Study design and setting
Secondary data analysis was employed using the 2016 Ethiopian demographic and health survey.The study is conducted in Ethiopia (3o -14o N and 33o - 48°E), located at the horn of Africa (Fig. 1). The country covers 1.1 million Sq. km and has a great geographical diversity, which ranges 4550 m above sea level down to the Afar depression to 110 m below sea level. There are nine regional states and two city administrations subdivided into 68 zones, 817 districts and 16,253 kebeles (lowest local administrative units of the country) in the administrative structure of the country (17).
Data source and measurements
Every five years, the Demographic and Health Survey program of the country (EDHS) has collected data on national representative samples of all age groups and key indicators including anemia among reproductive-age women. The sociodemographic, socioeconomic, child health and maternal related variables were included in the questionnaire.
A stratified two-stage cluster sampling procedure was employed to select study participants. In 2016 survey 645 EAs (202 urban and 443 rural) were selected. From these enumeration areas, 18008 households and 16583 eligible women were selected. The hemoglobin level was measured for those eligible mothers after having consent and it was adjusted for altitude (17). In the current study, 4657 lactating mothers breastfeeding.
The hemoglobin level was measured for those eligible mothers after having consent and it was adjusted for altitude. Therefore, the current study was based on the altitude adjusted hemoglobin level which was already provided in the EDHS data. Lactating mothers were considered to be anemic if their hemoglobin level was <12 g/dL. Hemoglobin level was measured in g/dL, operationalized as a categorical variable by predeﬁned cut-off points for mild, moderate and severe anemia recommended by the WHO for women above the age of 15 years. For this analysis, we recategorized anemia level as anemic and non-anemic from prior classiﬁcations in levels (no, mild, moderate, severe) because of very small numbers of cases in the categories of severe and mild anemia, Therefore, women with hemoglobin level <120 g/L were considered as anemic and coded as “1” whereas those nonanemic were coded as “0” for further analysis.
From the 2016 EDHS datasets, the mothers’ age, educational status of mother and husband, parity, wealth status, sex of household head, maternal BMI, ANC visit, cesarean delivery, history of a terminated pregnancy, smoking, health insurance, maternal occupation, religion, marital status, perception of distance from the health facility, source of drinking water, type of toilet facility, place of delivery, iron supplementation, use of current contraceptive, duration of breastfeeding, births in the past five years and birth interval were considered as individual-level variables.
Whereas community poverty, community media exposure, community illiteracy level and place of residence were community-level variables. The aggregate community level explanatory variables were constructed by aggregating individual-level characteristics at the community (cluster) level. They were dichotomized as high or low based on the distribution of the proportion values computed for each community after checking the distribution by using the histogram. If the aggregate variable was normally distributed mean value and if not, normally distributed median value was used as a cut-off point for the categorization. Community poverty level was categorized as high if the proportion of women from the two lowest wealth quintiles in a given community was 50–100 % and low if the proportion was 0 - 49%. Community media exposure was categorized as low if the proportion of women exposed to media in the community was 0–28.60 % and categorized as high if the proportion was 27–100 %. Community illiteracy level was categorized as high if the proportion of illiterate women per cluster was 83.3-100% and low if it was less than 83.30%.
Four models were fitted. The first was the null model containing no exposure variables which was used to check variation in community and provide evidence to assess random effects at the community level. The second model was the multivariable model adjustment for individual-level variables and model three was adjusted for community-level factors. In the fourth model both individual and community level variables were fitted with the outcome variable.
Parameter estimation method
The fixed effects (a measure of association) were used to estimate the association between the likelihood of anemia and explanatory variables at both community and individual levels and were expressed as odds ratios with 95% confidence interval. Regarding the measures of variation (random-effects) intracluster correlation coefficient (ICC), Proportional Change in Community Variance (PCV) and median odds ratio (MOR) were used.
The aim of the median odds ratio (MOR) is to translate the area level variance in the widely used odds ratio (OR) scale, which has a consistent and intuitive interpretation. The MOR is defined as the median value of the odds ratio between the area at the highest risk and the area at the lowest risk when randomly picking out two areas. The MOR can be conceptualized as the increased risk that (in median) would have if moving to another area with a higher risk.
It is computed by; MOR=exp[√(2×Va)×0.6745] (18)
Where; VA is the area level variance, and 0.6745 is the 75th centile of the cumulative distribution function of the normal distribution with mean 0 and variance 1. See elsewhere for a more detailed explanation (18). Whereas the proportional change in variance is calculated as PCV=[(VA-VB)/ VA]*100; (19)
Where; where VA=variance of the initial model, and VB=variance of the model with more terms
Ethical clearance was approved by an Institutional ethical Review committee of the Institute of Public Health, College of Medicine and Health Sciences, University of Gondar. The approval letter for the use of the EDHS data set was also gained from the Measure DHS (ORC MACRO). No information obtained from the data set was disclosed to any third person.