Variation in the relationship between birth weight and subsequent obesity by household income

The well-known fetal origins hypothesis stresses the importance of fetal development in links between measures of fetal and infant health and later-life health outcomes [1, 2]. In relation to this, it has been suggested that intrauterine growth rate is closely linked to the fetal development of tissues and organs that in postnatal life control eating patterns, physical activity, and metabolism, such as the hypothalamus, pancreatic b-cells, fat tissue, and muscles [3‐6]. Hence, a question of interest is whether the prenatal period affects later risk of overweight and obesity. A vast number of studies, including a systematic review and a meta-analysis, have found an association between high birth weight and the risk of overweight in children, adolescents, and adults [7, 8]. The interpretation of these associations may be difficult, however, because many of the studies compared persons who were born to different mothers and brought up in different families. Hence, several researchers have examined the within-family association between birth weight and later body mass index (BMI) in siblings, using sibling-fixed effects designs, and found significant associations [6, 9‐11].

While the association between birth weight and subsequent obesity is established, little attention has been paid to variation in this effect between population groups. Such considerations are pertinent given that theory and evidence suggests that parents allocate their investments unequally among their children, and by this reinforce or compensate for initial differences [12]. A growing number of studies also suggests that investment responses vary by family socioeconomic status (SES).

Parental investment responses to a high birth weight may vary by a family’s SES because expectations for children’s BMI, parenting knowledge about diet and exercise, and the availability of resources to respond to childhood obesity. It has been suggested that parental resources may affect (a) preferences for equity in their children’s outcomes, (b) the productivity of investments in obese children, and (c) knowledge about potential ways to compensate for a high BMI, leading some parents to reinforce and others to compensate [13, 14].

This means that the relationship between birth weight (BW) and later BMI might vary by socioeconomic factors. The framework in Fig. 1 show that household environment, which captures parental behavior for individual i (Z _i ), may affect subsequent obesity (Y _i ) in three ways. It affects obesity indirectly via its effect on birth weight (B _i ), which in turn affects subsequent obesity (arrow 1); it affects obesity directly (arrow 2); and, it affects obesity by modifying the relationship between birth weight and obesity (arrow 3).

The aim of the following study is to measure the third effect (arrow 3), i.e., whether or not the relationship between BW and later BMI varies by household environment, which in this study is measured by SES. A small but growing literature has examined whether a family’s SES mitigates the effects of poor child endowments on child outcomes [15‐20]. These studies find that poorly endowed children achieve worse outcomes and that the negative impacts of a poor endowment are often larger for children born in disadvantaged families. For example, Almond, Edlund [15] examine the effect of pollution from the Chernobyl disaster on the Swedish cohort that was in utero at the time of the disaster. Those who suffered the greatest radiation exposure were less likely to qualify for high school, and had lower math grades. However, the damage was much larger among those children whose parents were less educated. We have not found any studies that have looked at whether or not the impact of high birth weight on obesity varies by the family SES.

To investigate whether or not the relationship between birth weight and subsequent obesity varies by SES we used data from the Health Survey for England (HSE). The HSE contains measures of children’s anthropometrics, household characteristics (like SES) and other individual characteristics. The survey is representative for English children and includes the BW of children aged 2–16 years. This dataset provides us with a unique opportunity to examine the effects of high BW on obesity by SES.

We modelled childhood obesity for individual i as:where Y is a binary measure of obesity for individual i; B is a measure of BW; Z is a measure of household income; and X is a vector of individual, maternal and household characteristics. u is an error term and c and γ are coefficients to be estimated. To allow for the effect of birthweight to vary by income we ran a second model including an interaction between BW and income:were the ZxB is an interaction between household income and BW. If the Eq. [2] has a better fit than Eq. [1] it suggests that the impact of BW varies significantly by household income. We tested this by a likelihood ratio test.

