Spatio-temporal distribution and determinants of home delivery in Ethiopia. Multilevel and spatial analysis of EDHS 2005- 2016


 Background: Globally, only 80% of live births occurred at health facilities assisted by skilled health personnel. In Ethiopia, only 26% of live births attended by skilled health personal. The aim of this study was to assess the spatial patterns and determinants of home delivery in Ethiopia from 2005 to 2016. Method: A total of 34,348 women who gave live birth in the five years preceding each survey were included for this study. ArcGIS version 10.7 software was used to visualize spatial distribution for home delivery. The Bernoulli model was applied using Kilduff SaTScan version 9.6 software to identify significant purely spatial clusters for home delivery in Ethiopia. Result: Home delivery was declined from 94.78% in 2005 to 90.05% in 2011, and 73.44% in 2016 in Ethiopia. Among the three surveys, consistently high clustering of home delivery was observed in Amhara and Southern Nation Nationalities and People Regions of Ethiopia. In spatial scan statistics analysis, a total of 128 clusters (RR= 1.04, P-value < 0.001) in 2005, and 90 clusters (RR = 1.11, P-value < 0.001) in 2011, and 55 clusters (RR = 1.29, P-value < 0.001) in 2016 significant primary clusters were identified. Educational status of women and husband, religion, distance to the health facility, mobile access, antenatal care visit, birth order, parity, wealth index, residence, and region were statistically associated with home delivery. Conclusion: The spatial distribution of home delivery among the three consecutive surveys were non-random in Ethiopia. Low educational status of women and her husband, long distance to the health facility, poor wealth index, rural residence, multiparity, have no mobile access, living in Amhara and SNNP region, and had no antenatal care visit were significant predictors of home delivery in Ethiopia. Therefore, An intervention needs to improve the coverage of antenatal care, women and her husband's education, health care facilities and mobile access. Special attention should give women live in Amhara and SNNPR regions. Key Words: Home delivery, EDHS, Spatial Distribution, Ethiopia.

facilities and mobile access. Special attention should give women live in Amhara and SNNPR regions. Key Words: Home delivery, EDHS, Spatial Distribution, Ethiopia.

Background
Maternal mortality reduction remains a priority agenda under goal three in the UN Sustainable Development Goals (SDGs) agenda through 2030 (1). Worldwide, about 295 000 maternal deaths occurred in 2017 which is 38% reduction since the year 2000 an average reduction of just under 3% per year. Even though a significant decline in maternal mortality in the last 25 years, still maternal mortality is unacceptably high (2). Every day, about 810 women died from preventable and related causes to pregnancy and childbirth, which is the vast majority of these deaths (94%) that occurred in low-resource settings (3).
Maternal mortality in Ethiopia fell from 1250 deaths per 100 000 livebirths in 1990 to 353 deaths per 100 000 livebirths in 2015, declined by 71.8% which is below the target of Millennium Development Goals (MDGs) related to maternal mortality (4,5). Sustainable Development Goal (SDG) goal 3 calls for the ambition of maternal mortality ratio reduction less than 70 per 100 000 live births between 2016 to 2030 (6).
About 73% of all maternal deaths were due to direct obstetric cases and deaths due to indirect causes accounted for 27·5% of all maternal deaths (7). Nearly one-quarter of maternal death occurred in the antepartum period, another quarter occurred in the intrapartum and immediate postpartum periods, one-third occurred in the subacute and delayed postpartum periods, and 12% occurred in the late postpartum period (8).
Globally, only 80% of live births occurred at health facilities assisted by skilled health personnel between 2012-2017. However, only 59% of the births were attended by skilled health personal in the sub-Saharan Africa region, where maternal mortality is the highest (9). Even though, skilled childbirth before, during and after can save the lives of women, still in Ethiopia, 94.5% in 2000, 93.1% in 2005, 87.9% in 2011 and 73.6% in 2016 birth attended at home which is unacceptable high (10,11).
In Ethiopian several studies evidenced that, women's low attainment of educational status, cultural factors, communal factors, limited access to health facilities, poor quality of care, lack of transportation, and poor wealth status were the significant factors that lead to low maternal health services utilization (10,(12)(13)(14). So far different studies in Ethiopia done to identify the factors for the choice of place of delivery (15)(16)(17). The spatial distribution of home delivery was unclear in regions of Ethiopia. Identifying the spatial distribution of home delivery in Ethiopia can help health planners and policymakers for intervention to decrease home delivery. The aim of this study was to assess the spatial patterns of home delivery and determinants for home delivery in Ethiopia from 2005, 2011, and 2016 Ethiopian demographic and health survey datasets.

