Health disparities in the USA are notable among groups of differing socio-economic status (SES) and race( 1 – Reference Go, Mozaffarian and Roger 4 ). While various explanations have been given for these disparities, both biological and non-biological( Reference Baum, Garofalo and Yali 5 – Reference Snowden 12 ), a recurring and aetiologically relevant theme has been differences in diet( Reference Aggarwal, Monsivais and Drewnowski 13 – Reference Olmedo-Requena, Fernández and Prieto 20 ). In order to ameliorate diet-mediated health disparities, it is necessary to identify diets that are associated with given health outcomes (e.g. CVD) and to recognise where disparities exist in the selection and adherence to the mentioned dietary patterns.
Previous studies investigating the associations between diet, race and SES have assessed diet in various ways. Earlier work has focused on examining individual nutrients, as well as groupings of nutrients, both macro and micro( Reference Aggarwal, Monsivais and Drewnowski 13 , Reference Bjørke-Monsen, Roth and Magnus 14 ). While scientifically and physiologically relevant, this approach does not reflect the manner in which most individuals select and consume foods. In order to assess dietary intake more holistically, other researchers have used hypothesis/investigator-driven (a priori) dietary indices in their analyses, which take into consideration patterns of foods consumed together( Reference García-Arenzana, Navarrete-Muñoz and Peris 16 , Reference Hiza, Casavale and Guenther 18 , Reference Olmedo-Requena, Fernández and Prieto 20 ). More recently, data-driven/hypothesis neutral (a posteriori) dietary patterns have been used to analyse these relationships, with dietary patterns serving as a proxy for a variety of factors that influence how individuals consume foods, including social, cultural and financial contributors( Reference Cutler, Flood and Hannan 15 , Reference Hare-Bruun, Togo and Andersen 17 , Reference Nobre, Lamounier and Franceschini 19 ). However, limited research has been conducted on how an individual's racial group and SES may affect adherence to dietary pattern, especially patterns demonstrated to be related to health outcomes.
The present team of investigators have previously shown that dietary pattern adherence is associated with the risk of incident stroke, contributing to an excess risk in black Americans compared with white Americans( Reference Judd, Gutiérrez and Newby 21 ). The present study seeks to advance this investigation by using the dietary patterns from our previous study to examine the relationships between SES and food choice and how these relationships may differ according to race. Although adherence to only some of the dietary patterns has been significantly associated with the risk of incident stroke (increased risk for southern and reduced risk for plant-based and sweets/fats patterns), all previously derived patterns were used in the present study as our team's examinations of potential associations between dietary patterns and CVD are ongoing, and may yield further insights germane to the findings of the present study. The population used for the present study is especially well suited to this endeavour as it is quite large, nationwide, includes persons of varying levels of SES (both individual-level and community-level indicators) and includes a percentage of black participants (34·6 %) large enough to explore racial differences.
Methods
Study design
The REasons for Geographic And Racial Differences in Stroke (REGARDS) study is observational in nature and comprised a population-based random sample (n 30 239) of men and women of black and white racial groups over the age of 45 years( Reference Howard, Cushman and Pulley 22 ). The original intent of this study was to elucidate the factors that increase an individual's risk of having a stroke. Recruitment occurred between January 2003 and October 2007, and the baseline data used in the present analyses were cross-sectional in nature; however, the study is ongoing. A related list from the same vendor used by the US Behavioral Risk Factor Surveillance System was employed to recruit individuals. The study intentionally oversampled both blacks and residents of the stroke belt, a region of the southeastern USA where individuals have a substantially increased risk for stroke (Alabama, Arkansas, Georgia, Louisiana, Mississippi, North Carolina, Tennessee and South Carolina). Initial contact with participants was conducted via a mass mailing to inform participants that they would be contacted via telephone to discuss about their participation. This call lasted approximately 45 min, and data on self-reported race, risk factor characterisation, SES and medical history were collected. An in-home visit by a trained health professional followed the phone session, during which they obtained written consent and blood and urine samples. The present study was conducted according to the guidelines laid down in the Declaration of Helsinki, and all procedures involving human subjects were approved by the Institutional Review Board at all participating institutions. Written informed consent was obtained from all participants.
