Abstract
Background. Improving the diet of communities experiencing health inequities can be challenging given that multiple dietary components are low in quality. Mississippi Communities for Healthy Living was designed to test the comparative effectiveness of nutrition education using a single- versus multiple-message approach to improve the diet of adult residents in the Lower Mississippi Delta. Method. The single-message approach targeted discretionary calories while the multiple-message approach also targeted vegetables, fruits, whole grains, and lean protein. Delta food frequency questionnaires were used to measure participants’ diet, while the Healthy Eating Index–2005 (HEI-2005) was used to generate diet quality scores. Generalized linear mixed model regression was used to test for significant time, treatment, and time × treatment interaction effects in HEI-2005 component and total score changes. Results. The majority of participants in the single- and multiple-message arms (n = 114 and 127, respectively) were female (88% and 96%, respectively), African American (90% and 98%, respectively), overweight or obese (92% and 87%, respectively), and 41 to 60 years of age (57% and 43%, respectively). Significant time effects were present for HEI-2005 total and component scores, with three exceptions—whole fruit, total grains, and saturated fat. Significant treatment effects were present for two components—total and whole fruit; scores were higher in the multiple-message approach arm as compared to the single-message approach arm across time points. No interaction effects were significant for any of the HEI-2005 scores. Conclusion. Focusing nutrition education on the discretionary calories component of the diet may be as effective as focusing on multiple components for improving diet quality.
Health inequities observed among racial and ethnic groups are numerous and wide as evidenced by the 20 times greater heart failure rate in African Americans younger than 50 years as compared to European Americans (Bibbins-Domingo et al., 2009; Robert Wood Johnson Foundation, 2011). Rural and lower socioeconomic status populations also experience greater health inequities than their counterparts (U.S. Department of Health and Human Services, 2014). When compared to urban areas, rural areas generally have older, poorer, and more underemployed residents (Colocousis & Rogers, 2010) and larger concentrations of minorities, all of which experience inequities associated with greater vulnerability to overweight/obesity (Befort, Nazir, & Perri, 2012; Flegal, Carroll, Kit, & Ogden, 2012) as well as elevated prevalence of high blood cholesterol (Tran, Pullen, Zimmerman, & Hageman, 2015), diabetes, and hypertension (Brewer & Langerman, 2013; Lee, Liu, & Sales, 2006; Lower Mississippi Delta Nutrition Intervention Research Consortium, 2004; Mainous, King, Garr, & Pearson, 2004).
Consistent lack of access to (and therefore consumption of) healthy foods contributes to and may exacerbate the high rates of chronic health conditions found in residents of the rural Lower Mississippi Delta (LMD; Brewer & Langerman, 2013). However, the food landscape in the LMD may be changing as legislation is enacted to increase access to fresh and healthy foods in underserved communities (Koprak & Lang, 2012). This is encouraging given increased consumption of fruits, vegetables, and fiber has been shown to lower cholesterol, blood pressure, and body mass index (BMI) in low–socioeconomic status populations, while consumption of lower sodium foods and reduced saturated fat intake have been effective in optimizing health outcomes (Brewer & Langerman, 2013, Lower Mississippi Delta Nutrition Intervention Research Consortium, 2004, Strazzullo, D’Elia, Kandala, & Cappuccio, 2009). Recognition of these dietary health benefits has led to attempts to modify diets in order to reduce chronic disease risk, with recent reviews highlighting the importance of using multicomponent and theoretically based interventions when targeting communities experiencing health inequities (Stuart-Shor, Berra, Kamau, & Kumanyika, 2012).
