Income Inequality in Obesity among US Adults 1999-2016 : Does Sex Matter ?


 Background

Obesity is a major public problem that varies by income and sex. However, there is little evidence that exists to determine the association between income and obesity may be moderated by sex and family income after controlling for potential covariates.
Objectives

To examine the association between income and obesity in adults ages 20 years and older and to test whether this relationship differs by sex.
Methods

Data for this study came from the 1999–2016 National Health and Nutrition Examination Surveys (NHANES). Obesity was defined by using Body Mass Index ≥ 30 kg/m2, and Gini Coefficient was calculated to measure income inequality by using Poverty Income Ratio (PIR). We categorized GC in five quintiles to examine the relationship between income-inequality and obesity. For the first set of analysis, Modified Poisson regression was used in a sample of 40,917 adults with almost equal number of men and women (women’s ratio was 50.6%). Models included age, racial/ethnic groups, marital status, education, health behavior (smoking and drinking status and physical activities), health insurance coverage, self-reported health, and household structure (live alone and female-headed household).
Results

Adjusting for potential confounders, higher-income men had a higher prevalence of obesity (PR: 1.11; CI: 1.00-1.25) compared to those with a lower-income, but women with upper middle income had a lower prevalence of obesity (PR:0.84; CI:0.77–0.93) compared to those with a lower income.
Conclusions

The association between income inequality and obesity was consistently significant among middle-income and higher-income men than lower-income men. The same association was not found for women, women more likely to suffer from income-inequality than men.

Despite a large body of knowledge of obesity and socioeconomic status, there is relatively little known about income and income-inequality impacts on obesity and its association with sex. Studies showed that obesity prevalence varies by income [1]. Income in uences health outcomes by shaping health behaviors to reduce behavioral risk factors, decreasing barriers to accessing care, moderating environmental factors or reducing inequalities [2]. Lower income population has higher rates of behavioral risk factors including smoking, drinking, obesity, and low levels of physical activity [3].
Isolating the contribution of income to health can be very di cult __ if not impossible __ many social risk factors including race, ethnicity, sex, geography, and educational status are connected to the income [4], people with higher incomes tend to live in healthier neighborhoods and have higher educational attainment and more social capital [5]. Income has direct and indirect effects on material conditions necessary for biological survival, social participation and opportunity to control life circumstances [6]. In spite of the aforementioned research, there is not enough literature to show how obesity changes by income inequality.
The spread of income-income inequality-opens a new argument that distribution of income may have higher impact on people and communities, with a higher income inequality in a society we expected to see a higher proportion of people in poverty [7,8], and it may explain why distribution of income within and between communities are as important as the level of income [9].
Consider obesity, one of the well-known health outcome measures the prevalence of obesity in 2017-2018 was 42.4% in US adults; from 1999-2000 to 2017-2018, the age-adjusted prevalence of obesity increased from 30.5% to 42.4% [10]. Obesity prevalence uctuates by income and socioeconomic status such as education level and maybe sex [1].
While numerous studies have investigated direct relationships between income and obesity, there is limited knowledge on income inequality and its association with obesity and between men and women.
This association may operate differently by men and women, for instance income inequality is negatively associated with weight status in men who are highly integrated, but is positively associated with weight status among men who have low integration [11]. In this study, we investigate the relationship between obesity and income inequality and how this association changes between men and women.

