Menu
Menu
Menu
Menu
Background: In Ethiopia, child undernutrition remains to be a major public health challenge and a contributing factor for child mortality and morbidity. To reduce the problem, it is apparent to identify determinants of child undernutrition in specific contexts to deliver appropriately, targeted, effective and sustainable interventions. Methods: An agroecosystem linked cross-sectional survey was conducted in 3108 children aged 6-59 months. Multistage cluster sampling technique was used to select study participants. Data were collected on socio-demographic characteristics, child anthropometry and on potential immediate, underlying and basic individual and community level determinants of child undernutrition using the UNICEF conceptual framework. Analysis was done using STATA 13 after checking for basic assumptions of linear regression. Important variables were selected and individual and community level determinants of child height-for-age Z score were identified. P values less than 0.05 were considered the statistical level of significance. Results: In the intercept only model and full models, 3.8% (p < 0.001) and 1.4% (p < 0.001) of the variability were due to cluster level variability. From individual level factors, child age in months, child sex, number of under five children, immunization status, breast feeding initiation time, mother nutritional status, diarrheal morbidity, household level water treatment and household dietary diversity were significant determinants of child height for age Z score. Also from community level determinants, agroecosystem type, liquid waste disposal practice and latrine utilization were significantly associated with child height-for-age Z score. Conclusion: In this study, a statistical significant heterogeneity of child height-for-age Z score was observed among clusters even after controlling for potential confounders. Both individual and community level factors, including the agroecosystem characteristics had a significant role in determining child height-for-age Z score in the study area. In addition to the existing efforts at the individual levels to improve child nutritional status, agroecosystem and community WASH related interventions should get more attention to improve child nutritional status in the study area.
This study was conducted in East Gojjam Zone, Amhara Regional State, Ethiopia. The area consists of different climatic zones from Choke Mountain (Blue Nile highlands) to Blue Nile depressions [19]. According to the 2015 Amhara Regional Bureau of Finance and Economics Development Report, a total of 381,309 under five children were reported in East Gojjam Zone [21]. East Gojjam Zone has a total of four town administrations and 16 rural districts. The area includes the Choke Mountain watersheds found in the Blue Nile Highlands of Ethiopia, which extends from tropical highland of over 4000 m elevation to the hot and dry Blue Nile Gorge, below 1000 m from sea level [19]. Based on different parameters such as farming system, temperature, rainfall, soil type, climate change vulnerability and climate adaptation potential, the area is divided into six agroecosystem with its respective characteristics [19]. The lowlands of Abay valley (Agroecosystem one) is characterized by low land areas with unfavorable agro ecological conditions with extensive land degradation [19]. The midland plains with black soil (agroecosystem two) is characterized with a considerable high agricultural productivity potential [19]. The midland plains with brown soil (agroecosystem three) is suitable for its agricultural productivity, since it has a good potential to use mechanized agriculture and irrigation schemes [19]. The midland sloping lands with red soils (agroecosystem four) is characterized by low natural fertility with high level of soil acidity, slope terrain and higher rate of water runoff with soil erosion making the crop production potential very low [19]. Hilly and mountainous highlands (agroecosystem five) is found in the hilly and mountainous highlands with constrained crop productivity due to erosion and deforestation [19]. The last agroecosystem is in the top of the mountain with relatively low agricultural productivity, due to its low temperature and since it is a conserved area [19]. This study was conducted from January to April 2015. A multistage stratified cross-sectional survey by agroecosystem was conducted to identify individual and community level determinant factors of child height for age Z score (stunting) among children aged 6–59 months. Sample size was calculated using double population formula [22, 23] for means using the mean of height for age Z score across different agro ecological zones. The EDHS 2011 data which contains altitude information above sea level were used to categorize the study clusters in to agro ecological zones of highland, midland and low land. The mean height for age Z score was calculated with standard deviation for each agro ecology zone assuming that agro ecology is one of the main determinants of child undernutrition in the study area. The height for age Z score was calculated for important variables to check the adequacy of the calculated sample size and found that sample size determined using agro ecology was found to be the maximum one. The computation was made with the following inputs in to Open Epi, Version 3 [24]: height for age Z score of −1.21 for low land areas and −1.40 for highland areas, 95% confidence level (Z α/2 = 1.96), design effect of 1.5, 80% power of the study and one to one ratio between at higher risk population (from highland areas) and lower risk population (from lowland areas). Each of the group’s population height for age Z score standard deviations were calculated from the data set and found to be 1.70 and 1.6 for low (lowland areas) and high (high land areas) risk groups, respectively. Then, the calculated sample size was found to be 2379 under five children (1185 for each group). However, to increase the power and apply multilevel model, we used the larger sample size of 3225, which was calculated to address another component of the study. A multistage cluster sampling procedure was used to select those study participants from 38 clusters. From each agroecosystem, sample districts from the East Gojjam Zone, kebeles from each district and clusters from each Kebele were selected using multistage cluster sampling technique. In the initial phase, five districts which represent one agroecosystem each were selected. Lowlands of Abay valley area (agroecosystem one) was represented by sample taken from Dejene District. The midland plain with black soil area (agroecosystem two) was represented by sample taken from Awabel District. The midland plains, with red soil area (agroecosystem three) were represented by sample from Debre Eliyas District and midland plains, with brown soil area (agroecosystem four) was represented by sample taken from Gozamin District. Finally, the hilly and mountainous area (agroecosystem five) was represented by sample taken from Sinan District. In the second phase, from each agroecosystem, potential rural kebeles were selected using simple random sampling technique. Since, a district may consist of more than one