Spatial Distribution and factors associated with childhood anemia in Ethiopia

The magnitude of childhood anemia has increased from 44% in 2011 to 56% in 2016. Thus, even if the Ethiopian government tried remarkable solutions, anemia among under-ve children still continues as a serious health issue. So, exploring spatial distribution and identifying factors associated with childhood anemia helps to design appropriate strategies for control and prevention. Methods For this study data from the recent 2016 Ethiopian Demographic Health Survey were employed. The sample size was 8602 children aged 6–59 months. They were selected by stratied two-stage cluster sampling techniques. Sat Scan version 9.4 was also used to identify childhood anemia by geographic clusters and ArcGIS version10.1 was used to show anemia cases through Regions of Ethiopia. Thus to declare factors that are statistically related with anemia among under-ve children a Mixed effect logistic regression model was utilized. This study showed there is spatial clustering of childhood anemia throughout Ethiopia (Moran’s I: 0.65, p<0.001). Statistically signicant clusters were detected in Somali, Afar, Harari and southern part of Oromia regions (P<0.001). Age of child, wealth index, mother’s current working status, maternal anemic status, number of living children in the family, history of fever, and stunting were signicant factors associated with anemia among under-ve children. providing have public


Introduction
Anemia is a disease which is characterized by decreased quantity of red blood cells or hemoglobin level that results in insu cient oxygen-carrying capacity of blood to meet cellular metabolic demand of the body (1). Nutritional de ciencies were the utmost cause of anemia. Among nutritional de ciencies, Iron de ciency was the major contributor of anemia globally. However, folic acid, vitamin B12 and vitamin A were also cause nutritional de ciency anemia. Other causes of anemia can be in ammations, infections due to parasites, and disorders which distress survival or production of red blood cells and hemoglobin synthesis (1,2).
Anemia is a global community health threat upsetting both high and low income nations (3). The most important effects of anemia were poor socio-economic development and increased mortality and morbidity (3). It occurs in all population groups of human being. But young children and pregnant women had higher risk of developing anemia. In children, besides increasing child mortality, severe anemia resulted in impaired cognitive and physical development (2,3).
Globally, majority of preschool children and 1.6 billion people in Africa and south East Asia were expected to be anemic (3). The national prevalence of anemia among preschool children was predictable above 40% in developing nations including Ethiopia (4)(5)(6).
According to World Health Organization (WHO); anemia considered as a major public threat when prevalence was greater than 40%, as a moderate public threat when prevalence was from 20-40%, and as a mild threat when prevalence was from 5-20%. In our country, the magnitude of childhood anemia increased from 44% in 2011 to 56% in 2016 (4,5). Even if the Ethiopian government and different stakeholders applied tremendous efforts to control the disease, childhood anemia still continues as threat (7). Therefore, understanding the spatial variation and determinants of anemia among this group of children is important to design effective interventions and to manage program resources fairly.
However, previously there were studies conducted to identify determinant factors of anemia in children.
But the majority of studies were based on data collected in speci c localized areas of the country where ndings were not representative (7)(8)(9)(10)(11). Although a study done by using 2011 Ethiopian demographic and health survey (EDHS) data which was representative of the entire country, its analysis for determinants of childhood anemia does not account for clustering and hierarchical nature of the EDHS data (12). Furthermore, spatial distribution of childhood anemia was not studied previously. Thus, by considering the above limitations, this study was designed to explore spatial clustering of childhood anemia in Ethiopia.

Study design and period
This study was conducted from January 18 to June 27, 2016 and a community based cross sectional study design was used.

Study setting
The study was done in Ethiopia which is located in the eastern horn of Africa with (3 o -14 o N and 33 o − 48°E). The country is divided into nine regional states and two city administrations. There are 68 zones, 817 districts and 16,253 kebeles (lowest local administrative units of the country) (5). It has an space of 1.1 million km 2 (square kilometer) and has abundant topographical variety, extending from 4,550 meter above sea level depressed to 110 m below sea level (Fig. 1).