Targeting communities experiencing health inequities can be challenging given their generally overall poor diet quality associated with low intakes or unhealthy food choices for multiple dietary components (Satia, 2009). Although multiple factors at the individual, social, and environmental levels influence food choices (Booth et al., 2001; Nestle et al., 1998; Popkin, Duffey, & Gordon-Larsen, 2005), key informants representing eight different community sectors in the LMD identified individual level factors, including not knowing how to make dietary changes, as important contributors to poor intake (Yadrick et al., 2001). These concerns, along with simulation modeling on actual diets of Delta residents to improve diet quality (Connell et al., 2015; Thomson, Onufrak, et al. 2011; Thomson, Tussing-Humphreys, et al. 2011), led to consideration of an intervention comparing two education approaches to dietary change. Hence, the primary aim of this study was to examine the comparative effectiveness of Mississippi Communities for Healthy Living (MCHL), a two-arm (single- vs. multiple-message approach; SMA and MMA, respectively), 6-month, theory-based, nutrition education intervention designed to improve the diet quality of adult LMD residents. Secondary aims included determining participant baseline characteristics associated with observed changes in diet quality.
Method
Sample and Setting
The study protocol was approved by the Institutional Review Board at The University of Southern Mississippi. All participants gave written informed consent during enrollment data collection events. Adult members of civic and social groups as well as faith-based organizations (n = 16) were recruited for study participation. Due to the proximity of organizations within counties and the overlapping social networks of individuals outside the community, two geographic clusters of organizations were identified in an attempt to reduce the risk of treatment cross-contamination. Organizations between clusters were then paired by member size and educational attainment. Organizations within each pair were randomly assigned by coin toss to the SMA or MMA. Further details regarding assignment to treatment arm are published elsewhere (Connell et al., 2015).
Intervention
The intervention was a 6-month pre–post study design with measures at enrollment, baseline, and postintervention. MCHL was constructed using the RE-AIM model (Reach Effectiveness Adoption Implementation Maintenance; Glasgow, Vogt, & Boles, 1999; Huye, Shoemaker, Landry, & Connell, 2015). The nutrition education program was developed using the diffusion of innovations theory (Huye, Connell, Crook, Yadrick, & Zoellner, 2013; Rogers, 2003). Diffusion of innovations theory suggests that an innovation (in this case new dietary behavior) that has perceived attributes of relative advantage, (low) complexity, compatibility, trialability, and observability will more likely be adopted. Each education session included components that were based on these five attributes and is more fully described in Connell et al. (2015). The intervention was designed as a two-arm, 6-month, pre–post study with measures collected at enrollment (Month 0), baseline (Month 3), postintervention (Month 9), and maintenance (Month 15). Changes between baseline and post intervention are the focus of this article. The SMA treatment arm included five 60-minute education sessions focusing on discretionary calories but with each session targeting a distinct aspect of solid fats and added sugars. The MMA arm also included five 60-minute education sessions; however, each session targeted a key message related to consumption of vegetables, fruits, whole grains, lean proteins (to reduce saturated fat), solid fats, and added sugars. A final sixth session summarized the educational content of each approach and provided participants an opportunity to share their own modified recipes at a potluck-style event. All education session messages and related session activities were developed using healthy adaptations and substitutions based on the usual diet in the region (Connell et al., 2015; Thomson, Onufrak, et al., 2011). Monthly sessions were held for both arms at locations identified by participating organization champions as convenient for group members. The champion, as described in Connell et al. (2015), was an opinion leader in the organization who encouraged participation and acted as a liaison between the organization members and research staff. After enrollment, one participant was hired as a study staff member because of her community involvement and the researchers’ need for an engaged leader. The education sessions were conducted by four nutrition educators local to the LMD, who were trained to lead sessions following designated lesson plans. Interactive sessions for both arms included discussions, games and activities, food demonstrations, and home challenges to encourage dissemination of the dietary changes to participants’ family and friends. After each monthly education session, participants received a newsletter tailored to each arm that reinforced the previous session’s key message, and a telephone call from the organization’s champion. All data were collected between June 2011 and October 2012 by trained study staff at locations determined by the community organizations to be convenient for participants. Further details regarding the intervention are published elsewhere (Connell et al., 2015).
