Conceived and designed the experiments: DBA TIAS KRF HTR. Performed the experiments: HTR RD CH DBA TIAS. Analyzed the data: HTR RD CH DBA. Contributed reagents/materials/analysis tools: TIAS. Wrote the paper: KRF DBA TIAS AJS.
The authors have read the journal's policy and have the following conflicts: Dr. Allison has received grants, honoraria, donations, and consulting fees from numerous food, beverage, pharmaceutical companies, and other commercial, government, and nonprofit entities with interests in obesity and nutrition, and receives royalties from obesity-related books, Vivus Pharmaceuticals, and Kraft Foods Frontiers Foundation. Dr. Sørensen has had industrial collaborations indicated at
We attempt to elucidate whether there might be a causal connection between the socioeconomic status (SES) of the rearing environment and obesity in the offspring using data from two large-scale adoption studies: (1) The Copenhagen Adoption Study of Obesity (CASO), and (2) The Survey of Holt Adoptees and Their Families (HOLT). In CASO, the SES of both biological and adoptive parents was known, but all children were adopted. In HOLT, only the SES of the rearing parents was known, but the children could be either biological or adopted. After controlling for relevant covariates (e.g., adoptee age at measurement, adoptee age at transfer, adoptee sex) the raw (unstandardized) regression coefficients for adoptive and biological paternal SES on adoptee body mass index (BMI: kg/m2) in CASO were -.22 and -.23, respectively, both statistically significant (p = 0.01). Controlling for parental BMI (both adoptive and biological) reduced the coefficient for biological paternal SES by 44% (p = .034) and the coefficient for adoptive paternal SES by 1%. For HOLT, the regression coefficients for rearing parent SES were -.42 and -.25 for biological and adoptive children, respectively. Controlling for the average BMI of the rearing father and mother (i.e., mid-parental BMI) reduced the SES coefficient by 47% in their biological offspring (p≤.0001), and by 12% in their adoptive offspring (p = .09). Thus, despite the differing structures of the two adoption studies, both suggest that shared genetic diathesis and direct environmental transmission contribute about equally to the association between rearing SES and offspring BMI.
In its strategic plan for obesity research, the National Institutes of Health of the United States proposed, “Socioeconomic status is also related to the incidence and prevalence of obesity, such that the poor are disproportionately affected by obesity, regardless of race/ethnicity. Research is needed to further understand the impact of socioeconomic status on the development of obesity.” The inverse association observed in developed societies between obesity and indicators of socioeconomic status (SES), such as occupational status, income, and education has been well established
There are multiple plausible causal relations underlying the SES-obesity association
Although these hypotheses are neither mutually exclusive nor exhaustive, it is important to distinguish between hypothesis (a) and the others. That is, if it is true that low SES increases the likelihood of obesity through a direct causal effect then efforts to identify and subsequently modify the causative aspects of a low SES rearing environment would be justified. Alternatively, if the association is due to obesity causing a decline in SES then policies designed only to modify SES may not be effective.
Voluntary adoptions, in which young children are randomly assigned to rearing environments differing in SES, can be thought of as ‘natural experiments’ that can be used to estimate whether there are potential causal effects of the SES of the rearing environment on subsequent obesity. A Danish adoption study that implemented this notion found that the parental SES of both biological and adoptive parents was inversely associated with BMI in the adult offspring
Specifically, if the SES of the rearing parents does causally affect the child's BMI and there were no other ways of generating this association, then a correlation should exist between the rearing parents' SES and the BMI of the children they raise, independent of biological parental BMI. Moreover, the magnitude of correlation should be the same whether the child is a biological or adopted offspring. In contrast, if the correlation is partly or fully due to a shared genetic diathesis between SES and BMI, then (i) the correlation between the rearing parents' SES and the child's BMI should be smaller for the adopted offspring than for the biological offspring, (ii) a correlation should exist between the SES of the biological parents and the BMI of the adopted-away offspring, and (iii) the correlation between parental SES and biological offspring BMI should be reduced in absolute value when parental BMI is controlled (see
In summary, our primary goal was to investigate whether the association between rearing parent SES and adoptee BMI was statistically significant, and whether it remained so even after controlling for rearing parents BMI. We then sought to disentangle the respective contributions of environmental and biological components of the association of SES of the rearing environment and obesity in the offspring (
This study was declared non-human subject's research by the Institutional Review Board of the University of Alabama at Birmingham.