Population and sample
All children aged 6 to 59 months in Ethiopia in the selected households from January 18 to June 27 in 2016 were study population for this study. A total of 8602 children were included in the study.
As sampling strategy strati ed two-stage cluster sampling technique was utilized. The sampling units were Enumeration areas (EA) and households for the rst stage and second stage respectively. There were 645 clusters in which 202 from urban and 443 from rural were carefully sampled with a probability proportional allocation to cluster size with independent sampling in each sampling stratum. Among these, 18,008 households and 16,583 eligible women were incorporated. The detailed sampling procedure was found in the full EDHS report (5).

Study variables and data collection procedure
In this study the response variable was anemia status among children age 6-59 months. The predictor variables were grouped as: (1).socio-demographic factors: sex of child, age of child, residence, educational status of mother, maternal age, husband's educational status, religion of mother, wealth index, working status of mother, and number of children in the house hold.
(3). Clinical factors: maternal anemia status, diarrhea, fever and cough two weeks prior to data collection, and (4). Medical related factors: taking of iron pills or sprinkles or syrup, and drugs for intestinal parasites, and vitamin A in the last six months.
Adjusted concentration of blood hemoglobin less than 11 mg/dl was considered as an anemic (1). Hence anemic status was determined based on blood hemoglobin concentration relative to altitude.
An approval letter for the use of the data was available from the Measure DHS.
The dependent and independent variables were obtained from the EDHS 2016 child data set. Both latitude and longitude coordinates was taken from selected clusters.
As a data collection tool a structured and pre-tested questionnaire was utilized. For 2016 EDHS tablet computers were used to record responses during interview. They were armed with Bluetooth device which enable remote electronic transfer of assignment sheets from supervisors to interviewers and transfer of completed copies back from interviewers to supervisors (5).

Data management and analysis
Interviewers were trained before data collection, and Interviews were performed using local languages (5). Descriptive and summary statistics were done using STATA version 14 after extracting and editing of data from EDHS 2016 child data set. Then Cross tabulations and summary statistics were used to de ne the study population.
Children found in one cluster are more similar as compared to those children found in other clusters (regions) of Ethiopia in case of EDHS data. This disrupts the assumption of independence of observations and equality of variance across clusters. This leads the use of advanced models to consider the between cluster variability. Thus, a mixed effect logistic regression model was considered. Hence the outcome variable was dichotomous (non-anemic child = no, anemic child = yes), logistic regression and GLMM (generalized linear mixed model) was tted. Then the GLMM was selected based on result of AIC (Akaikie Information Criteria) and BIC (Bayesian information criteria), and ICC (intra class correlation) values. Thus variables having p-value up to 0.2 in the bi-variable analysis were taken to t the model in the multi variable analysis. Finally, p-value less than 0.05 in the multivariable model of mixed-effects logistic regression was used to select variables which had statistically signi cant association with anemia. Goodness of t was checked by using deviance and ICC. Spatial scan statistical analysis Spatial scan statistic used a scanning window which travels through the study area. It was explored to detect the presence of statistically signi cant spatial clusters of childhood anemia by using Kuldorff's SaTScan version 9.4 software (23). Children with anemia were taken as cases and those without the disease as controls to t the Bernoulli model was tted i.e. anemic childhood as cases and non-anemic childhood as controls. To detect both small and large clusters, and overlooked clusters which contained more than the maximum limit with elliptical shaped window the default maximum spatial cluster size of < 50% of the population was used as an upper limit. To decide if the number of observed anemia cases within the cluster was signi cantly greater than the expected or not a Likelihood ratio test statistic was employed. Likelihood ratio tests based on the 999 Monte Carlo replications and using p-values were indicated Primary and secondary clusters (23).

Ethical consideration
Ethical clearance was gained from the ethical review board of Institute of Public Health, College of Medicine and Health Sciences, University of Gondar. Written consent was gotten from Measure DHS International Program which approved the data-sets. All the data used in this study are publicly available, aggregated secondary data with not having any personal identifying information that can be linked to particular individuals, communities, or study participants. Con dentiality of data was maintained anonymously.

Results
Socio -demographic characteristics of respondents A total of 8602 under-ve children were incorporated in the 2016 EDHS survey. Among these children, 7815 granted the consent statement for hemoglobin. Out of 7815 children who granted the consent form, hemoglobin level was determined for 7794 children. About half, 4010(51.45%) children were males and 6467 (82.97%) rural residency. About two third, 5,084 (65.23%) of mothers had no formal education. Considering age of children, the mean age was found to be 2.64 years (S.D ± 0.18 months). Regarding of wealth index, 2854(36.62%) of respondents belongs to the poorest families (Table 1). The highest and the lowest prevalence of anemia among under-ve children were identi ed in Somali (81.94%) and Amara (42.40%) regional states respectively.