Measures
The semiquantitative, 142-item Delta food frequency questionnaire (FFQ) was developed and validated using interviewer-administered methods (Carithers et al., 2009; Tucker et al., 2005). Based on the researchers’ previous survey experiences with the targeted population, the complexity of the Delta FFQ, and lack of studies on the reliability and validity of a self-administered Delta FFQ, study staff were trained to administer the FFQ to the participants. Participants used prompts like balls, cups, and decks of cards to assist in reporting portion sizes and eating habits over the previous 3 months at each data collection time point. Scannable FFQs were analyzed by the dietary assessment center at Northeastern University (http://www.northeastern.edu/dac/delta-niri-ffq/), and nutrient, food group servings, and Healthy Eating Index–2005 (HEI-2005) component and total scores were generated for each participant at each time point. The HEI-2005 is a diet quality tool that measures adherence to the 2005 Dietary Guidelines for Americans. It consists of 12 components: total fruit (including 100% fruit juice); whole fruit, total vegetables, dark green and orange vegetables and legumes (DGOVL), milk, meat and beans, total grains, whole grains, oils, sodium, saturated fat, and solid fats, alcoholic beverages, and added sugars (SoFAAS). Component scores are summed to create a total diet quality score that ranges from 0 to 100. Total scores of 80 and above indicate that recommended daily values of each component are being met (Guenther, Reedy, & Krebs-Smith, 2008). The HEI-2005 has been validated as an indicator of diet quality (Guenther et al., 2008) and used to determine effectiveness of nutrition education interventions in similar populations (Dynesen, Haraldsdottir, Holm, & Astrup, 2003; Mistry & Kasperson, 1998).
Height was measured to the nearest 1 cm using a stadiometer (Perspective Enterprise Adult Infant Measure model 101). Weight and BMI were measured concurrently to the nearest 0.1 kg using a body composition analyzer (Tanita model 310). Demographic characteristics and medical history were collected for each participant. Additional measures, including medication, health status, and psychosocial constructs related to diet self-efficacy, decisional balance, and social support that were assessed but are not relevant to the current article, are discussed elsewhere (Connell et al., 2015; see Table 1). All study procedures were tested in a feasibility study prior to implementation of MCHL and adjustments to procedures were made as needed.
Delta Healthy Eating Attitudes Scale: Mississippi Communities for Healthy Living, 2011-2012.
Data Analyses
Statistical analyses were performed using SAS software, Version 9.4 (SAS Institute Inc., Cary, NC). The significance level of the tests was set at .05. Generalized linear mixed models, using maximum likelihood estimation, were used to test for significant baseline differences and time, treatment, and time × treatment interaction effects in HEI-2005 component and total score changes. Maximum likelihood estimation is a method that can be used to handle missing data in repeated measures (Dempster, Laird, & Rubin, 1977). However, because results from previous analyses found no significant changes between enrollment and baseline (with the exception of an increase in total vegetables diet quality), enrollment values were imputed for missing baseline values in 22 participants. A sensitivity analysis in which these 22 participants were excluded also was conducted.
Time (baseline [3 months] and postintervention [9 months]) was modeled as a repeated measure (participant nested within social group) using variance components covariance matrix structure. Social group membership (nested within treatment) was modeled as a random effect using variance components covariance matrix structure. Diet quality was first modeled using time, treatment, and their interaction term as predictor variables. Subsequently, demographic characteristics and baseline measures (i.e., age, marital status, educational attainment, income, and baseline BMI) were included in change models to determine if treatment differences retained or attained significance in the presence of covariates. For comparison and modeling purposes, age was collapsed into three categories (18-40 years vs. 41-60 years vs. ≥61 years); marital status was collapsed into two categories (married vs. other [widowed, divorced, separated, and never married]); and educational attainment was collapsed into two categories (≤high school graduate/GED vs. ≥some college). Income was used in continuous form due to the large number (n = 11) of $5,000 increment categories. Gender, race, and smoking status were not included in these models due to the low proportions of males, races other than African American, and current smokers in the data set. Interpretation of results based on such small subsets of participants is difficult and can be misleading. Least squares means were computed to estimate and compare diet quality changes with Tukey–Kramer adjusted p values used for multiple group comparisons.