Analysis variables
Socio-economic status: measurements utilised
The present study assessed SES using education, income and a community-level index( Reference Diez-Roux, Kiefe and Jacobs 23 ). Education is reasonably well stable into adulthood( Reference Shavers 24 ) (matching the composition of the REGARDS population), and has been shown to capture lifestyle/behavioural practices( Reference Shavers 24 ). Income can influence access to goods such as healthful foods( Reference Shavers 24 ), and has been shown to affect diet independently of education( Reference Galobardes, Morabia and Bernstein 25 ). The community-level index used in the present study included measures of assets, home value and occupational prestige (in addition to income and education)( Reference Diez-Roux, Kiefe and Jacobs 23 ). These additional measures are informative in that: (1) assets may help an individual successfully weather financial storms better than income alone( Reference Shavers 24 ); (2) one's home is typically the largest asset and its value has been shown to be associated with health behaviours and outcomes( Reference Coffee, Lockwood and Hugo 26 ); (3) occupational prestige can be demonstrative of an individual's standing in society and allostatic load independently of income( Reference Shavers 24 , Reference Adler and Stewart 27 ). Including all these measures into a composite index at the community level offers an additional insight by revealing where characteristics of an individual's community may uniquely influence health behaviours vis-à-vis an individual's personal/family characteristics( Reference Robert 28 ); furthermore, such indices are being increasingly used in attempts to capture the polyfaceted nature of SES( Reference Oakes and Rossi 29 ).
Assessment of individual education and household income
The level of education achieved and annual household income were self-reported during the initial phone interview session. Annual income was categorised into four groups: up to $20 000; $20 000–$34 999; $35 000–$74 999; $75 000+. The four levels of education response options were less than high school (non-high school graduate), high school graduate or equivalent certification, some college and college graduate or above. Participants could refuse to provide information on both income and education.
Assessment of community-level socio-economic status
The geographical area used to assess community-level SES was the US census block group (units averaging 1000 individuals). Using the 2000 US census data, the method developed by Diez-Roux et al. ( Reference Diez-Roux, Kiefe and Jacobs 23 ) was employed to assess community-level SES through six representative measures of education, wealth/income and occupational prestige. Two of these measures are given as dollar values: Median household income and median value of housing units. The other four measures are given as percentages: households with interest, dividend or rental income; adults who completed high school; adults who completed college; and residents employed in executive, managerial or professional occupations. z Scores were calculated for each measure and then added together to create a comprehensive score, with the lower number representing lower SES. To reflect the structure of the individual-level indicators of income and education, this variable was transformed into a categorical variable with four quartiles.
Dietary assessment
The Block98 FFQ was given to the participants during the in-home visit with detailed directions on how to complete it and instructions to mail all forms to the REGARDS Coordinating Center at the University of Alabama at Birmingham. The Block98 FFQ (www.NutritionQuest.com) is a well-validated instrument for measuring a majority of nutrients( Reference Patterson, Kristal and Tinker 30 ), and other versions of the FFQ have been further validated in diverse populations( Reference Block, Woods and Potosky 31 ). The Block98 FFQ used in REGARDS study was developed by Block Dietary Data Systems and distributed by NutritionQuest. The Block98 FFQ includes 150 multiple-choice questions to obtain information about the frequency and portion of food consumed, as well as preparation of 107 food items. Forms were collected, scanned and verified by at least two trained personnel at the REGARDS Coordinating Center, the results of which were sent to NutritionQuest for scoring and quantification of intake.
