The CDAH study is a population-based, prospective cohort study established to examine childhood predictors of adult CVD and diabetes. The CDAH study design and procedures have been described in detail elsewhere but will be summarized here
[17, 18]. Extensive lifestyle and biological data were collected in 1985 (baseline) on a representative sample of 8498 school children aged 7 to 15 years as part of the Australian Schools Health and Fitness Survey
. A sub-sample of 2809 children (9, 12 and 15 year-olds) underwent additional measurements, including blood pressure, blood lipids, and further fitness tests. These additional measurements were restricted to a sub-sample owing to economic and time constraints. The follow-up study, CDAH, was performed from May 2004 to May 2006. A total of 2410 were re-measured at one of 34 field-work clinics across Australia. The population was then aged 26–36 years. During the follow-up survey, a random sample of 204 participants (approximately 1 in 3 participants with relevant childhood measures as 9, 12 and 15 year-olds) had M-mode echocardiography performed. Complete data for these analyses were available on 181 participants. Echocardiography was restricted to a sub-sample owing to time constraints of field-clinics and the need to reduce respondent burden. Participants who received echocardiography were similar to those clinic attendees who did not have the examination with respect to BMI, total cholesterol, low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), triglycerides, insulin, and glucose, but had higher mean (SD) systolic (121 (13) vs 117 (13)) and diastolic (75 (9) vs 72 (9)) blood pressure, and waist circumference (86.0 (12.5) vs 83.6 (12.3)). The childhood characteristics of those with cardiac measures were also very similar to those who did not attend the follow-up study as adults on the following measures: BMI, waist circumference, total cholesterol, LDL-C, triglycerides, systolic and diastolic BP. Those with cardiac measures, however, had higher mean (SD) HDL-C (1.52 (0.31) vs 1.45 (0.28)).
At baseline, consent from both parent and child was required for inclusion in the study; at follow-up, all participants gave written informed consent. The baseline study was approved by the State Directors General of Education and the follow-up survey was approved by the Southern Tasmania Health and Medical Human Research Ethics Committee.
Weight was measured with participants in light clothes without shoes using regularly calibrated bathroom scales that recorded to the nearest 0.5 kg in 1985, and with a digital Heine portable scale that recorded to the nearest 0.1 kg at follow-up. Standing height was measured to the nearest 0.1 cm using a portable stadiometer, with the participant in bare feet. BMI was calculated using the formula: BMI = weight (kg)/height (m)
. Childhood overweight and obesity for BMI were defined using age and sex specific cut-points
. Waist circumference was measured to the nearest 0.1 cm on the skin or over light clothing. At baseline, the measurement was taken at the level of the umbilicus with a constant tension tape. At follow-up, a non-stretch tape was used to obtain the measurement at the narrowest point between the lower costal border and iliac crest.
Fat mass and percent body fat at baseline and follow-up was estimated from established regression equations that incorporate four measures of skin fold thickness. In 1985, tricep, bicep, subscapular, and suprailiac skin folds were measured at locations determined by reference to anatomical landmarks
 using Holtain Calipers to the nearest 0.1 mm. Three readings were taken from each site, and the readings were then averaged. At follow-up, tricep, bicep, subscapular, and iliac crest skin folds were measured using Slim Guide calipers to the nearest 0.5 mm. Measurements were repeated a maximum of three times, or discontinued if the first two readings were unchanged. The average of the two closest readings was used as the location specific score. As we have previously detailed
, skin fold values exceeding 40 mm at follow-up were imputed from BMI and waist circumference. Body density was estimated using regression equations for 9
, 12 and 15 year old children
 at baseline, and for adults at follow-up
. Body density was then used for the calculations of fat mass and percent body fat
Cardiorespiratory fitness was estimated sub-maximally at baseline and follow-up as physical working capacity at a heart rate of 170 beats per minute (PWC170) on a friction-braked bicycle ergometer (Monark Exercise AB, Sweden). The test protocol comprised three successive workloads of three minutes duration at baseline, or four minutes duration at follow-up. The workloads were selected on an individual basis to induce steady-state heart rate responses from the participant at the end of each workload. Heart rate was recorded during the final 20 seconds of each workload. Physical work capacity at 170 beats per minute was estimated by extrapolating the line of best fit from the heart rates recorded at each sub-maximal workload
[26, 27]. Cardiorespiratory fitness is expressed in relative terms as watts per kg (W/kg) of lean body mass.
