Determinants and spatial distribution of institutional delivery in Ethiopia: evidence from Ethiopian Mini Demographic and Health Surveys 2019

Background Over the past few decades, maternal and child mortality had drawn the attention of governments and policymakers. Institutional delivery has been among the implementations needed to reduce maternal and child mortality. The fact that the problem was persisted intensified studies to research for more factors. Thus, the current study was intended for further analyses of EMDHS to identify the magnitude, spatial patterns, and predictors of institutional delivery. Methods A cross-sectional survey data from EMDHS 2019 was analyzed involving 5488 reproductive-age women regarding institutional deliveries. We presented descriptive statistics using mean, standard deviations, and proportions. To check the nature of the distribution of institutional delivery, we applied the global Moran’s I statistics. Getis-Ord Gi statistics was applied to detect spatial locations, and we applied spatial interpolation to predict unknown locations of institutional delivery using the Ordinary Kriging method. Kulldorff’s SatScan was also applied to identify the specific local clustering nature of institutional delivery using the Bernoulli method. We applied multilevel binary logistic regression for the scrutiny of individual and community-level factors. We applied P < 0.25 to include variables in the model and P < 0.05 to declare associations. AOR with 95% CI was used to describe variables. Results The prevalence of institution/facility delivery was 2666.45(48.58%) in the survey. The average number of children was 4.03 ± 2.47, and most women in this survey were in the age range of the 25-29 years (31.84%) and 30–34 years (21.61%). Women who learned primary education (AOR = 1.52; 95% CI 1.20–1.95), secondary education (AOR = 1.77; 95% CI 1.03–3.07), and higher education (AOR = 5.41; 95% 1.91–15.25), while those who can read and write sentences (AOR = 1.94; 95% 1.28–2.94), Rich (AOR = 2.40 95% CI 1.82–3.16), and those followed 1–2 ANC (AOR = 2.08; 95% CI 1.57–2.76), 3 ANCs (AOR = 3.24; 95% CI 2.51–418), and ≥ 4 ANCs (AOR = 4.91; 95% CI 3.93–6.15) had higher odds of delivering at health institutions. Conclusion The institutional delivery was unsatisfactory in Ethiopia, and there were various factors associated differently across the different regions. Pastoralist regions showed high home delivery than institutions which invites further interventions specific to those regions. Factors like age, highest education level achieved, preceding birth interval, literacy status, wealth status, birth order, regions, and rural residences were all affected institutional delivery so that interventions considering awareness, access, and availability of the services are vital.


Background
The safety of the mother and newborn is always the target during delivery and can be improved by institutional delivery. Developing countries have many constraints in reducing maternal and newborn problems due to access, availability, and awareness [1]. As a country in the Sub-Saharan region, Ethiopia got similar limitations regarding maternal and child cares. Henceforth, the country planned to increase institutional delivery from 20% in 2011 to 60% in 2015 [2]. Because of the various factors, the country wasn't able to achieve as planned. Many studies indicated that there was a high proportion of home delivery in the country [3][4][5][6][7][8][9]. There were various factors cited affecting women to prefer home over institutions for delivery. A study conducted in the Farta district indicated that the magnitude of institutional delivery was 64.4%. Family size, accessibility of transportation, planned pregnancy, information about the place of delivery, participation of women in monthly health conference (PWMHC), information about exempted service, and having antenatal care (ANC) follow up during their last pregnancy were predictors [10,11]. In other studies, institutional delivery was 38.9% and influenced by focused antenatal care, multiple gestations, urban residence, and formal education of the women [12,13].
Residence, region, maternal education, wealth status, ANC visit, preceding birth interval, and community media exposure were all predictors of the institutional delivery from other studies [14][15][16]. Institutional delivery was 18.2% in Oromia regional state, where urban residence, maternal education, pregnancy-related health problems, previous history of prolonged labor, and the decision made by the husbands or relatives put forth a positive correlation [17]. There was high variation throughout different administrative levels in the country. For instance, institutional delivery was 72% in Bule Hora in Oromiya [18]; 89.1% in Damot district in SNNP [19]; 61.5% in North-Western part of the country [20]; 76% in Southwest Ethiopia [21]; 71.7% in Dejen Woreda; 78.8% in Bahir Dar [22]; 18.3% in Dangila woreda; 47% in semi-urban parts of the country [23]; 38% in Mandura district [24]; 53% in Hossana town [25]; 74% in Dallocha town [26] and 49% in Hetosa district [27]. The rate of increase was 5.4% in 2000 and 11.8%,2011 [28]. Ethiopia showed a good achievement toward the millennium development goal five [29], although the contextual facts look unchanged significantly. The plan to upsurge institutional delivery from 20% in 2011 to 60% in 2015 was not adequately achieved until recently from the above literature. In the countries like Ethiopia where resources are very limited, there were limited chances to collect country representative data; however, analyzing the data from EMDHS-2019 enabled us to provide the countrylevel figures. And it is paramount to identify modified and persistent factors for unsuccessful for the next health plans. Therefore, our study was aimed at identifying the prevalence and factors through spatial, descriptive, and multilevel analyses to support further policies.
Health Surveys (EMDHS) conducted in 2014 and 2019. EMDHS usually conduct between the standard EDHS (two to three years) after the EDHS conducted. The 2019 EMDHS is the second nationwide mini survey conducted in the country. In Ethiopian DHS, data has been collected using a two-level multistage stratified cluster sampling to pick eligible respondents from rural and urban areas. For the current analysis, we used Ethiopia Mini Demographic Health Survey (EMDHS) 2019 data. All nine regions and two city administrations were involved in the data collection. The regions were further categorized as agrarian (Benishangul-Gumuz Amhara, Southern Nations, Nationalities, and People Gambela, Oromia, Harari, Region (SNNPR), and Tigray), pastoralists (Afar and Somali), and city administrations (Addis Ababa and Dire-Dawa) contextually. We retrieved the data from the DHS website: (www. dhspr ogram. com) after the measure program allowed us to download the datasets. The weighted sample became 5488 women who had live births in the last five years before the survey. They conducted the interview on the permanent residents and visitors who stayed the day before the survey in the residences, and it was a face-to-face manner [30].

