Study design and sample
This cross-sectional study was conducted from August to November 2016 in several public secondary schools that were randomly selected within three states in the southern region of Peninsular Malaysia namely Negeri Sembilan, Melaka and Johor. A probability proportionate sampling design was used for selecting these schools (16). This was done by obtaining a complete list of public secondary schools and the estimated number of adolescents aged 13 years old in each school (sampling frame) from the Ministry of Education (MOE) Malaysia. In Malaysia, adolescents usually enter public secondary schools at the age of 13 years. This age is important as it represents a key transition period from primary to secondary education levels that correlates with major pubertal changes and increases in personal autonomy (e.g. less parental control and more social freedom), both of which may affect food choices (17). A total of 24 selected schools were approached and out of these, only 21 schools agreed to participate.
Sample size for this study was estimated using a formula for single cross-sectional survey. The calculation for minimum sample size was based on the prevalence of obesity in children (20.9%) (18). After taking into account a 20% response rate and a design effect of 2, a total number of 1000 adolescents was estimated for this study. Out of the estimated 1000 adolescents, a total of 933 agreed to participate (93% of response rate). Information on anthropometric, physical activity, dietary assessment and biochemical measurement was provided by 930, 793, 585 and 507 adolescents, respectively. Out of these, 336 adolescents provided both cardiometabolic and valid dietary data while 582 adolescents provided both the anthropometric and dietary data. Supplementary figure 1 illustrates the number of adolescents who provided anthropometric, biochemical, dietary and physical activity data in this study.
Full ethical approval was obtained from the Ethics Committee for Research Involving Human Subjects (JKEUPM) of Universiti Putra Malaysia (UPM) (Reference number: FPSK (EXP16) P031). Approval to conduct this study in the selected schools were obtained from the Ministry of Education Malaysia, state education departments and selected schools. The study respondents and their parents provided written consent before the commencement of the study.
Patient and public involvement
Prior to the recruitment, adolescents in the selected schools were screened for their eligibility. Adolescents aged 13 years old during the data collection comprised of both sexes, with adequate ability to read and understand Malay or English language were eligible to participate in the study. However, adolescents with physical disabilities and chronic conditions were not eligible to participate in this study. Eligible adolescents were invited to take part in this study and were given a copy of study information sheet. Interested adolescents were asked to sign an assent form after being clearly briefed on all the procedures required for the study. An informed consent form was sent to their parents for further approval. Parents who agreed with the study involvement were asked to complete the parental questionnaire at home and return it to the study researcher through their children. During the second visit to the school, a research team comprised of a dietitian, a medical doctor, and two research enumerators collected various data i.e. socio-demography, dietary, physical activity, anthropometric and biochemical measurements from the study adolescents.
Dietary assessment
A validated adolescent food frequency questionnaire (FFQ) was used to assess dietary intakes in the past 12 months and details of this FFQ has been previously described (19). In brief, the MyUM FFQ was originally developed by a group of researchers from a contemporaneous study to that of this study in Universiti Malaya (UM) (17). It comprised of 195 food items and is a self-administered questionnaire designed especially for Malaysian secondary school adolescents aged between 13 to 18 years. Before administering the FFQ, step-by-step instructions on how to fill in the questionnaires were given to all participants by the study researchers. As an aid to estimate food intake, the participants were provided with a flipchart on household measurements. The food intake frequency and portion size of each food item were recorded by the adolescents. The average frequency for consumption of each food item over the past year was recorded as ‘never’, ‘1-3 times per month’, ‘one time a week’, ‘2-4 times per week’, ‘5-6 times per week’, ‘one time a day’, ‘2-3 times per day’, ‘4-5 times per day’, or ‘≥6 times per day’. The study researchers checked all the questionnaires upon submission to ensure that all fields were filled in.
The process of converting the FFQ raw data to daily energy and other nutrients were conducted manually using a standard conversion factor to estimate daily food intake based on the frequency of food consumption (20). The estimation of daily food intake was analysed using the Nutritionist Pro software version 3.1 (Axxya Systems, USA). The dietary data of the food items which were included in the FFQ were derived from the Malaysian Food Composition (21,22). Adolescents whose overall dietary energy intake were outside the range of 400 – 8000 kcal (or 1674 – 33472 kJ) were excluded from the DP analysis (23).
Dietary misreporting was estimated using the Goldberg equation (2000) according to the ratio of energy intake (EI) to basal metabolic rate (BMR) (24). BMR was calculated using a sex-specific formula for Malaysian adolescents aged 13 years (25). Physical activity level (PAL) was set at 1.55 as majority of adolescents had been reported to have low physical activity levels (26). The cut-off values for dietary misreporting was calculated based on the confidence limit of agreement between the ratio of EI to BMR and PAL. Adolescents with the ratio of EI to BMR ranging from 1.09 to 2.21 were considered as plausible reporters, otherwise, they were considered as under-reporters or over-reporters, accordingly. Variable on dietary misreporting was included as a potential covariate in statistical models.
Dietary patterns
Reduced rank regression (RRR) analysis was used to derive empirical DPs using SAS software version 9.4 (SAS Institute, Cary, NC). The RRR is a statistical method to determine linear function of predictor variables (food groups) by maximising the explained variation in nutrients (response variables) related to the disease of interest (27). All the food items from the FFQ were categorised into 13 food groups (g/d) based on their nutritional characteristics and were used as predictor variables “Supplementary Table 1” (13). DED, percentage of energy from total fat intake, percentage of energy from total sugar intake and fibre density were selected as response variables. These selected response variables showed significant associations with obesity and other cardiometabolic risk factors in previous prospective studies in Australia and UK (11,12).
