A number of studies have found significant associations between rapid infancy weight gain and later overweight(Reference Baird, Fisher and Lucas1), leading to the suggestion that prevention(Reference Paul, Bartok and Downs2, Reference Daniels, Magarey and Battistutta3) and even treatment of childhood obesity(Reference Cole4, Reference Chomtho, Wells and Williams5) should begin as early as the first year of life. However, weight gain in infancy reflects growth in bone and muscle as well as fat and some infants will be showing rapid gain in height or lean mass rather than adiposity(Reference Wells, Hallal and Wright6). Thus while on average rapid weight gain may predict later adiposity, what is not clear is how well it would prospectively identify individual children at risk. There is also recent research that suggests there are distinctive childhood eating behaviours related to overweight which may reflect an inherent tendency to overeat(Reference Carnell and Wardle7), so eating behaviour in infancy could predispose to, or protect against, later obesity(Reference Carnell and Wardle7, Reference Taveras, Scanlon and Birch8). Apart from studies examining how the type and style of milk feeding relates to later obesity(Reference Owen, Martin and Whincup9–Reference Agras, Kraemer and Berkowitz11), we currently know little about eating behaviour in infancy and even less about how it tracks on to later adiposity or eating style. We hypothesised that eating avidity, a global term to denote enthusiasm and hunger for food, might be a useful predictor of gain in fat.
The Gateshead Millennium Study was set up in order to examine infant growth and weight gain and how this relates to eating behaviour, prospectively measured from birth(Reference Parkinson, Pearce and Dale12). These children have now been followed into childhood where measures of body composition have been collected. We could thus use these data to explore the extent to which infancy weight gain and eating avidity predict adiposity later in childhood and the extent to which these can specifically identify children at risk of later adiposity.
The Gateshead Millennium Study aimed to recruit a whole population cohort of 1029 Gateshead-resident children within one week of birth between June 1999 and May 2000(Reference Wright, Parkinson and Drewett13) and the mothers of 81 % agreed to take part. The children were studied prospectively using parent report questionnaires shortly after birth, at 6 weeks and at 4, 8 and 12 months. The cohort has since been re-traced at school entry, parent report questionnaires completed at age 5–8 years, and a range of anthropometric and body composition measures collected at age 7–8 years(Reference Parkinson, Pearce and Dale12). The cohort was predominantly (98 %) white Caucasian, but the present analyses excluded a small number of Haredi (ultra orthodox Jewish) children who had follow-up data at age 7 years, as these children had very different growth and feeding patterns in the first year(Reference Wright, Stone and Parkinson14). Ethical approval for all phases of the study was granted by Gateshead and South Tyneside Local Research Ethics Committee.
Growth and body composition measures
Routinely collected clinic weights were returned with questionnaires by parents throughout the first year. At age 13 months, children were seen by research nurses who measured length (Raven rollameter) and weight (Seca scales). At age 7–8 years children were visited at school where research staff measured height (Leicester portable measure) and weight and leg-to-leg bioelectrical impedance (BIA) using the Tanita TBF-300MA. Measurements were also taken of triceps and subscapular skinfolds, using Holtain skinfold callipers, and waist circumference using a non-stretchable tape measure. Length, height, weight and BMI were converted into Z-scores compared with the UK 1990 reference(Reference Freeman, Cole and Chinn15). Change in weight Z-score from birth to 12 months conditional on birth weight(Reference Wright, Parkinson and Drewett16) was calculated to give a figure for conditional infancy weight gain (CWG). Waist and skinfold thicknesses were also converted into Z-scores using the best available external references(Reference McCarthy, Jarrett and Crawley17, Reference Tanner and Whitehouse18) and the mean skinfolds’ Z-score taken. The BIA data were converted into Z-scores for fat mass and lean mass, standardised for height, gender and age, using reference data from the ALSPAC (Avon Longitudinal Study of Parents and Children) cohort(Reference Sherriff, Wright and Reilly19).
Factor analysis, with a promax rotation, on the logged data was undertaken for thirteen measures of size and/or adiposity: height, width of shoulders, diameters of elbow, wrist, hip and knee, waist circumference, skinfold measures of the subscapular, triceps, biceps and suprailiac, and the impedance-based measures of fat and lean mass, as well as age. This produced a three-component model which explained 79 % of the overall variance. The first component consisted mainly of measures of fat mass (waist circumference, the four skinfold measurements and BIA fat mass), while the second consisted mainly of measures of size (height, shoulders, elbow, wrist, knee and lean mass) and the third consisted mainly of age. The first factor results were thus used to create an adiposity index.
Internal centiles adjusted for height were calculated separately for girls and boys. Children with values above the 75th and 90th internal centiles for any measure were defined as having raised or high values, respectively.
