Background Although Ethiopia had made a significant change in maternal morbidity and mortality over the past decades, it remains a major public health concern. World Health Organization designed maternal continuum of care to reduce maternal morbidity and mortality. However, majority of the mothers didn’t utilize the maternal continuum of care. Therefore, this study aimed to assess the spatial distribution of incomplete utilization of maternal continuum of care and its associated factors in Ethiopia. Methods This study was based on 2016 Demographic and Health Survey data of Ethiopia. A total weighted sample of 4,772 reproductive aged women were included. The study used ArcGIS and SaTScan software to explore the spatial distribution of incomplete utilization of maternal continuum of care. Besides, multivariable Generalized Estimating Equation was fitted to identify the associated factors of incomplete utilization of maternal continuum of care using STATA software. Model comparison was made based on Quasi Information Criteria. An adjusted odds ratio with 95% confidence interval of the selected model was reported to identify significantly associated factors of incomplete utilization of maternal continuum of care. Results The spatial analysis revealed that incomplete utilization of maternal continuum of care had significant spatial variation across the country. Primary clusters were detected at Somali, North-Eastern part of Oromia, and East part of Southern Nation Nationalities while secondary clusters were detected in the Central Amhara region. In multivariate GEE, rural residency, secondary education, higher education, Protestant religious follower’s, Muslim religious follower’s, poorer wealth index, richer wealth index, richest wealth index, currently working, having barriers for accessing health care, and exposure for mass media were significantly associated with incomplete utilization maternal continuum of care. Conclusion Incomplete utilization of maternal continuum of care had significant spatial variations in Ethiopia. Residence, wealth index, education, religion, and barriers for health care access, mass media exposure, and currently working were significantly associated with incomplete utilization of maternal continuum of care. Therefore, public health interventions targeted to enhance maternal service utilization and women empowerment in hotspot areas of incomplete utilization of maternal continuum of care are crucial for reducing maternal morbidity and mortality.
The present study used 2016 EDHS data. The survey was collected every 5 years to assess population and health indicators at the national and regional levels using a structured, validated, and standardized questionnaire. It was also conducted for four times in Ethiopia. Hence, the 2016 EDHS is the latest and the fourth survey conducted in the country. Ethiopia is an East African country with an estimated population of 109.2 million that makes second most populous country in Africa [18]. Ethiopia is federally decentralized in to nine regions and two city administrations and the regions are further divided into zones, and zones into administrative units called districts [19]. The district again subdivided into kebele which is the lowest administrative unit. Regarding to the health care system in Ethiopia, the fourth health sector development plan introduced a three-tier health-service delivery system. This system were arranged by including Primary health care unities (i.e. health posts and health centers) and primary hospitals at primary level, general hospitals at secondary level, and specialized hospitals at tertiary level [20]. All reproductive aged women who were booked for ANC service and giving birth within 5 years preceding the 2016 survey in Ethiopia were the source population, while all reproductive aged women who were booked for ANC service and giving birth in the selected Enumeration Areas (EAs) within 5 years before the 2016 survey were the study population. The most recent birth characteristics was used for those who give more than one birth within five years preceding the survey. A two stage stratified cluster sampling technique were employed to select study participants. Stratification of regions into urban and rural areas were considered. In the first stage, 645 EAs (202 from urban area) were selected using probability sampling proportional to the EAs size and with independent selection in each sampling stratum. In the second stage, 28 households from each cluster were selected with an equal probability of selection from the household listing [21]. A total of 4,772 weighted reproductive aged women were included in the study. The response variable for this study was maternal continuum of care. Maternal continuum of care is a series of cares provided for mothers during the three periods of maternity [11, 17]. It is a composite variable obtained from ANC, institutional delivery, and Post Natal care (PNC) services. The response variable for the ith mother from jth cluster (EAs) was represented by a random variable Yij, with two possible values coded as 1 and 0. The outcome variable of the ith mother in the jth cluster (Yij) = 1 if ith mother had incomplete maternal continuum of care or if the women had not utilize one of the three maternity services (i.e. 4 and above ANC visits, institutional delivery or postnatal checkup) and Yij = 0 if the mother had complete continuum of maternal care (if the women’s had utilize all the three maternity services). Age of the women, residence, marital status, religion, maternal education, wealth