Interventions promoting healthy eating as a tool for reducing social inequalities in diet in low- and middle-income countries: a systematic review

Following the PRISMA guidelines, studies were identified by searching PubMed, Scielo and Google Scholar electronic databases. No restrictions were set for year of publication. The last search was run on the 8^th of March 2016. The free search terms included in the search are summarized in Additional file 1: Table S1. Additionally, we included the following MeSH terms in Pubmed: Chronic Disease, Diabetes Mellitus, Cardiovascular Diseases, Obesity, Hypertension, Coronary Disease, Cerebrovascular Disorders, Neoplasms, Respiration Disorders, Chronic Disease/diet therapy, Nutrition Policy, Nutrition Therapy, Diet Therapy, Developing Countries, Asia, Africa, Latin America, South America and soecioeconomic factors. Reference lists of included studies were also searched to identify potential articles.

Exclusion criteria were: 1) pharmacological interventions; 2) prevalence or qualitative studies; 3) studies conducted on HICs; 4) studies conducted on undernourished or diseased individuals, children, adolescents or non-representative populations (e.g. studies in prisioners or caregivers); 5) articles written in languages other than English, French, Spanish or Portuguese, and 6) non-peer-reviewed studies, duplicate publications or articles restricted to an abstract. Interventions conducted in children and adolescents were excluded as the target population were adults and dietary practices differ according to age [29]. Studies including individuals with a body mass index ≥ 25 kg/m² were included as such participants were not considered as diseased.

Inclusion criteria were: 1) studies conducted in LMICs, defined according to the World Bank classification [30]; 2) cross sectional or quasi-experimental, peer-reviewed studies; 3) focusing on interventions promoting healthy eating among adults (≥18 years) or households; 4) measuring at least one healthy eating outcome [20] (i.e. fruit and vegetable intake or changes in food behavior) and 5) showing results stratified by at least one SES indicator (i.e. household assets, education, income, occupational position or area-based SES indicators).

For all articles initially selected, the title and abstract were separately assessed by two reviewers for eligibility (ALM and COVD). Disagreements were resolved by discussion. Data extraction for each study included country where the study was conducted, period of intervention, sample size, target population, SES categories, communication tools, and duration of the nutrition education sessions. Total duration of interventions, individual health behavior change model used, expected outcome measures and effects of healthy eating were also exctracted.

All interventions were categorized according to two criteria: 1) those focused on disadvantaged populations categorized as such by each included study; or 2) those directed to the entire population but stratifying results by SES. In this study, social inequalities in diet refer specifically to differences in dietary intake between socioeconomic groups, not by other social stratifiers such as age, sex or ethnicity. The effect of each intervention on social inequalities in diet was assessed using a similar approach to the one used in a previous study [20]. Interventions were defined as: 1) likely to reduce social inequalities if healthy eating outcomes had improved more in low vs. high SES individuals or households (e.g. more servings per day of fruit and vegetable consumption in low vs. high SES groups); 2) likely to increase social inequalities if healthy eating outcomes had improved more in the high vs. low SES individuals or households (e.g. more servings per day of fruit and vegetable consumption in high vs. low SES groups), and 3) likely to have no impact on social inequalities if healthy eating outcomes reduced or increased similarly in individuals and households of all SES groups.

The study selection procedure is shown in Additional file 2: Figure S1. The search of the three databases plus hand searched articles provided a total of 249 studies. Most excluded studies were conducted in children, adolescents or diseased individuals, assessed no intervention, or did not stratify results by SES. A total of seven articles were included; their main characteristics are shown in Table 1. All studies used nutrition education to promote healthy eating. Studies assessed interventions conducted between 2002 and 2014 in seven LMICs: five in the Americas and the Caribbean (Brazil [31], Chile [32], Colombia [33], Panama, Trinidad and Tobago [34]), and two in the Middle East and Africa (Iran [35], Tunisia [36, 37]). One study focused on families [31], one on private sector employees [37], and four on women [32, 34, 35, 38]. One study included individuals with risk factors (body mass index ≥ 25 kg/m²) [32].

