Dietary patterns and associated risk factors among school age children in urban Ghana

Study design, population and setting

A cross-sectional design was used to assess the dietary behaviour and food consumption patterns of basic school pupils aged 9–15 years. Participants were enrolled from both private (n = 12) and public (n = 12) schools in the Ga-East Municipality of the Greater Accra Region of Ghana. Participating schools had both primary and Junior high school departments. The Ga-East Municipality is one of 16 administrative districts of the Greater Accra Region which contains the nation’s capital, Accra.

Sampling procedure and technique

The ability of children to adequately identify foods and recall dietary information is age-dependent. Thus, the minimum age for participating pupils was 9 years in accordance with recommendations regarding age-appropriateness of available dietary assessment methods [39]. The minimum sample size of 448 estimated was based on a pupil population of 50,200; expected frequency of obesity of 10%, assuming a least expected prevalence of 7% and a non-consent rate of 15%. A sample proportional to the size of each school was selected to make up a total sample size of 500 pupils. A total of 487 pupils eventually participated in the dietary assessment.

The full list of basic schools within the district was obtained from the Ghana Education Service (GES). Simple random sampling (balloting technique) was used to select 12 private and 12 public schools from a total of 80 (47 public and 37 private) registered basic schools (having both primary and Junior High School (JHS)) in the municipality. Within each school, balloting without replacement was used to select eligible pupils: Ballot bowls (separate coloured for boys and girls) containing pre-labelled (‘yes’ or ‘No’) ballot papers equivalent to the number of pupils on roll for specific research day were moved round primary classes 4 to JHS 3 by research assistants for pupils to pick. Selected pupils were given a written informed consent form to obtain parental consent. Completed consent forms were submitted to class teachers and subsequently collected by research team. Pupils who received parental consent also needed to complete an assent form to indicate child’s willingness to participate. As part of the assent process, the objectives, risks/benefits, costs and procedures of the study were explained to the children in their preferred language (English, Twi or Ga).

Study measures and data collection procedures

Data analysis

Data were entered using a Microsoft access (2007 version) database, with activated checks to reduce errors. IBM SPSS version 20 software was used for data analysis. Frequency distributions were summarized in tables for categorical demographic characteristics. Arithmetic mean was determined for respondent age. Preliminary tests were carried out on data to check for normality/distribution (z- scores for outliers, Kolmogorov Smirnov test, Q-Q plots) and also reliability test (Cronbach’s alpha test statistic) performed on items included in Principal Component analyses (PCA).

Age (years) of respondents was grouped into the following categories: 9 to 10 years, 11 to 12 years and 13 to 15 years age groups); sex of pupil (either male or female); school administration type and household socio-economic status (SES) of respondents. Household SES was ascertained using principal component analysis. The factor score based on ownership of 8 items (car/ vehicle, house, television, refrigerator, gas/electric cooker, DVD/VCR/video player, air conditioner, and satellite dish) was used to group respondents into rich, middle and poor households [42]. Respondents were classified as overweight/obese or non-overweight/obese based on their BMI-for-age and sex (BMA) z-score. Overweight and obesity was defined as BMI for age-and –sex z-score (BMA) > 1 SD and ≥ 2 SD respectively; normal (non-obese/overweight) was defined as BMA ≤1 SD [43]. For the purposes of this study, pre-obese and obese groups were considered together in one group known as “overweight” and used for tests of associations.

Dietary patterns were assessed using PCA based on data collected with FFQ. In order to ensure that > 5% of respondents consumed each food item included in PCA, food items were categorized into several sub-groups based on shared nutritional value/significance. Sub-groups of foods (see Additional file 2) created for PCA are: processed meats; spreads and toppings; fried foods; cocoa beverages and Dairy; high caloric foods; poultry and eggs; starchy roots& tubers; sugar sweetened beverages; fruits and fruit juices; vegetables and soups; meats (all non-poultry meats); cereals/grains; legumes & nuts; and fish and sea foods. Thus, a total of 14 items were subjected to PCA in order to identify foods highly correlated with each other and possibly combined to describe the major patterns in respondent diets. PCA used orthogonal rotation (varimax) and number of components determined by Kaiser’s stopping rule (factors with Eigen values over 1.0 are considered in analysis) (Kaiser [44]) and examination of scree plots. Varimax rotation was used to maximize the factor loadings onto each component to allow ease of interpretation of patterns. Patterns were defined based on factor loadings > 0.4 and < − 0.4. In order to assess suitability of data for factor analysis, acceptable cut-offs for the Kaiser-Meyer-Olken (KMO) Measure test and Bartlett’s test of sphericity were used. Dietary patterns derived from PCA were appropriately labeled to reflect the constituents of each isolated component accounting for variation in diets. A dietary pattern, thus, refers to a distinguishable set of highly correlated food items as extracted in PCA.

