Supplementary information

Model for Annual weight (BMI) gain across age and sex quantiles

We use a synthetic cohort approach using two cross-sectional health surveys to estimate longitudinal BMI change in the period 1995 to 2012 by matching members of the two national level surveys by birth year (1). The first was the National Nutrition survey of 1995 (2), and the second was the Australian Health Survey of 2011/12 (3). Specifically, the following 10 year age-groups for men and women in 1995 (20-29, 30-39, 40-49, 50-59) were matched to the 10 year age groups in 2012 (37-46, 47-56, 57-66, 67-76) respectively, thus matching by birth cohorts. To examine patterns of BMI change across the BMI spectrum we determined BMI change across each of 20 quantiles of BMI from 1995 to 2012 for four synthetic cohorts. Mean BMI was determined for each quantilewithin age/sex groups using survey analysis (‘svy’ suite of commands in Stata) which takes account of the survey design and allows estimates to be population representative. Annual weight gain based on age, sex and position on the BMI spectrum was then determined by assuming a fixed annual rate within the 17 years span. Using age and BMI from the mid –point of the matched surveys, this provided 80 estimates (20 from each of 4 synthetic cohort) of BMI change across matched quantiles within each synthetic cohort.Finally using multiple linear regression, we developed separate models for men and women for annual BMIgain based on age and position on the BMIspectrum. Polynomial splines were used to account for the plateauing of BMI gain at the upper part of the BMI spectrum, for women.

Model for annual BMI gain – women

Coef. / Std. Err. / P
Age ( years) / -0.005 / 0.0002 / <0.001
BMI, under 30 / 0.0185 / 0.0007 / <0.001
Constant / -0.0861 / 0.0205 / <0.001

e.g for a woman aged 30, whose BMI is currently 28

Annual weight gain =-0.0861-0.005*30 + 0.0185* 28 =0.28units BMI

e.g for a woman aged 30, whose BMI is currently 35

Annual weight gain =-0.0861-0.005*30 + 0.0185*30 =0.41 units BMI

This model captures the negative relationship of weight gain with age, and higher weight gain at the upper end of the BMI spectrum, but plateauing above BMI of 30.

.

Figure 1 Model fit for 4 synthetic cohorts:cohort 1= 20-29 years in 1995; cohort 2 = 30-39 years in 1995; cohort 3=40-49 years; cohort 4 =50-59 years . Grey circles = data from 1995 -2012 period; black line = regression model ( R2= 0.92); dotted lines= 95% confidence interval investigated in sensitivity analysis

Model for annual BMI gain – men

Coef. / Std. Err. / P
Age / -0.0043 / 0.0002 / <0.001
BMI / 0.0126 / 0.0005 / <0.001
Constant / -0.02405 / 0.0150 / <0.001

e.g for a man aged 30, with BMI of 28;

Annual weight gain =-0.02405-0.0043* 30 + 0.0126*28=0.20 units BMI

and for a man aged 50 with BMI of 28 :

Annual weight gain =-0.02405-0.0043* 50 + 0.0126*28=0.114 kg/m2

Figure 2 Model fit for 4 malesynthetic cohorts: cohort 1= 20-29 years in 1995; cohort 2 = 30-39 years in 1995; cohort 3=40-49 years; cohort 4 =50-59 years. Grey circles = data from 1995 -2012 period; black line = regression model ( R2= 0.94); dotted lines= 95% confidence interval investigated in sensitivity analysis

Modelling age and BMI specific annual mortality rate

We used age and sex-specific mortality rates (qx) from the ABS life tables of 20011/12to calculate a life table to incorporate the impact of BMI on mortality. The age-specific association of BMI with mortality was based on a large meta-analysis (4), as shown in Table 1. The model assumed an increase in mortality for individuals in a higher weight categories, compared with healthy weight. At the end of each simulation cycle, deaths were determined from the number of people alive multiplied by mortality rates for each weight category.

Example 1 shows how mortality rates are calculated to account for this association.

