¹Menzies Centre for Population Health Research, University of Tasmania, Hobart, Tasmania, Australia

²National Centre for Epidemiology and Public Health, Australian National University, Canberra, Australian Capital Territory, Australia

³University of Melbourne Department of Paediatrics and Murdoch Childrens Research Institute, Royal Children's Hospital, Melbourne, Victoria, Australia

Correspondence to: Professor T Dwyer, Menzies Centre for Population Health Research, University of Tasmania, GPO Box 252-23, Hobart 7001, Australia. E-mail: T.Dwyer@utas.edu.au

^†A statistical appendix further detailing the methods used in the factor analysis is a available from the website. See supplementary information

OBJECTIVE: Syndrome X (clustering of insulin resistance, dyslipidaemia and hypertension) in adults with central obesity has been suggested to be a consequence of poor foetal development. We investigated clustering of syndrome X factors in a sample of 8-y-old Australian children, and whether the clusters were associated with size at birth and childhood obesity.

DESIGN: Longitudinal, 1997 follow-up of children enrolled as singleton-born neonates in 1989.

SUBJECTS: A total of 298 healthy Australian children (208 boys, 90 girls, age range 7.4-8.9 y).

MEASUREMENTS: Anthropometry at birth and at 4 weeks. In 1997, at 8 y of age: fasting insulin and glucose, total and HDL-cholesterol, triglycerides and blood pressure.

RESULTS: Adverse levels of insulin and glucose, cholesterol and triglycerides co-existed more often than expected by chance (P<0.01). Three factors were identified in factor analysis: one loading on systolic and diastolic blood pressure ('blood pressure'); a second loading on insulin and glucose ('insulin resistance'); and a third loading negatively on HDL-cholesterol and positively on triglycerides ('dyslipidaemia'). The blood pressure factor was correlated with fatness at age 8 y (eg fat mass estimated from skin folds, r=0.11) and, after adjustment for current size, with birth weight (r=-0.15). Fat mass was also correlated with both 'insulin resistance' (r=0.24) and 'dyslipidaemia' (r=0.19). The increase in 'insulin resistance' (P=0.03) and 'dyslipidaemia' (P<0.01) per category of fat mass was greatest for subjects with higher-than-median subscapular-to-triceps ratio of skin folds. Neither 'insulin resistance' nor 'dyslipidaemia' was associated with anthropometry at birth.

CONCLUSIONS: The Syndrome X risk variables clustered among children who had a tendency to deposit fat on the trunk. There was no evidence in this sample that infant size predicts development of the insulin resistance or dyslipidaemic components of the syndrome by age 8.

International Journal of Obesity (2002) 26, 1301-1309. doi:10.1038/sj.ijo.0802111

birth weight; obesity; fat distribution; syndrome X; children

Introduction

Clustering of insulin resistance, dyslipidaemia (with hypertriglyceridaemia and low levels of high-density lipoprotein cholesterol) and hypertension has been identified in adults and referred to as Syndrome X.¹ Components of Syndrome X are most common and most severe in individuals with truncal or central obesity, and the combination of Syndrome X with central obesity has been referred to as the Metabolic Syndrome.² At issue is whether there exists a unique syndrome that explains the coalescence of the multiple risk variables.

One approach to detecting a common aetiology, if one exists, is to use factor analysis to analyse correlations between measured risk variables known to predict diabetes and cardiovascular disease. The purpose is to explain those correlations using a theoretical model in which values of each risk variable are determined by hypothetical, unmeasured 'factors'. This statistical method has been used in more than a dozen studies of adults³ and children⁴ since 1994. It should attribute the observed correlations to a single underlying factor if a single shared pathogenesis underlies the clustering of the risk variables. In the one study of children to date, Chen et al⁴ identified two factors that explained risk factor clustering observed for 5-11-y-olds from Bogalusa independently of sex or race: a metabolic process (insulin/glucose, dyslipidaemia and obesity) and a haemodynamic process (insulin and blood pressure). Their results suggest that the syndrome is a pair of related disorders that are linked through the action of elevated insulin, and that it can be observed in children.