All included interventions are summarized in Table 2. Four of the seven interventions focused on disadvantaged populations only [31, 32, 34, 35, 38]. Two used printed material for education purposes [32, 38] and one used media and face to face training [34]. Their total duration ranged from 6 weeks in Panama and Trinidad and Tobago [34] to 6 months in Chile [32]. Two of the four interventions used a behavior change model to plan the nutrition education sessions [34, 38]. One intervention was a healthy eating intervention only [31] while the others aimed at improving healthy eating plus increasing physical activity [32, 34, 38], decreasing smoking [38] or increasing cancer screening [34].

Three interventions were directed to the entire population [35‐37]. All used printed material and media for nutrition education purposes, and two used face to face training [35, 37]. Their total duration ranged from 3 to 5 years and aimed at improving healthy eating, physical activity and decreasing tobacco use.

The effect of interventions promoting healthy eating and their impact on social inequalities are described in Table 3. Two of the four interventions focusing on disadvantaged populations reported an increase in fruit and vegetable consumption after the intervention [31, 38], engendering a decrease in the overall inequalities in diet while assuming that all conditions were kept equal for the advantaged group. One study reported a significant decrease in fruit and vegetable intake after the intervention [34], being likely to increase social inequalities in diet. Another study reported food behavior change after the intervention (i.e. increased intake of skim milk and whole bread), being likely to reduce social inequalities in the intake of these specific food items [32].

Two of the three interventions focusing on the entire population showed a significant increase in fruit and vegetable intake in the entire population after intervention [36, 37], being likely to have no impact on social inequalities in diet. Another intervention showed a significant decrease of the global dietary index in the entire population after intervention (i.e. improved healthy eating) [35], also being likely to have no impact on social inequalities in diet.

Most interventions were focused on women only (four interventions), as they are usually in charge of food preparation and household food security [39]. However, further research is needed to assess the effectiveness of interventions directed to men in LMICs, who frequently hold the power of decision making and usually have a major role in the distribution of food portions in households. Also, evidence on the impact of healthy eating interventions by gender is scarce, which highlights the need for research on the differences between interventions on men and women’s dietary habits. Most interventions used printed material, including two of the three interventions which were likely to reduce social inequalities in diet. As both effective interventions targeted only disadvantaged populations, we cannot conclude on the effect of using equity-sensitive printed materials in structural interventions and further research is suggested.

The three interventions that were likely to reduce social inequalities in diet had the shortest duration (4 to 6 months), suggesting that duration is not related to the effect of an intervention on social inequalities in diet. This result has very important implications for prioritizing the allocation of resources in interventions [40]. Moreover, duration of interventions promoting healthy eating that successfully improved dietary patterns in HICs [41‐43] and LMICs [44] varies considerably.

Only two of the seven interventions used an individual health behavior change model to plan the nutrition education sessions, and only one of these showed a reduction in social inequalities in diet. Thus, we cannot firmly conclude if the inclusion of behavior change models results in more effective interventions and greater benefit for participants [45].

Evidence on interventions to reduce social inequalities in diet in LMICs remains scarce. In HICs, interventions to reduce health inequalities have been conducted for more than a century [15] and usually include structural or agentic approaches [16]. Structural approaches, which are those that promote health by altering the structural context in which health is produced [17], have usually targeted the entire population [46‐49]. They may refer to nutrition-specific policies (i.e. policies aimed to influence food supply or consumption) or nutrition-sensitive policies (i.e. policies implemented outside of the health and food sectors). Some examples include food labeling, food reformulation, food taxes and subsidies, income and social protection policies [22]. Agentic approaches have usually targeted the entire population or disadvantaged populations only [50, 51] (e.g. nutrition education and health information campaigns) [22]. However, whether interventions should target only disadvantaged individuals, the entire population or combine both approaches is a matter of debate. Advocates of high-risk strategies highlight the need to direct resources towards individuals at high risk [52], while those in favor of population strategies argue that many people at a small risk may lead to more cases of disease than a few people with high risk, making these strategies temporary and palliative [53]. Thus, in HICs there is no “one-fits-all” approach. Several authors have proposed the two perspectives are compatible [15, 16] and should not be mutually exclusive but complementary [25] because focusing only on agentic interventions does not reduce the unequal population distribution of social and economic resources which cause dietary inequalities [54].