Dietary pattern derivation
Using the original 107 food items, fifty-six food groups were constructed based on the criteria such as nutrient similarities, culinary use and previous studies( Reference Judd, Gutiérrez and Newby 21 ). These fifty-six food groups were then used to derive the dietary patterns in the REGARDS cohort with FFQ data (n 21 636) obtained via a form of exploratory factor analysis, i.e. principal component analysis (the detailed description of the dietary pattern derivations in the REGARDS study has been published( Reference Judd, Gutiérrez and Newby 21 )). A random split sample method was used to validate the dietary patterns and ensure their ability to be replicated. The second half of the sample was used to carry out confirmatory factor analysis to validate the principal component analyses, and only food groups with absolute value loadings exceeding 0·20 for a given factor were included in the initial confirmatory factor analysis model. From these analyses, five dietary patterns emerged: convenience, plant-based, sweets/fats, southern and alcohol/salads (see Table 1 for the top ten factor loadings, see online Supplementary Table S1 for a more comprehensive list). For each participant in the study, the factor loading of each food group was multiplied by the average consumption of each food group in order to calculate their factor scores for each dietary pattern. Adherence to dietary pattern was determined by splitting individual factor scores at the median. For a given dietary pattern, participants with a factor score above the median were categorised as high adherers, whereas individuals with a factor score below the median were categorised as low adherers. The adherence of a participant to any given dietary pattern did not preclude the participant from being a high or low adherer to any other dietary pattern.
Table 1 Dietary patterns generated by principal component analysis from the REasons for Geographic And Racial Differences in Stroke (REGARDS) study population (2003–7) and their top ten constituent food/beverage items, as ranked by factor loadings
Veg., vegetable; SSB, sugar-sweetened beverages.
* Includes items such as jelly, jam, syrup and sugar in coffee/tea.
† Added fats include shortening, lard, vegetable oil, olive oil, gravy and mayonnaise.
‡ ‘Tea’ includes unsweetened tea or tea sweetened with non-nutritive sweeteners, SSB includes all beverages that use nutritive sweeteners other than 100 % juice, ‘Soda’ is a subset of SSB and includes only carbonated SSB.
§ High-fat dairy include dairy products made from whole fat milk, whereas Milk – Highfat only includes whole fat liquid milk.
Analytic sample
The total sample size used in the analyses was 17 062 (Table 2). The total sample size of the original REGARDS study (2003–7) was 30 183, but was reduced for our analyses due to unavailable data on FFQ (n 8547), as well as measures of SES (n 4574). When compared with the full REGARDS cohort, the analytic sample was not significantly different in terms of age, sex or region. Due primarily to lower rates of return for FFQ among black participants, there was a significantly lower percentage of black participants in the analytic sample (34·6 %) than in the full REGARDS cohort (41·5 %), but still enough to detect racial differences, and still a percentage at least that of the general US population (13·2 %)( 32 ) (see online Supplementary Table S2). All results (descriptive statistics and logistic regressions) presented in the following section make use of the analytic sample.
Table 2 Descriptive statistics of the REasons for Geographic And Racial Differences in Stroke (REGARDS) study population (2003–7)* †
SES, socio-economic status.
* Values were significantly different between both races for all variables shown (P< 0·05; χ2 tests were performed for all categorical variables, and Student's t test for continuous variables).
† Unless otherwise indicated, values given are percentages.
Statistical analyses
Logistic regression was used to assess the associations between adherence to dietary patterns and measures of individual education, household income and community-level SES. Analyses were run with only one dietary pattern per model (the dependent variable in each model), and were first run in the overall population sample, and then by racial group. In the overall model, covariates included age, racial group, sex and region (stroke belt, stroke buckle, defined as 153 counties in the coastal plains of North Carolina, South Carolina and Georgia, and non-stroke belt), whereas in the models analysed by racial group, covariates included age, sex and region. Three interaction variables were created to test for interactions between racial group and socio-economic variables: race × individual education; race × household income; race × community-level SES. When the interaction terms were statistically significant, analyses were stratified by racial group. A significant interaction was considered to be P< 0·05.
All assessments of the significance of associations were taken from the Type III analysis of effects/Wald χ2 test, and OR estimates were obtained from the point estimate and 95 % Wald confidence limits. The SAS version 9.3 (SAS Institute, Inc.) was used for all analyses.