There is some evidence that early life factors play a part in the development of this syndrome. Barker et al⁵ observed that Syndrome X among men and women born in England between 1920 and 1945 was associated with small head circumference and low weight at birth. Based on associations observed with birth size, they suggested that syndrome X should be renamed 'the small-baby syndrome'. To date, no study of this outcome in children has tested that hypothesis directly.

We had the opportunity to use the factor analytic approach in a sample of Australian children for whom we had collected relevant anthropometric measures in infancy and had measured blood pressure and biochemical parameters at age 8 y. Our objective was to identify any clustering of risk variables, and to determine whether the clusters were associated with birth weight and infant anthropometry independently of measures of growth and current size.

Methods

Subjects

The study population consisted of children from singleton births in Tasmania during 1989 who were eligible for inclusion in the Tasmanian Infant Health Study.⁶ Eligibility was based on a scoring system⁷ comprising six risk factors for sudden infant death syndrome (young maternal age, male sex, low birth weight, autumn or winter season of birth, maternal intention to bottle feed, and shorter duration of second stage labour).

There were 6813 live births in 1989, and 1256 were singletons eligible for inclusion. Of the 1256 infants, 696 were born in southern Tasmania (a defined geographical region). Parents of 96.3% (670/696) of those infants agreed to an in-hospital interview, and 86.5% (602/696) participated in a follow-up at one month.

In 1997, 545 of the 696 participants were identified at schools in southern Tasmania. After seven exclusions principally for remoteness of location, 538 were invited to take part in a study of HDL-cholesterol requiring venepuncture, and 73.7% (388/538) did so. Laboratory results of lipid and glucose analyses were returned for 379 of them, and insulin analyses were available for 340. Blood pressure measurements were obtained for 387 and, in total, 336 subjects had a complete set of measurements of insulin, glucose, cholesterol, HDL-cholesterol, triglycerides, systolic blood pressure and diastolic blood pressure.

The ethics committee of the University of Tasmania approved each stage of the study.

Measurements

At birth and in early infant life: Birth and infancy data were collected in three stages.⁶ Firstly, routine obstetric data (gestational age, placental weight, birth weight, crown-to-heel length, head circumference) were copied from hospital records. Secondly, direct measurements of the infants (mid-upper-arm circumference, subscapular and triceps skin-fold thicknesses) were made at an in-hospital interview soon after birth. Thirdly, weight, length, girths and skin-fold measurements were repeated at a home interview conducted at around 4 weeks of infant age. Skin-fold measurements were made using the methods of Tanner and Whitehouse.⁸

At age 8 y: Children for whom we had parental informed consent and child agreement underwent measurements of anthropometry and blood pressure, and had a 5 ml fasting blood sample taken.

Height was measured using a stadiometer with the subject in bare feet. Weight in light clothing was measured using bathroom scales that were calibrated daily using known weights. Skin-fold thicknesses were measured with calipers at six sites (triceps, biceps, subscapular, suprailiac, mid-abdominal and medial calf). Mid-upper-arm, hip and waist circumferences were measured with a plastic tape.

Blood pressure was measured three times using a Critikon Dinamap Adult/Pediatric Vital Signs Monitor with the cuff size selected in accordance with the mid-upper-arm girth measurement and the child seated with the left arm resting comfortably on a pillow with the elbow approximately level with the heart.

Blood samples were analysed by a laboratory accredited by NATA and participating in RCPA/AACB external quality assurance programmes. The Vitros analyser was used for biochemical estimation of serum glucose (Ortho Clinical Diagnostics Vitros Analyser test method) and triglycerides (Vitros Ektachem Test Methodologies manual). Insulin was measured by double antibody solid phase radioimmunoassay using the Phadebas Insulin kit. Standard enzymatic methods were used to measure serum cholesterol, and serum high density lipoprotein cholesterol (HDL-cholesterol), the latter after precipitation of very-low-density lipoprotein and low-density lipoprotein with dextran sulphate.