We define a family unit to be a collection of any analyzable combination of adopted offspring, biological offspring of parents in rearing household, adoptive/rearing parents, and biological parents as defined in the text. We used data from adoption studies that met the following criteria: (1) the study provides information on the weight, height, and/or BMI (measured or self-reported) of persons raised in adoptive families (adoptees); (2) the study provides information on the weight, height, and/or BMI (measured or self-reported) of either adoptive siblings (defined here as the biological offspring of the adoptive parents) or both the adoptive rearing parents and the biological parents of the adoptees; (3) the study provides information on the SES of the rearing environment (variables such as income, occupational prestige, and education); (4) the data is publicly available or readily obtainable; and (5) if data were available on the biological offspring of the adoptive parents, the regression of the adoptee BMI on their parents' SES indicators had to be negative in sign and statistically significant. The reason for including criterion #5 is because we did not wish to assess whether the SES-obesity association is present in every available dataset (the association does not necessarily exist in all populations
To obtain data from adoption studies that met the aforementioned criteria, we searched the following electronic sources: Inter-University Consortium for Political and Social Research (
Our search yielded eight datasets, two of which (The Survey of Holt Adoptees and Their Families, 2005 [HOLT] and The Copenhagen Adoption Study of Obesity [CASO]) met our inclusion criteria. The 6 other datasets (the National Health Interview Survey, 1987 Adoption Supplement; the National Child Development Study, UK; the Colorado Adoption Project; the Iowa Adoption Project; the Family Life Project; and the Panel Study of Income Dynamics (PSID) Childbirth and Adoption History) did not meet all criteria (i.e., did not contain weight/height or BMI information on adoptive siblings or the adoptive rearing parents and the biological parents of the adoptees).
The two datasets used, the Copenhagen Study of Obesity (CASO) and the Survey of Holt Adoptees and Their Families (HOLT), are described below and in
Characteristic | Copenhagen Adoption Study of Obesity (CASO) | Survey of Holt Adoptees and Their Families (HOLT) |
Dates of Study | 1924–1947 | 2004–2006 |
Adiposity Indicator | Body mass index (BMI) | Body mass index (BMI) |
Socioeconomic Status (SES) Measures | Occupational prestige scores | Income and education |
Country | Denmark | USA |
Total number of family units in the analyses | 831 | 1,207 |
Race of adoptees | European/Danish | Korean-American |
Adopted Offspring (N) | 831 | 1,690 |
Biological offspring of parents in rearing household (N) | 0 | 1,196 |
Adoptive/Rearing parents (N) | 1,637 | 2,414 |
Biological Parents (N) | 1,493 | 0 |
Dataset | Variable | Adopted Offspring | Biological offspring of parents in rearing household | Adoptive/Rearing parents | Biological Parents |
|
N | 831 | 0 | 827 mothers 811 fathers | 817 mothers 723 fathers |
Age, mean (sd) | 45.1 (8.3) | N/A | 33.5 (5.7) – mother 36.0 (6.5) – father | 24.3 (5.5) – mother 29.3 (8.6) – father | |
Sex, % female | 56.4% | N/A | 50% | 50% | |
BMI, mean (sd) | 25.0 (5.5) | N/A | 24.1 (4.0) – mother 25.3 (3.3) – father | 23.9 (4.4) – mother 25.1 (3.6) – father | |
Obesity (BMI ≥30) | 21.0% | N/A | 7.8% - mother 7.4% - father | 7.9% - mother 8.2% - father | |
|
N | 1690 | 1196 | 2414 | 0 |
Age, mean (sd) | 28.2 (4.6) | 32.3 (5.1) | 59.6 (6.3) – mother 62.1 (7.0) -father | N/A | |
Sex, % female | 70.5% | 37.8% | 50% | N/A | |
BMI, mean (sd) | 23.1 (3.7) | 24.0 (4.0) | 25.6 (4.9) mother 27.4 (4.2) father | N/A | |
Obesity (BMI ≥30) | 5.8% | N/A | 16.0% - mother 21.5% - father | N/A |
*For CASO, because of the sampling procedure used, the sample proportions reported in this row are valid descriptors of the sample utilized, but not of the population from which the sample was drawn. In the population overall, the prevalence of obesity was roughly 4% at the time the data were collected
Based on the Danish Adoption Register, the CASO
The study focused on families who adopted a Korean-American child through Holt International Children's Services from 1970 to 1980. The adoptees were quasi-randomly assigned to families in infancy using a queuing policy (i.e., on a first come, first served basis). The agency conducted a follow-up survey, HOLT, for family members when the adoptees had grown into adulthood
