[Obesity Discussion]

RESULTS
As shown in Table 1, the mean age of women in 1988 was 35.4 years. Their mean BMI was 29.6, with nearly 43% classified as obese and an additional 30% classified as overweight. More than 80% of the women grew up in low SEP (i.e., unskilled or farm laborer) households. In adulthood, 81.3% were in low SEE Almost 70% (n = 487) of the women were classified as low/low SEP, or relatively disadvantaged in both childhood and adulthood; 14.1% (n = 79) were classified as low/high SEP, relatively disadvantaged in childhood but not in adulthood; 12.4% (n = 88) were classified as high/low SEP, relatively advantaged in childhood but not in adulthood; and 4.6% (n = 25) were classified as high/high SEP, relatively advantaged in both childhood and adulthood.

Table 2 contrasts selected demographic, socioeconomic, and behavioral characteristics of the sample by membership in life-course SEP categories. On average, women in the low/ low SEP category were about 3 years older than their high/high SEP counterparts. Women in the low/high and high/high SEP categories were most likely to be married (67.6% and 55.1%, respectively) and least likely to be formerly married (17.2% and 18.7%, respectively). These 2 groups of women also had the highest mean scores (36.1 and 35.3, respectively) on the Rand Corporation's Current Health Scale,(n28) a measure of self-perceived health. Alcohol consumption varied little across life-course SEP categories; however, cigarette smoking was most prevalent among high/low SEP women (40.3%) and least prevalent among high/high SEP women (16.8%). Low consumption of fruits and vegetables was characteristic of all women.

Not surprisingly, women in the low/high and high/high SEP categories were least likely (5.0% and 1.2%, respectively) to report high levels of financial strain, defined as "very hard" to pay for food, housing, heat, and medical care. Household crowding also tended to be less common for these 2 groups of women (4.7% and 5.7%, respectively). Although childhood material deprivation as measured by food insecurity and absence of household heat, electricity, and plumbing was least common among women in the high/high SEP category, only the difference for no indoor plumbing (21.4%) was statistically significant.

To illustrate in absolute terms the differential burden of obesity in this population of African American women, Table 3 summarizes the unadjusted prevalence of obesity by childhood, adulthood, and life-course SEE Nearly 46% of the women from low childhood SEP backgrounds were obese compared to 27.6% of women who grew up in less impoverished households. When stratified by adulthood SEE obesity was more common among women in the low (44.2%) than in the high SEP category (36.2%). When further stratified by life-course SEP, obesity characterized 47.2% of low/low SEP women, 38.7% of low/high SEP women, 27.2% of high/low SEP women, and 28.5% of high/ high SEP women.

The relative odds of obesity by childhood SEP, adulthood SEP, and life-course SEP are presented in Table 4. The odds ratios in the first column are adjusted for age only; those in the second column are adjusted for age plus the indicated covariates. For parsimony's sake, odds ratios describing associations between obesity and the potential confounders, as well as the potential behavioral mediators, are not shown. Before and after multivariable adjustment, women from low childhood SEP backgrounds had a 2-fold greater odds of obesity (multivariable adjusted OR = 2.21; 95% CI = 1.32, 3.68) than women who grew up in less disadvantaged households. These increased odds were not diminished when adulthood SEP was added to the model. Nor was there any evidence of mediation by strenuous physical exercise or fruit/ vegetable consumption.

As shown in Table 4, women in the low adulthood SEP category had a nonsignificant 25% greater odds of obesity (multivariable adjusted OR = 1.25; 95% CI = 0.74, 2.11) than did their high SEP counterparts. Nonsmokers and formerly married women (in contrast to those never married) were also more likely to be obese (data not shown). Neither self-reported fruit/vegetable consumption nor strenuous physical exercise mediated these results.

For all life-course SEP comparisons, these associations between obesity and potential confounders on the one hand, and potential mediators on the other, remained unchanged from the main effects models. Therefore, in the interest of parsimony, only findings from the life-course SEP comparisons will be summarized here. As shown in Table 4, for both age-adjusted and multivariable adjusted models, women in the low/low SEP category had a 2-fold but not statistically significant greater odds of obesity (multivariable adjusted OR = 2.12; 95% CI = 0.75, 6.00) than their counterparts in the high/high SEP category.