index, currently working, mass media exposure, number of births/parity, contraceptive use, barriers for accessing health care (women reported at least one challenge of health care access (money, distance, companionship, and permission) considered as having barriers of for accessing health care while if a woman didn’t report none of the above challenges were considered as no barriers for accessing health care) [22], wanted pregnancy were included as independent variables. After accessing the data, the variables of the study were extracted from birth recorded data set of EDHS data, data cleaning, and recoding were conducted in STATA version 14.1. The data were weighted using sampling weight and complex survey design was used to adjust for unequal probability of selection due to the sampling design employed in EDHS data. The spatial analysis was done using ArcGIS V.10.7 and SaTScan V.9.6 software. These study conducts the spatial autocorrelation, hot spot analysis, spatial interpolation, and SaTScan analysis of incomplete utilization of maternal continuum of care. Spatial autocorrelation was conducted to test whether the spatial distribution of incomplete utilization of maternal continuum of care was randomly distributed or not. The Global Moran’s I statistics which ranges from −1 to +1 was used to measure whether the distribution of incomplete utilization of maternal continuum of care was dispersed, random, or clustered in the study area [23]. The statistic values close to −1 indicate spatial distribution of incomplete utilization of maternal continuum of care is dispersed, a statistic close to value 0 indicates incomplete utilization of maternal continuum of care is randomly distributed, and a statistic close to +1 means the spatial distribution of incomplete utilization of maternal continuum of care was clustered [24]. Getis-Ord Gi* statistics was used to identify areas with higher rates of incomplete utilization of maternal continuum of care (significant hot spots areas), and areas with lower rates of incomplete utilization of maternal continuum of care (cold spot areas) [25]. To assess the presence of statistically significant spatial clusters of incomplete utilization of maternal continuum of care, Bernoulli based spatial scan statistical analysis with circular window was done. Women with incomplete utilization of maternal continuum of care were taken as cases and women with complete utilization of maternal continuum of care was taken as controls to fit the Bernoulli based model. The default maximum spatial cluster size of less than 50% of the population was used as an upper limit for the identification of both small and large clusters. Log Likelihood Ratio (LLR) test was used to the significance of the clusters and the 999 Monte Carlo replications were used to calculate p values and to rank using their LLR test. Finally, the primary cluster was chosen as the spatial window when it has greatest LLR test [26]. To predict incomplete utilization maternal continuum of care in unsampled areas in the country based on the data in sampled clusters /EA, spatial interpolation technique was employed. Although various spatial interpolation techniques are available, this study used an Empirical Bayesian Kriging (EBK) technique which are considered the best methods since it incorporates spatial autocorrelation and statistically optimize the weight [27]. In the EDHS data, women are nested within a cluster/EAs and those who reside within the same clusters have similar characteristics compared to those from another clusters. This violates the independence and equal variance assumptions of the ordinary logistic regression model. Thus, Intra-class Correlation Coefficient (ICC) was computed to measure the variability between clusters after fitting a model without any covariate. It quantifies the degree of heterogeneity of incomplete utilization of maternal continuum care between clusters (the proportion of variance explained by the between cluster variability). It also computed as; ICC=σμ2σμ2+π23; Where: the ordinary logit distribution has variance of π23, σμ2 indicates the cluster variance [28]. The calculated ICC was 36.37%, showed that about 36.37% of the variation in incomplete continuum of care was explained by the between cluster variation. This implies the need to take into account between-cluster variability by using advanced modelling techniques. Therefore, Generalized Estimating Equation (GEE) model was fitted to identify the associated factors of incomplete utilization of maternal continuum of care among reproductive aged women [29]. It is a marginal model that considers working correlation structure among clusters that estimates a robust standard error and also controlled for within-cluster correlations. Generalized Estimating Equation (GEE) model was fitted with a logit link function and binomial family with independent and exchangeable working correlation structures. Quasi Information Criteria (QIC) was used to select the best-fitted model. The model with exchangeable correlation structure was selected as the best fitted model since it had smaller QIC value. Variables with p-value <0.2 in the bi-variable GEE were considered for the multivariable GEE model. To assess the strength of association between outcome variable and independent factors both crude and adjusted odds ratio with a 95% Confidence Interval (CI) were computed. Variables having less than 5% p-value in the multivariable GEE model were considered as the associated factors with the incomplete utilization of maternal continuum care.