Results
In total, 319 individuals were enrolled in the study and subsequently randomized to a treatment arm (see Figure 1 for CONSORT diagram). Both baseline and post intervention measures were provided by 219 participants, and 22 provided both enrollment and postintervention measures, resulting in an analytic sample size of 241 participants (76% retention postintervention). Table 2 contains comparisons between the SMA (n = 114) and MMA (n = 127) treatment arms at baseline. The majority of participants in the SMA and MMA treatment arms were female (88% and 96%, respectively), African American (90% and 98%, respectively), overweight or obese (92% and 87%, respectively), and 41 to 60 years of age (57% and 43%, respectively). Prevalence of diagnosed high blood glucose was 19% and 27% and diagnosed high blood pressure was 64% and 65% for SMA and MMA groups, respectively. Baseline demographic comparisons between study completers (participants providing postintervention measures) and noncompleters revealed that noncompleters were more likely to be male as compared to completers (18% vs. 8%, p = .011).
CONSORT 2010 flow diagram.
Baseline Demographic and Anthropometric Measures by Treatment Arm: Mississippi Communities for Healthy Living, 2011-2012.
Note. SMA = single-message approach; MMA = multiple-message approach; BMI = body mass index.
p value based on generalized linear mixed models with social group treated as a random effect and nested within treatment arm. bCategories collapsed to African American versus other (European American, American Indian/Alaskan Native, Native Hawaiian/Pacific Islander, and multiracial). cCategories collapsed to underweight/healthy weight versus overweight/obese for comparison purposes. dCategories collapsed to married versus other (widowed, divorced, separated, never married) for comparison purposes. eCategories collapsed to ≤high school graduate/GED versus ≥some college(including trade or vocational school) for comparison purposes. fTreated as a continuous variable due to large number (11) of $5,000 increments.
Table 3 details the HEI-2005 component and total scores by treatment arm within time point. Significant time effects were present for HEI-2005 total and component scores, with three exceptions—whole fruit, total grains, and saturated fat (i.e., no change over time). For the total fruit, milk, and sodium components, postintervention scores were significantly lower than baseline scores (i.e., worsened diet quality). For the total vegetables, DGOVL, whole grains, meat and beans, oils, and SoFAAS components and total diet quality, postintervention scores were significantly higher than baseline scores (i.e., improved diet quality). Significant treatment effects were present for two components—total and whole fruit; scores were higher in the MMA treatment arm as compared to the SMA arm across time points. No interaction effects were significant for any of the HEI-2005 total or component scores. These results did not change substantively in the sensitivity analysis, which excluded the 22 participants for which enrollment values were imputed for missing baseline measures. That is, changes in significance were not observed. Hence, the imputed data set was used for subsequent analyses.
Mixed Model Linear Regression Analysis for Time and Treatment Effects on Diet Quality: Mississippi Communities for Healthy Living, 2011-2012.
Note. HEI = Healthy Eating Index; SMA = single-message approach; MMA = multiple-message approach; LSM = least squares mean; Trtmt = treatment; DGOVL = dark green and orange vegetables and legumes; SoFAAS = solid fats, alcoholic beverages, and added sugars.
p value for time by treatment interaction not significant for any HEI-2005 component or total score.
