined by computing Cronbach’s alpha.
After the measurement models were refined, two structural equation models tested the conceptual model linking the home food environment to adolescent dietary outcomes: FV, Fat, SSB, Desserts/treats, and Snacks. For the analyses, servings of FV were expressed per 1000 kcal (to account for energy intake and to maintain a scale comparable with the other dietary variables). First, all of the independent variables were regressed on each dietary outcome to determine direct effects. Second, the independent variables were regressed on dietary outcomes as well as on home availability variables. Covariates included adolescent age, sex, maternal education and household income. The Means- and Variance- adjusted Weighted Least Squares (WLSMV) method of estimation was used to handle a combination of continuous and dichotomous outcome variables. WLSMV has been proposed as the best estimator when categorical data are present [29], was designed specifically for use with small and moderate sample sizes, and is fairly robust to non-normality [30, 31]. Model fit was assessed using the indices described earlier as well as the Weighted Root Mean Square Residual (WRMR). When using the WLSMV estimator, the RMSEA and WRMR are the best indices of model fit, with a WRMR of less than 1.0 and a RMSEA of less than 0.6 suggesting a good fit [28].
Missing data were handled using pairwise deletion (< 5% missing). All conceptual paths were included in the model and were considered significant at p-value< 0.05. All statistical analyses were conducted using MPlus® (version 7, Los Angeles, CA).