Background: Despite of the existing intensive efforts to improve maternal health in Ethiopia, the proportion of birth delivered at home remains high and is still the top priority among the national health threats. Objective: The study aimed to examine effects of individual women and community-level factors of women’s decision on place of delivery in Ethiopia. Methods: Data were obtained from the nationally representative 2011 Ethiopian Demographic and Health Survey (EDHS) which used a two-stage cluster sampling design with rural-urban and regions as strata. The EDHS collected data from a big sample size but our study focused on a sample of 7,908 women whose most recent birth was within five years preceding 2011 and 576 communities in which the women were living in. The data were analyzed using a two-level mixed-effects logistic regression to determine fixed-effects of individual-and community-level factors and random-intercept of between-cluster characteristics. Results: In the current study, 6980 out of 7908 deliveries (88.3%) took place at home. Lower educational levels (OR=2.74, 95%CI:1.84,4.70; p<0.0001), making no or only a limited number of ANC visits (OR=3.72,95%CI:2.85, 4.83; p<0.0001), non-exposure to media (OR=1.51, 95% CI 1.13, 2.01; p=0.004), higher parity (OR=2.68, 95%CI:1.96,3.68; p< 0.0001), and perceived distance problem to reach health facilities (OR=1.29, 95%CI:1.03,1.62; p=0.022) were positively associated with home delivery. About 75% of the total variance in the odds of giving birth at home was accounted for the between-community differences of characteristics (ICC=0.75, p<0.0001). With regard to community-level characteristics, rural communities (OR=4.67, 95%CI:3.06,7.11; p<0.0001), pastoralist communities (OR=4.53, 95% CI:2.81,7.28; p<0.0001), communities with higher poverty levels (OR=1.49 95% CI:1.08,2.22; p=0.048), with lower levels of ANC utilization (OR=2.01, 95%CI:1.42,2.85; p<0.0001) and problem of distance to a health facility (OR=1.29, 95%CI:1.03,1.62; p=0.004) had a positive influence on women to give birth at home. Conclusions: Not only individual characteristics of women, but also community-level factors determine women's decision to deliver at home.
We used data from the EDHS 2011, particularly data on individual women. Ethiopia Demography and Health Survey used a two-stage cluster sampling design with rural-urban and regions as strata. In EDHS 2011, a sample of 624 clusters was drawn by the Ethiopian Central Statistical Agency from its master sampling frame of census 2007. Cluster (community) was defined as a randomly selected area, which contained 150–200 households. In total, 17,817 households and 16,515 reproductive women age 15–49 were sampled using random selection from these clusters (Fig 1). With respect to structure of the data, women are nested within household and household are nested within clusters. The survey was conducted from December 27, 2010 to June 3, 2010 in all the nine regions and two administrative councils of Ethiopia. We included individual data of 7,908 women (weighted) whose last birth was alive and delivered within five years preceding 2011 and community characteristics of 576 clusters (weighted). For mothers with more than one births, we used the most recent birth for the study (Fig 1). The DHS captures a wide scope of data, generally concerning the health of women, men and children. However, for the current study, we used specific data, related with maternal health. The primary entry criterion to this study was women having a live birth within five years preceding the DHS. For the analysis, we included individual variables of socioeconomic and demographic characteristics, obstetric, fertility, perception of women to access health facilities, access to health facility, mass media, and others. The outcome of interest for the study was place of delivery and was grouped into two categories: home and facility based delivery. Home delivery was defined as any birth that had taken place in the women’s or others’ home; while deliveries that occurred in governmental health post, health center, hospital and private clinic and hospital and NG health facilities were grouped as facility-based delivery. In Ethiopia if a birth takes place at home, it is unlikely skilled health professionals assist it. In this context, no home delivery in the EDHS survey was assisted by a nurse, doctor or midwife. The study also focused on community characteristics. We took place of residence as urban versus rural without changing the original coding in the DHS dataset. However, the regions in Ethiopia are divided into eleven for administrative purpose; but, the delineation of the regions may not necessarily be related to the health status of their population. For this study, we have classified the regions into three contextual—agrarian, pastoralist and city dwellers—based on the characteristics of their population in relation to maternal health, particularly place of delivery. Based on their living ways, Ministry of Health of Ethiopia has clearly identified which regions are agrarian, pastoralist or city dwellers so as to make a contextual intervention for each region [19]. Except place of residence and geographic regions, the EDHS did not capture variables that can describe the characteristics of the communities. Yet, we created more community characteristics by aggregating the individual mothers’ characteristics within their clusters. The aggregates for clusters were computed using mean of the proportions of women in each category of a given variable. We categorized the aggregate of a cluster into groups based on the National Median Values. We used median since all distributions of the aggregates were not normally distributed. For the community ANC utilization, for example, we computed the proportion of ANC utilization in each cluster. Finally, we categorized these aggregate values into lower and higher ANC utilization based on the National median of ANC utilization. Ultimately, we used these individual and community level factors to answer why Ethiopian women still deliver at home. The data were analyzed using STATA 11 (Stata Corporation, College Station, TX, USA). The different characteristics of women and communities were described using descriptive statistics. The proportions and frequencies were estimated after applying sample weights to the data to adjust for disproportionate sampling and non-responses. Since DHS data are hierarchical, i.e. mothers are nested within households, and households are nested within clusters, use of flat models could underestimate standard errors of the effect sizes, which consequently can affect decision on null hypothesis. In such data, mothers within same cluster may be more similar to each other than mothers in the rest of the country. This violates the assumption of flat models—independence of observations and equal variance across the clusters. Thus, we used two-level mixed-effects logistic regression model to test the effect sizes of individual and community factors on women’s decision to place of delivery and estimate the between-cluster variability of odds of home delivery. We ran four models: Empty model, model containing only individual factors, model containing community—level factors and model combining both individual and community-level factors. We fit the data into the model: Where The distribution of u 0j is normal with mean 0 and variance σ2 u0. The Intra-Class Correlation (ICC) was calculated using between-cluster variance and within cluster variance [ICC = σu 2 /(σu 2 + π 2 /3)]. In log distribution, the residual variance of women within a cluster is zero but variance is considered constant at π2/3. This helped to show the level of between-cluster correlation within a model and to compare the successive models by looking at the decline of the ICC. The Proportional Change in Variance (PCV) was also computed for each model with respect to the empty model to show how much of variability on the odds of home delivery be explained by the successive models. The PCV was calculated as PCV = (V e—V mi)/V e where Ve is variance in women’s decision in the empty model and Vmi is variance in successive models. We used Variance Inflator Factor (VIF) to scrutiny high multicollinearity among the explanatory variables. The fixed effect sizes of individual and community-level factors on place of delivery were expressed using the Odds Ratio (OR) and the population effect sizes were estimated using 95% Confidence Interval (CI). We accessed the data from MEASURE DHS database at http://dhsprogram.com/data/available-datasets.cfm. We retrieved data of women only. As the data were obtained from records, we could not consent women for accessing their records. However, the records were anonymized and de-identified prior to analysis. MEASUSRE DHS governs the DHS data of all countries and researchers can use the data obeying the data sharing policy. The organization accessed us the data after reviewing our proposal. We accepted the terms and conditions attached to data sharing policy; i.e, we need to keep the data confidential and we would not use the data for purposes other than the current study.
N/A