Study population
The Australian Diabetes, Obesity and Lifestyle (AusDiab) study methods and response rates have been described previously [10]. In brief, a stratified cluster sample of 11, 247 adults aged ≥25 years was drawn from 42 randomly selected census collector districts across Australia in 1999/2000. A total of 10, 659 participants were included in this analysis. 588 participants (5.2 %) were excluded due to missing data: BMI (n = 180); WC (n = 194); education (n = 110); country of birth (n = 2); smoking status (n = 212); TV viewing time (n = 92); CVD (n = 75); dyslipidaemia (n = 2); diabetes status (n = 169); and hypertension status (n = 73) (numbers are not additive). There was no difference in mean age between those with and without missing data. Women were more likely to have missing data than men (p < 0.001). This study was approved by the International Diabetes Institute Ethics Committee and the Monash University Human Research Ethics Committee. All participants gave written informed consent.
Measurement of body mass index and waist circumference
Height was measured to the nearest 0.5 cm without shoes using a stadiometer. Weight was measured without shoes and excess clothing to the nearest 0.1 kg using a mechanical beam balance. BMI (kg/m2) was calculated as weight (kg) /height (m)2 and categorized as: (i) non-obese: <30 kg/m2; and (ii) obese: ≥30 kg/m2. WC was measured at the point midway between the iliac crest and the costal margin and the mean of two measures was calculated. WC was categorized as: (i) non-obese: <102 cm for men, <88 cm for women; and (ii) obese: ≥102 cm for men, ≥88 cm for women [6]. Adiposity categories were created using a combination of BMI and WC as follows: (i) BMIN/WCN; (ii) BMIN/WCO; (iii) BMIO/WCN; and (iv) BMIO/WCO, where N = non-obese and O = obese.
Measurement of metabolic outcomes
In all states except Victoria, blood pressure was measured using a Dinamap® oscillometric blood pressure recorder (General Electric Company, Milwaukee, WI, USA). In Victoria, blood pressure was measured using a standard mercury sphygmomanometer and adjusted accordingly [11]. The average of two measures was used in the analysis. Hypertension was defined as blood pressure >140/90 mmHg or the use of antihypertensive medication.
Blood samples were collected by venepuncture after an overnight fast (≥9 h). All samples were centrifuged on-site to separate plasma and serum, and were transported daily to a central laboratory where possible. If transport to a central laboratory was not possible, samples were stored on-site in a freezer at −20 °C and then transferred to a −70 °C storage facility within 1 to 2 weeks following collection [12]. Diabetes was defined on the basis of fasting plasma glucose ≥7.0 mmol/l or two-hour plasma glucose ≥11.1 mmol/l, or current treatment with insulin or oral hypoglycaemic agents. Dyslipidemia was defined as triglycerides >2.0 mmol/l or high density lipoprotein (HDL) cholesterol <1.0 mmol/l. Cardiovascular disease (CVD) status was self-reported and was defined as previous angina, stroke, and/or coronary artery disease.
Measurement of covariates
Covariate data was collected using interviewer-administered questionnaires. Educational attainment was categorized as: (i) low: secondary school qualification or lower; (ii) middle: attained trade or technician’s certificate, associate or undergraduate diploma, or nursing or teaching qualification; and (iii) high: attained a bachelor degree or post-graduate diploma. Country of birth was dichotomised into Europid and non-Europid. Physical activity was self-reported and was categorised as: (i) inactive (0 min/week); (ii) insufficient (1–149 min/ week); and sufficient (≥150 min/ week). Smoking status was categorised as: (i) current smoker; (ii) ex-smoker; and (iii) non-smoker. Time spent watching TV was used as a measure of sedentary behaviour and was categorised as: (i) <2 h/day; (ii) 2–3.9 h/day; and (iii) ≥4 h/day [13]. Information on alcohol consumption and energy intake were collected using the Food Frequency Questionnaire [14]. Alcohol consumption was dichotomised into ≤10 g/day and 10 g/day [15]. Energy intake was analysed as a continuous variable.
Statistical analysis
To account for the clustering and stratification of the survey design, and to adjust for non-response, the data was weighted to match the age and sex distribution of the 1998 estimated residential population of Australia aged ≥25 years. The weighting factor was based on the probability of selection in each cluster.
Population attributable fraction (PAF) was calculated to determine the proportion of hypertension, diabetes, dyslipidaemia and CVD that is attributable to obesity, with obesity defined two ways: (i) obese according to either BMI or WC; and (ii) obese according to BMI alone. PAF was calculated using the formula:
$$ PAF={P}_E\frac{\left( OR-1\right)}{\left(1+{P}_E*\left( OR-1\right)\right)} $$
where PE = prevalence of obesity and OR = odds ratio.
Discrimination for obesity defined as obese BMI or obese WC and obesity defined as obese BMI alone, with each metabolic outcome, was determined using area under the receiver operating characteristic curve (AUC) and compared using Wald chi-squared tests [16]. An AUC of 1.0 indicates perfect discrimination and AUC of 0.5 indicates that the discriminatory power of the predictor is no better than chance.
Logistic and linear regressions were used to explore the relationships of adiposity categories with hypertension, dyslipidaemia, diabetes and CVD, and with systolic blood pressure, fasting total cholesterol and fasting plasma glucose, respectively. Adjustments were made for age, sex, education, country of birth, alcohol consumption, smoking, and sedentary behaviour. Physical activity and energy intake were initially included in the multivariate models, however, they did not alter the relationship between adiposity measures and metabolic variables in this study and were excluded from the final model to avoid over adjustment [17]. An interaction term was considered for age (dichotomised using the approximate sample mean age as <55 years and ≥55 years) and sex with the adiposity categories to test whether the relationship between adiposity and our chosen metabolic variables differed by age or by sex. As an interaction was found with both age and sex, we analysed men and women, and those aged <55 and ≥55 years separately.
To test whether any differences between BMI and BMI plus WC classification was due to the larger proportion of individuals classified as obese using BMI plus WC compared to BMI alone, we shifted the BMI cut-point to define obesity, such that the same proportion of people would be identified as obese using BMI plus WC and using BMI alone. The proportion identified as obese using BMI plus WC (BMI ≥30 kg/m2 and/or WC ≥102 cm for men and ≥88 cm for women) was 32 %, thus we examined the alternate BMI cut-point of ≥28.5 kg/m2 which also identifies 32 % of the study population as obese.
All analyses were performed using STATA® 11.2 (STATA, college Station, TC, USA).
Sensitivity analysis
To test for any potential effect of different BMI and WC cut-points for obesity for different countries of birth, a sensitivity analysis was performed excluding 1007 non-Europid participants.