Blood pressure measurements were obtained from the left brachial artery using a standard mercury sphygmomanometer at baseline, and from the right brachial artery using a digital automatic monitor (Omron HEM907, Omron Healthcare Inc, Kyoto, Japan) at follow-up. Blood pressure levels are reported as the mean of two measurements at baseline and the mean of three measurements at follow-up.
Blood samples were collected at baseline and follow-up from the antecubital vein after an overnight fast. In 1985, serum total cholesterol and triglycerides were determined according to the Lipids Research Clinic Program
, and HDL-C analyzed following precipitation of apolipoprotein-B containing lipoproteins with heparin-manganese
. In 2004–2006, serum total cholesterol, triglyceride, and HDL-C concentrations were determined enzymatically
. LDL-C concentration was calculated using the Friedewald formula
. At follow-up only, plasma glucose levels were measured enzymatically using the Olympus AU5400 automated analyser. Two methods of insulin determination were used during the follow-up study. Plasma insulin was measured by a microparticle enzyme immunoassay kit (AxSYM, Abbot Laboratories, Abbot Park, Illinois, USA) initially, before a change in kit by the laboratory to measure serum insulin determined by electrochemiluminescence immunoassay (Elecsys Modular Analytics E170; Roche Diagnostics, Mannheim, Switzerland). Insulin levels assayed using the first methodology were corrected to levels in participants assayed using the second methodology (as per correction factor equation of the laboratory).
The data analysis was performed with Stata software (Stata Inc., 2009, Texas, USA). Means and standard deviations (or medians and interquartile ranges) were used to summarize quantitative variables and percentages were used to summarize categorical variables.
In order to provide a graphical illustration of the associations between adiposity measures and LVMI, separate categorical variables describing pattern of change between childhood and adulthood in adiposity status were created. For each adiposity measure (BMI, waist circumference, fatmass and skinfold thickness) those in the highest quartile (age and sex specific) were defined as overweight/obese. For this analysis the top quartile of each adiposity measure was used to allow direct comparison across the adiposity measures. The resulting categories of change were: (1) normal weight as child and normal weight as adult; (2) overweight/obese as child and normal weight as adult; (3) normal weight as child and overweight/obese as adult; (4) overweight/obese as child and overweight/obese as adult. We present means on the LVMI outcome for each adiposity change category.
The relationship between quantitative measures of adiposity and LVM/LVMI was estimated separately for males and females using linear regression. LVM was used to assess BMI as LVMI is indexed to height. This was undertaken as the association between adiposity and CVD risk is linear. Relationships were estimated separately for the four indicators of adiposity: BMI, waist circumference, fat mass and skinfold thickness. A life course epidemiologic approach was used that simultaneously takes account of the contribution of adiposity in both childhood and adulthood
. Adiposity and change in adiposity were modeled as continuous predictors in the regression models. Change is based on taking the difference of continuous scores.
Models were fitted in which childhood adiposity and change in adiposity between childhood and adulthood were used as predictors (Model 1); the first model was extended to adjust for potential confounding factors in adulthood: fitness, age, triglyceride levels and total cholesterol (Model 2); finally a third model also included potential mediating factors in adulthood: systolic BP and FPG (Model 3). In these models the regression coefficient for the child adiposity variable is the difference in the mean cardiac outcome between subjects who are one unit apart in their adiposity level at both childhood and adulthood (e.g. subject A is one unit higher in BMI than subject B at both time points). If the association were causal it would represent the effect of an increase of 1 unit in adiposity at both childhood and adulthood. The coefficient for the change in adiposity variable quantifies the mean increase in the cardiac outcome associated with a one unit difference in change (e.g. comparing cardiac outcome between two children whose changes in adiposity are 1 unit apart, say an increase of 2 units versus an increase of 1 unit).