We cannot rule out that omitted variables bias the relationship between BW and obesity. As children are brought up in different families both early life conditions and parental background factors can affect BW and obesity. To mitigate this we ran the following model using sibling-fixed effects specifications: where ij denotes individual i in family j, and ε represents a family fixed effect. This means that we compared only siblings within each family, and X is a vector of control variables that are not shared between siblings. These are: birth order; maternal age at delivery; whether or not the child was born prematurely; and, the child’s age interacted with gender. Household income does not vary across siblings, which means that we cannot include household income in itself in the models. However, the interaction between household income and BW (ZxB) might still be unique for each sibling and is included in Eq. 4.

Our outcome of interest is a limited dependent variable and our primary models are linear probability models (LPM) with heteroscedasticity robust standards errors. This will yield the best least squares approximation of the true conditional expectation function and we can interpret the coefficients as marginal effects. There are two reasons for choosing LPM over, e.g., logit or probit models. Firstly, the interpretation of interactions in non-linear models is less clear as the coefficients are multiplicative [36]. Secondly, non-linear fixed effects models has been shown to provide biased results [37], [Greene W, Han C, Schmidt P. The bias of the fixed effects estimator in nonlinear models. Unpublished Manuscript, Stern School of Business, NYU. 2002;29]. However, we reran part of our regressions with logit models and calculated marginal effects for interactions according to Ai and Norton [36]. The conclusions did not change.

We apply survey weights reported in the HSE to each observation. The individual survey weights were generated separately for adults and children. For children (aged 0 to 15), the weights were generated from the household weights and the child selection weights – the selection weights corrected for only including a maximum of two children in a household. The combined household and child selection weights were adjusted to ensure that the weighted age/sex distribution matched that of all children in co-operating households. The survey weights were not used in the regressions with sibling-fixed effects.

It is possible that observations are independent across households, but not within households. We therefore also controlled for clustered sampling within household using unique household identifiers that produced Huber/White/sandwich robust variance estimators that allowed for within-household dependence [38].

Roughly, two thirds of the full sample was also included in the sibling analyses (Table 1). Although the sociodemographic characteristics were comparable, the obesity prevalence was higher in the full sample (11.2%), than in the sibling sample (10.5%).

Predicted smoothed values of subsequent obesity by BW are in Fig. 2. The predictions are based on locally weighted regressions, with a bandwidth of 0.8. They show that obesity prevalence increased with BW. This pattern was found both in the children with a household income above and below the median. However, the figures suggest that the obesity prevalence increased relatively more in the children with a household income below median, compared with the children in families with an income above median. The findings in the full sample and in the sibling sample were similar, though the differences by household income appear to be more pronounced in the sibling sample.

Linear BW was associated with increased risk of obesity, while the log of household income was associated with reduced risk of obesity in regressions with covariates (Table 2). The interaction between BW and income was significant and the LR-test suggests a significantly better fit in the model with the interaction. These results were found in the full sample and the sibling sample. The BW-coefficient was significant in the sibling fixed effects specification. In addition, the interactions between income and BW remained significant in the sibling fixed effects models. This suggests that the impact of BW on subsequent obesity varies significantly with household income. The effect of BW was larger in children from low-income families.

When we stratified the children into tertiles by income, we also observe a more pronounced effect of BW on obesity, in the lowest income tertile (Table 3). Similar findings were done in the sibling fixed effects analysis, where the effect of BW, on obesity, was 0.067 in the lowest income tertile while it was 0.036 in the highest income tertile.

As shown in Fig. 2, the effect of BW on subsequent obesity might not be linear. Hence, we reran the analysis with a categorical BW variable. The findings, which are presented in Appendix 1, supports the results, presented in Tables 2 and 3 using linear BW. In low-income children, the association between high BW and obesity was significantly more pronounced, compared with high-income children.

The aims of this study were to investigate the relationship between birth weight and subsequent obesity, and whether or not this relationship varies by household income. Our main findings were that of a significant correlation between high birth weight and subsequent obesity, and that the correlation varied significantly by household income. The effect of BW on subsequent obesity was more pronounced in children from low-income households compared with children from high-income households.