Simple linear regression, controlling for age, was used to test the association between dietary pattern factor scores and child sex, weight status, school type and household SES. Sensitivity analyses showed that age used in continuous form or categorical form showed no significant changes in level of significance of associations between DPs and covariates. Thus, age was used in the categorical form in analyses. Multiple linear regression was performed to determine significant factors associated with each of the four dietary patterns. An alpha level of 0.05 was used to determine statistical significance.

Demographic characteristics

Emerging dietary patterns

Associations between major dietary patterns and pupils characteristics

The second pattern, “starchy roots &vegetable-based diet”, which is ‘typically Ghanaian’, explained 13.7% of variation in pupil diets and was negatively associated with overweight status [F(2, 482) = 7.369, p = 0.001]. It was also associated with older age groups [F(1, 483) = 10.091, p = 0.002], public school attendance [F(2, 482) = 12.702, p < 0.001] and lower household SES [F(2, 482) = 8.725, p < 0.001] categories. After controlling for age, sex, school type and household SES, consumption of the ‘starchy root & vegetable diet’ was no longer associated with child weight status. However, the starchy roots &vegetable-based diet was significantly associated with public school attendance and older age groups after controlling for household SES, age and weight status [F(5, 479) = 5.868, p < 0.001].

The third DP was negatively associated with private school attendance at the bivariate stage [F (2, 482) =3.059, p < 0.05]. Both the third and fourth identified dietary patterns were not significantly associated with any of the child background characteristics examined in this study in multivariate analyses.

A multiple logistic regression examining the relationship between child weight status and the four identified dietary patterns controlling for age, sex, school type and household SES showed a positive association between energy dense pattern and child overweight (AOR = 1.34, 95%CI = 1.04–1.70). Male sex was significantly associated with reduced odds of being overweight (AOR = 0.41, 95%CI = 0.24–0.69) after controlling for age, school type, household SES and dietary pattern. Childhood overweight was not significantly associated with any of the other three dietary patterns.

Major dietary patterns emerging from consumption data

The first major dietary pattern (DP), “energy dense” identified in this study was characterized by the consumption of sugar sweetened beverages, candies, ice creams, fried foods, processed meats, spreads, toppings, and snacks like cakes, pies, doughnuts and other savory foods. This diet pattern typically referred to as the “westernized diet” has been identified in other non-western societies including Ghana, probably due to industrialization and global food market expansion [3, 45, 46]. This DP has also been seen in almost all studies that sought to examine dietary patterns among children and adolescents elsewhere [47–49]. Food items associated with the energy dense pattern have been reported elsewhere, although sometimes split into different but related components. For instance, Pala et al. [47] identified two energy dense patterns labeled “snacking” and “sweet and fat” which together explained most (16%) of the variability in the diets of European children. Also, Correa et al. [50] described a similar pattern as “Brazilian industrialized”. Findings of this study agrees with the body of evidence linking energy-dense DP consumption and overweight/obesity among children [48, 51, 52]. However, Pala et al. [47] found no association between energy dense DPs and overweight status among European children in a longitudinal study. Energy dense DP, was significantly associated with higher SES and its’ proxy, private schools; the relationship remained significant after controlling for all background characteristics. The association between higher SES groups and ‘unhealthy’ dietary behaviours has been typically reported in Ghana and other developing countries, but contrary to developed country situations. For instance, Craig et al. [48] reported energy dense DP among lower SES groups while healthier DPs were associated with children belonging to higher SES groups. This disparity between developing and developed countries regarding the link between SES and health outcomes is well documented as a function of countries’ overall GDP [53] and will eventually shift towards lower SES groups over time. Nevertheless, the negative effects of energy dense diets on child weight status in Ghana, irrespective of SES, suggests need for interventions to control consumption of energy dense DPs.