Methods.
Data for this project came from the 1999-2016 National Health and Nutrition Examination Surveys (NHANES) [12]. NHANES is a cross-sectional survey, provides nationally representative estimates of health and nutritional status for the US population, with response rate of 73.2 between 1999-2016 [13,14]; and a multistage probability sampling design, that makes the sample representative of "each of the four regions of the United States" [13]. The original sample between 1996-2016 was 90,394 individuals, with women ratio of 50.6%. For this study we have included participants who are 20 years old and above, we dropped all pregnant women (1,667) and all missing observations for Poverty Income Ratio (PIR) from the analysis, this yielded an analytic sample of 20,341 men and 20,576 women.
Outcome Variable. Using Body Mass Index (BMI)-derived by dividing weight in kilograms by height in meters squared (kg/m^2) -we created a binary variable to identify participants who were obese (if BMI>=30) as the outcome variable [15].
Main Independent Variable. The main independent variable of interest is income inequality. Gini Coe cient was calculated as a measure of income inequality using Poverty Income Ratio (PIR) using Poverty Income Ratio (PIR) -a ratio of family income to poverty threshold -and Family Poverty Level.
The Gini Coe cient (GC) is one of the most well-known measures of income inequality. The GC is based on the Lorenz Curve (LC) -ratio of two integration calculations on a graph of two curves, one curve representing actual inequality and other, perfect income equality. The GC is de ned as A/(A+B) if 'A' equal zero then GC will be zero, which means perfect equality and if 'B' was zero then the GC will be one, which means complete inequality [16] and Lorenz curve represents the actual distribution of income in given society [17].
We calculated the Gini Coe cient (GC) using the following equation [16].
Using the income quartile approach, we have de ned a categorical variable with ve quintiles; lowest quintile, second quintile, middle quintile, fourth quintile and top quintile.
Covariates. For the demographic variables we included age (years) and socioeconomic (SES) variables. SES was assessed using a binary variable for marital status (1=married, o=otherwise), education categories (less than high school graduate, high school graduate or general equivalency diploma equivalent recipient, more than high school education or some college and above). Health-related characteristics included having health insurance (1 = yes; 0 = no) and self-reported health (excellent-very good, good-fair and poor). Health behavior was measured by three variables including smoking (never smoked, former smoker and current smoker), drinking (never drink, former drinker and current drinker) and physical activity, a binary variable to show have not participated in vigorous activities (1 = yes; 0 = no) during a typical week. Household structure is an important element to predict individual/household income [18], so we used a binary variable to present living alone (1 = yes; 0 = no) and female-headed household (1 = yes; 0 = no).
Analytic Strategy. For the rst set of analysis, the mean and proportional differences between men and women for obesity, demographic, socioeconomic (SES), health-related characteristics, health behavior were evaluated using unequal t-tests. In our sample the prevalence of obesity was greater than 10%; therefore, we used a weighted modi ed Poisson regression analysis [19][20][21] that produced prevalence ratios (PR) and corresponding 95% con dence intervals(CI) [19][20][21]. Six models were estimated. Model 1, was an unadjusted model examining the relationship between income level and obesity for men and women. Model 2 was an adjusted model 1 with including sex (=1 if female) as a covariate, we ran model 3 with an interaction term of income quintile and sex. Because the interaction between income quintile and sex was signi cant (p<0.001), we strati ed the analyses by sex. These resulted to model 3 and model 4 as unadjusted and adjusted models for women and model 5 and model 6 for men, respectively.
To make our estimates as a representative of the U.S. civilian non-institutionalized population, all analyses were weighted using the NHANES individual-level sampling weights for 1999-2016 (8 waves of data) [22]. We considered P-values <0.05 as statistically signi cant and all tests were two-sided. We used STATA statistical software, version 15 to perform all analysis.