We opted to use traditional regression modeling, as opposed to structural equation modeling, for the following reasons: (1) ease of implementation, (2) ease of communication, and (3) robustness. Regarding robustness, structural equation modeling is more sensitive to violations in assumptions or normality than ordinary least squares (OLS) regression. Moreover, OLS regression also afforded us the opportunity to perform sensitivity analyses and diagnostics (i.e., bootstrapping, residuals analyses) to examine robustness. In each dataset, missing data were handled by multiple imputation
The dependent variable modeled was adoptee BMI. Residuals from models with adoptee BMI as the dependent variable were not normally distributed. Hence, adoptee BMI data were normalized by a log transformation and then re-scaled to have the same mean and variance as the original data. Covariates included adoptee age at measurement, adoptee age at transfer, adoptee sex, and age of all parents (adoptive and biological) at the time of the adoptee's birth. The primary independent variables were adoptive and biological paternal SES. The difference between the regression coefficients for adoptive versus biological paternal SES were tested as described by Neter et al.
Because data on multiple children from the same family were available, a linear mixed model (LMM) with correlated residuals in a compound symmetric structure was used to account for within-family effects. HOLT included 2,886 children from 1,207 families. The predictor variables for HOLT included the child's age and gender, the adoptive family's SES, the adoptive mother's BMI, and the adoptive father's BMI. The child's BMI was regressed on these variables. The adoptive family's SES was computed through Principal Components Analysis (PCA). The PCA incorporated the mother's education (highest completed grade), the father's education (highest completed grade), and household income. Z-scores from the standardized first principal component were used as predictors.
In the regression of adoptee BMI on biological and adoptive paternal SES, after controlling for covariates (i.e., adoptee age at measurement, adoptee age at transfer, adoptee sex, and age of all parents (adoptive and biological) at the time of the adoptee's birth), the joint regression coefficient of adoptive and biological paternal SES (that is the value if the regression coefficient estimated for both adoptive and biological parents' SES is constrained to be equal) was statistically significant (p = 0.011). The difference between the regression coefficients for the biological and the adoptive father's SES was not significant (p = 0.982) and the two coefficients were very similar (-.22 and -.23, respectively). This supports the notion of a possible causal effect of the SES of the rearing environment in that the correlation between the genetic influence on BMI and its influence on SES each accounted for roughly half the correlation between rearing environment SES and BMI (see
Controlling for parental BMI (both adoptive and biological) reduced the coefficient for biological paternal SES by roughly half (reduced by 44% in absolute value). Given that biological parent BMI is significantly correlated with both parental SES and with adoptee BMI, this reduction was statistically significant (p = 0.034; cf. 25). In contrast, controlling for parental BMI (both adoptive and biological) hardly changed the coefficient for adoptive paternal SES at all (reduced by 0.7% in absolute value). This is consistent with the idea that controlling for biological parental BMI controls for the correlated genetic diathesis, but not for any causal effect from SES (See
The rearing parents' BMI was positively associated with the BMI of biological children, but not with adopted children, which is consistent with a genetic influence on BMI
Among biological offspring, when mid-parental BMI (the average of mother and father BMI) was added to the model, the absolute value of the coefficient fell 47%, from -.42 to -.20 (p = .10 after adjusting for mid-parental BMI; p<.0001 for the reduction in value). In contrast, for the adopted offspring, when the mid-parental BMI of rearing parents was added to the model, the coefficient was virtually unchanged, from -.25 to -.22 (p = .09 for the change), and remained statistically significant (p = 0.02). In other words, the effect of SES was reduced in the presence of mid-parental BMI for biological children but was not significantly reduced for adopted children. After adjusting for mid-parental BMI, the apparent effect of rearing-parent SES was essentially the same in biological and adopted children (see
Our results, from two well-characterized adoption datasets, are consistent with a model in which both the rearing parents' SES and the genetic influence on BMI and its relation to SES contribute equally to the association between rearing parents' SES and the adiposity of their offspring.
