Menu
Menu
Menu
Menu
Background: Skilled birth attendant (SBA) delivery is vital for the health of mothers and newborns, as most maternal and newborn deaths occur at the time of childbirth or immediately after birth. This problem becomes worsen in Ethiopia in which only 28% of women give birth with the help of SBA. Therefore, this study aimed to explore the spatial variations of SBA delivery and its associated factors in Ethiopia. Methods: A secondary analysis was carried out using the 2016 Ethiopian Demographic and Health Survey. A total weighted sample of 11,023 women who had a live birth in the 5 years preceding the survey was included in the analysis. Arc-GIS software was used to explore the spatial distribution of SBA and a Bernoulli model was fitted using SaTScan software to identify significant clusters of non-SBA delivery. The Geographic Weighted Regression (GWR) was employed in modeling spatial relationships. Moreover, a multilevel binary logistic regression model was fitted to identify factors associated with SBA delivery. Results: In this study, SBA delivery had spatial variations across the country. The SaTScan spatial analysis identified the primary clusters’ spatial window in southeastern Oromia and almost the entire Somalia. The GWR analysis identified different predictors of non- SBA delivery across regions of Ethiopia. In the multilevel analysis, mothers having primary and above educational status, health insurance coverage, and mothers from households with higher wealth status had higher odds of SBA delivery. Being multi and grand multiparous, perception of distance from the health facility as big problem, rural residence, women residing in communities with medium and higher poverty level, and women residing in communities with higher childcare burden had lower odds of SBA delivery. Conclusion: Skilled birth attendant delivery had spatial variations across the country. Areas with non-skilled birth attendant delivery and mothers who had no formal education, not health insured, mothers from poor households and communities, Primiparous women, mothers from remote areas, and mothers from communities with higher childcare burden could get special attention in terms of allocation of resources including skilled human power, and improved access to health facilities.
We used the Ethiopian Demographic and Health Survey (EDHS) 2016 to conduct this study. The EDHS is a survey collected across the nine regional states and two city administrations of Ethiopia every 5 years. The latest EDHS (EDHS 2016), was conducted from January 18, 2016, to June 27, 2016. The sample was stratified and selected in two stages. In the first stage, a total of 645 EAs (Enumeration Areas) (443 in rural areas) were selected with probability proportional to EA size (using 84,915 EAs created for the 2007 Ethiopian population and housing census as a sampling frame). A fixed number of 28 households per cluster were selected in the second stage with an equal probability systematic selection after the household listing was done in all of the selected EAs (the lists of households used as a sampling frame for the selection of households in the second stage). Any additional information about data collection, sampling, and questionnaires used in the surveys are described in detail in the 2016 EDHS report [2]. For our study, women aged 15 to 49 years who gave birth within 5 years preceding the survey were included. For those women with two or more live births during the preceding 5 years, data from the most recent birth was used. Accordingly, a total weighted sample of 11,023 women was used in the final analysis. The outcome variable for this study was delivery by SBA. Skilled attendant delivery in this study refers to births delivered with the assistance of doctors, nurses/midwives, health officers, and health extension workers [2]. After searching of literatures, both individual and community level factors were incorporated as independent variables (for the multilevel analysis). The individual level factors include maternal education, maternal age, religion, parity, birth order, household wealth status, access to mass media, and health insurance coverage. The community-level factors were community level of women education, community poverty level, community level media exposure, community childcare burden, perception of distance to the health facility, region, and place of residence (Table 1). Definition/description and measurement of independent [both individual and community level] variables The above four community level factors [community level of women education, community poverty level, community level media exposure, and community level of child care burden] were not directly found in the EDHS data. As a result, they were created by aggregating their respective individual level factors (Table (Table11). Moreover, different explanatory variables were considered in modeling spatial relationships. The candidate variables were proportions of women with no education, the proportion of Primiparous women, proportion of women from poor household wealth status, proportion of women with no health insurance, and proportion of women who perceives distance from the health facility as a big problem, and proportion of women with no media exposure. Both Arc GIS version 10.3 and Kuldorff’s SaTScan version 9.6 software were used to explore the spatial distribution of SBA and to identify significant hotspot areas/clusters of non-SBA respectively. The global spatial autocorrelation was done using the Global Moran’s I statistic, which is used to ascertain whether the spatial distribution of SBA is clustered, dispersed, or random across the country [36, 37]. The spatial interpolation technique was employed to predict the prevalence of non-SBA delivery on the un-sampled/unmeasured areas based on the sampled measurements. The Kriging spatial interpolation method was used in this study for predicting non-SBA in unobserved areas since it had a small mean square error and residual. In addition, hot spot and cold spot analysis were done to identify specific significant hot spots areas (areas with higher rates of non-SBA delivery) and cold spot areas (areas with lower rates of non-SBA delivery) using Getis-Ord Gi* statistics, relative to the mean SBA rate across the country. Moreover, we conducted a spatial scan statistical analysis to identify significant primary and secondary clusters. In the SaTScan analysis, the Bernoulli based spatial scan statistical analysis which requires information about the location of a set of cases (deliveries that were not attended by SBA) and controls (those who delivered by SBA), as well as the coordinate files (latitude and longitude) was used. The default maximum spatial cluster size of < 50% of the population was used as an upper limit for detecting both small and large clusters and ignored clusters that contained more than the maximum limit. The primary and secondary clusters were identified and p values were assigned and ranked using their LLR test based on the 999 Monte Carlo replications. The circle with the highest LLR test statistic was defined as the most likely (primary) cluster, the cluster that is least likely to have occurred by chance. For each identified cluster, the location, radius/size, log-likelihood ratio (LLR) test statistic with its p-value, and the relative risk (RR) were reported. The RR represents how much more common non-SBA delivery with a value of greater than one is used to indicate an increased risk of non-SBA delivery in a specified spatial window as compared to outside the window. We also used Arc GIS version 10.3 for assessing spatial relationships/spatial regression. Spatial regression modeling was performed to identify predictors of the observed spatial patterns of non-SBA delivery. We conducted both ordinary least square (OLS) and geographically weighted regression (GWR) analysis. Findings from ordinary least squares (OLS) regression are only reliable if the regression model satisfies all of the assumptions that are required by this method. While conducting the OLS regression the assumptions to be fulfilled, the model performance, as well as the model significance were checked [38, 39]. In addition, a certain independent variable may be a strong predictor in one cluster and it may not be in another cluster. This is non-stationarity and can be identified using GWR [40–42]. These two (OLS and GWR) models were compared using different parameters. Finally, the coefficients which were created using GWR were mapped. Stata 14 software was used for analysis. To avoid geographical strata selection variability and non-responses, as well as to assure representativeness and have better estimations of parameters, sampling weight was done throughout our analysis. Both bivariable and multivariable multilevel logistic regression analyses were performed. Because of the hierarchical nature of EDHS, we used the multilevel logistic model for the appropriate estimation of parameters. To do so, four models have been fitted; the null model- a model without explanatory variables, the model I- a model with individual-level factors only, model II- a model with community-level factors, and model III- a model with both individual and community-level factors simultaneously. Among the four fitted models, the model with the lowest deviance was selected as the best-fitted model. The intraclass correlation coefficient (ICC), a proportional change in variance (PCV), and median odds ratio (MOR) were also used to examine the clustering effect and the extent to which community-level variability explains the unexplained variance of the null model.