Data analysis

For data analysis, the sample was reduced to 298 subjects (208 boys, 90 girls) by excluding 38 children who reported that they had not fasted.

Derived anthropometric and Syndrome X risk variables: Relative birth weight was calculated as birth weight divided by the 50th percentile of weight at the same gestational age in a large representative sample of singleton births in Australia.⁹ Infant ponderal index (kg/m³) and head circumference-to-length ratios were calculated.

Body composition at age 8 y was assessed from four skin-fold thicknesses (triceps, biceps, subscapular, suprailiac) as previously described.¹⁰ Intra-abdominal adipose tissue area of these Caucasian subjects was estimated from abdominal and subscapular skin-fold thicknesses (mm) using the prediction equation of Goran et al.¹¹ Ponderal index (kg/m³) and body mass index (BMI, kg/m²) were calculated as indicators of body fatness.

Low-density lipoprotein cholesterol (LDL-cholesterol) was calculated from total cholesterol, HDL-cholesterol and triglycerides.¹² The HOMA¹³ equation (insulin (mU/l)´ glucose (mmol/l)/22.5) was used to calculate an index of insulin resistance. The average of three measurements was used for systolic and diastolic blood pressures.

Medians and interquartile ranges are presented because distributions of the anthropometric and Syndrome X risk variables were not symmetric.

Comparisons of infant anthropometry and maternal characteristics were made using analysis of variance and chi-squared tests of independent proportions to determine whether the 298 study subjects differed from the remaining 372 members of the source cohort who either did not participate in follow-up (n=334) or participated but were excluded because they had not fasted (n=38).

Correlation between Syndrome X risk variables: Product-moment correlation coefficients (r) were calculated as a measure of the association between the risk variables, which were log-transformed to improve the symmetry of their distributions.

Clustering of adverse levels of Syndrome X risk variables: Adverse levels of the risk variables were defined as values in the lowest 25% for HDL-cholesterol and highest 25% for insulin, glucose, LDL-cholesterol, triglycerides, systolic pressure and diastolic pressure. The observed number (O) of subjects with adverse levels of each pair of risk factors was compared with the expected number (E) calculated under the null hypothesis of independent risk factors. The cluster with the greatest O/E ratio is reported in each case. Significance testing was based on the one-sample binomial test (two-sided) using exact methods.¹⁴

Interpretation of correlation between risk variables: factor analysis: Factor analysis¹⁵ by the principal factor method, a standard option in the statistical package¹⁶ used, was undertaken to interpret relationships between the risk variables. Oblique transformations were used in interpretation, resulting in factors that were correlated. The matrix of correlations between the risk variables and factors (the 'factor structure') is reported. Further details are available in a statistical appendix on the website.

Measures of birth size and fatness at age 8: associations with metabolic factors: Spearman correlation coefficients were calculated from ranks of the data. The correlation coefficients were routinely corrected by adjusting for mother's age, baby's gender, month of birth, maternal intention to breastfeed and duration of second stage labour. These are five of the six cohort selection factors. The sixth selection factor was a study factor (birth weight). Age at the time of each measurement was also routinely adjusted for.

Fat distribution as an effect modifier: Factor analysis was repeated for groups of subjects classified by their subscapular-to-triceps skin-fold thickness ratio (STR). In regression analysis of metabolic factors on fat mass, which was categorized into thirds, linear trends were estimated using predictors taking integer scores (1, 2, 3) for category of fat mass. To compare the trend in mean scores between groups of subjects defined by their STR, a binary (0/1) term for STR and a (fat mass´STR) product term were included in the regression. Probability values reported are those from the test of significance of the coefficient of the product term.