We provide evidence to show that children born with high birth weight had significantly higher obesity prevalence, than those of normal birth weight. Variables shared between siblings did not explain our findings, according to the sibling fixed effects models. These results are qualitatively similar to those in other studies, which have also shown that high birth weight is associated with increased risk of subsequent overweight and obesity [7, 8]. Our findings also support earlier studies that have used sibling fixed effects and found a significant impact of birth weight on obesity [6, 11]. However, direct comparisons of the coefficients are difficult as most other studies report their findings in odds ratios, while we reported marginal effects. A few studies have shown an association between birth weight and subsequent obesity using UK data [39, 40]. However, this study is, to our awareness, the first study using UK data and a sibling fixed effects design.

Our main finding is however, that the association between birth weight and subsequent obesity varied significantly by household income; the association was more pronounced in children from lower income households. Children born with high birth weight in low-income households were more likely to be obese than those of normal birth weight in low-income household, and were more likely to be obese than those in higher income households who was born with the same birth weight. This trend was also observed after controlling for individual and household characteristics and in sibling fixed effects models. While a number of studies have investigated the impact of birth weight on subsequent obesity controlling for socioeconomic variables in multivariate analyses, we are not aware of any published studies that have stratified their analyses by socioeconomic status to investigate whether or not the association between birth weight on obesity varies by socioeconomic status.

Our findings are consistent with the theories proposed in the introduction. For example, the theory on strategic investment propose that parental allocation decisions largely are driven by conscious investments strategies of the parents [24]. This results in low-income families concentrating their limited resources on the higher-ability children, to maximize their human capital returns. While high-income parents adopt compensatory strategies because they can afford this, i.e. they devote more resources to the less endowed children, while still ensuring enough resources to their other children to maximize their probability of success.

An alternative explanation might be that low income parents might be more likely to lack the psychological and material resources to handle high need children [14]. Caring for high birth weight children might be more burdensome, not only are they at increased risk for being obese, they are also more likely to suffer from a number of chronic illnesses [41, 42] and have worse academic outcomes [43, 44]. Successful obesity interventions does not only demand a persistent effort from the child, they are also more likely to be successful if the parents are involved [45]. In this case, our results would not be a result of conscious investments, but from immediate responses to the current situation. Although our study cannot separate out one of these alternative explanations, it offers a strong empirical finding: the impact of birth weight on subsequent obesity varies by household income.

There are a number of implications of these findings. The fetal origins literature emphasizes fetal development and its consequences for later life health outcomes [1, 2]. However, this study finds that environmental and social factors may alter the biological effect of high BW on later life obesity. This suggests that the conditions in utero is less important for the development of obesity than the epidemiological literature has suggested as the home environment may—depending on whose home it is—promote or prevent the development of subsequent obesity in those being born with a high birth weight. If this is true, studies that do not account for heterogeneity in the effect of high BW on obesity may, on the one hand, overestimate the effect in lower SES families, but, on the other hand, underestimate the effect for children born into high SES families. Our findings also suggest that socioeconomic inequalities in obesity might be related to conditions in utero.

The present study has limitations. First, our measure of obesity was BMI, which has been criticized, e.g., because it does not incorporate body fat, which is an independent predictor of ill health [46]. In addition, earlier studies have suggested that birth weight is positively associated with both lean body mass and fat in adults [47]. Although we used age and gender specific cut-off values for obesity, caution is necessary when BMI is used as children and adolescents can experience growth in height and weight during brief periods [48]. Second, the children, in the sibling fixed effects models, were members of families with more children and were less obese than were the children in the total study population. Thus, the results of the sibling comparisons may not be fully representative for the total population of English children.

Our study has shown that, as in previous studies, high BW was associated with increased risk of subsequent obesity. In addition, we have shown that the association between BW and obesity was more pronounced in children from lower income households. High household income buffers the effect of birth weight on subsequent obesity.