Higher obesity in US racial/ethnic minorities has been linked to their disproportionate representation in low socioeconomic status (SES) groups.1 2 However, the simple inverse relationships between SES and obesity do not necessarily apply in racial/ethnic minorities,3 4 suggesting disparities in the health benefits conferred from high SES.5 6 Racial/ethnic disparities in obesity are of particular interest during the transition from adolescence to adulthood, characterised by dramatic obesity incidence and divergent racial/ethnic trends,7 8 but have been largely understudied because of challenges in defining SES during this period.

The complex relationship between race/ethnicity, SES and obesity during the transition to adulthood can be better understood using the life course approach, which posits that factors acting early in life accumulate and/or interact with later life factors to influence adult health.9^–11 Parental SES is a reasonable proxy for SES during adolescence, as most offspring are still dependent on their parents at this time point. However, most studies use parental SES as a proxy for SES exposure over the entire transition and thus miss the nuances of intergenerational SES transmission, which may differ by race/ethnicity. Measures of life course SES that combine parental SES in adolescence with the SES the respondents define for themselves as young adults can capture variations in SES exposure of critical importance for obesity development.

A common approach to combining SES information in the life course literature uses a “social mobility” framework, defining groups by their change in SES over time.12^–15 However, the number of groups required to exhaustively represent all combinations of more than one or two SES variables per time point can become prohibitively large for analytical purposes.16 Thus, although most researchers agree that the SES construct comprises multiple dimensions,17^–19 social mobility measures are typically restricted to traditional, single indicators of income, education or occupation at each time point. Although such measures may reasonably capture the SES of parents in the relative stability of middle adulthood, these measures are problematic during the less stable young adult period when misclassification is possible. For example, delayed transitions in schooling or employment may result in misclassification as downwardly mobile, while early transitions into the workforce may lead to misclassification as upwardly mobile.20

We examined the relationship of SES with obesity in a racially/ethnically diverse sample of adolescents followed over 5 years into adulthood. Profiles of SES exposure over this early life course period were identified using latent class analysis (LCA), a model-based technique that defines groups with distinct patterns of responses to multiple observed indicators.21 22 This method was well-suited for characterising a cumulative, multidimensional measure of early adult life course SES based on the young adult’s own social position and the SES of their parents. We hypothesised that there would be racial/ethnic differences in the associations of life course SES groups with obesity development across the transition from adolescence to adulthood as a result of structural barriers preventing minorities from translating higher SES into health benefits.

METHODS

Study population and design

The National Longitudinal Study of Adolescent Health (Add Health) is a longitudinal, nationally representative, school-based study of US adolescents in grades 7–12 in Wave I (1994–1995), followed by multiple interview waves into young adulthood (Wave III: 2001–2002) and collected under protocols approved by the Institutional Review Board of the University of North Carolina. A sample of 132 high schools and middle schools was selected to be representative of US schools with respect to region of country, urbanicity, school size, school type and race/ethnicity. Adolescents were sampled from school enrolment registers to participate in detailed, in-home surveys; in Wave I, a parent or guardian was also interviewed. The survey design and sampling frame have been described elsewhere.23

The analysis sample was drawn from the pool of young adults interviewed in Waves I and III with longitudinal, post-stratification sample weights (n = 14 322), excluding pregnant (n = 292) or seriously disabled (limited to walking one block or having a physical condition that keeps them from working; n = 188) respondents at Wave III because of the impact of these conditions on height and weight, as well as those missing data on the longitudinal obesity outcome (n = 747) or model covariates (n = 155). Our final sample included 12 940 respondents (47% female) comprising four major racial/ethnic groups: non-Hispanic whites (n = 7144), non-Hispanic blacks (n = 2693), Hispanics (n = 2113) and Asians (n = 990), aged 18–28 years (mean = 21.8 years) at Wave III. The excluded sample had a higher proportion of females and blacks.