Results

Anthropometric and Syndrome X risk variables

There were more boys (n=208) than girls (n=90) in this sample of 8-y-old singletons. This reflects the positive weight given to male sex in selection of babies for inclusion in the source cohort. Because low birth weight was also a selection factor, 20% (42/208) of boys and 24% (22/90) of girls had birth weight below the 10th percentile of Australian birth weights at the same gestational age. The full range of birth weights was nevertheless represented. For example, 8.2% (17/208) of boys and 8% (7/90) of girls had birth weights exceeding the 90th percentile.

Table 1 shows median levels of measures of size and fatness in infancy and at 8 y of age for the 298 subjects. Boy participants in this study had smaller (P<0.01) crown-to-heel length at birth (median 48.0 cm) than the (n=257) boys from the source cohort not included here (median 51.0 cm), but otherwise study participants were statistically indistinguishable from non-participants in respect of infant measures in Table 1. The two groups were alike also in maternal characteristics, including mother's age and education and the proportion with private health insurance. Table 1 also shows levels of Syndrome X risk variables at 8 y for study participants.

Correlation between Syndrome X risk variables

The correlation matrix is shown in Figure 1 for boys (below the diagonal) and girls (above the diagonal). Because of the similarity of the correlations for boys and girls, we combined them in subsequent analyses.

Clustering of adverse levels of Syndrome X risk variables

Table 2 shows that risk variables were clustered in the same individuals more than expected by chance. The observed-to-expected ratio increased with the number of risk variables involved in each cluster, but there were few subjects in the more extreme clusters.

Interpretation of correlation between risk variables: factor analysis

Factor analysis was undertaken to determine whether observed data on seven Syndrome X risk variables (insulin, glucose, HDL-cholesterol, LDL-cholesterol, triglycerides, systolic pressure and diastolic pressure) could be interpreted as being derived from a small set of conceptually meaningful but unmeasured 'factors'. Neither one (P<0.001) nor two (P<0.001) factors were sufficient to explain the correlations between the measured variables, but three (P=0.525) were. Results of estimating three factors are shown in Table 3. Factor F1 ('blood pressure') was correlated with systolic pressure and diastolic pressure, but not with any other risk variable. Factor F2 ('insulin resistance') was correlated with insulin and glucose, but not with any other risk variable. Factor F3 ('dyslipidaemia') was correlated most strongly with triglycerides and HDL-cholesterol, and less so with LDL-cholesterol, but not with any other risk variable.

Also shown in Table 3 are correlations between the factors. Reflecting the correlations between risk variables in Figure 1 'insulin resistance' and 'dyslipidaemia' were modestly correlated, but neither was correlated with 'blood pressure'.

Measures of birth size and fatness at age 8: associations with metabolic factors

The 'blood pressure' factor was correlated with measures of fatness at age 8 y, including fat mass estimated from skin folds (r=0.11, P=0.05). After adjustment for fat mass and lean body mass at 8 y, the 'blood pressure' factor was correlated with birth weight (r=-0.15, P=0.01). These associations are not further detailed here because we¹⁷ have previously reported the associations of infant and childhood size with systolic pressure for the children in this sample.

Table 4 shows that the other two factors, 'insulin resistance' and 'dyslipidaemia', were in general only weakly correlated with infant size. In contrast, measures of fatness at age 8 y were positively associated with both factors. For 'dyslipidaemia', correlations with measures of fatness at age 8 were stronger when calculated from the raw data (Pearson correlations), thereby allowing large values to have more influence (eg logarithm of fat mass, r=0.16, P=0.03).

Also shown in Table 4 is the effect of adjusting associations of infant size with 'insulin resistance' and 'dyslipidaemia' for fat mass at age 8 y. The weak associations were reduced because fatter children at age 8 tended to have been larger infants. For example, fat mass was positively correlated with weight at birth (r=0.20, P<0.001) and at 4 weeks (r=0.26, P<0.001).