Study variables
The outcome variable for this study was the health institutions/facilities delivery, which was coded as "0" if the women gave birth at home and "1" if the women gave birth at a health facility. Institutions/facilities delivery was stated as the births at health institution/facility within five years afore the survey.

Individual-level (covariates) variable
Maternal education, maternal age, religion, ANC followup, sex of household head, literacy, the total number of children, birth order, preceding birth interval, the timing of 1st ANC visit, wealth index, and marital status were the variables.

Community-level variable
Region and place of residence.

Statistical analysis Descriptive statistics
Before conducting the descriptive data analysis, we weighted the data to adjust the non-proportional allocation of samples to strata and regions. Then, descriptive statistics were presented using weighted and unweighted frequencies, mean ± (standard deviations), and percentage, while all analyses were performed using STATA version 15 (STATA Corporation. IC., TX, USA). The mean-variance inflation factor also was checked to be 3.53, which was in the acceptable range.

Spatial analysis
For spatial analysis, we used ArcGIS 10.7 that determined the clustering, dispersion, and random distribution nature of the institutional delivery. Moran's I output lies between (− 1 to + 1). The values close to − 1 indicated dispersed institutional delivery, and those closes to + 1 indicated clustering distribution. After discovering significant global autocorrelation, we tested the local Getis Ord statistics to identify the areas with high and low institutional deliveries [31].

Spatial interpolation
For statistical optimization of the weight, the Ordinary Kriging spatial interpolation method was applied, and enabled us to make the prediction of institutional delivery for un-sampled areas of the country.

SaTscan analysis
SaTScan Version 9.6 software was used for the local cluster detection. A circular window that moves systematically throughout the study area was used to identify a significant clustering of institutional delivery. We presented the results of primary and secondary observed clusters using log-likelihood (LL) and p-value < 0.05.

Multilevel binary logistic regression
Since the data from country representative surveys are usually clustered or have hierarchical structure, we applied multilevel analysis. We went through four consecutive models building strategies to identify felicitous predictors of institutional delivery in the country. Model 0 is an empty/null (the intercept only model) existed before addition of the predictors. Model 1 (fixed effect model) included all individual-level variables that were initially significant at p-value of < 0.25 to determine the level of variance explained by the model. Model 2 (random effect model) included cluster-level (community -level) variables and model 3 (the mixed effect model) was the final model in which both the individual and community level variables introduced to judge final model performance. The log of the probability of the institutional delivery was modeled using multilevel binary logistic regression as: log where, i and j are the level 1 (individual) and level 2 (community) units; X and Z refer to individual and community-level variables, in sequence. Πij is the probability of the institutional delivery for the i th mother in the j th community. We resolute random effect using Intra-community Correlation (ICC), ICC = σ 2 a σ 2 a +σ 2 b ; where, σ 2 a is the community level variance and σ 2 b indicates individual level variance. The variance (σ 2 b ) is equal to π 2 3 which is the fixed value. Likelihood Ratio (LR) test for model comparison and deviance (−2LL) for the goodness of fit check were calculated, while Median Odds Ratio (MOR) and Proportional Change in Variance (PCV) were also estimated [32].
Finally, the mixed effect model, which included both fixed and random effect variables were fitted. To include variable in the model p-value < 0.25 and to declare association p-value< 0.05 were used.
AOR with 95% CI was also used to articulate the results.