Spatial distribution of childhood anemia
In this study the distribution of childhood anemia across the country among children age 6-59 months were clustered with Global Moran's I 0.65 (p < 0.001). The clustered patterns (on right sides) depicted that high number of anemia cases in the study setting. A z-score of 9.7 indicated that there was less than 1% likelihood for this clustered pattern due to a random chance. An increased level of signi cance was demonstrated by the bright red and blue colors in both side ends (Fig. 2).
Each spot (point data) on the map denotes one census enumeration area which contains a number of anemia cases. In the Map more cases depict anemia risk regions. Consequently, the red color shows areas had high number of anemia cases (Fig. 3).

Spatial scan statistical analysis
In the study about 114 signi cant clusters (80 primary and 34 secondary) were recognized. The spatial window for primary clusters was positioned in south-eastern part of oromia, Harari, Dire-Dawa and  Fig. 4). But the relatively small secondary clusters window was found in afar regional state. It was xed at 11.430282 N, 40.918452 E with 145.46 km radius, and a RR (Relative Risk) of 1.50 and LLR of 6.84, with p < 0.001. It revealed that children found inside the spatial window were 1.5 times more risky for anemia compared with children found outside the window ( Table 2, Fig. 4).
The most statistically signi cant spatial windows contained primary most likely (primary) clusters of childhood anemia was represented by bright red colors. Childhoods found inside the spatial window (cluster) have a greater risk for anemia compared with those childhoods found outside the spatial window.