Fat distribution as an effect modifier

The modest correlation between the two metabolic factors in Table 3, 'insulin resistance' and 'dyslipidaemia', was not consistent with the clustering between risk variables seen in Table 2. To investigate, we stratified subjects by STR as a measure of fat distribution. The two metabolic factors were more highly correlated among the 149 subjects with higher-than-median STR (r=0.39) than among the 149 subjects with lower-than-median STR (r=0.07). Similar but less pronounced differences were found with the subjects categorized by waist circumference, or by any other measure of fatness, but not when subjects were categorized by waist-to-hip ratio, for which correlations for the higher-than-median and lower-than-median subjects were very similar (r=0.26 and r=0.24, respectively).

Furthermore, the association of every measure of fatness at 8 y with each of 'insulin resistance' and 'dyslipidaemia' was greater for subjects with higher-than-median STR than for subjects with lower-than-median STR. Results are depicted in Figure 2 for fat mass categorized in thirds. The tests of difference in the trend per category of fat mass gave P=0.03 ('insulin resistance') and P<0.001 ('dyslipidaemia').

There was no evidence that distribution of body fat was a modifier of the weak association of infant size with either 'insulin resistance' or 'dyslipidaemia', however.

Discussion

We examined several risk factors for diabetes and CVD (fasting insulin and glucose, HDL-cholesterol, LDL-cholesterol, triglycerides, systolic and diastolic blood pressure) in this study of 8-y-old Australian children. Clustering of insulin resistance, dyslipidaemia and hypertension has been identified in adults and referred to as Syndrome X.¹ In factor analysis of the correlations between the risk variables, however, we found no evidence of a single metabolic abnormality. Instead the risk variables appeared to resolve into three independent groups. The first ('blood pressure') related only to systolic and diastolic blood pressure, the second ('insulin resistance') related only to insulin and glucose, and the third ('dyslipidaemia') related to HDL-cholesterol negatively and to LDL-cholesterol and triglycerides positively. The finding that lipids and blood pressure were associated with different factors is consistent with results of the previous study of children by the Bogalusa group.⁴ In that study, however, insulin was associated with both factors. Factors in this study were not as clearly linked in that way.

The reason for attempting to identify clusters via factor analysis was to gain insights into potential causal pathways. We were particularly interested in identifying clusters of risk factors that might be associated specifically with either measures of birth size (a summary measure of foetal growth) or childhood obesity. The two metabolic factors we identified¾'insulin resistance' and 'dyslipidaemia'¾were not strongly associated with birth weight or with head circumference, or with any of a number of other measures of infant anthropometry that we examined. On the other hand, childhood obesity¾whether measured as BMI or using skin fold thicknesses¾was strongly associated with both metabolic factors. In contrast to the study⁴ of children from Bogalusa using the factor analytic method, we found no link between blood pressure and the metabolic factors other than an indirect link through a common influence of obesity.

Adverse concentrations of insulin and lipids have been observed¹⁸ to be most common in children with a preponderance of body fat in the abdomen, upper body and trunk. Consistent with this, stronger associations between 'insulin resistance' and 'dyslipidaemia' became evident in our study when children with higher-than-median ratio of subscapular-to-triceps skin fold thickness were examined separately.

Our finding that no infant factor was an important predictor of either 'insulin resistance' or 'dyslipidaemia' before or after adjustment for current fatness is contrary to the finding of some other studies. Some researchers have reported that low birth weight^{19,20,21,22,23} or thinness at birth²⁴ is associated with one or more metabolic abnormalities independently of current size of children. Of particular interest are the findings by Bavdekar et al²¹ because their sample of 8-y-old Indian children included a high proportion with low birth weight. The Indian children at highest risk of metabolic abnormalities were low birth weight infants who became fat at age 8 y. This was interpreted as possibly reflecting a 'thrifty phenotype' adaptation to poor intra-uterine environment and subsequent over-nutrition. We were interested to test this in a sample of children from a Caucasian population. There was no evidence of elevated risk in the fattest 8-y-old children of low birth weight. Conversely, there was some suggestion in our sample that the 8-y-olds most at risk for insulin resistance were the fattest children who had been the fattest infants.