Results
The average age of the study sample was 64·7 years, with 54·4 % being females (Table 2). In the sample, 34·6 % of the population was black, 38·4 % were in the highest bracket of education and 19·4 % were in the highest bracket of income.
The associations between SES measures and adherence to dietary patterns for the overall sample, i.e. both black and white participants, are presented in Table 3 as OR with CI. Those participants in the highest v. lowest tiers of all SES measures – household income, individual education and community-level SES – were significantly more likely to adhere to plant-based and alcohol/salads dietary patterns (OR>1), but less likely to adhere to either sweets/fats or southern dietary patterns (OR < 1). Additionally, participants in the highest v. lowest tiers of both individual education and community-level SES, but not household income, were more likely to adhere to the convenience dietary pattern.
Table 3 Logistic regression of dietary patterns v. socio-economic status (SES) indicators in the REasons for Geographic And Racial Differences in Stroke (REGARDS) study population (2003–7) (Odds ratios and 95 % confidence intervals, n 17 062)
Ref, reference.
* Values were significantly different by race (P< 0·05); see Table 4.
† Covariates for income models include age, sex, race, region and education; covariates for education models include age, sex, race, region and income; covariates for community SES models include age, sex, race and region.
‡ Community SES z scores were obtained using the six-component method developed by Diez-Roux et al. ( Reference Diez-Roux, Kiefe and Jacobs 23 ), and then stratified into quartiles. Components included median household income, median value of housing units, percentage of households with interest, dividend, or rental income, percentage of adults who completed high school, percentage of adults who completed college and percentage of residents employed in executive, managerial or professional occupations.
Interactions between racial group and socio-economic variables were tested in the overall sample using the following interaction terms in the logistic regression models: race × individual education; race × household income; race × community-level SES. As denoted by asterisks in Table 3, statistically significant differences between black and white participants were observed in the associations between household income and adherence to alcohol/salads, individual education and adherence to plant-based and sweets/fats, and community SES and adherence to convenience patterns. Given the significant racial differences in the associations between measures of SES and adherence to dietary patterns, these associations were analysed for each racial group. Although the magnitude of adherence differed significantly by racial group for some dietary patterns and measures of SES, the direction of adherence was the same for both black and white participants (Table 4).
Table 4 Logistic regressions of dietary patterns v. socio-economic status (SES) indicators by race: high v. low SES indicator and adherence to dietary patterns in the REasons for Geographic And Racial Differences in Stroke (REGARDS) study population (2003–7) (Odds ratios and 95 % confidence intervals, n 17 062)
* Statistically significant association of the interaction of race × SES indicator with adherence to the given dietary pattern in the overall model (P< 0·05). Interaction models for individual SES are for race × income or race × education, respectively. Covariates for these interaction models include age, sex, race, region, education, income and the other interaction term. Interaction models for community SES are race × community SES and covariates include age, sex, race, region and community SES.
† Covariates for income models include age, sex, race, region and education; covariates for education models include age, sex, race, region and income; covariates for community-level SES models include age, sex, race and region.
‡ Community SES z scores were obtained using the six-component method developed by Diez-Roux et al. ( Reference Diez-Roux, Kiefe and Jacobs 23 ), and then stratified into quartiles. Components included median household income, median value of housing units, percentage of households with interest, dividend or rental income, percentage of adults who completed high school, percentage of adults who completed college, and percentage of residents employed in executive, managerial or professional occupations.
When further including lifestyle/demographic variables such as smoking, physical inactivity and BMI (online Supplementary Tables S3 and S4), all of these associations were maintained except that those participants in the highest v. lowest tiers of income (overall sample only) were now more likely to adhere to the convenience pattern, but no longer or less likely to adhere to the plant-based dietary pattern. These shifts are not surprising given the close proximity of the previous CI to 1, and the fact that health behaviours tend to be related to one another( Reference Gillman, Pinto and Tennstedt 33 , Reference Blair, Jacobs and Powell 34 ).
