Introduction

Many populations worldwide are experiencing a marked increase in childhood obesity.1, 2 At the simplest level, obesity can be attributed to a dietary intake in excess of energy requirements. However, the irrefutable mathematics of the energy balance equation detracts from its ability to identify the behavioural or physiological causes of weight gain.3 There is increasing interest in the hypothesis that experience during early life may contribute to the risk of subsequent obesity.

A recent review discussed considerable evidence linking higher birthweight and later obesity.4 However, such studies have used body mass index (BMI) as the outcome. Since BMI is correlated with both lean and fat mass (FM), it is a poor index of fatness,5 and may be a misleading outcome in obesity research.6 Recent studies showed that birthweight is related to later lean mass (LM).7, 8, 9 Using a different approach, others reported a positive link between low birthweight and subsequent abdominal adiposity,10, 11, 12, 13, 14 although not all studies are consistent.15

Rapid growth rate in the first year of life has also been associated with greater BMI16 or skinfold thicknesses17 in mid-childhood. Rapid infant growth has also been linked to other components of the metabolic syndrome including insulin resistance.18 More recently, it was reported that rapid growth even in the first four months was related to adult obesity.19, 20 Again, however, it not clear whether fatness itself is associated with early growth, as skinfold data were not adjusted for body size.

These studies suggest that the tendency to childhood obesity may in part be ‘programmed’ in utero or during infancy. However, it is still unclear whether prenatal or postnatal growth rate is of greater importance for later obesity. In much research on nutritional programming, relationships between birthweight and later outcomes are only found after adjustment for current size. There is concern that this approach may lead to misinterpretation, attributing to fetal life events that are actually due to postnatal growth.21

Few studies on the development of obesity have been conducted in areas where malnutrition is common.11 Many non-western populations are undergoing nutritional transition, such that a high prevalence of low birthweight occurs in combination with exposure to an energy-dense childhood diet. The implications of this transition for childhood obesity are just beginning to receive attention. Weight gain in the first year of life was positively associated with childhood BMI in the Seychelles,16 and with skinfold fatness and high BMI in Brazilian adolescents.22 Again, however, body composition itself has not been studied.

Increased infant growth has previously been linked to decreased morbidity and mortality in a cohort from Brazil, where low birthweight is common.23 Thus, before altering the apparently beneficial strategy of promoting growth in those born small in such populations, it is important to obtain more detailed evidence concerning the relationship between early growth and later obesity.

Our aim was to test the hypotheses that indices of (1) prenatal growth (birthweight or ponderal index) and (2) postnatal weight gain (during infancy or childhood) are associated with body composition in later childhood in Brazilian boys.

Methods

The investigation was carried out in Pelotas, a southern Brazilian city. In this city, 99.7% of all infants born in 1993 participated in a longitudinal birth cohort study (n=5249). Measurements of birthweight and length were made in the whole sample, with further measurements of weight and length made at 6 months, 1 and 4 y in a systematic subsample. From 650 subjects with comprehensive data on early growth, a subsample of 172 boys was investigated at 9 y. The sample was selected to include high variability in both birthweight and postnatal growth. The sample was divided evenly by low birthweight (<2500 g) yes/no. In each of these groups, children were further sampled by above/below the median rate of weight gain from birth to 4 y. The Federal University of Pelotas, affiliated with the National Medical Research Commission, gave ethical approval.

Measurements comprised height (locally manufactured anthropometer), and weight and percentage fat (Tanita BF 350). This bio-electrical impedance instrument records body weight while simultaneously predicting %fat from impedance measurements obtained foot–foot, using integral predictive equations. Precision24 was 0.23 cm (height), 0.07 kg (weight), and 0.26 units (%fat). All measurements were made in the home or school of the subject, by a single investigator (JW). Ponderal index at birth was calculated as weight in kilograms divided by the cube of HT in metres. Data on weight and % fat were used to calculate FM, and hence LM by difference from weight.