Study design, period and setting
The repeated cross-sectional study design was conducted in Ethiopia using 2005, 2011 and 2016 Ethiopian Demographic and Health Survey (EDHS). Ethiopia is located in the Horn of Africa and has 9 Regional states and two administrative cities.

Source and Study Population
The source population was all reproductive age group women

Data collection tools and procedures
The data was obtained from children's records (KR) in each were survey year using www.dhsprogram.com website. The web provided the data only for authorized users. Data also contained longitude and latitude coordinates. Ethiopian Demographic and Health Survey data were collected by two-stage stratified sampling. Each region of the country was stratified into urban and rural areas.

Variables
Outcome variable: The outcome variable taken as binary response woman gave birth at home and others home coded as home delivery, and women gave birth different governmental health facilities, private health facility, and non-governmental health facilities taken as intuitional delivery.

Predictor variables
From the EDHS dataset all sociodemographic and obstetric characteristics (Individual and Community level) taken as a predictor variable in the threeconsecutive survey.
Data management and analysis: The data was cleaned by STATA version 14.1 software and Microsoft excel. Sample weighting was done for further analysis.
Spatial autocorrelation and hot spot analysis: We used Arc GIS 10.7 software for spatial autocorrelation and detection of hot spot areas analysis. Spatial autocorrelation (Global Moran's I) statistic measure was used to assess whether home delivery was dispersed, clustered, or randomly distributed in Ethiopia.
Moran's I values close to −1 indicates home delivery, close to +1 indicates clustered, and if Moran's I value zero indicates randomly distributed (18). Hot Spot Analysis (Getis-Ord Gi* statistic) of the z-scores and significant p-values tells the features with either hot spot or cold spot values for the clusters spatially.

Spatial interpolation:
The spatial interpolation technique is used to predict home delivery for unsampled areas based on sampled EAs. For the prediction of unsampled EAs, we used deterministic and geostatistical Ordinary Kriging spatial interpolation technique using ArcGIS 10.7 software.

Spatial scan statistics:
We employed Bernoulli based model spatial scan statistics to determine the geographical locations of statistically significant clusters for home delivery using Kuldorff's SaTScan version 9.6 software (19). The scanning window that moves across the study area in which women gave birth at home were taken as cases and those women who gave birth at health facility taken as controls to fit the Bernoulli model. The default maximum spatial cluster size of < 50% of the population was used as an upper limit, allowing both small and large clusters to be detected, and ignored clusters that contained more than the maximum limit with the circular shape of the window. Most likely clusters were identified using p-values and likelihood ratio tests on the basis of the 999 Monte Carlo replications.

Model Building:
We fit four models, the null model without predictors, model I with only individual-level variables, model II with only community-level variables, and model III both individual-level and community-level variables. These models were fitted by a STATA command "xtmelogit" for the identification of predictors with the outcome variable. For model comparison, we used the log-likelihood ratio (LLR) and Akakian Information Criteria (AIC) test. The highest log-likelihood and the lowest AIC wins the best fit model.