Discussion
Although a number of studies have investigated the associations between diet, race and SES, the majority of research has assessed diet either according to its component nutrients( Reference Huffman, Vaccaro and Zarini 35 – Reference Darmon and Drewnowski 37 ), or using prescribed, a priori dietary indices( Reference García-Arenzana, Navarrete-Muñoz and Peris 16 , Reference Hiza, Casavale and Guenther 18 , Reference Olmedo-Requena, Fernández and Prieto 20 , Reference Monsivais and Drewnowski 38 ). These assessments, while informative in their own right, may not entirely incorporate or reflect underlying patterns of food choice. By using a posteriori dietary patterns, researchers are able to see how foods group together in the diet and which foods make up a large portion of the diet( Reference Newby and Tucker 39 ), allowing for more effective behavioural and policy-based interventions. Using a large nationwide sample and a posteriori dietary patterns, some of which have previously been demonstrated to be associated with the risk of incident stroke( Reference Judd, Gutiérrez and Newby 21 ), the present study examined how SES, both individual and community level, as well as racial group, may be associated with the dietary patterns that individuals choose to follow.
All measures of SES were positively associated with adherence to a plant-based dietary pattern; however, household income showed the weakest association with adherence to this pattern, leading to the conjecture that education may be a more important determinant of what could be described as a ‘healthy’ diet than income alone. This finding is echoed by recent work in children demonstrating maternal education to be a more consistent predictor of diet than household income( Reference Zarnowiecki, Ball and Parletta 40 ).
While both racial groups demonstrated positive associations between individual education and adherence to the plant-based dietary pattern, education in whites was more strongly associated with adherence to this pattern compared with blacks. This finding brings to mind existing racial disparities in educational attainment, namely that the percentage of persons educated at the collegiate level and beyond is higher among whites than blacks( 41 , Reference Aud, Fox and Kewal Ramani 42 ). The racial disparity in plant-based diet adherence, compounded by disparities in educational attainment, may contribute to racial health disparities since consumption of fruits, vegetables and legumes appears similarly beneficial for both black and white Americans( Reference Kolonel, Hankin and Whittemore 43 , Reference Brown, Swanson and Gridley 44 ). Although initiatives to close the higher education gap are currently in place, our research evidences the importance of implementing such initiatives as a potential tool to prevent diet-mediated racial disparities in health.
A unique pattern that emerged from the factor analyses was the alcohol/salads pattern. Adherence to this pattern was positively associated with all indicators of SES. The race by household income interaction was significant in this pattern, with greater adherence seen in whites in the highest income category than blacks in the same income category. This racial difference is not entirely surprising given the fact that the percentage of the white population who are current drinkers exceeds that of the black population by 15 %( 45 ). Furthermore, heavy alcohol consumption such as binge drinking has been shown to increase relative to income more in whites than blacks( Reference Spiegler 46 , Reference Orcutt and Rudy 47 ). Although green salads are a positive component of the alcohol/salads pattern, the results of these analyses, taken in the aforementioned context, reveal the potential need for public health initiatives to encourage responsible consumption of alcohol among higher-earning white Americans.
Our data show that adherence to dietary patterns in both blacks and whites was in the same direction in all significant relationships, but that the magnitude of adherence between the racial groups was different in some instances. This demonstrates that while an individual's racial group may be an influential factor in determining food choice, stark dichotomies are not observed merely due to race.
An important message that can be gleaned from the results of the present study is that SES may supersede culinary traditions in some instances. This is most clearly demonstrated when looking at the southern dietary pattern. In a previous study examining black and white older adults in the southeast, education has been negatively associated with a preference for typical southern foods( Reference Yang, Buys and Judd 48 ). Our findings support this study, since in every indicator of SES used in the present study, and in both racial groups (no race × SES indicator interactions were significant), SES was negatively associated with adherence to the southern dietary pattern. The ability of education, income and/or SES to overcome the ties of cultural lifestyle observances, especially when those observances could be deleterious to an individual's health, may prove hopeful in public health campaigns (e.g. encouraging those individuals in the lower tiers of SES who are strongly adherent to this pattern to reduce their consumption of fried and processed meats, as well as sugar-sweetened beverages). However, in order for increasing an individual's education to have an independent effect on diet choice, education itself must be the driving force behind this association, perhaps exerting its effect by increasing an individual's health literacy. This relationship, however, remains to be conclusively known, and perhaps educational attainment may be, to some extent, a marker for familial characteristics, or even greater self-efficacy.