Descriptive statistics Institutional delivery
We analyzed the data on 5488 weighted reproductive-age women who gave birth in the last five years earlier to the survey and found 2666.45(48.58%) of them delivered at health institution/facility.

Individual-level characteristics
The average number of children in families in the survey was 4.03 ± 2.47, and the average preceding birth interval was 40.61 ± 26.30 months. Most women in this survey were in the age group of the 25-29 years (31.84%) and 30-34 years (21.61%). In other words, 53.55 and 35.47% were learned no education and primary education, respectively. Similarly, 47.45% of the women had no ANC follow-ups, while 30.52% of women were followed ANC of ≥4 visits. Economically, 45.53% of the women were poor, while 35.57% were in the rich category. Nearly two-thirds of the women cannot read and write (64.34%), and those who can read or write the complete sentence were 20.83%.

Spatial analysis Moran analysis
Examining spatial autocorrelation, we recognized that the distribution of institutional delivery was non-random in Ethiopia. We also learned that the global Moran's I test was 0.66(p < 0.0001). As global Moran's I is significant and greater than zero, we concluded that the distribution was clustered. By this, the null hypothesis which stated random distribution of institutional delivery was rejected at Moran's I test of 0.66(p < 0.001) (Fig. 1).

Spatial distribution of institutional delivery
The geo-distribution of institutional delivery was clustered in some parts of the country. Based on the Gettis-OrdGi statistical analysis, Addis Ababa city, Dire Dawa city, Hawassa town in SNNP, some places in Benishangul Gumuz, and few places in the Oromia region displayed the highest prevalence of institutional delivery showing a significant Z-score with 90% and above confidence levels (Fig. 2).

Ordinary kriging interpolation
To determine the occurrence of institutional delivery throughout the un-sampled areas, Ordinary Kriging interpolation measured the distance from the known point to predict unknown points/areas and indicated the point in the ranges of the event occurrences. The evidence from Fig. 3 showed that the possible areas for the event happening were Addis Ababa, Dire Dawa, Benishangule, Tigray, and some parts of Gambella.

SaTscan statistics
In all spatial analyses above, we tried to identify the nature of institutional delivery distribution in the country; furthermore, SaTscan statistics enabled us to identify specific local clusters, which are necessary for specific local interventions. According to Table 2 and Fig. 4, there were one primary/most likely and six secondary and significant clusters in the country. Totally 27 locations with Coordinates/radius -(6.639662 N, 44.465853 E) / 390.28 km) with relative risk (RR) of 1.80 and LL of 187.82 were included in the primary cluster. The cluster was located in the Somali region including some borders of Oromia. It means these areas were 1.80 (p < 0.001) times more at risk of home delivery. The remaining clusters are secondary and presented in Table 2 and Fig. 4.

Multilevel modeling of the institutional delivery
During multilevel modeling: age, highest education level achieved, preceding birth interval, literacy status, wealth status, and birth order at the individual level and place of residence and region at community level were significant associated with institutional delivery.  Table 3). The application of multilevel binary logistic regression was proved vital by abridging the ICC (the variation observed only due to the differences among clusters) from 61 to 28%. The remaining 28% unexplained intercommunity variation can be reduced by including   (Table 4).

Institutional delivery
Of the 5488 women involved in the current study, 48.58% were given births in health institutions. Evidence from other studies in the country unveiled institutional delivery was 71.7% in Dejen [33], 78.8% in Bair Dar [22], 38% in Mandura district [20], and 38.9% in 2016 EDHS [12]. There was a high variation from pocket studies but it showed good improvement from EDHS 2016. All the inconsistencies might be an indication of inconsistent interventions that need consideration during the next interventions (Table 1).