Whereas women in the low/high SEP category had a 55% greater odds of obesity (multivariable adjusted OR = 1.55; 95% CI = 0.50, 4.83) than women in the high/high SEP category, those in the high/low SEP category had a non-statistically significant 14% lower odds of obesity (multivariable adjusted OR = 0.86; 95% CI = 0.27, 2.70).

DISCUSSION
For African American women in the Pitt County Study who grew up in the most economically disadvantaged households, the odds of obesity in adulthood were twice as high as were those for women from less impoverished backgrounds. This association was independent of the SEP women had attained in adulthood. The association was also independent of the woman's marital status and health behaviors such as smoking, alcohol consumption, strenuous exercise, and fruit and vegetable consumption. Lower SEP in adulthood, however, was associated with only a 25% greater odds of obesity after control for childhood SEP, marital status, and health behaviors.

When the data were analyzed from a life-course perspective, women who were relatively disadvantaged in both childhood and adulthood had twice the odds of obesity as women who were relatively advantaged at both points in time. Upwardly mobile women--those who were relatively disadvantaged in childhood but not in adulthood--had a 55% greater odds of obesity than women who were relatively advantaged across the life course. Finally, women who were relatively advantaged in childhood, but not in adulthood, had a slightly lower odds (14%) of obesity than women who maintained their relatively advantaged position from childhood into adulthood. Because of small numbers in some of the life-course SEP categories (especially high/high SEP), the 95% confidence intervals for all life-course comparisons included the null value.

The findings for childhood SEP were especially robust. They conform with findings from a nationally representative study that reported a strong, inverse relationship between parental education and adult obesity in African American women,(n15) but they contrast with null findings from 2 other studies.(n10,n16) Differences across regions, data collection periods, and other methodologies (e.g., the choice of parental occupation vs. education, use of EHC vs. conventional questionnaire methods) for ascertaining childhood SEP may have contributed to these inconsistent results.

There are several potential pathways through which socioeconomic deprivation in childhood could increase the odds of obesity in adulthood, independent of adulthood SEP. First, it has been suggested that adverse intrauterine conditions or adverse early postnatal physiological disturbances, both socially patterned by maternal SEP, can predispose the developing fetus/child to an array of health difficulties (some of which are obesity-mediated) later in life.(n29,n30) Second, a growing number of studies suggest that individuals who experience socioeconomic deprivation in childhood are more likely to engage in detrimental health-related behaviors, such as poor nutrition and low levels of physical activity in both childhood(n31-n34) and adulthood.(n35,n36) Though we included measures of adulthood physical activity and fruit/ vegetable consumption, measurement error in these variables plus lack of data on physical activity and nutrient intake at earlier points in the life cycle make it impossible to completely rule out an explanatory role for these behavioral pathways.

An inverse relationship between adulthood SEP and obesity is a common finding in studies of European(n7-n9,n37,n38) and White American women;(n10,n39) however, findings for African American women are more mixed.(n10,n19) The inconsistent findings for African American women could be attributable to differences in geographical settings, differences in how SEP is measured across studies, or, as suggested by this study, intergenerational social mobility dynamics that effectively determine whether the inverse relationship between adulthood SEP and obesity will be large, small, or nonexistent.

In the current study, for example, the modest 25% excess odds of obesity for women in the low, compared with high, adulthood SEP category seems largely attributable to intergenerational social mobility dynamics. As shown in Table 3, upwardly mobile women (low/high SEP) had a lower prevalence of obesity than did women who remained relatively disadvantaged over their life course (low/low SEP). In contrast to this pattern of diverging risks over time, women who grew up in relatively advantaged households had roughly comparable odds of obesity in adulthood regardless of their childhood SEE Hence, for women in the Pitt County Study, upward mobility mattered, but not enough to wholly negate the long reach of childhood socioeconomic disadvantage on the odds of obesity in adulthood. Of course, the degree to which weak versus strong intergenerational social mobility effects underlie the mixed findings in the literature concerning adulthood SEP and obesity in African American women is an empirical question that only additional research can answer.