SES variables

Our desire to capture the inherent multidimensionality of the SES construct and the heterogeneity of SES combinations across generations was balanced by convergence difficulties of the latent class model with increasing numbers of items. A large set of indicators reflecting the wide range of parental and young adult SES-related variables measured in Add Health was iteratively reduced to select variables with the potential to influence weight status through intermediate behavioural factors (diet, physical activity), representing three domains of SES that uniquely identify social status: (1) material endowments, including earned and investment income, property and freely exchangeable assets; (2) skills and knowledge; and (3) the status, power and abilities of one’s social network, or material, human and social capital respectively.19 We ultimately included 11 indicators of parental SES from parental and adolescent surveys (Wave I) and 14 indicators of young adult SES (Wave III) (see Appendix A). Our prior research guided the inclusion of variables of particular relevance to young adulthood (eg, current enrolment in higher education, social capital and economic hardship) (Scharoun-Lee, personal communication).

Other variables

Obesity

Measured and self-reported height and weight were available to define obesity in Wave III (young adulthood), whereas only self-reported measures were available for Wave I (adolescence). To maximise sample size and for comparability across waves, we used self-reported values at both time points,24 which also allowed inclusion of missing data resulting from weight measurement refusal or those in excess of scale capacity. Add Health self-report values correctly classify a large proportion of the respondents.25 Young adult obesity was defined using the adult body mass index (BMI) obesity cut-point (30 kg/m2).26 Adolescent obesity was defined using the International Obesity Task Force (IOTF) reference, which statistically links childhood and adolescent BMI centiles to the adult BMI cut-point and provides the only comparative reference data for defining longitudinal obesity patterns across the transition from adolescence to adulthood.27

Demographics

Self-designated race/ethnicity from Wave I was used to classify respondents into mutually exclusive categories of non-Hispanic white, non-Hispanic black, Hispanic and Asian or Pacific Islander, or white, black, Hispanic and Asian. Gender and age (at last birthday) were self-reported at Wave III.

Statistical analysis

We hypothesised that there would be racial/ethnic differences in associations between life course SES and obesity development because minorities do not experience the same returns as whites on higher SES achievement.5 6 We tested this hypothesis by assessing the interaction of race/ethnicity with SES in models of obesity.

Latent class analysis (LCA)

LCA was used to identify the number, size and characteristics of latent classes necessary to capture the heterogeneity of cumulative life course SES experience in young adults based on a diverse array of SES traits of the parental household in adolescence (Wave I) and the SES that the young adult created for themselves (Wave III).28 29 We used Mplus version 4.0,30 correcting for clustering of respondents and using post-stratification sample weights to account for unequal probability of selection.31

The final LCA model was selected using several criteria, including: (1) the visual plot of log-likelihoods of similarly specified models across numbers of classes, selecting models where the log-likelihoods no longer show a substantial improvement in fit;32 (2) the Bayesian information criterion (BIC), a widely used goodness of fit criterion for comparing models regardless of underlying distribution, with smaller values representing more parsimonious models;33 and (3) interpretability of model solution parameters, with attention to the meaningful interpretation of the response profiles for each class, uniqueness of classes and non-triviality of class size.32 33 These criteria supported a five-class model solution, correctly classifying 90–95% of individuals in each of the five latent classes. The parameters from this final model solution were used to classify respondents into “life course SES groups” who share a common profile of parental and young adult SES characteristics. For ease of presentation, we used distinguishing characteristics to assign brief labels to these groups, ie “persistent disadvantage”, “disadvantage with autonomy”, “material advantage”, “educational advantage” and “highest overall advantage”.

Multivariate modelling of obesity patterns

Multinomial (polytomous) logistic regression models were used to estimate the likelihood of: (1) becoming obese (incidence); (2) staying obese (persistence); or (3) becoming non-obese (reversal; n = 227, results not shown), relative to (4) staying non-obese associated with life course SES group membership using Stata, version 9.2.34 Coefficients from these models were exponentiated to obtain estimates of the relative risk ratio (RRR) of each obesity outcome as a function of the SES group.34 Posterior probabilities of membership in each latent class or “group” were entered into the model as categories of a nominal exposure variable, omitting the most advantaged group (referent).35 Given gender differences in obesity and the influence of SES on obesity, all models were gender stratified. Within gender, we assessed effect modification by race/ethnicity to explicitly test for racial/ethnic differences in the association of each SES group with specific obesity patterns. Only significant SES group by race interactions (p<0.10 for Wald test of interaction terms) were retained. Model coefficients were then used to predict age-adjusted obesity patterns for each life course SES group. Survey procedures in Stata were used to correct for unequal probability of selection, underestimation of variance due to clustered sample design and non-response bias due to attrition from Wave I to Wave III.