Accuracy of Tanita for body composition assessment was investigated in a subsample of 18 boys. Data on % fat and weight were used to calculate LM and hence total body water, assuming hydration of LM to be 0.754.25 Total body water was measured by deuterium dilution, as described previously.26 Agreement between methods was evaluated.24 The mean bias was 0.4 (s.d. 0.8) litres of TBW (P=0.04), and there was no significant correlation between the bias and the magnitude of body water (r=0.11, P=0.67). Therefore, Tanita measures body composition in this population with a small bias, but this bias is not heterogeneous within the population.

Both FM and LM (kg) were adjusted for HT (m) in order to calculate the lean mass index (LMI) and the fat mass index (FMI). Following log-log regression analysis,26 LMI was calculated as LM/HT2, and FMI as FM/HT6. These variables are independent indices of FM and LM adjusted for body size.27 FMI is preferred to percentage fat as an index of fatness, as HT is a more appropriate aspect of body size than weight to adjust for. However, it is also useful to consider fatness in relation to the mass of metabolically active tissue,28 so using log–log analysis the ratio FM/LMn was also calculated, where the power of n was 2.7. These values were multiplied by 10 000 to improve clarity.

Data on weight and length or height were converted into standard deviation score (SDS) format, using UK reference data.29, 30 Obesity was categorised as BMI >95th centile. The weight SDS values were used to calculate change in SDS (ΔSDS) over the periods: birth (zero) to 6 months, 6–12 months, 1–4 y and 4–9 y.

The sample was categorised into quartiles for birthweight SDS, ponderal index, Δ weight SDS birth–6 months, 6–12 months, 1–4 y and 4–9 y. The association between early growth and childhood body composition was explored using ANOVA tests for trend and regression analysis controlling for confounding factors. Nonparametric analyses were conducted when variance of the quartiles was not homogenous. Several variables (breastfeeding duration, maternal education, BMI and weight gain during pregnancy, and familial monthly income) were tested as potential confounders. As the sampling scheme did not give equal probability of selection to all cohort boys, all analyses were repeated after weighting for the different sampling fractions. The weighted results were virtually identical to the unweighted values, which are reported below.

Results

A description of the sample is given in Table 1. Birthweight and length had negative standard deviation scores relative to the UK reference data; however, because of the wide variability between individuals induced by our study design, these differences did not achieve statistical significance. Monthly family income in the sample in US$ was as follows: <33 (16.2%); 33–99 (40.3%); 100–199 (26.7%); 200–667 (9.3%); >667 (7.5%). Using the BMI-based definition, 31 boys (18%) were categorised as obese at 9 y.

Table 1 Description of the sample (n=172a)
Full size table

The pattern of growth in terms of weight SDS is illustrated in Figure 1, showing the average rise through the UK centiles between birth–6 months, and between 1–4 y. The figure also shows the trend for all boys in the cohort from birth to 4 y, indicating that over this time period, the major increase in weight SDS in this population occurred after 1 y. The difference between these two lines reflects the oversampling of low birth weight infants in our sample. In the subsample of 172 subjects, the s.d. scores for height (birth −0.88 (s.d. 1.25); 6 months −0.30 (s.d. 1.43); 1 y −0.39 (s.d. 1.43); 4 y 0.53 (s.d. 1.42); 9 y 0.25 (s.d. 1.27)) were very similar to those for weight given in Table 1.

Figure 1
figure 1

The thick line shows mean weight SDS for the sample of 172 boys in this study between birth and 9 y, with the standard deviation shown as the error bar. The thin line shows mean weight SDS in all boys from the 1993 cohort, measured at birth, 1 and 4 y.

Full size image

Table 2 presents correlation analyses for Δ weight SDS between different time periods. Birthweight SDS was significantly negatively correlated with Δ weight SDS 0–6 months, but was not significantly related to later growth periods. Likewise, Δ weight SDS 0–6 months was significantly negatively related to Δ weight SDS 6–12 months, Δ weight SDS 1–4 y, and Δ weight SDS 4–9 y. After infancy, most correlations were not significant, although there was a weak negative correlation between Δ weight SDS 6–12 months and Δ weight SDS 4–9 y.