Infancy eating avidity measures
The parent report questionnaire at 12 months included twenty-five questions drawn from previous research and clinical practice selected to describe enthusiasm and appetite for or aversion to food, as well as any oro-motor feeding difficulties. Many of these correlated significantly with both weight gain and appetite at different ages, but even the most predictive of these variables (appetite) individually predicted only a tiny proportion of the variance in weight to 1 year (4 %)(Reference Wright, Parkinson and Shipton20). We thus needed some sort of data reduction and summarising process to form a composite measure of infancy avidity eating behaviours (avidity score).
We first explored these variables using principal components analysis, but while this produced apparently coherent factors, they were unrelated to weight gain, even when component variables were known to be individually predictive of weight gain(Reference Wright, Cox and Le Couteur21). We therefore adopted a new approach and used general linear regression modelling to identify which variables independently predicted conditional weight gain (CWG) from birth to 1 year. All variables that showed a borderline univariable association with CWG (P ≤ 0·2) were added to a multivariable general linear model. Variables with insignificant (P > 0·05) coefficients were removed from the model successively until only variables that were independently predictive (P ≤ 0·05) remained. These six items (see Table 4) were then combined by summing each regression coefficient, multiplied by the response to each item, to produce each child's avidity score.
The association of CWG and avidity with measures of stature and adiposity at age 7–8 years was assessed using Pearson's bivariate and partial correlations and the association of high weight gain or avidity with high adiposity using the χ 2 test. Post hoc analysis would suggest that with about 500 subjects the minimum detectable correlation would be about 0·18. This number would give 80 % power to detect a statistically significant relative risk of 2·5 between any child with values in the top 10 % of centiles compared with the remainder.
The original cohort comprised 1029 infants (51 % male), of whom 764 had growth data at age 1 year and 585 had body composition data at age 7–8 years, after exclusion of seven Haredi children. The growth characteristics of the cohort members in infancy and at follow-up when aged 7 years are shown in Table 1. Infancy CWG showed a significant association with adiposity at age 7 years, but a stronger association with height, BMI and lean mass (Table 2). Children with the most rapid CWG in infancy were nearly three times more likely than the remainder to go on to have high adiposity at age 7 years (Table 3). Children with high BMI at age 1 year were also more likely to go on to have high adiposity, but the two effects were not additive. Having relatively high weight gain and BMI was also associated with some increased risk of going on to have high adiposity (relative risk = 1·95).
CWG, conditional weight gain; SDS, standard deviation score; BIA, bioelectrical impedance.
*Compared with UK 1990 reference(Reference Freeman, Cole and Chinn15).
†Compared with McCarthy reference(Reference McCarthy, Jarrett and Crawley17).
‡Compared with Tanner and Whitehouse skinfold reference(Reference Tanner and Whitehouse18).
§Compared with ALSPAC (Avon Longitudinal Study of Parents and Children) reference(Reference Sherriff, Wright and Reilly19).
BIA, bioelectrical impedance.
Values are Pearson's correlation coefficients (r) of each measure, expressed as a Z-score, with conditional weight gain from birth to 12 months.
Twelve individual eating variables univariably predicted CWG across the first year at P ≤ 0·2 and of these, six remained independently predictive in a multiple linear regression model (Table 4). These variables were used to construct the avidity score available for 561 eligible children who also had weight data at 1 year. This score explained 8 % of the variability in CWG from 0 to 12 months (Fig. 1). There was substantial clustering of scores in the centre of the distribution, with 219 (37 %) children having the median value, but then a much wider spread in both directions; the 26 % of children with values above the median were thus categorised as having a raised avidity score (Fig. 1).
*Including all other variables in table. R 2 for whole multivariable model = 0·08.
†This variable was significant P < 0·2 in univariate analysis before recoding from five to three response categories.
At age 7–8 years, although correlation coefficients were low, the avidity score was significantly associated with both height and BMI in boys. A weak and non-significant correlation with adiposity became weaker still after adjustment for height (Table 5). Children with a raised avidity score in infancy showed only a borderline tendency to be in the high adiposity range (above 90th centile) at 7 years (Table 3). Children with avidity score data did not differ from the cohort as a whole at age 1 year (see Table 1) or in adiposity score at age 7–8 years (data not shown).
BIA, bioelectrical impedance.
Values are Pearson's correlation coefficients (r) of each measure, expressed as a Z-score, with infant avidity score.
Our aim in the present analysis was to explore the extent to which weight gain and eating behaviour in infancy can predict the risk of future obesity in childhood. While CWG did significantly relate to later adiposity, it was much more strongly related to size than to adiposity. A number of studies have now shown a relationship between early weight gain and later adiposity(Reference Chomtho, Wells and Williams5, Reference Wells, Hallal and Wright6, Reference Demerath, Reed and Choh22–Reference Ong, Emmett and Northstone25), which demonstrates that many children destined to become obese are probably already laying down excess fat in infancy. However, those studies which also considered the relationship between adiposity and height or lean mass also generally found a stronger association with later height than with fat mass(Reference Chomtho, Wells and Williams5, Reference Wells, Hallal and Wright6, Reference Cameron, Pettifor and De Wet23, Reference Ong, Emmett and Northstone25). If a majority of children with rapid weight gain in infancy are growing fast, rather than becoming fat, this would be a non-specific way of predicting future risk.