Parameter Estimation Methods
In the multilevel multivariable logistic regression model, fixed effect estimates measure the association between the odds of home delivery of individual and community level factors with a 95% confidence interval. The random effect measures variation in-home delivery across clusters expressed by Intraclass Correlation (ICC) quantifies the degree of heterogeneity of home delivery between clusters, Percentage Chane in Variance (PCV) the proportion of the total observed individual variation in-home delivery that is attributable to between cluster variations and Median Odds Ratio (MOR) median value of the odds ratio between the cluster at high-risk home delivery and cluster at lower risk of home delivery when randomly picking out two clusters (EAs)

Background characteristics of individual women:
A total of 33482 women (10,721 in 2005, 11,872 in 2011, and 10, 889 in 2016) were included for this study. Overall, 94.78%, 90.05%, and 73.44% of women gave birth at home in each survey respectively. From the three consecutive surveys, more than 60% of the mothers were in the age group of 20-34 years and had the same mean ± SD age of 29 ± 6.6 years. Among the three surveys, a significant number (48%) of the female household head was observed in the 2011 EDHS survey. All most all (>90%), of the women were married in five years preceding the survey in three consecutive surveys. The educational status of the women was 79%, 69%, and 66% were unable to read and write in each survey year respectively. As well, 31%, 47%, and 56% of women were had not any work in the consecutive surveys respectively. In EDHS 2016 survey, personal mobile and health insurance status was an interview but not in EDHS 2005 and EDHS 2011. In five years, preceding the survey 16.4% of women were had personal mobile and 3.46 % of women were insured for health insurance ( Table 1)

Characteristics of the cluster:
The unit of analysis for the community factors on home delivery was a cluster. In Ethiopian Demographic Health and Survey 2016, 645 clusters were selected; from these 643 clusters were eligible in which the women give birth preceding five years the survey.
The maximum number of households selected per cluster was 28. Among the total number of clusters, 69% were rural in residence and almost half (49%) of the clusters were had a big problem accessing any health institution. Regarding the aggregate community, ANC utilization rate half of the clusters were had low community utilization. From the total of clusters, half of them were had low community women educational attainment and high community poverty status. ( Table 2). Spatio-temporal distribution of home delivery in Ethiopia.
The spatial distribution of home delivery in Ethiopia was non-random in the three In EDHS 2011, a total of 127 most likely clusters were identified in spatial scan statistics which is located at Southeastern Oromia and Sothern part of the Somali region of Ethiopia.
Among the most likely cluster, 90 of them were primary clusters located at 5 (Figure 3).

Prevalence of home delivery in Ethiopia in the three EDHS surveys:
For the prediction of home delivery prevalence for unsampled areas we used ordinary

Multilevel analysis (random effect analysis)
The home delivery prevalence rate was not similarly distributed across the communities.
About 67.32% of the variance in the odds of home delivery in women could be attributed to community-level factors, as calculated by the ICC based on estimated intercept component variance and also the variation was statistically significant (p-value <0.001).
After adjusting for individual-level and community-level factors, the variation in-home delivery across communities remained statistically significant. About 89 % of the odds of home delivery variation across communities was observed in the full model (model 4).
Moreover, the MOR indicated that home delivery was attributed to community-level