Table 2 Correlation of change in weight SDS between different periods of growth
Full size table

Table 3 displays the relationships between birthweight SDS or ponderal index and childhood height, BMI and body composition. Birthweight SDS was significantly related to childhood height, BMI and LMI. Ponderal index was positively related to childhood height. However, neither was significantly related to childhood FMI or the FM/LMn index. These relationships were unchanged if adjustment was made for socioeconomic status and maternal BMI. Weight gain 0–6 months, but not birthweight, was associated with later obesity prevalence (P=0.02).

Table 3 Relationships between prenatal and postnatal growth and childhood body composition
Full size table

Table 3 also presents the relationships between Δ weight SDS during or after infancy and childhood height, BMI and body composition. From 0–6 months, Δ weight SDS was positively associated with childhood height, BMI and LMI, but not with FMI or FM/LMn. There was a borderline association between Δ weight SDS 6–12 months and childhood height (P=0.08), but not for any of the other outcomes. From 1–4 y, Δ weight SDS was positively associated with all the outcome variables. However, from 4 to 9 y, Δ weight SDS was significantly related only to BMI, FMI and FM/LMn. These relationships were unchanged if adjustment was made for socioeconomic status and maternal BMI. Δ weight SDS 1–4 y and 4–9 y were significantly associated with the prevalence of obesity at 9 y (P<0.001).

Figure 2 summarises the relationships between the different growth periods and the outcomes. All analyses presented in Table 3 were repeated for height (data not shown); however, in general, the results were negative except that birth length, Δ length 0–6 months and Δ height 4–9 y were positively associated with height at 9 y. Analyses were also repeated for BMI, and here the findings were all similar to those for weight.

Figure 2
figure 2

Schematic diagram illustrating statistically significant associations between weight gain in the five growth periods and the outcomes at 9 y.

Full size image

Discussion

Growth during both fetal life and infancy has previously been linked to indices of later obesity. These findings have generated the hypothesis that early life is an important period during which later obesity may be programmed. However, the majority of previous studies have assessed whole-body obesity by measures wholly or largely dependent on relative weight, usually BMI. In the present study, we focused on both LM and fatness at 9 y, and aimed to adjust these outcomes appropriately for body size. This approach allowed us to consider discrete periods of growth in relation to both the fat and lean components of weight.

In Brazilian boys, we found that size at birth and infant growth rate were associated with later height, BMI and LMI, but not with later fatness. Neither total fat, adjusted for size, nor the ratio of FM to LM, were associated with fetal or early infant growth. These relationships between early growth and later relative lean size appeared to disappear by the second 6 months of life, as growth in weight during this period was not significantly related to any outcome.

The ability to detect associations between different growth periods and outcomes is influenced by both the length of the growth period, and the proportion of total growth that occurs during a given period. The potential effect of measurement error on calculated weight gain is also greater when the measurements are closer in time, and the weight gain smaller. These factors may in part explain why significant associations were found for Δ weight SDS 1–4 y, but not for Δ weight SDS 6–12 months. Nonetheless, it is notable that those periods during which there was a substantial change in weight SDS (see Figure 1) were also those in which growth rate was significantly related to later LM and/or fatness.

The results for birthweight and ponderal index are consistent with recent studies linking birthweight with LM rather than fatness in adolescence and adulthood.7, 8, 9 Our findings indicate that the period during which LM is programmed extends into the first 6 months of infancy. We found no such relationship for the second 6 months of infancy, which may reflect either the closure of a critical window during which early growth programmes later LM, or the reduced ability of our study design to detect such an association as mentioned above.

Although fetal and infant weight gains were not related to later fatness, an association was found between both birthweight and early infant growth rate and later BMI. Early infancy growth rate also predicted later obesity prevalence, if the condition is categorised using BMI. Although consistent with several other recent reports,16, 17, 19, 20, 22 this apparent link between early growth and later obesity is paradoxical, given the lack of any direct association with fatness itself.