Individual and community level characteristics
Although there was no adequate evidence to support, the average number of children was 4.03 ± 2.47 in the current study. In Mizan Tapi SNNP, 40% of the women had 3 to 4 children [34]. It was 1 to 4, 2 to 4, and 4 to 6 children in other studies in the country [15,34,35]. There were many inconsistencies, but our study might show the representative national figure. In other words,   [36]. In Ethiopia and even in the East of Africa, the average ranged between 24 and 48 months that is a good improvement over decades. However, the numbers of uneducated (53%) and economically poor (45.53%) women in this study were still very high. One study indicated that 79.2% of women in Ethiopia were uneducated [36] and resides in poor households [12,13]. The consistent finding might be an indication of unsuccessful previous efforts and need more considerations recently. In the same way, a large proportion of women had no ANC follow-up (47.45%) and 64.34% of them were unable to read and write. Evidence from other studies also showed that 54% of the women giving birth in Ethiopia were unable to read and write [37] and 56% of them followed no ANC [38]. Overall, women's literacy status and ANC follow-up have remained below par and invite more intervention. The fact that more than 75% of women residing in the rural parts of the country might also contribute to the above findings as indicated in similar studies [15,37] (Table 1).

Spatial distributions
The attempt of localizing the findings to the specific regions for enhanced interventions was successful; the distribution of institutional delivery could guide interventions as it showed clustering in some parts of the country. The significant Moran's test of positive and greater than zero, the hot/cold spot areas identified with significant z-score, the predictions made from known spatial location of event occurrence (spatial interpolation) of ≥0.56%, and spatial scan statistics which encircled significant areas were enabled us to identify the regions needed more attention of improving institutional delivery in the country and were supported with other studies [14,16]. From Table 2 and Figs. 1, 2, 3, and 4, we recognized the clustering nature of institutional delivery in Ethiopia. Accordingly, Addis Ababa, Hawassa, Dire Dawa, some parts of Benishangule Gumuz, and Oromia were found to be areas with a high proportion of institutional deliveries. However, Somali, Afar, some parts of SNNP, Oromia, and Amhara also showed low institutional delivery. This finding is also consistent with a systematic review done in the country [8], EDHS 2016 [4], spatial variation study in Northern Ethiopia, and EDHS 2005-2011 [38][39][40].

Multilevel analysis of predictors
According to multilevel binary logistic regression analysis, women in the age group of 20-24, 25-29, and 30-34 years had higher odds of institutional delivery. Similar evidences were found in other studies in the country [37,41]. The consistent information might indicate institutional delivery increased at the time women might take more responsibilities for their births in the middle ages of the childbearing period. In other words, women's education and literacy were independent predictors of institutional delivery. This is a very global finding and common in most studies [35,42,43] and might indicate that improving mothers' educational and literacy status might be very vital to robust institutional delivery. Mothers who had any ANC visits showed a high correlation of delivering in institutions. The finding was also supported with plenty of evidence from similar studies [26,35,42]. The exposure of mothers to health professional counseling might be a factor that spurred the utilization; however, it might also be because of they were the only mothers who had access. Additionally, mothers with rich wealth status had a higher tendency of delivering in institutions as is also the case in other studies [36,44]. This might show that in addition to access, availability, and affordability could be considered during interventions. In this study, mothers tend to give birth at health institutions for later pregnancies which were also common in other studies [45] (Table 3). Contextual analysis indicated that except Oromia, all agrarian regions and city administrations showed good institutional deliveries; however, pastoralists regions (Somali and Afar) showed high home deliveries; henceforth, the home delivery was also higher in rural areas. Other studies also supported this finding [4,16]. During this analysis, we recognized many limitations. Among them, disproportionate sampling, using third party data, secondary nature of data, hierarchical nature of the data, lack of GPS coordinates for few records, and cross sectional nature of the study were some of the limitations. To handle disproportionate sampling, we applied weighting to the data before analysis. To deal with the third party nature of the data, we applied all the necessary protocols which are internationally standards to all countries and described all methods and manipulation of the data under method section. Due to the hierarchical nature of the data, we conducted multilevel binary logistic analysis. We also followed all standard methods to comprehend whether each captured data fit to the standard set by international DHS authorities. For the data without coordinates, we removed the records so that each event can be located on the test map.

Conclusions
After 2016 EDHS, EMDHS was the first country representative data to make an impression about distribution and factors associated with institutional deliveries in Ethiopia. According to our analysis, although there was good progress from some previous evidence, institutional delivery was still low and requires enhanced further interventions. Factors that contributed to low prevalence were age, highest education level achieved, preceding birth interval, literacy status, wealth status, birth order, regions, and rural residences. The positive correlation of institutional delivery with educational/literacy status and ANC might invite further interventions in the area of awareness, access, and availability of the service. Regional disparities need specific interventions as pastoralist regions showed poor institutional delivery, while economic support for women in the rural area needs to be in the equation as the part of the further policy interventions.

Acknowledgments
We are very grateful to EMDHS program for the permitting us to access the data and our gratitude should also goes to the other stakeholders who involved directly or indirectly.