Associations of childhood, maternal and household dietary patterns with childhood stunting in Ethiopia: Proposing an alternative and plausible dietary analysis method to dietary diversity scores

Background: Identifying dietary patterns that consider the overall eating habits, rather than focusing on individual foods or simple counts of consumed foods, better helps to understand the combined effects of dietary components. Therefore, this study aimed to use dietary patterns, as an alternative method to dietary diversity scores (DDSs), and investigate their associations with childhood stunting in Ethiopia. Methods: Mothers and their children aged under 5 years (n = 3788) were recruited using a two-stage random cluster sampling technique in two regions of Ethiopia. Socio-demographic, dietary and anthropometric data were collected. Dietary intake was assessed using standardized dietary diversity tools. Household, maternal and child DDSs were calculated and dietary patterns were identified by tetrachoric (factor) analysis. Multilevel linear and Poisson regression analyses were applied to assess the association of DDSs and dietary patterns with height-for-age z score (HAZ) and stunting, respectively. Results: The overall prevalence of stunting among children under-five was 38.5% (n = 1459). We identified three dietary patterns each, for households (“fish, meat and miscellaneous”, “egg, meat, poultry and legume” and “dairy, vegetable and fruit”), mothers (“plant-based”, “egg, meat, poultry and legume” and “dairy, vegetable and fruit” and children (“grain based”, “egg, meat, poultry and legume” and “dairy, vegetable and fruit”). Children in the third tertile of the household “dairy, vegetable and fruit” pattern had a 0.16 (β = 0.16; 95% CI: 0.02, 0.30) increase in HAZ compared to those in the first tertile. A 0.22 (β = 0.22; 95% CI: 0.06, 0.39) and 0.19 (β = 0.19; 0.04, 0.33) increase in HAZ was found for those in the third tertiles of “dairy, vegetable and fruit” patterns of children 24-59 months and 6-59 months, respectively. Those children in the second (β = -0.17; 95% CI: -0.31, -0.04) and third (β = -0.16; 95% CI: -0.30, -0.02) tertiles of maternal “egg, meat, poultry and legume” pattern had a significantly lower HAZ compared to those in the first tertile. No significant associations between the household and child “egg, meat, poultry and legume” dietary patterns with HAZ and stunting were found. Statistically non-significant associations were found between household, maternal and child DDSs, and HAZ and stunting. Conclusion: A higher adherence to a “dairy, vegetable and fruit” dietary pattern is associated with increased HAZ and reduced risk of stunting. Dietary pattern analysis methods, using routinely collected dietary data, can be an alternative approach to DDSs in low resource settings, to measure dietary quality and in determining associations of overall dietary intake with stunting. A cross-sectional study was conducted in the South Nations, Nationalities and People (SNNP) and Tigray (northern Ethiopia) regions between June and September 2014. The two regions are geographically located at opposite ends of Ethiopia, in the south and north, with differences in agroecology, subsistence farming being the most common occupation in both regions. The SNNP is a larger geographic area and has a greater population size compared to the Tigray region. This study was part of a larger project of the Alive and Thrive’s (A&T) impact evaluation for community-based interventions. The major objectives of the evaluation included assessment of infant and young child feeding (IYCF) practices and stunting prevalence. The baseline and progress evaluation of the project were conducted between June and September 2010 and 2013, respectively. The sample size was calculated based on the 2010 baseline and 2013 progress evaluation surveys’ estimates. It took into account an intracluster correlation of 0.03–0.04. A total of 75 clusters (enumeration areas [EAs]), a one-sided test, a power of 80%, and a significance level of α = 0.05 were included in the calculations. Based on these considerations, a minimum sample size of 2950 was required (Table 1). However, in total, data were collected from 3788 children and their mothers. Sample size calculation, 2014 A two-stage cluster sampling technique was used to select households with children under five. In the first stage (primary sampling unit), EAs were selected from 89 districts (the second smallest administrative units in Ethiopia). The EA is a geographical unit devised by the Central Statistical Authority (CSA) of Ethiopia, which consists of 150–200 households. This is the smallest cluster used in Demographic and Household Surveys and roughly coincides with the kebele (the smallest administrative units) boundaries (Fig. 1). A total of 75 EAs (26 from Tigray and 49 from SNNP), from 56 districts (19 from Tigray and 37 from SNNP) were selected using probability proportional to size (PPS) sampling in relation to the population of the EAs. Sample description In the second stage, children