One hypothesis is that hormonal programming, induced by early nutritional experience and growth rate, might influence subsequent metabolism and hence growth rate. However, our study provides no evidence of such a mechanism, as infant growth rate did not correlate positively with childhood growth rate (Table 3). Correlations between growth in different periods are expected to be negative due to both measures sharing a common term, and due to regression to the mean. However, our findings were also negative for growth periods separated by an intervening period. In the longer term, birthweight was not significantly associated with childhood growth rate. Thus, the association between early fetal and infant growth and later outcomes could not be attributed to consistency in growth rate.

A second hypothesis is that early growth rate programmes insulin resistance, which then interacts with environmental factors such as childhood diet and physical activity, making such individuals more prone to obesity. Recent studies in preterm infants have shown that slower early growth decreases the risk of subsequent insulin resistance.18 However, the lack of information on dietary intake and activity level at any time in childhood prevented us from considering this hypothesis in the present study.

A third hypothesis is that, if obesity is defined in terms of BMI, increased LM that has tracked from infancy onwards will increase weight and hence increase the likelihood of exceeding the obesity cutoff. The hypothesis that physique rather than obesity is programmed in early life therefore merits further attention. In the present study, this hypothesis is strongly supported by our finding that early weight gain, like birth weight, is associated with later LM but not with FM.

We have commented previously on the unsuitability of BMI for research into the aetiology of paediatric obesity.5, 6 Cross-sectional variability in BMI may be due to variability in relative lean deposition as well as fatness, and in childhood there is a two-fold variation in fatness for a given BMI value.5 Obese children have increased LM as well as fatness,31 and one hypothesis is that early excess lean deposition might precede subsequent excess fat deposition. Such a hypothesis is not supported by recent evaluations of children's body composition, which showed that contemporary children in general have lower levels of LM than in previous generations.32 Thus, increases in fatness, especially central fatness,33 appear to precede the increases in LM that occur in childhood obesity, with much of the apparent increase in LM attributable to increased height, and to changes in hydration and mineral mass that follow increases in weight.31

It is possible therefore that our finding that early rapid growth was associated with BMI, but not with fatness, may be due to two issues: first, variability in LM may adversely influence the categorisation of obesity; second, diet and activity levels may mediate possible relationships between early hormonal programming and later fatness. These issues require further investigation in order to understand their relative importance for public health strategies.

Despite the positive findings in relation to early growth and later obesity, the strongest influence on later fatness and obesity risk was weight gain from 1 to 4 y. Weight gain after infancy was strongly associated with later BMI, LMI, FMI and the FM/LM ratio. Weight gain from 4 to 9 y was also strongly related to fatness at 9 y. The rapid average rise through the weight centiles shown in Figure 1 likewise supports the notion that early childhood represents an important period in the aetiology of obesity. Alhough our sample was not representative of the Pelotas population, the results were essentially unchanged if sample weighting was taken into account.

Our finding that childhood, rather than infant, weight gain is most strongly related with later obesity and body fatness is consistent with a previous Brazilian study,22 highlighting the importance of directing public health programmes to environmental factors that influence childhood energy balance. Whether this is true for other populations undergoing nutritional transition and modernisation requires further research. However, we note that low birth weight is common in many such populations, and rapid childhood weight gain is also seen increasingly in modernising populations as they are exposed to the western lifestyle.

Body composition in this study was measured using Tanita, a bioelectrical impedance device. This technique is known to be population-specific,34 and we therefore assessed its accuracy in 18 boys from our study population. We found only a small mean bias, and no evidence of a relationship between bias and body size. We therefore believe that our results are not adversely influenced by the body composition method used. Furthermore, one of the main findings of our study (that obesity status can be attributed primarily to weight gain in childhood rather than fetal or infant life) is clear from the data on weight and height alone.

Conclusion

Our findings are consistent with other studies in associating rapid infant weight gain with an increased risk of later obesity categorised by BMI. However, increased early growth was associated with later LM rather than FM, suggesting that early weight gain might programme physique as opposed to fatness per se. In this population early childhood remained the dominant period for the development of obesity. The generalisability of our findings is likely to depend on whether the pattern of growth found in this cohort is observed in other populations also undergoing nutritional transition.