Beyond the aforementioned publication examining the CASO study, we know of no other studies which have taken the approach we have to consider adoption as a form of ‘natural randomization’ to better assess the extent to which the SES association with obesity represents an effect of SES causing obesity. However, there is an existing literature discussing multiple alternative causal hypotheses and the fact that the nature and direction of causation is not known
With respect to obesity potentially leading to lower SES, obesity appears to reduce wages given equal qualifications
Our study has several strengths. First, the study made new use of adoption data to address a set of important questions that could not be ethically investigated via a randomized experimental trial (c.f.,
The limitations of this study include: the datasets relied upon self-reported height and weight as opposed to direct measurements, and moreover, some of the parental weight and height data were derived by proxy self-report by the children; the age at which adoptee BMI was assessed was quite different in the two studies; the two populations studied had relatively low rates of obesity; certain biases may have been introduced by the use of mail questionnaires; the adoptees in the CASO study did not come from abroad, raising the possibility that some adoptees were, in fact, familial adoptions (however, while we cannot eliminate this possibility entirely, the Danish Adoption Register sought to filter out all adoptions where there was some relationship between the child and the adoptive parents); and we did not investigate the role additional variables such as neighborhood factors (e.g., stress, crime, violence) might have had upon the results.
Given these limitations, despite the fact that the two study populations we analyzed samples from were quite different (e.g., HOLT study parents were better educated, offspring BMI was measured in the CASO study many years after they left their rearing environment), it should be examined whether our finding holds in other groups and time periods because the relation between SES and obesity does not appear to be constant across populations or within populations across time
In comparison with the previous analyses conducted with CASO
The finding that the association between the SES of the rearing environment and offspring adiposity has a component that is independent of the BMI of the rearing parents strongly suggests that the mechanism is not to be found in the frame of what is considered ‘cultural transmission’ of obesogenic factors in the family environment. It is possible that the SES or related psychosocial factors (e.g., cognition and educational proficiency) of the offspring are mediating the effects of parental SES by being related to both the SES of the rearing parents and the subsequent development of adiposity in the offspring
In conclusion, across two different datasets collected during two different time periods, in two different countries, and for two different ethnicities, we found remarkably similar results. These results suggest that roughly half of the association between the SES of the rearing parents and the subsequent BMI of their biological offspring whom they rear is due to a potential causal influence of the rearing parents' SES and that roughly half is an association due to a genetic correlation between BMI and SES. On the one hand, implication of some degree of causation is positive because it suggests that if we can identify the specific aspects of low SES that predispose to obesity, we may be able to influence such factors to achieve reductions in obesity risk. On the other hand, the results suggest that the effects of any such manipulations should be expected to, at most, have an effect equivalent to half of that which would be expected if the association were all causal.
(DOC)