between 0 to 59.9 months (n = 3788) were selected. A complete household listing with the number of children residing in each household, in each selected cluster, was developed in collaboration with the local health and administrative offices. This list included identification of all eligible candidates for the survey (mothers of those children under 60 months of age). From this list, three sampling frames were developed: children aged 0–5.9 months, 6–23.9 months, and 24–59.9 months. From each sampling frame, study subjects were selected using a systematic random sampling (SRS). Households selected to participate in one age category were not included in the other sampling frames, even if there were other eligible children in a household. All data (interview and anthropometric) were collected by trained data collectors who had bachelor degrees or above. To maintain the data quality, demonstrations and pilot testing were conducted during the training period. Trained supervisors oversaw and monitored the data collectors during the field work. The supervisors also checked 5–10% of the anthropometric and interview data to ensure reliability. The length or height of children was measured to the nearest 0.1 cm using the United Nations International Children’s Emergency Fund (UNICEF) recommended wooden board with an upright wooden base and movable headpieces. Children ≥24 months were measured while standing upright while those less than 24 months in the recumbent position. Weight was measured to the nearest 0.1 kg using UNICEF’s scale [22]. Immunization cards or home records of the date of birth, if available, were used to determine the age of the children. In the case of absence of these documents, mother’s recall was taken using the local calendar and then converted to the Gregorian calendar. Adult weighing and height scales were used to measure maternal weight and height, respectively. Mothers were asked to remove shoes and heavy cloths before weight and height were measured. Weight and height were recorded to the nearest 0.1 kg and 1.0 cm, respectively. Dietary data for children [7] and women [8] and household dietary and food insecurity data [9, 23] were assessed using standard tools. Dietary intake was assessed for the preceding day (24 h). For children aged 6–23 months, seven food groups (grains, roots and tubers; legumes and nuts; dairy products (milk, yogurt, cheese); flesh foods (meat, fish, poultry and liver/organ meats); eggs; vitamin-A rich fruits and vegetables; and other fruits and vegetables) were included [7]. For children 24–59 months, an additional two food groups (oils and fats and independent categories of other fruits and vegetables) were included. The maximum dietary diversity for women of reproductive age (MDD-W) assessment includes 10 food groups (grains, white roots and tubers, and plantains; pulses (beans, peas and lentils); nuts and seeds; dairy; meat, poultry and fish; eggs; dark green leafy vegetables; other vitamin A-rich fruits and vegetables; other vegetables; other fruits). Consumption of food by any of the household member from any of 12 food groups in the last 24 h was also assessed and the household DDS was determined. The food groups were cereals; roots and tubers; vegetables; fruits; meat, poultry, offal; eggs; fish and seafood; pulses/legumes/nuts; milk and milk products; oils/ fats; sugar/honey and miscellaneous [23]. Data including socio-demographic (such as maternal age, maternal education, sex of the head of the household and paternal education), economic (household asset), environmental factors (such as water source and latrine type), health service utilization (such as place of delivery for index child), and household characteristics (such as the number of under-five children living in a household) were collected. Different socio-economic indicators were combined using principal component analysis to construct household wealth. The factor scores were divided into quantiles (poorest, poorer, middle, richer and richest) to indicate the relative socio-economic status of the participants. The highest level of education achieved was categorized into no education, primary, and secondary and above. Water source was classified as piped, other improved and unimproved. The type of functional latrine used in the household was categorized into traditional pit latrine, improved latrine and no facility/bush/field. Height-for-age z score (HAZ), an indicator of linear growth, was compared with reference data from the World Health Organization (WHO) Multicentre Growth Reference Study Group, 2006 [24] using the ENA (Emergency Nutrition Assessment) SMART (Standardized Monitoring and Assessment of Relief and Transitions) 2011 software. Children whose HAZ is <−2 SD from the median of the WHO reference population were considered stunted (short for their age). In our analysis, both HAZ (continuous) and stunting (categorical; stunted = 1/not stunted = 0) were used as outcome variables. Maternal body mass index (BMI) was calculated based on the measured weight (kg) and height (meters) (weight [kg]/(height[meters])2). For children aged 6 to 23, and 24 to 59 months, the minimum acceptable DDS was defined as consuming food from four or more of the standardized set of seven (6 to 23 months) or nine (24 to 59 months) food groups on the preceding day of the survey [7, 25]. For women (mothers/caregivers of the index children), a threshold of at least 5 food groups of the 10 was considered acceptable [8]. The scores of household DDS were continuous, ranging from 0 to 12, based on whether any of the members of the household consumed any of the 12 food groups. Minimum household DDS was not determined because a dichotomous indicator has not been developed [8, 23]. However, we assumed that consumption of food groups above the median number as adequate. Nine Household Food Insecurity Access Scale (HFIAS) generic questions were used with a dichotomized response (0 = no/1 = yes) to assess food insecurity [9]. Each of the questions were asked with a recall period of four weeks (30 days). If a respondent answers “yes” to any of the above nine questions, frequency-of-occurrence questions were asked to determine whether the condition happened rarely (1 = once or twice), sometimes (2 = three to ten times) or often (3 = more than ten times). The insecurity status was categorized into four groups (secured, mild, moderate, and severe) using the Food and Nutrition Technical Assistance (FANTA) algorithm [9]. Dietary patterns were identified by polychoric (tetrachoric) analysis—a family of factor analysis which uses a tetrachoric correlation matrix to construct latent variables from dichotomized (binary) observed data [26]. For each of the dietary patterns, factor scores were assigned for all study participants. Factor scores show the relative position of the study participants in each of the identified patterns, thus reflecting adherence to the patterns. Pattern-specific factor scores are calculated as the sum of the products of the factor loading coefficients and standardized daily consumption of food and nutrient groups related to the pattern. The factor scores were orthogonally (varimax) rotated to create less correlation among the patterns and to facilitate their interpretability. Participants were then assigned into tertiles (first [lowest adherence]; second; and third [highest adherence] tertiles) based on their factor scores. Eigenvalues (>1), scree plots, and interpretability of the factors were used to determine the number of dietary and nutrient patterns. Factor loadings (the correlation between each pattern and the food and nutrient groups) were calculated. Percentages of variances (the variations that were explained by the identified dietary and nutrient patterns) were also computed. The chi-square (categorical variables), ANOVA (normally distributed continuous variables) and Kruskal-Wallis (continuous but not normally distributed) tests were used to compare differences of proportions, means and medians, respectively, between groups. Principal component analysis (PCA) was used to compute economic status (in quintiles) of households. To assess the associations of household, maternal and child dietary diversity and patterns with HAZ and childhood stunting, β coefficients and the prevalence ratio (PR) with their corresponding 95% confidence intervals (CIs) were determined using multilevel linear and Poisson regression models, respectively [27]. Since the data were collected using a multi-stage cluster sampling technique, stunting could potentially be correlated in clusters (EAs). We, therefore, used a two-level model with individual factors as level 1 and geographical areas (EAs) at level 2 (random effects). A stepwise backward elimination of covariates in the models was conducted and potential factors were retained at p-value < 0.20. This method was used for both individual and community level factors. Dietary diversity and pattern scores were treated as categorical (model 1) and continuous (model 2) variables. Estimates of associations were adjusted for socio-demographic factors (child age, sex, maternal age and education, number of under-five children in a household), maternal anthropometry (height and BMI), infant and young child feeding practices (exclusive breastfeeding) and household food security at level 1. At level 2, water source was included. Model fit was assessed using Akaike’s (AIC) and Bayesian (BIC) information criteria. We tested interactions between DDSs, dietary patterns, other covariates with HAZ and stunting using multiplicative terms. We conducted sensitivity analysis: 1) by labelling missing values of covariates as “missing” and including in the models; 2) by including and excluding covariates (such as household wealth, paternal education, place of delivery and latrine type). Further, the association between joint classifications of tertiles of dietary patterns and HAZ was explored. Statistical analyses were performed using Stata version 14.1 (Stata Corporation, College Station, TX, USA). A 2-sided t-test value of P < 0.05 was considered statistically significant.