DED was calculated by dividing total food energy (kJ) with total food weight (g) by excluding beverages (13). Meanwhile, fibre density was determined by absolute fibre intake (g/d) divided by total daily energy intake (MJ) (13). Percentages of energy from total fat and sugar intakes were expressed by dividing total energy intake from fat (kJ) or total energy intake from free sugar (kJ) by total energy intake (kJ), followed by a multiplication of 100 (13,14). In this study, dietary sugar was defined as short-chained carbohydrates known as monosaccharide and disaccharides presented naturally in foods such as fruits or in manufactured products such as refined sugar (28).
A separate RRR analysis was applied specifically to investigate any gender variations between the DPs. However, as the DPs derived for both males and females were similar in their factor loading of food groups, thus DPs derived for the total respondents were taken for further analysis. Each adolescent obtained an individual z-score for each DP derived; a higher z-score corresponded to a higher adherence to the identified DP. Z-scores for the identified DP were analysed continuously and categorically (tertiles) with the lowest tertile set as the reference category.
Measurement of cardiometabolic risk factors
A digital scale (Tanita HD319, Japan) was used to measure the adolescents’ body weight in kilogram, while their body height in centimetre was measured using a stadiometer (Seca 206, Germany). A measuring tape (Seca 201, Germany) was used to measure the adolescents’ waist circumference (WC) at the midpoint between their lower border of the ribs and their upper border of the pelvis. All measurements were repeated twice to obtain a mean value for each variable which were then used in the analysis.
Adolescents’ BMI was estimated by using a standard formula, body weight (kg) divided by the squared measured body height (m2). In addition, BMI z-score for age and gender was calculated using the WHO Anthro Software (29). Overweight and obesity were defined using the BMI z-scores, whereby values of more than one and two standard deviations indicated possible risk of overweight and obesity, respectively. Computed WC z-scores were utilised for the analyses of abdominal obesity and was defined according to the Malaysian WC centile (30).
Adolescents who consented for blood withdrawal were asked to fast overnight before venepuncture by a phlebotomist. A total of 10 ml of fasting blood was withdrawn from each adolescent for insulin, fasting blood glucose (FBG), total cholesterol (TC), triglycerides (TG), HDL-C and low-density lipoprotein cholesterol (LDL-C) analyses. Insulin resistance was estimated using Homeostasis Model Assessment (HOMA) (31). FBG concentration was measured by hexokinase assay using reagent by ADVIA Chemistry Glucose Hexokinase_3 and insulin concentration was measured using two-site sandwich immunoassay by ADVIA Centaur Insulin assay (Siemens Healthcare Diagnostics Inc., Tarrytown NY, USA). TC and HDL-C concentration were measured by enzymatic endpoint method and elimination/catalase method, respectively using reagent by ADVIA Chemistry (Siemens Healthcare Diagnostics Inc., Tarrytown NY, USA). TG concentration was measured by enzymatic reaction method with Trinder endpoint using ADVIA TRIG_2 reagent (Siemens Healthcare Diagnostics Inc., Tarrytown NY, USA). LDL-C and HOMA was calculated using the Friedewald formula (1972) and standard formula by Metthews et al. (1985), respectively (32,33).
Dyslipidaemia during adolescence was defined when either the study adolescents’ TC level was greater than or equal to 5.2 mmol/L or their LDL-C level was greater than or equal to 3.4 mmol/L (34). Adolescents with abnormal biochemical values were classified if their biochemical parameter values were ≥5.60 mmol/L for blood glucose, ≥5.20 mmol/L for total cholesterol, ≤1.03 mmol/L for HDL-cholesterol, ≥4.12 mmol/L for LDL-cholesterol, ≥1.70 mmol/L for triglycerides, ≥25.0 uIU/mL for serum insulin and ≥4.0 unit for HOMA-IR level (34–37).
Covariates
A set of parents’ and study adolescents’ socio-demographic information was collected in this study. Parents were requested to complete a self-administered parental questionnaire comprised of information on educational level, occupation and monthly income. Meanwhile, adolescent questionnaire included questions such as date of birth, ethnicity, religion and gender. Self-reported physical activity in the past seven days was assessed using Physical Activity Questionnaire for Older Children (PAQ-C)(38). PAQ-C has been validated and showed acceptable validity and good internal consistency among Malaysian adolescents (39,40).
Statistical analysis
Descriptive data were presented in mean ± standard deviation (SD) for continuous data and in frequency (n) and percentage (%) for categorical data. Comparisons between genders were performed using independent t-test and chi-square test. Binary logistic regression analysis was conducted to evaluate association between tertiles (2nd tertile vs. 1st tertile and 3rd tertile vs. 1st tertile) of the DP z-scores and overweight or obesity, abdominal obesity, dyslipidaemia, elevated values of FBG, TC, LDL-C, TG, insulin, HOMA-IR and low HDL-C. Regression models were conducted for all the cardiometabolic parameters in male and female, separetely due to sex dimorphism and puberty-related differences in growth (41). All the models were adjusted for covariates including gender, school location (rural and urban), mother’s educational level (no formal education or primary level, secondary school level and tertiary level), dietary misreporting, physical activity and BMI z-score (for biochemical parameters only). All the above-mentioned analyses were ran using IBM SPSS Statistics software version 23 (IBM Corporation, New York, US) and a p-value of <0.05 was considered as statistically significant.