The second (starchy roots and vegetable) and third (cereal-grain and poultry) DPs identified in this study both include the consumption of staples typically available in Ghana. However, there are differences in consumption due to the changing diets of Ghanaians as documented by Goody and Goody [9]. Foods in the third DP are likely more ‘prestigious’ and less labour intensive compared to those in the third DP. Additionally, foods in the grain-cereal and poultry DP are increasingly made available in school canteens and sold by food vendors as they are generally more appealing to children. Meanwhile, foods in the starchy roots and vegetables’ DP are generally more affordable and can be accessed easily from vendors outside of the school. It is therefore not surprising that the consumption of the second DP was associated with lower SES groups, older age of child, and public-school attendance. Considering that only 5% of pupils in a previous study brought food daily from home [54], sourcing of food from the school and its’ environs remains inevitable. Also, the likelihood that parents will pack traditional meals (like kenkey, banku, kokonte, gari, eba, rice ball,³fufu), which are served with soups and vegetables is low due to the inconvenience (possible spills and requires skill to eat without stains) and low prestige associated with traditional foods. The soups and stews served with these typically Ghanaian foods are also highly perishable and there are food safety concerns about packing it for children. Consequently, younger aged children will consume more foods in the third DP, grains and chicken compared to the older children who will may patronize local foods they have developed preference for at school. While this study did not assess the usual sources of staple foods, anecdotal evidence suggests that these typically Ghanaian foods are more likely to be consumed at home with family during supper on school days or anytime on weekends. Thus, children living in homes that do not cook meals frequently, will generally record lower consumption of DP 2 and a higher consumption of DP 3.

Although studies among European children found a protective effect of DP based on whole grains, legumes and vegetables [47] on weight, this study did not find significant associations between the third DP and SES, age, school type and weight status of children. This finding must be interpreted with caution since the relationship is likely influenced by nature of data collected. Our study questionnaire did not discriminate between whole and refined cereal/grain consumed by the children. Again, the poultry consumed in this DP with cereals is most likely fried and or stewed, considering how grains-based meals are typically served (both at home and school) or sold in Ghana.

The last DP, “fish and sea foods” explained 7.4% of the variation in the diets of the children surveyed and was not associated with any child characteristic. While it is not normal for fish and seafoods alone to be consumed in any diet, this sub-food group recorded the highest communality (.882) compared to the remaining 15 items used in PCA, suggestive of relevance. Craig and colleagues [48] identified “fish and sauce” DP among children 5–11 years old in Scotland. The consumption of fish as part of a healthy diet is beneficial [55]. Fish is widely consumed in Ghana [8, 56] and most often served whole (with head and bones) irrespective of cooking method. It is possible that the skill required to eliminate bones as well as other factors may influence both parent and children’s’ from choice of fish compared to ‘easier to consume’ options like eggs or poultry. An assessment of fish consumption by school-age children may provide evidence required to understand its inclusion in diets and how it relates with weight status.

To our knowledge, this is the first study to identify dietary patterns in the diets of school-age children in Ghana. We agree that this study is cross-sectional and thus, causality cannot be assumed for the observed association between energy dense dietary pattern consumption and child overweight/obese status. Nevertheless, the identification of dietary patterns similar to those found in other studies is a contribution to knowledge regarding the public health goal of identifying optimal dietary patterns across settings.

A limitation of this study is the inability of data collected on cereal grains to distinguish between refined and unrefined grains consumed. The researchers felt that it would be nearly impossible for children to provide this detail for all grains eaten in and out of the home. This anticipated difficulty is also due to the possibility of these grains to be masked when cooked in a composite meal like “jollof rice” or ‘⁴Wakye’, which are popularly consumed forms among children. Also, grains available in Ghana are mostly whole grain with the exception of rice and maize which have polished versions. Nevertheless, we believe that this had no significant influence on the major dietary patterns identified among the children. Secondly, our classification of households based on possession of 8 assets may have oversimplified or poorly ranked households. Nevertheless, the use of household assets in ranking households in this setting is reliable [42]. Lastly, we acknowledge that no single dietary assessment method, including the FFQ, is able to perfectly assess dietary exposure. Nevertheless, the FFQ when specifically designed for target groups, remains the tool of choice in measuring usual intakes of populations [57].