Though small numbers in the referent category for the current study prevented the associations between life-course SEP and odds for obesity from reaching conventional levels of statistical significance, the associations nonetheless followed a plausible gradient of risk. Women who were relatively disadvantaged in both childhood and adulthood had the greatest odds for obesity, followed by upwardly mobile women, and then by women who grew up in relatively advantaged households, irrespective of their adulthood SEP.

This particular gradient of risk is consistent with both cumulative burden and critical period explanatory models of adult chronic diseases,(n39) here extended to the study of obesity in African American women. Some manifestations of a cumulative burden model, which emphasizes the health-damaging effects of the accumulation of risks over the life course, can be seen in the generally elevated pattern of material and psychosocial disadvantages to which women in the low/low SEP group were exposed during both childhood and adulthood (Table 2). Likewise, some manifestations of a critical period model, which emphasizes the long-lasting effects on adult health of environmental exposures encountered during developmentally sensitive epochs, can be seen in the contrasting childhood material and psychosocial advantages for low/low SEP women compared to their high/high SEP counterparts (Table 2). The contrasts in childhood material life conditions were less sharp between low/ low and high/low SEP women, suggesting that other unmeasured childhood advantages underlie the lower odds of obesity observed for the latter group of women.

Our decision to require study participants to possess 3 out of 4 key socioeconomic resources to qualify for membership in the high adulthood SEP category very likely minimized misclassification on this exposure. However, this decision also reduced statistical power for the life-course SEP comparisons because it severely limited the number of women who could be considered "relatively advantaged" in both childhood and adulthood. We believe that our decision to use multiple indicators of adulthood SEP and to set a fairly high threshold for membership in the high adulthood SEP category was justified on 4 grounds. First, as already noted, education and occupation were weak predictors of 1988 hypertension outcomes(n18,n24) in this population. Second, SEP is a multidimensional construct and multiple indicators are being used increasingly in epidemiological studies in order to better capture its multiple facets.(n40) Third, setting the threshold at 3.0 isolated a subset of individuals who possessed conventional "middle-class" credentrials, adding credence to our labeling the low/high SEP women in this study "upwardly mobile." Fourth, lowering the threshold from 3.0 to 2.5 points (the median) increased the number in the referent category by only 9 women, with essentially no changes in study findings.

Finally, our study findings are potentially subject to both recall and loss-to-follow-up bias. Childhood SEP was assessed retrospectively in 2001; hence, recall bias would in-rate estimates of the association between low childhood SEP and obesity if women classified as obese differentially (and incorrectly) reported that their parents were farm laborers or unskilled workers. Though possible, such differential overreporting seems unlikely. Loss-to-follow-up bias would exist if the observed associations between exposures and outcomes for the 679 women comprising the analysis sample differ greatly from what would have been observed had all 1112 women in the 1988 baseline survey been studied. Since analyses in the current study were weighted both for oversampling of middle-class individuals in 1988 and for nonresponse to the 1988 and 2001 surveys, our findings can be generalized to the entire 1988 cohort. That said, future advances in life-course research into the social determinants of obesity in African American women, or any other population at high risk for this condition, will require prospective study designs, adequate sample sizes, and high quality information on both personal and neighbor hood-level resouces(n41,n42) known or suspected to influence dietary practices and physical activity across the life course.

This article was accepted June 22, 2005.

Contributors
S.A. James originated the study and led the writing. A. Fowler-Brown conducted the literature review and assisted with the writing. T. E. Raghunathan supervised the data analysis. J. Van Hoewyk constructed the study variables mad conducted the data analyses.

Acknowledgments
This study was funded by the National Institutes of Health (grant HL 65645).

Human Participant Protection
This study satisfied all criteria for the ethical treatment of human participants and was approved by the University of Michigan's human subjects institutional review board.