RESULTS

Salient characteristics of the five life course SES groups identified using latent class analysis are summarised in table 1. Findings show the great variation and depth of the life course SES profiles, ranging from “persistent disadvantage”, which was most likely to be economically deprived, coming from poor, single-mother households in adolescence and maintaining the cycle of hardship into adulthood, to “highest overall advantage”, which was most likely to have professional, high-income, two-parent households in adolescence and the highest education, occupation and financial access in young adulthood.

While whites and Asians were most likely to be in advantaged groups, blacks and Hispanics were most represented in the “persistent disadvantage” and “disadvantage with autonomy” groups respectively (table 2). Obesity incidence exceeded 10% for all except Asians; persistence was also considerable. Blacks had the highest obesity trends, followed by Hispanics.

We observed no significant interactions between life course SES and race/ethnicity for obesity incidence in males (p = 0.38) or females (p = 0.60). Racial/ethnic-specific estimates suggest similar conclusions for obesity persistence, but sparse data in Hispanics and Asians precluded statistical interaction assessment across the total sample. However, we observed no significant differences in obesity persistence between blacks and whites (p>0.10), for whom we had sufficient data. Thus, we used racial/ethnic-pooled, multinomial logistic regression models to estimate the relative risk of obesity incidence and persistence versus staying non-obese (outcome referent) for each SES group relative to “highest overall advantage” (exposure referent) (table 3). We observed strong racial/ethnic-pooled associations between “disadvantaged” life course SES experiences and higher obesity risk, although the nature of high-risk disadvantage differed by gender. “Disadvantage with autonomy” was the only group significantly associated with higher risk of obesity incidence for males (RRR = 1.64; 95% CI 1.12 to 2.40), whereas, for females, the estimates for all groups excluded the null, with strongly positive associations for the “persistent disadvantage” group. In general, SES group membership had stronger associations with obesity persistence than incidence.

The model results are further illustrated in figure 1. For males, the “disadvantage with autonomy” group had the highest total obesity, whereas members of the “persistent disadvantage” group had the second lowest predicted incidence and persistence. For females, in contrast, membership in the “persistent disadvantage” group conferred the highest overall risk of young adult obesity, primarily due to substantial incidence.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 Predicted* total young adult obesity, split into incidence and persistence from Wave I (1994–1995) to Wave III (2000–2001) for males (left) and females (right) from the National Longitudinal Study of Adolescent Health, across life course SES groups. *Coefficients from multinomial logistic regression models estimating relative risk of longitudinal obesity patterns in males and females were used to predict the probability of obesity incidence (becoming obese from Wave I to Wave III) and persistence (staying obese from Wave I to Wave III) for a 22-year-old of specified race/ethnicity and life course SES, holding all other variables constant.

Although a large proportion of both white and black females in the “persistent disadvantage” SES group experienced obesity incidence and persistence across the transition to adulthood, the overall risk for blacks was clearly higher (fig 2). A similar pattern was seen for the privileged life course SES experience of the “highest overall advantage” group.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2 Predicted* longitudinal obesity patterns from Wave I (1994–1995) to Wave III (2000–2001) for white and black females from the National Longitudinal Study of Adolescent Health, across selected life course SES groups. *Coefficients from the multinomial logistic regression model estimating relative risk of longitudinal obesity patterns in females were used to predict the probability of obesity incidence (becoming obese from Wave I to Wave III) and persistence (staying obese from Wave I to Wave III) for a 22-year-old female of specified race/ethnicity and life course SES group, holding all other variables constant.

DISCUSSION

Using latent class analysis in US adolescents transitioning to adulthood, we captured the heterogeneity of cumulative SES exposure in five “life course SES” groups representing distinct profiles of parental and young adult SES. Although we observed a promising absence of racial/ethnic differences in the association of life course SES with obesity development during this period, substantial disparities endured, manifested as a consistently higher burden of obesity in racial/ethnic minorities across SES groups. In the pooled racial/ethnic sample, strongly positive associations between “disadvantaged” life course SES exposure and obesity risk were observed, with variation across gender. Notably stronger associations for obesity persistence versus incidence support an earlier influence of life course SES on obesity development.