Table 1 – Hazard ratio (HR) per 5 kg/m2 higher BMI between 25 and 50 kg/m2as determined by Whitlock et al. 2009 ,with the referent a BMI of 22.5kg/m2

Age at risk (years) / HR
35-59 / 1.37 (95% CI 1.31 – 1.42)
60-69 / 1·32 (95% CI 1·27–1·36)
70-79 / 1·27 (95% CI 1·23–1·32)
80-89 / 1·16 (95% CI 1·10–1·23)

Mortality from the life table refersto overall mortality; hence using information on known prevalence of different BMI classes, this was apportioned to each BMI class.

Example 1: Calculating mortality rates for males aged 40years, by apportioning to each BMI class. From the life table, overall qx =0.00174.

Mortality rates for the healthy(BMI<25) population:

Mortality rates for weight categories overweight, obese I, obese II, obese III and obese IV:

Table 2 - An example of mortality rates for men aged 40 years for different BMI classes, from overall mortality rate from the life table, and age-specific prevalence of different BMI classes in 2011/12.

Healthy
BMI<25 / Overweight
BMI 25-25.99 / Obese class I
BMI 30-34.99 / Obese class 2
BMI 35-39.99 / Obese class 3
BMI 40-44.99 / Obese class 4
BMI >45
HR for mortality cf healthy weight / 1 / 1.37 / 1.372 / 1.373 / 1.37 4 / 1.37 5
Prevalence
( 2011/12) / 28.2% / 51.3% / 16.3% / 3.2% / 0.8% / 0.1%
qx / 0.00124 / 0.00170 / 0.00237 / 0.00319 / 0.00437 / 0.00598

Projection of obesity prevalence by age cohort

Table 3 Model projections of obesity at age 55-64 years for different birth cohorts

Birth cohort / Age in 1995 / % obese at age 55-64
men / women / overall / year
1961-1970 / 25-34 / 47 / 43 / 45 / 2025
1951-1960 / 35-44 / 39 / 34 / 37 / 2015
1941-1950 / 45-54 / 29 / 25 / 27 / 2005
1931-1940 / 55-64 / 25.3 / 24.8 / 25.0 / 1995

Estimation of likely impact of migration on obesity prevalence in 2008

In an ad hoc analysis we estimated the level and direction of the potential bias in predicted obesity prevalence in 2008 as a result of not including migration in the model.

Between 1995 and 2008, total net migration was around 1.6m people (5).

Assuming that half of these migrants came from Asia/SE Asia, were in a healthy weight range and remained in a healthy weight range until 2008, these would contribute to a total additional 0.8m healthy weight people in the population.

If the remainder of the migrants had same BMI profile as Australian adults by 2007, there would be an additional 298k healthy weight, 301k overweight and 214k obese.

Simulated adult population in 2007 ( without migration) =13.8m (composed of 5.1m healthy weight, 5.2m overweight and 3.7m obese)

Adult population including migration = 13.8 +1.6 =15.4m

Obese prevalence including migration = (3.7m + 0.2m)/ 15.4m = 25.4%( cf 26.7% in simulation)

Overweight prevalence including migration = (5.2m + 0.3m)/ 15.4m = 35.6%( cf 37.6% in simulation)

Hence the model may overestimate obesity prevalence by around 1.5% and overweight prevalence by 2% compared to real population data because it does not account for migration.

References

  1. A Hayes, E Gearon, K Backholer, A Bauman and A Peeters. Age specific changes in BMI and BMI distribution among Australian adults using cross-sectional surveys from 1980 to 2008. International Journal of ObesityAug;39(8):1209-16 doi: 10.1038/ijo.2015.50
  1. ABS Statistics 1996. Information Paper: National Nutrition Survey 1995; ( cat No. 4805.0) Canberra: Australian Bureau of Statistics
  1. Australian Health Survey: First Results, 2011-12; ( cat no 4363.055.001). Canberra: Australian Bureau of Statistics
  1. Prospective Studies Collaboration. Whitlock G et al. Body-mass index and cause-specific mortality in 900 000 adults: collaborative analyses of 57 prospective studies. Lancet 2009; 373(9669): 1083-96.
  1. Table 4 Migration to Australia since federation : a guide to the statistics available at