Results.
Association between income-level and obesity. Table 1 compares the distribution of the sample's characteristics for the total sample and by sex. On average about one-third of the population were obese with a signi cant higher rate of obesity for women (P< .001). Income quintiles showed that the in comparison with men women population in our sample was in lower and middle quintiles (23.8% women vs. 20.2% men) and men was in top quintile (37.5% men vs. 33.5% women).
Overall, the sample age was 47.0 18.0 years old with a slightly older women population (47.9 (17.9) vs. 46.0 (18.0)). Women sample had higher percent of black non-Hispanic (NH) and men sample had higher percent of Mexican-American with non-signi cant differences between percent of white NH and other racial/ethnic groups in both samples. The majority of the women had more than a high school education (61.1%) and were covered by any kind of health insurance. Two-third of the women had never smoked (59.0%), currently drank (67.0%), were physically active (57.4), and were healthy (82.5%). In comparison 46.5% of the men had never smoked, 85.2% were currently drank, 61.5% were physically active and 84.2% were healthy. Women were live more alone than men (14.9% vs. 11.8%) and only 43.7% of them were head of household in compare with 72.3% of men. Table 1 details information on women and women. Table 1.
The association between income-inequality and obesity in women and men is displayed in Table 2. Based on the results of the unadjusted model, rich population were less obese (PR:0.85; CI:0.79-0.91); however, in the adjusted model, the association of income-level and obesity has disappeared (PR:0.94; CI:0.87-1.01), this gives us this idea to run strati ed model for women and men. Based on the full model (Model 2) results, the obese population was more women, black NH or Mexican American. They were married, had higher education, former smokers and drinkers, not physically active with poor health condition. Table 2.
Model 2 is an adjusted model 1 with including sex (=1 if female) as a covariate, we ran an interaction term of income inequality categories and sex. The association of income-level and obesity in the women sample is displayed in Table 3. Model 3 and model 4 present the association between income-inequality and obesity in women. Based upon unadjusted model women in 4 th and 5 th income-quintile group less likely to be obese (PR: 0.87; CI: 0.81-0.93) and (PR: 0.69; CI: 0.64-0.75), respectively. After adjusting model, the association between income-level and obesity in 4 th quintile has disappeared but remained signi cant in rich population (PR: 0.84; CI: 0.76-0.93). There is negative association between income-level and obesity in women. Model 5 and model 6 present the association between income-inequality and obesity in men. Interestingly, there is a positive association between income-inequality and obesity in men in unadjusted and adjusted models for example in 4 th quartile (PR: 1.20; CI: 1.08-1.32) and 5 th quartile (PR: 1.24; CI: 1.13-1.37) more likely to be obese. Table 3.
Our adjusted models for women and men showed that individuals with obesity were black NH or Mexican American NH, with High school graduates. They were former smoker or drinker, with no vigorous activities and fair-poor health condition.
As presented in the regression results there is different association between obesity and incomeinequality in men and women, to understand more about these differences in men and women we have used Lorenz curve and Gini Coe cient. Figure 1 -Lorenz curves -shows the Gini coe cient for income to poverty ratio in men and women population in the US between 1999-2016, to plot these curves we used average GC with jackknife standard errors. This gure compares GC between men and women and obese and non-obese population by sex.
In the panel A, the blue solid line plots distribution of income in non-obese women and the dash-red line shows distribution in non-obese men. The blue line stays over the red line for about 40% of population and closer to the perfect equality line that means smaller income inequality within women and greater inequality within men. Interestingly, there is a different pattern in the obese population. For all income quintiles, obese women suffer more from income inequality than obese men (See panel B). For example, in obese men lower 25% of population observed only 7.1% of income and 53% of income observed by 75% of population and rest of them observed by top 25% of population. In obese women these distributions changed to 6% and 48% of rst 25% and 75% of population and 52% of income observed by top 25% of population. The Gini coe cient rose from 0.295 (SE: 0.002) in men to 0.322 (SE: 0.002) in women, the jackknife standard errors for these estimators are very small and there is a signi cant difference between GC in men and women (P<0.001).
In panels C and D, we compared the GC between obese and non-obese men and women population. As presented obese women between the fth and tenth percentile of the population (middle and higher income population) suffer more from income-inequality (panel C), GC moved between 0.342 (SE: 0.003) in obese women to 0.0305(0.002) in non-obese women. For men there is a higher income inequality in non-obese population (Panel D), the red-dash-line the represent obese-men stays above the solid blue line (non-obese men) meaning lower income inequities for all groups of income, the GC moved from 0.285 (SE: 0.003) in obese men to 0.298 (SE:0.002) in non-obese men. Sensitivity analysis. As a sensitivity analysis and with using the CDC approach household income was de ned as a categorical variable with ≤130% as low-income, >130% to ≤350% middle income, and >350% as higher middle income household [1]. We found similar patterns with the original models, for example negative association between income-inequality and obesity in women with a positive association in men (See Table 4). Discussion.
In this study we investigate the relationship between obesity and income inequality and how this association changes between men and women. Our ndings showed that higher income inequality in men had positively associated with higher prevalence of obesity but in women income inequality has negatively associated with prevalence of obesity. For example, men in higher income groups experienced higher probability of being obese that can be explained by lower physical activities, but the pattern in women was different; increasing income in women resulted in reducing obesity -maybe because of better intake food, more exercise or changing the neighborhood.
Our literature search shows that in spite of fair amount of publication on impact of income inequality and health [23][24][25][26][27][28], and some critics that income inequality is not in fact associated with individual health outcomes [29][30][31][32], a little bit known about the impact on income inequality and obesity speci cally on men and women, we found only a few articles to address income inequality and in adult men or women, for example Campbell et al. (2019) reported that income is negatively associated with weight status for men who are highly social-integrated but is positively associated with weight status among men who have lower social integration [11] and Kim et al. (2018) reported that increased income inequality and lower poverty percentage signi cantly associated with lower obesity rates across for men in New York counties [33].
Most of published articles on obesity and gender differences discussed impact of geographical location [34], neighborhood factors [25,[35][36][37], behavior and physical activities [38][39][40] and incomeinequality [41], food security [42], access to healthy foods [43,44]. There is an essential need to nd more about differences between men and women health especially in different income inequality groups.
There are aspects of the study that deserve comment. The data is cross-sectional; therefore, ruling out the possibility of reverse-causation is not possible, evidence shows that extent bias because of reversecausation is largely indirect [45]. NHANES has some limitations with the income variable and did not report the real income instead a categorical variable, otherwise using household income as a continuous variable could give us a better opportunity to nd the impact of income differences instead of a proxy variable such as PIR.
There are also strengths to this study. To our knowledge, it is the rst study to examine the relationship of income inequality and obesity in wide range of NHANES data (1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016) and between sexes. In addition, we used weighted models that make our ndings nationally representative estimates, increasing the generalizability of these results. We need to notice that the Lorenz curves are unaffected by the mean of the distribution, and "they cannot be used to rank distributions in terms of social welfare, only in terms of inequality" [18]. To investigate the reasons behind the different association between men and women, we should control our models based on the geographical variables such as urban and rural area and neighborhood or at the county level to capture more variances, this is the next step in our study.

Conclusions.
The association between income and obesity operates differently for men and women. Our ndings showed that higher income positively associated with prevalence of obesity in men but negatively associated with women's obesity at the same income groups, this nding in Income inequality also plays different roles between men and women, women experienced higher income-inequality, and obese women in middle and upper middle-income groups, suffer more from obesity than men in the same income groups. In obesity treatment for men and women, policy makers need to consider the redistribution of income in communities alongside with prioritizing vulnerable population. With low-income women considered as one of the most vulnerable population, most public interventions are targeted to lowincome persons. Moreover, public health interventions should target low-income women of color and even minority individuals of very high income as well, especially African Americans [46].