Differences between our findings and those from other studies are unlikely to be explained by selection bias in this study. Our response rate was relatively high and respondents were similar to non-respondents in most respects. In addition, our sample provided a good opportunity to investigate the importance of birth size because children of low birth weight were over-represented. Subjects were not representative of all Australian children, but nor do they need to be for causal inferences to be valid.^25,26 In addition to having an adequate sample size and a well-defined study population, what is important in an analytical study such as ours is that the sample contains a wide distribution of study factors and their effect modifiers.²⁵ That was the case in our sample.

In summary, we found clustering of Syndrome X risk variables among children who were the fattest at age 8 y and who had a tendency to deposit fat in the upper body. However, infant anthropometry did not predict development of early stage metabolic syndrome by 8 y independently of current fatness. Based on this sample, we have to conclude that, despite the likely connection between childhood blood pressure and low birth weight, it is fatness in childhood, rather than infant size or postnatal growth, that is the important determinant of early stage insulin resistance and dyslipidaemia.

Acknowledgements

This study was funded by the National Health and Medical Research Council of Australia. The Tasmanian Infant Health Survey was funded by the National Health and Medical Research Council of Australia, US National Institutes of Health (grant 001 HD28979-01A1), Tasmanian State Government, Australian Rotary Health Research Fund, Sudden Infant Death Syndrome Research Foundation, National Sudden Infant Death Syndrome Council of Australia, Community Organizations' support programme of the Department of Human Services and Health, Zonta International, Wyeth Pharmaceuticals and Tasmanian Sanatoria After-Care Association. Dr Ponsonby was supported by a National Health and Medical Research Council PHRDC Fellowship. Dr Morley was supported by VicHealth (The Victorian Health Promotion Foundation). We thank the nurses who took blood samples and measured blood pressures and anthropometry, and especially thank the children who participated in the study, their families and the schools that provided facilities.

1 Reaven GM. Role of insulin resistance in human disease. Diabetes 1988; 37: 1595-1607. MEDLINE

2 Zimmet P. Diabetes epidemiology as a tool to trigger diabetes research and care. Diabetologia 1999; 42: 499-518. Article MEDLINE

3 Sakkinen PA, Wahl P, Cushman M, Lewis MR, Tracy RP. Clustering of procoagulation, inflammation and fibrinolysis variables with metabolic factors in insulin resistance syndrome. Am J Epidemiol 2000; 152: 897-907. MEDLINE

4 Chen W, Srinivasan S, Elkasabany A, Berenson G. Cardiovascular risk factors clustering features of insulin resistance syndrome (Syndrome X) in a biracial (black-white) population of children, adolescents and young adults. Am J Epidemiol 1999; 150: 667-674. MEDLINE

5 Barker D, Hales C, Fall C, Osmond C, Phipps K, Clark P. Type 2 (non-insulin dependent) diabetes mellitus, hypertension and hyperlipidaemia (syndrome X): relation to reduced fetal growth. Diabetologia 1993; 36: 62-67. MEDLINE

6 Dwyer T, Ponsonby A-L, Newman N, Gibbons L. Prospective cohort study of prone sleeping position and sudden infant death syndrome. Lancet 1991; 337: 1244-1247. MEDLINE

7 D'Espaignet E, Dwyer T, Newman N, Ponsonby A-L, Candy S. The development of a model for predicting infants at high risk of sudden infant death syndrome in Tasmania. Paediatr Perinat Epidemiol 1990; 4: 422-435. MEDLINE

8 Tanner JM, Whitehouse RH. Revised standards for tricep and subscapular skinfolds in British children. Arch Dis Child 1975; 50: 142-145. MEDLINE

9 Beeby P, Bhutap T, Taylor L. New South Wales population-based birthweight percentile charts. J Paediatr Child Health 1996; 32: 512-518. MEDLINE

10 Dwyer T, Blizzard C. Defining obesity in children by biological endpoint rather than population distribution. Int J Obes Relat Metab Disord 1996; 20: 472-480. MEDLINE

