The study was a descriptive cross sectional study and sample size was calculated using an equation by Fleiss, 1981 by considering 20 % difference between the intervention and control groups in relation to knowledge on micronutrients, an alpha error of 5 %, beta error of 10 % (power of 90 %) with 1:1 ratio of the two groups, where the minimum sample size required was calculated to be 550. The sample was stratified by urban (n = 300) and rural sectors (n =300), and a total of 600 participants were recruited.
According to available baseline data on the proportion of out of school adolescents in each area, six Public Health Midwife (PHM) areas were selected from each of these two districts and all school dropouts within each were recruited to the study. All participants were examined by a medical officer who excluded those with any known medical conditions. Exclusion criteria were pregnancy, lactation or having a child below the age of six months. Complete dietary data were available for 552 female adolescent (aged 15–19 years of age) school dropouts. All dietary, anthropometric, body composition and qualitative data included in the analysis are from the pre-intervention data and are not subject to bias due to intervention. This study was conducted according to the guidelines laid down in the declaration of Helsinki and all procedures involving human subjects were approved by the ethics review committee of the Faculty of Medicine, University of Colombo, Sri Lanka. Written informed consent was obtained from all study participants.
Anthropometric data
Weight was measured to the nearest 0.1 kg with a calibrated electronic scale (Seca813). Height was determined to the nearest 0.1 cm using a stadiometer (Seca225, telescopic height measurement) according to standard protocol. In order to avoid variability in results, all height and weight data were measured by one researcher using the same equipment and participants were attired in light indoor clothing. Body mass index (BMI) was calculated as weight (kg)/ height (m2), and categorized into underweight, normal weight and overweight on the basis of age specific World Health Organization classification of BMI for adolescent girls [11].
Percentage body fat calculation
A skin-fold thickness (SFT) equation [12] specific for 15–19 year old Sri Lankan adolescent girls was used for the calculation of percentage body fat. The triceps and supra-iliac skin fold thickness required for the equation was measured in triplicate on the left side of the body, in accordance with standard protocols [12], using a Harpenden caliper (gradation 0.2 mm, range: 80 mm, Seca, HSK-BI), to the nearest 1 mm. The average of three SFT recordings was used in the analysis. Percentage body fat (kg) was categorized into tertiles as low (<12 kg), moderate (12–27 kg) and excess (>27 kg) body fat.
Dietary data
An interviewer administered food frequency questionnaire was used for obtaining information regarding the weekly frequency of consumption of common food items. The questionnaire assessed the frequency of consumption and portion size of carbohydrates (breakfast, lunch and dinner), animal food, dairy, beverages, fruits and vegetables, sweets, snacks and instant food over a week (95 items). Portion size was determined using a pretested colored photographic atlas of selected cooked foods as well as by kitchen utensils where photos were not available (tablespoon, rice spoon, glass, tea cup).Weights of the cooked food were converted to the raw weights. The response options for each food item included never/rarely consumed, number of times consumed per day(1–3: breakfast, lunch, dinner) and number of days consumed per week (1–7).This food frequency questionnaire was pretested on 20 girls of similar age and area of residence in a pilot study.
Physical activity
Physical activity was assessed by obtaining data on the number of hours per day and days per week spent on walking, exercising and carrying out strenuous activities. The sum of hours / week spent on these (at home and at work) was used for analysis as physical activity per week. Physical activity/week was categorized in to tertiles as slow (<9 hrs/week), medium (9–34 hrs/week) and high (>34 hrs/week) physical activity.
Qualitative data
Separate focus group discussions were held with 15–19 year girls, their mothers, health volunteers and PHMs of the urban and rural areas. Discussions were on the following themes: foods consumed, preferences in diet, Cooking practices, perceived constraints to good nutrition and myths associated with nutrition and diet. Focus group discussions were held prior to carrying out other investigations, so as to avoid bias in the discussion, and were held until saturation point was reached. Sessions were hand written and also recorded for accuracy and data was cross checked by a second investigator for validity. The data collected from focus group discussions was classified according to identified themes and analyzed using the cut-and paste technique described by Stewart and Shamadasani and triangulated with quantitative data [13].
Identification of dietary patterns
Food items listed in the food frequency questionnaire were grouped into 17 food groups (Rice and rice based products, potatoes and tubers, vegetables, dark green leafy vegetables (DGLV), wheat flour based products, pulses, fresh fruits, pickled fruits with added sugar, fish, meat, processed meat, liver, egg, dairy, sweetened beverages, sweets and fried food) according to their nutritional characteristics. Dietary patterns were identified based on the frequency of consumption of each of the 17 food groups using exploratory factor analysis, which is an analytical tool. Factor scores (composite variables which provide information about an individual’s placement on the factor(s)). were computed by the statistical package for each of the dietary patterns, and these factor scores were used in subsequent analysis. The Bartlett test of sphericity (BTS) and the Kaiser-Meyer-Olkin measure of sampling adequacy (KMO) were used to assess data adequacy for factor analysis. The factor analysis model would be considered appropriate with a KMO greater than 0.6 and BTS of P < 0.05. A correlation matrix was then constructed. Principal component analysis was used for the extraction of factors. The first model was tested without setting the eigen value. The number of retainable factors was determined by the Cattel’s Scree Plot method. Then, a second model was tested setting the eigen value to 1.5. Factors with eigen values greater than 1.5 were retained. Food groups with factor loadings greater than 0.30 and communality over 0.20 were retained in the patterns identified [14].
Data analysis
The Statistical Package for the Social Sciences for Windows (v. 18.0, SPSS Inc., Chicago, IL, USA) was used in the data analysis. Body fat was compared with BMI and diet. The independent samples t test was used to assess the association between independent variables and factor score of the dietary patterns. BMI, body fat and physical activity of girls in urban and rural areas were compared using ANOVA, where BMI, physical activity and body fat were considered as dependent variables and area of residence was considered as the factor score. For the purpose of logistic regression, BMI and Fat mass were regrouped, where BMI was categorized as overweight and non overweight (by combining the underweight and normal weight), and fat mass was categorized as having excess body fat and low or moderate body fat (by combining the first and second tertiles). Logistic regression models were used to calculate the odds ratio. When identifying the likelihood of a girl residing in an urban area being overweight; and BMI (overweight vs. non overweight), Body fat (excess vs. low/moderate), physical activity(hrs/wk), and factor scores for pattern 1 and 2 were considered the independent variables when identifying the likelihood of an urban girl having excess body fat. Factor scores of the two patterns were compared with area of residence, BMI status and physical activity using ANOVA, Post Hoc tests were carried out where there were more than two categories in a variable (i.e. BMI status: underweight, normal weight, over weight and Physical activity: low medium, high). Multivariate linear regression models were used to assess the association between independent variables (physical activity, BMI, area of residence) and dietary patterns. Each model was mutually adjusted for confounding factors. Factor scores (continuous variables) of food patterns were the dependent variables. Linear regression analysis was carried out to determine the percentage variation in dietary-pattern scores, physical activity, BMI and body fat as explained by area of residence (urban vs. rural).