Table 4 details the mixed model linear regression results for diet quality score changes in which age, marital status, educational attainment, income, baseline BMI, and baseline value of outcome variables were included as covariates. Only nine diet quality components and the total score were included in these analyses due to their significant time effects. Of these nine components, three (total vegetables, whole grains, and oils) were not reported in the table because no covariates were significant in the models. Also, because treatment was not significant in these models, the changes reported in Table 4 represent mean changes across the two treatment arms (i.e., for the entire study cohort). Postintervention increases for DGOVL, meat and beans, SoFAAS, and total scores ranged from 0.5 points (meat and beans) to 2.0 points (SoFAAS). Conversely, postintervention decreases for total fruit, milk, and sodium scores ranged from 0.5 points (total fruit and milk) to 2.0 points (sodium). Marital status was a significant predictor of changes in meat and beans and sodium scores. The significant increase in the meat and beans score observed for participants who were not married was significantly larger than the nonsignificant increase observed for married participants. Conversely, decreases in the sodium score were significantly larger for participants who were not married as compared to married participants. Educational attainment was a significant predictor of changes in the SoFAAS score such that the significant increase observed for participants with more than a high school education was larger than the nonsignificant increase observed for participants with no more than a high school education. For the total and five of the six HEI-2005 component scores (all except SoFAAS), the baseline outcome value (e.g., baseline total score) was a significant negative predictor of change such that for every 1-point lower baseline diet quality score, respective postintervention diet quality scores increased 0.17 to 0.34 points. Baseline BMI was a significant negative predictor of change for total fruit and total diet quality such that for every 1-unit lower baseline BMI, the total fruit score increased 0.04 points and the total diet quality score increased 0.18 points at postintervention as compared to baseline. For both SoFAAS and total diet quality, income was a significant negative predictor of changes such that for every $5,000 increment decrease in income, SoFAAS and total diet quality scores increased 0.24 and 0.48 points at postintervention as compared to baseline.
Note. HEI = Healthy Eating Index; BVO = baseline value of outcome; BMI = body mass index; β = regression coefficient; DGOVL = dark green and orange vegetables and legumes; NS = not significant; HS = high school; SoFAAS = solid fats, alcoholic beverages, and added sugars.
Only HEI component scores with significant time effects were modeled and only those with significant covariates were included in the table (i.e., no covariates were significant in the total vegetables, whole grains, and oils models). bPredictor variables included time, treatment, age, marital status, education level, income, baseline BMI, and BVO. cValues in boldface indicate change is significantly different from zero.
Discussion and Conclusions
The purpose of this study was to evaluate the comparative effectiveness of a theory-based, two-arm, nutrition education intervention developed, based on formative research and the local diet in a rural region of the LMD experiencing health inequities. Results indicated that overall the intervention was effective at improving diet quality in this population of adults. However, meaningful differences between treatment arms were not apparent suggesting that targeting the discretionary calories component of the diet may be as effective as targeting multiple components of the diet for achieving improvements in diet quality. It should be noted that suggested methods for improving (or limiting) the SoFAAS dietary component involved substitutions that would consequently have a positive effect on other dietary components. For example, SMA messages included a “meal makeover” that prompted participants to replace fried vegetables within a meal with roasted vegetables, a substitution that may have effected a positive change in both the SoFAAS and vegetable components of diet quality. Other SMA messages included replacing fatty cuts of meat with leaner cuts and baking or steaming versus frying foods. Likewise, participants making these substitutions would have reported dietary intakes indicative of positive changes in SoFAAS as well as other diet quality components (e.g., meat and beans, oils). Hence, it is perhaps more accurate to conclude that targeting the discretionary calories component of the diet was as effective as targeting multiple components for improving diet quality because healthful changes in the discretionary calories component often positively affected other diet quality components as well.
It is not clear why the total fruit, milk, and sodium components of HEI diet quality significantly worsened over the course of the intervention. Participants were asked to recall consumption for the past 3 months; therefore, baseline intakes covered summer months, while postintervention intakes covered winter/early spring months. Although it is possible a seasonality effect resulted in the decreased total fruit diet quality, it does not explain the decreases observed in milk and sodium diet quality. Messages to decrease fat in the diet may have resulted in decreased intakes of full-fat milk and cheese products rather than substitution with lower fat versions. Furthermore, given that reduced-fat food substitutions were being made, these changes may have resulted in more sodium in the diet as such foods often add salt or sodium-based ingredients to improve the taste and mimic the sensory characteristics of full-fat products (Mistry & Kasperson, 1998).