We had hypothesised that avid eating behaviours might be more a specific predictor of later adiposity, but in fact if anything they were less specific and their only significant association was with attained height. Rapid growth requires high nutrient intake, so it is plausible that infancy eating avidity would be driven by rapid growth at least as much as by any tendency to overeat and become obese. There has been little previous work relating infant eating avidity to growth or weight gain, apart from one, which found associations between milk feeding vigour and concurrent skinfold thicknesses in healthy infants(Reference Agras, Kraemer and Berkowitz11) as well as with BMI in the same infants at age 6 years(Reference Agras, Kraemer and Berkowitz10). A number of studies in childhood have found relationships between child eating behaviour and overweight or adiposity(Reference Carnell and Wardle26–Reference Johnson and Birch29), but these were not prospective, making it difficult to distinguish cause and effect. One study has shown that child eating behaviours track through mid childhood(Reference Ashcroft, Semmler and Carnell30). Recent work in this area has suggested that childhood eating behaviours associated with adiposity are quite strongly heritable(Reference Wardle and Carnell31) and thus one might expect that there would be behavioural associations with adiposity present at the earliest age. However, the apparent degree of heritability increases with the age of the children studied(Reference Haworth, Carnell and Meaburn32), suggesting that a heritable tendency to overeat may emerge later than the age studied here, once children are exposed to the family and wider food environment.
A weakness of the present study is that under half of the cohort had full data at the age of 7 years, but this level of attrition is comparable to other studies of this long duration when examining multiple measures(Reference Sherriff, Wright and Reilly19), and there were no systematic differences between those with full data and those without. While this may lessen overall representativeness, it should not invalidate the internal relationships revealed.
While BMI is highly correlated with measures of adiposity and is often a useful proxy, it can nevertheless mislead, particularly if children have unusual body composition(Reference Wright, Sherriff and Ward33) or activity levels(Reference Wells34). In the present study the avidity score was associated with BMI at age 7 years, but not with adiposity, probably reflecting the combined association of BMI with height and lean mass, rather than fat mass. One strength of the study is the wide range of body composition measures used. Measures of body composition are all relatively imprecise and prone to a range of biases(Reference Parker, Reilly and Slater35), but using different techniques should allow a ‘triangulation’ of the estimates. The adiposity index summarises that process, with the aim of arriving at an average value that should be more precise and accurate. A limitation is that the adiposity index requires complete data for six different measurements, which means that it was available for even fewer children than the single measures.
While rapid weight gain in infancy does predict later adiposity, both infancy weight gain and eating avidity predict subsequent height more strongly than adiposity. Those infants who both gained weight rapidly and had a relatively high BMI at the age of 1 year did have a nearly threefold greater risk of high adiposity at age 7 years; but even among these infants, a majority did not go on to have high adiposity at age 7 years. So what does this imply for directing interventions to ‘high-risk’ infants? Even parents of older children are often unaware that their children are overweight(Reference Parkinson, Drewett and Jones36) and unwilling to institute change(Reference Towns and D'Auria37). As infancy is a time when rapid growth and weight gain have great priority for parents, much more specific indicators than weight gain or eating behaviour are going to be needed if targeted interventions are to be either acceptable or effective.
Sources of funding: The Gateshead Millennium Study was supported by a grant from the National Prevention Research Initiative (incorporating funding from the British Heart Foundation; Cancer Research UK; the Department of Health; Diabetes UK; the Economic and Social Research Council; the Food Standards Agency; the Medical Research Council; the Research and Development Office for Northern Ireland Health and Social Services; the Chief Scientist Office, Scottish Government Health Directorates; the Welsh Assembly Government; and the World Cancer Research Fund) and the present analysis was undertaken with further funding from the Chief Scientist Office, Scottish Government UK. The cohort was first established with funding from the Henry Smith Charity and Sport Aiding Research in Kids (SPARKS) and followed up with grants from Gateshead NHS Trust R&D, Northern and Yorkshire NHS R&D, and the Northumberland, Tyne and Wear NHS Trust. Conflict of interest: None of the authors have any conflicts of interest. Author contributions: C.M.W. ran the infancy phase and helped plan the childhood phases of the study, participated in the analysis and drafted the paper. A.J.A. planned and ran the childhood phase of the study. M.S.P. produced the adiposity index. K.M.C. undertook the main analyses, supervised by A.S., and helped draft the paper. All authors have seen and commented on the manuscript. Acknowledgements: Thanks are especially due to the Gateshead Millennium Study families and children for their participation in the study and to the research team for their efforts. The authors appreciate the practical support of the Gateshead Health NHS Foundation Trust, Gateshead Education Authority and local schools, and the support of an External Reference Group.