Determinants of childhood anemia
Model comparison The calculated AIC and BIC for Mixed-effects Logistic regression was smallest as compared with Logistic regression (Table 3). In addition to this, the ICC value was 0.13 which implies to use mixed effect logistic regression model over the traditional logistic regression model. Age of child, number of under-ve children in the house, age of child, stunting and wasting status, residence, maternal and husbands educational level, educational level, history of diarrhea, fever and cough in the last two weeks, family wealth index, taking of iron pills or sprinkles or syrup, maternal working status, maternal anemic status, size of child at birth, age of mothers, drugs for intestinal parasites in the last six months, and vitamin A supplement in the same period were statistically signi cant in the bi-variable analysis at p-value < 0.05.
But age of child, religion, wealth index, mother's current working status, maternal anemic status, number of under ve children in the house hold, fever in the last two weeks, and stunting remained signi cant predictors of childhood anemia in the multivariable mixed effect logistic regression model. The likelihood of developing anemia among children who had working mothers were decreased by 13% (AOR = 0.87, 95%CI = 0.76-0.99) as compared with children whose mothers were not working currently.
Regarding maternal anemic status, children who had anemic mothers were 53% (AOR = 1.53, 95%CI = 1.35-1.73) more likely to develop anemia as compared with those children who had non anemic mother.  The eastern and south-eastern parts of the country mainly Somali, Afar, Harari and Dire Dawa were the hotspot regions of childhood anemia. The most probable explanation for geographic variation in the risk of anemia might be related with the dynamics of the soil content of minerals since the identi ed high risk regions were categorized to eastern and south-eastern part of the country. Meaning high risk regions share similar environmental condition as evidenced by boundary formation to each other. In addition, epidemiological factors such as fever and stunting which were identi ed as determinants of anemia in different studies including our study were more common in high risk regions. Similarly the nding from Nigeria indicated that Northern parts of the country were at a greater threat of anemia (13). Furthermore, according to a study done in sub-Saharan Africa which reported the distribution of anemia was exacerbated by factors acquired from the environments (6).These all implied that the dynamics of mineral content of the soil was the most probable explanation for observed geographical variation in the risk of anemia.
This study showed that children age between 12-59 months was less affected by anemia. This nding was consistent with majority of studies conducted a cross the world (8,9,11,12,(14)(15)(16)(17)(18)(19)(20)(21)(22). This might be because of a wide gap between high iron demand for fast growth (3) and low iron supply because of inappropriate initiation of complimentary feeding (11,24) and highest depletion of prenatal iron store starting at six months of age (25). However, this nding is different from the study conducted in southern Ethiopia that demonstrated statistically non-signi cant effect for anemia (7). This can be explained by the difference in sample size. The previous study's sample size (399 children) was very small as compared with current study.
This study also showed wealth index was signi cant predictors of childhood anemia. Consequently, children from poor, middle, richer and richest family were lowered by 25.8%, 40.4%, 35.2% and 43.3% to develop anemia respectively as compared with those children from the poorest family. The nding was consistent with studies conducted in brazil (20,21), Nigeria (13) and Ethiopia (8,11). The possible explanation for this could be, poorest families had no competence to pay for diversi ed foods and additionally suffering from health getting medical services which causes leads to anemia (3,26).
Children who had working mothers were lower compared with children who had mothers not working currently. It is supported by a study done in Ethiopia (12). The possible justi cation could be related with increased empowerment of working mothers to childcare and other health related actions (27,28).
Maternal anemic status was also signi cant predictors of childhood anemia in the study. This nding is concordance with reports from Cuba (18), Burma (20) and Ethiopia (12). This association might be explained by in uence of poor maternal iron reserve during pregnancy and breast feeding on iron store of their child (29,30). Nevertheless, the association might partly re ect the effect of confounding because of unmeasured observation on shared socioeconomic, genetic and biological environment.
Those children from households who had three or more under-ve children were higher in anemia. It was concordant with majority of ndings done from Brazil (15), Lao People's Democratic Republic (17) and Ethiopia (12). The possible reason for the observed association is that higher number of children imposes high demand on household not only for food but also for clothing and health care services. Again it exacerbates quality to care for children, and thereby worsen anemia. However, this nding was different from studies conducting in Nigeria (13) and southern Ethiopia (7) which showed that; number of children in the household had no impact on anemia (7). The discrepancy might be related with inclusion of all children in the household in the previous studies. This mixed independent family members with dependent family members which pools the real association towards the null. In addition the study conducted in southern Ethiopia had very small sample size (399 children) as compared with current study.
Children who had fever were more affected by anemia in this study when compared with the counterparts. The result was concordant with reports from Burma (20), Nigeria(13) and southern Ethiopia (7). This might be related with the infectious cause of childhood fever mainly malaria, septicemia, tuberculosis and leshimaniasis which causes anemia by infecting and destructing red bloods cells or other mechanisms (31).
Stunting was also found to be signi cant predictors of childhood anemia in the study. The result was similar with majority of previous ndings conducted in Burma(20), Kenya (14) and Ethiopia (7)(8)(9)(10)(11)(12). This association could be explained by both pathophysiological mechanism and confounding effect of inadequate dietary intake (both share the common cause). Pathophysiological explanation implied reverse causation between stunting and anemia. In malnourished children, gastrointestinal epithelium disturbance leads to development of anemia by impairing absorption of nutrients (31). To the reverse, anemia during period of rapid growth leads to irreversible growth retardation (3,29).
As strength ndings could be generalized to all under-ve children found in the country. In addition the study used advanced spatial techniques which could demonstrate consistent and statistically signi cant high burden clusters for childhood anemia. But as a limitation the study didn't show variations of childhood anemia seasonally. Furthermore, for some independent variables like diarrhea, cough, and fever within two weeks there might be misclassi cation of exposure status because of unable to remember the event (recall bias).
The result would provide valued policy suggestions for targeted interventions and designing related programs. Overall, the study has paramount relevance for the policy makers and stakeholders for appropriately intervening childhood anemia.

Conclusion
This study showed that the distribution of childhood anemia varied in Ethiopia. More risk areas of childhood anemia were detected in the eastern and south-eastern parts of the country, while low risk areas of childhood anemia were noted in the central, North Western, western and, south-western parts of Ethiopia. Being in the early age category, having poorest family, having mother who is not currently working, having anemic mother, being in the house-hold that had three or more under ve children, having fever in the last two weeks, and having moderate or severe stunting were factors increased the odds of developing childhood anemia.  Most likely (primary) and secondary clusters of under-ve pneumonia across regions in Ethiopia, 2016.