The secondary objectives for this article, determining participant baseline characteristics associated with observed changes in diet quality, yielded interesting results. Similar to other studies (Mistry & Kasperson, 1998; Robinson et al., 2004), we found larger improvements in diet quality, specifically the SoFAAS component, in participants with greater than a high school education as compared to those with no more than a high school education. However, we also found that as income decreased, improvements in SoFAAS and total diet quality increased. Although these results are somewhat contradictory, we could find no evidence of collinearity (with education modeled in its original seven categories) or an interaction between education and income. Hence, it would appear that the messages related to discretionary calories had the largest effect on more educated participants as well as those with lower income.
Another interesting finding was the larger increase in meat and beans diet quality coupled with a larger decrease in sodium score observed in unmarried participants as compared to married participants. It may be that these two components are related given that many preparations for meat and bean dishes can be high in sodium (Tucker et al., 2005). Indeed we did find a significant negative relationship between scores for these two dietary components (r = −.35, p < .001). It is not clear why these differential changes occurred in the two marital groups as very little literature exists either refuting or supporting this finding. Regardless, these findings suggest that care should be taken when suggesting improvements in the protein component of the diet so that such improvements are not coupled with higher sodium intakes.
Finally, the negative association between BMI and changes in fruit as well as total diet quality was concerning given that overweight/obese individuals in general may benefit the most from improved diet quality. In a representative sample of U.S. adults, specifically in men aged 30 to 59 years and women aged 50 to 59 years, an inverse association between weight status and diet quality (as measured by the HEI-2005) was found (Pate, Taverno Ross, Liese, & Dowda, 2015). It may be that healthful dietary changes are harder or take longer to have an effect in more overweight/obese individuals as compared to those closer to a healthy weight. It also is possible that our intervention was simply not as effective with participants at higher BMI. In fact, determining what factors affect adherence to a diet plan within differing weight groups is the objective of an on-going clinical trial (https://clinicaltrials.gov/ct2/show/NCT01862796).
A strength of this study is the fact that the approaches for each intervention arm were developed specifically for the cultural and dietary milieu, building on the presence of local social groups and adapting healthier food choices from existing diets. This tailoring also may limit the generalizability of results. The inclusion of peer leaders or local champions, including the participant who was hired as research staff, is also a strength, given the promise of community-engaged research for improved health outcomes (Coughlin & Smith, 2016), as well as a history of mistrust of health research among African Americans (Scharff et al., 2010). A potential weakness is recall bias, given the retrospective nature of FFQs for dietary data collection.
The results of MCHL suggest that a theory-based, community nutrition education intervention can improve specific aspects of, as well as overall diet quality in, adult residents of the LMD. Furthermore, targeting the discretionary calorie component of diet quality appears as effective as targeting multiple components of the diet. This may be an important consideration for programs focused on nutrition education to improve dietary intake, such as SNAP-Ed (Supplemental Nutrition Assistance Program educational project). Nutrition education programs targeting community members may also need to consider tailoring that addresses differences in education level and weight status of enrolled participants, as well as including adaptations that address differences in household composition.
Footnotes
Acknowledgements
The authors would like to acknowledge The University of Southern Mississippi research team who was instrumental in the development and implementation of Mississippi Communities for Healthy Living, including LaShaundrea Crook, Briauna Elam Perryman, Martha Resavy, Jamie Zoellner, and Karen Zynda.
Authors’ Note
Research was conducted at The University of Southern Mississippi. Jessica L. Thomson was with United States Department of Agriculture–Agricultural Research Service when she completed this study.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Support for the study was provided by the U.S. Department of Agriculture, Agricultural Research Service, Delta Obesity Prevention Research Unit, under Project 58-6251-8-043.