agroecosystem; care was taken to avoid misclassification bias during kebele selection. List of all kebeles that can represent agroecosystems were listed within a district and then using lottery method, 6–9 were selected randomly. In the third step, from each selected kebele, one got (lowest administrative level) was selected using simple random sampling. The total number of clusters, included were 38 from the five agroecosystems and all eligible under-five children were considered for the survey. Data were collected by trained data collectors on socio demographic characteristics, using interviewer administered questionnaire and child anthropometry using height and weight measuring scales. In addition, data were collected on the potential immediate, underlying and basic determinants of child undernutrition, using interviewer administered questionnaire. From the immediate determinants, data were collected on childhood illness 2 weeks prior to the survey and on dietary intake data, including breast feeding practice, complementary feeding practice, child dietary diversity, and maternal nutritional status. The underlying determinants of child undernutrition, including household food insecurity, environmental health conditions and maternal and child care practices data were collected. Latrine utilization was assessed using four main indicators as a check list: presence of foot path to the latrine, not using the latrine as store, presence of fresh feces around the hole of the latrine and absence of feces around the compound. Also household level waste management practice was assessed using a check list on the presence of the pit and current utilization. Household food insecurity status was measured using Household Food Insecurity Access Scale (HFIAS) of Food and Nutrition Technical Assistance (FANTA) questionnaire developed in 2006 with a recall period of 30 days. Observational check lists were used to collect data on Environmental health conditions. Also maternal health service utilization, wealth status, and child care practice data were collected using interviewer administered questionnaire. Agroecosystem type data were accessed using previously published article on the area [25]. Weight of the child was measured using a digital scale designed and manufactured under the guidance of UNICEF with 100 g precision to measure body weight. Length/height measurements were taken using a locally produced UNICEF measuring board with a precision of 0.1 cm. Children below 24 months of age were measured in a recumbent position, while standing height was measured for those who were 24 months and older. Weight and height of the child were taken twice and variations between the two measurements of 100 g were accepted as normal for weight and 0.1 cm in height/length of the children. However, repeated measurements were carried out upon significantly larger variations which were above 100 g in weight and 0.1 cm in height/length. The weight scale was calibrated before measuring the child weight (Table 1). Definition and measurements of variables used in the dissertation, East Gojjam Zone, Amhara Regional State, Ethiopia, 2015 a MUAC Mid Upper Arm Circumference Intensive training with pretesting was given for 5 days to data collectors and supervisors to ensure all research team members can administer the questionnaires properly, read and record measurements accurately. The pretesting was done in none selected with have similar characteristics to the study community. Then, all necessary corrections were made based on the pretest, before the actual data collection. Repeated measurements were taken during weight and height measurements to check the consistency of measurements by two measurers independently. At the end of every data collection day, each questionnaire was examined for its completeness and consistency and pertinent feedbacks were given to the data collectors and supervisors to correct it in the next data collection day. Data were cleaned using frequencies for logical and consistency errors before further analysis. Data were cleaned, coded and entered into EPI Info version 3.5 [26] and exported to STATA (Stata Corp LP, College Station TX) [27]. Child nutritional status was calculated using WHO Anthro version 3.2.2 [28]. Since the outcome is a continuous variable, normal distribution of the dependent variable assumption was checked using graphical and formal statistical tests. Multicollinearity was checked using the variance inflation factor (VIF) and variables with VIF less than 10 were considered for the analysis. Descriptive statistics was used to present frequencies, with percentages in tables and using texts. Multilevel linear mixed effects regression analysis was used to identify individual and community level factors after selecting important variables using simple linear regression analysis using P < 0.05 as an entry criteria to the model. As justifications for using a multilevel linear modeling is related to the determinants of child stunting are found at different level and factors some have neighborhood effects which impose negative/positive externality in the surrounding community [29, 30]. As a result analyzing variables from different levels at one single common level using the classical linear regression model leads to bias (loss of power or Type I error). This approach also suffers from a problem of analysis at the inappropriate level (atomistic or ecological fallacy). Multilevel models allow us to consider the individual level and the group level in the same analysis, rather than having to choose one or the other. Secondly, due to the multistage cluster sampling procedure in the current study, individual children were nested within clusters/villages/; hence, the probability of a child being stunted is likely to be correlated to the cluster level factors. As a result, the assumption of independence among individuals within the same cluster and the assumption of equal variance across clusters are violated in the case of nested data. Hence, the multilevel analysis is the appropriate method for such kind of studies [29, 30]. Assuming the continuous responses variable Yij depend on individual level explanatory variable Xij and community level explanatory variable Zj, if deviation from the average intercept and slope due to cluster (community level factors) effect are represented by u0j and u1j, the two models are given in the following way. The intercept-only model = Yij = γ00 + u0j and the full model = Yij = γ00 + γ01Zj + γ10Xij + u0j + u1jXij. The intercept γ00 and slopes γ01 and γ10 are fixed effects, whereas uoj and u1j are random effects of level-2. The intercept-only model allows us to evaluate the extent of the cluster variation influencing child height for age Z score [31]. The intra-class correlation coefficient (Rho) refers to the ratio of the between-cluster variance to the total variance and it tells us the proportion of the total variance in the outcome variable that is accounted at cluster level. Mathematically, Rho (ICC), is given as ρ=δo2δo2+δ2 Where σ2 refers to individual level variance and σ0 2 refers to community level variance. The regression coefficients were interpreted and p values less than 0.05 were considered as level of significance [31].
N/A