11 Goran MI, Gower BA, Treuth M, Nagy TR. Prediction of intra-abdominal and subcutaneous abdominal adipose tissue in health pre-pubertal children. Int J Obes Relat Metab Disord 1998; 22: 549-558. Article MEDLINE

12 Assmann G (ed.). Lipid metabolism disorders and coronary heart disease MMV Medizin: Munich, 1993.

13 Mathews D, Hosker J, Rudenski A, Naylor B, Treacher D, Turner D. Homeostasis model assessment: insulin resistance and beta-cell function from fasting glucose and insulin concentrations in man. Diabetologia 1985; 28: 412-419. MEDLINE

14 Rosner B. Fundamentals of Biostatistics 3rd edn PWS-Kent: Boston, Mat, 1990, pp 228-230.

15 Bartholomew DJ, Knott M. Latent variable models and factor analysis 2nd edn Arnold: London, 1999, pp 41-76, 173-180.

16 SAS Institute Inc. SAS/STAT User's Guide, Version 8 SAS Institute: Cary, NCI, 1999, pp 1121-1192.

17 Dwyer T, Blizzard L, Morley R, Ponsonby A-L. Within pair associations between birth weight and blood pressure at age 8 in twins from a cohort study. Br Med J 1999; 319: 1325-1329.

18 Freedman D, Serdula MK, Srinivasan S, Berenson GS. Relation of circumferences and skinfold thicknesses to lipid and insulin concentrations in children and adolescents: the Bogalusa Heart Study. Am J Clin Nutr 1999; 69: 308-317. MEDLINE

19 Whincup P, Cook D, Adshead F, Taylor S, Walker M, Papacosta O, Alberti K. Childhood size is more strongly related than size at birth to glucose and insulin levels in 10-11-year-old children. Diabetologia 1997; 40: 319-326. Article MEDLINE

20 Crowther NJ, Cameron N, Trusler J, Gray IP. Association between poor glucose tolerance and rapid post natal weight gain in seven-year-old children. Diabetologia 1998; 41: 1163-1167. MEDLINE

21 Bavdekar A, Yajnik C, Fall C, Bapat S, Pandit A, Deshpande V, Bhave S, Kellingray SD. Insulin Resistance Syndrome in 8-year-old Indian children. Small at birth, big at 8 years, or both? Diabetes 1999; 48: 2422-2429. MEDLINE

22 Cowin I, Emmett P. Cholesterol and triglyceride concentrations, birthweight and central obesity in pre-school children. Int J Obes Relat Metab Disord 2000; 24: 330-339. Article MEDLINE

23 Fewtrell MS, Doherty C, Cole TJ, Stafford M, Hales CN, Lucas A. Effects of size at birth, gestational age and early growth in preterm infants on glucose and insulin concentrations at 9-12 years. Diabetologia 2000; 43: 714-717. MEDLINE

24 Law CM, Gordon GS, Shiell AW, Barker DJ, Hales CN. Thinness at birth and glucose tolerance in seven-year-old children. Diabet Med 1995; 12: 24-29. MEDLINE

25 Miettenen O. Design of the Study Base. In: Theoretical Epidemiology: Principles of Occurrence Research in Medicine John Wiley: New York, 1985, pp 46-68.

26 Rothman KJ. Modern Epidemiology Little, Brown: Boston, MA, 1986.

Figure 1 Correlations between Syndrome X risk variables in 8-y-olds.

Figure 2 Mean levels of the two metabolic factors by level of fat mass, for subjects classified as having low truncal fat (lower-than-median subscapular-to-triceps ratio of skin-fold thickness) or high truncal fat (higher-than-median subscapular-to-triceps ratio).

Table 1 Infant size, current size and fatness, and metabolic variables in a sample of 8-y-olds

Table 2 Clustering of Syndrome X risk variables in 8-y-olds

Table 3 Factor analysis of the correlations between Syndrome X risk variables in 8-y-olds

Table 4 Correlations of measures of infant and child size with two metabolic factors, 'insulin resistance' and 'dyslipidaemia' at age 8 y