The present study was unique due to a confluence of elements. By using dietary patterns previously shown to be associated with the risk of incident stroke( Reference Judd, Gutiérrez and Newby 21 ), the present study was able to address how SES and/or an individual's racial group relate to diet selection, and sheds light on how such disparities in diet may affect disparities in health outcomes observed among such groups. The use of a nationwide data set allows for generalisability beyond the southeastern USA, a region with increased cardiometabolic risk and the original impetus for the REGARDS study. Furthermore, the population included individuals of varying levels of SES (both individual-level and community-level indicators) as well as a substantial number of minority participants (34·6 % black), especially compared with previous studies in this arena. The large number of black participants strengthened the study by providing the power to detect significant interactions between variables within racial groups.
Despite the many strengths of the study, a number of limitations persist. While the FFQ used has been validated in a population similar to that of the present study( Reference Block, Woods and Potosky 31 ), no self-reported diet measures can perfectly represent actual food consumption, especially when the recall period is an entire year. Self-report was also utilised in the collection of measures of household income and education. Since data on household size were not available, it was not possible to ensure that household income was assessed homogenously across all study participants, and this may have introduced some uncertainty in the analyses using household income. However, given that the relationships between household income and adherence to the dietary patterns mirrored the relationships between the other measures of SES and adherence to the dietary patterns (except for the convenience pattern, which approached significance), this does not appear to be of great concern. Although diet cost has been associated with diet quality in the literature( Reference Drewnowski 49 – Reference Darmon, Briend and Drewnowski 52 ), data pertaining to food purchases were not available in this sample. As a result, the putative impact of income on food accessibility, and therefore dietary pattern selection, unfortunately could not be directly assessed in this population. Although the sample size for the present study was quite large (n 17 062), almost half of the REGARDS cohort did not have sufficient SES or nutrition data for inclusion in the study (see online Supplementary Table S1). Additional research needs to be conducted in order to more fully understand the underlying reasons behind the associations revealed in the present study, such as accessibility to food outlets, time constraints in food preparation, taste preferences, and among others.
The present study provides insight into how groups differing by SES and/or race currently consume food, and serves as a springboard for further elucidation of diet-mediated health disparities. Given the complex nature of diet selection, intervention studies should be designed to more fully isolate the effects of specific SES components. Based on the knowledge gleaned from the present study, as well as future studies, interventions can be more effectively designed to shift individuals' dietary patterns and practices towards those that are healthier. Such interventions may not only ameliorate observed health disparities among groups, but also raise the overall level of health for the general population.
Supplementary material
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/S0007114515000938
Acknowledgements
This research project is supported by a cooperative agreement U01 NS041588 from the National Institute of Neurological Disorders and Stroke, National Institutes of Health, Department of Health and Human Service. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of Neurological Disorders and Stroke or the National Institutes of Health. Representatives of the funding agency have been involved in the review of the manuscript but not directly involved in the collection, management, analysis or interpretation of the data. The authors thank the other investigators, the staff and the participants of the REGARDS study for their valuable contributions. A full list of participating REGARDS investigators and institutions can be found at http://www.regardsstudy.org.
The first author was supported by the following NIH grant mechanisms: NIDDK P30DK056336 and NHLBI T32HL105349.
None of the authors has any conflict of interest to declare.
The authors' contributions are as follows: K. P. K. was involved in the conception and design, analysis and interpretation of the data, drafting and revision of the article, and final approval of the version; S. E. J. was involved in the conception and design, analysis and interpretation of the data, revision of the article, and final approval of the version; K. E. P., J. M. S. and J. R. F. were involved in the interpretation of the data, revision of the article and final approval of the version.