We expected that racial/ethnic minority respondents would not receive the same health benefits from SES as whites, given the societal discrimination that devalues SES attainment in minorities.5 36 Institutionalised racism, in particular, can create structural barriers that restrict the benefits of SES achievements for racial/ethnic minorities.37 Thus, we hypothesised that societal discrimination would restrict the ability of racial/ethnic minorities to translate advantaged early life course SES into better health. However, based on our findings, we could not reject the possibility that the association between SES and obesity is similar across race/ethnicity. Although our complex measure of early life course SES may have minimised differences in obesity associations by capturing a substantial proportion of the heterogeneity between racial/ethnic groups, the observed similarities in the health benefits of SES across race/ethnicity during the transition to adulthood provides a promising contrast to the literature in adults. The minimal impact of discrimination during this life course stage suggests that this may be an opportune time to halt further widening of racial/ethnic disparities in health.

However, the overall burden of obesity was consistently higher in historically underprivileged racial/ethnic minorities versus whites across several life course SES groups, including those heavily represented by minorities (eg, “persistent disadvantage”). SES factors captured in our measure may influence racial/ethnic differences in obesity earlier than our baseline adolescent assessment. Furthermore, the stronger relationships for persistence versus incidence in the pooled racial/ethnic sample suggest that SES started influencing the process of obesity development prior to adolescence and helped maintain obesity during the transition, thus supporting a “cumulative” life course framework.38 39 We need to look prior to this transitional period to understand the higher rates of obesity in racial/ethnic minorities, supporting the need for early preventive efforts.

Gender differences in the influence of disadvantage on obesity merit further comment. Elevated obesity risk for both sexes growing up in working-poor households and transitioning early into marriage and lower status jobs (ie, “disadvantage with autonomy”) highlights this SES profile as particularly high risk for all young adults. In contrast, “persistent disadvantage” was associated with strongly increased obesity incidence among females only. Given the influence of maternal SES on weight change in young adult women,12 it is possible that impoverished single mothers transmit improper weight loss or dieting attempts to their female offspring, resulting in dramatic obesity incidence. A full understanding of the complex dimensions of SES across generations can help to identify groups at risk and the timing of effects.

Our study had many strengths, including the use of the complex LCA methodology to classify life course SES exposure, a diverse sample and detailed parental and adolescent data from an understudied life cycle period known for high obesity risk. However, this work was not without limitations. Although posterior probabilities from the LCA model can be used to allocate individuals to their most likely group, the majority of individuals have a non-zero probability of membership in every group identified. Further, individual variability within LCA groups may result in less accurate classifications for a small number of individuals in each group. Timing of measures was another limitation. As our parental SES indicators were assessed during adolescence and thus could only approximate earlier exposure to socioeconomic conditions, we were unable to fully capture the influence of early SES experiences on obesity development prior to the baseline adolescent assessment. However, our examination of longitudinal obesity patterns did disentangle influences that occur during the transition to adulthood versus before adolescence.

The impact of excluding pregnant females at Wave III on sample selectivity was a concern. SES exposure patterns increased the likelihood of “persistent disadvantage” classification, while pregnancy increased the likelihood of “obesity incidence” classification in these excluded females (data not shown). Ultimately, this exclusion reduced the potential for bias as a result of an artificial inflation of the association between persistent disadvantage and obesity incidence. Furthermore, the excluded pregnant sample did not differ from the analysis sample in racial/ethnic composition, and thus their exclusion was not likely to influence conclusions about disparities. We also considered excluding females with a history of pregnancy before Wave III. Given the weight gain and retention associated with pregnancy,40^–42 females with a history of earlier pregnancy were likely to become obese in adolescence and maintain obesity into adulthood. However, results from a secondary analysis (data not shown) suggest that excluding the 25% of females with earlier pregnancies would reduce the precision of estimates and bias our analysis in an unpredictable manner, supporting our original exclusion based on pregnancy at Wave III only.

Overall, we observed an absence of racial/ethnic differences in the association of complex profiles of life course SES with obesity development during the transition to adulthood, a promising contrast to the literature in older adults. However, disparities persisted, manifested in higher obesity trends in racial/ethnic minorities across SES groups. Gender differences in the type of disadvantaged SES exposure most associated with obesity risk underline the importance of complex characterisations of life course SES for detecting group differences in the study of health. The stronger associations with persistence and enduring racial/ethnic disparities in obesity risk across SES groups suggest that the characteristics captured by these groups play a larger role in disparities earlier in the life course.