In LMICs, interventions to improve diet have mainly taken the form of nutrition education programs (i.e. agentic interventions) but most did not assess social inequalities in diet. Even though some studies assessed structural interventions (e.g. subsidies of fruit and vegetables), they could not be included in our review as results were not stratified by SES. Thus, our results are limited to agentic interventions. In our review, three interventions focusing on disadvantaged populations were likely to reduce social inequalities in diet. One educational intervention specifically targeting disadvantaged populations yielded the unexpected result of a decrease in fruit and vegetables consumption (although a causal effect of the intervention could not be proved). A qualitative study conducted by the authors showed that a likely explanation was the high price of fruit and vegetables and a preference for fried foods, carbohydrates and sweetened beverages among the participants. Several intervention trials in HICs have shown that nutrition education alone is insufficient to increase the intake of fruit and vegetables, but that the combination of nutrition education plus decreased prices is [22, 55]. This was the case in one of the interventions in our study which additionally included increased access to fruit and vegetables [31]. Thus, we were unable to disentangle the individual effects of the nutrition education sessions and accessibility to healthy foods on social inequalities in diet in LMICs. Similarly, two other studies in our review combined nutrition education sessions with social support [34, 38], and we could not disentangle their effects on social inequalities in diet. Further research is needed on interventions promoting healthy eating in LMICs, especially regarding increased purchasing power and social support to reduce social inequalities in diet.

The three interventions targeting the entire population improved healthy eating outcomes in all socioeconomic strata, thus being likely to have no impact on social inequalities in diet. These results are in line with evidence in HICs showing no impact on social inequalities in diet of interventions such as user-friendly food labeling (e.g. traffic light schemes), mandatory and standardized front-of-pack labeling, and population-wide control of unhealthy food marketing through mass media [22].

In our study, as all agentic interventions targeting the entire population failed to reduce social inequalities in diet, our results may suggest that structural interventions targeting the entire population can tackle the unequal distribution of factors that hinder the opportunity to eat a healthy diet [22]. One example in a HIC (New Zealand) is a targeted food pricing policy which was effective to improve the diet of the entire population and reduce dietary inequalities [56]. In LMICs, some examples of structural interventions to reduce social inequalities in diet and improving health and nutrition outcomes include “Bolsa Familia”, a social welfare program in Brazil, and “Oportunidades”, a conditional cash transfer program in Mexico. Even though they did not specifically target inequalities, they approached underlying drivers of social inequalities which often lead to the reduction of dietary inequalities [22].

Our study has some limitations. First, most interventions were conducted in the Americas and the Caribbean, and implementation of interventions promoting healthy eating might differ in other countries, especially in low income countries. However, we believe that our main conclusion is applicable to all low- and middle-income countries similar to the ones where the studies were conducted, although no intervention is effective in all settings or considered as “one-size fits all”. Second, two interventions targeted several settings in communities (i.e. neighborhoods, schools, and workplaces) [35, 37]. Thus, we could not separate the results obtained in one setting from the others. However, none of these interventions had an impact on social inequalities in diet. Third, due to the small number of studies reporting results by SES, only seven studies could be included in our review, and further research on the effect of interventions on social inequalities in dietary intake in LMICs is required. Fourth, three of the interventions were quasi-experimental [32, 36, 37] so causality between the intervention and the behavioral change can be inferred. Fifth, no study assessed long term dietary behavior change and the reduction of social inequalities in diet and further research is suggested.