Community-level predictors for home delivery.
In the multivariable multilevel logistic regression model residence, region, community ANC utilization rate, community women education, and distance to any health institution were significantly associated with community-level factors for place delivery.  )). Women who live in a high community ANC utilization were 50% less likely to give birth at home than in low community ANC utilization in five years preceding the survey (AOR =0.50, (95% CI, 0.39, 0.65)). Women in a cluster had no problem to access any health institution were less likely to deliver at home by 29% those women had a problem to access health facilities(AOR= 0.69, (95%CI,0.53, 0.90)). Furthermore, high women education status in the cluster (community) were 28% less likely deliver at home than low women education attainment at the cluster in five years preceding the survey (AOR = 0.78, (95% CI, 0.60, 0.99)) ( Table 3).  (21)(22)(23)(24)(25)(26). The discrepancy might be study area setting deference, cultural attitude for health facility delivery, and infrastructure. In this study, most of the women live in rural areas, which results in a higher prevalence of home delivery. This finding was lower than the 2011 Ethiopian demographic health survey report (90%), and rural community of Nigeria (95.3%) (24,27).
The possible reason might be time difference, and multidimensional strategies are taken to enhance health facility delivery in Ethiopia through health extension workers.
The spatial distribution of home delivery in Ethiopia was non-random in the three successive EDHS surveys. The spatial distribution of home delivery consistently high in Amhara, and SNNPR regional states of Ethiopia. In 2005 EDHS, spatial scan statistics identified the most likely significant cluster at Southwestern Tigray, Amhara, Northern Benishangul, and some part of Oromia regional state of Ethiopia. In 2011 and 2016 EDHS most likely significant cluster located at Somali and Oromia regional states of Ethiopia.
This study evidenced that nearby access to a health facility with a reasonable distance (less than 1 km) positively affects the choice of place of delivery at the individual and community level. Similar findings reported from different studies in Ethiopia, Ghana, and Nigeria (24,26,28,30,33). In line with this finding, this study also evidenced that mothers living in a rural area of Ethiopia were more likely giving birth at home than those living in urban, which is also supported by other studies (30,32,34) The possible justification is the fact that mothers living in a rural area far from the health facility. Ethiopia works to expand primary health care facilities through health extension workers at the health post level. Even though accessing services close to the community, still the topography and infrastructure of Ethiopia are difficult to reach for ambulance service. Therefore, mothers living in a rural area didn't reach the health facility on time for delivery and would influence by the first and second delays for health facility delivery (35).
Another predictor variable that affects the choice of place of delivery was a mobile phone.
Women who had a mobile phone or access to mobile phones from their families were 40% less likely to deliver at home than those who had no mobile phone. The number of mobile users and service accessibility increases in Ethiopia (36). Women having own mobile or the family member can access ambulance on time and service for transportation to the health facility.
Furthermore, women who had at least one Antenatal care (ANC) visit at the health facility was 82% less likely to deliver at home than those who had no ANC visit. Similar evidence reported from a meta-analysis in Ethiopia (37), Kenya (38), Nepal (39), and Akordet town, Eritrea (22). Women during ANC follow up got health education about the choice of place of delivery and the benefit of health facility delivery. Therefore, women during ANC follow up will got behavioral change towards health facility delivery.
The uptake of health facility delivery decreased with high birth order and parity of 2-5, which is consistence with the study conducted in different countries (26,34,40,41). This might be the service quality given in previous births. Even if the Ethiopian health system has improved in the previous decade, still there were critical shortages of health personnel, inconsistent supplies of drugs and equipment. This could discourage women from utilizing health services in later pregnancies for delivery.
Similar to other studies evidenced (34,40) that this study revealed that the poor wealth status of a household increases the likelihood of delivering at home. In a resource-limited setting like Ethiopia, the cost of health service is not affordable. Transportation cost was a big problem in resource limiting settings. Hence in such a situation household wealth was the main invaluable factor for health facility delivery.
Furthermore, mothers live in Afar, Amhara, and Somali regional state of Ethiopia highest odds of giving birth at home than in Addis Ababa, which consistence another study has done Ethiopia (13). This might be the difference in access to health services, infrastructure, and social and cultural attributes. This might be also due to better availability and accessibility of maternal health facilities around Addis Ababa as compared to other regions (13,42).
This study tried to assess the spatial variation and determinants of home delivery in regions of Ethiopia. Identifying the high-risk area of home delivery in regions of Ethiopia could be used to target intervention for the home delivery reduction in high-risk areas.
Therefore, identifying the spatial patterns and determinants of home delivery would help health planners and policymakers in Ethiopia.

Strength and limitation of the study
This study has strengths of having large dataset include thee EDHS survey and were nationally representative. Multilevel multivariable analysis was used to account for cluster correlations. The spatiotemporal analysis was also used for identifying hotspot areas, most likely clusters and the prediction was performed to predict unsampled/unmeasured areas in the country. However, the limitation of this study is the cross-sectional nature of the study design may affect causality.

Conclusions
The Availability of data and materials The data was available from the corresponding 21 author and we can provide upon request. We, authors, declare that we had no competing interests.

Funding Statement
We didn't receive any funds for this study.