Individual-and community-level determinants of neonatal mortality in the emerging regions of Ethiopia: a multilevel mixed-effect analysis

Background Unlike infant and child mortality, neonatal mortality has declined steadily in Ethiopia. Despite the large-scale investment made by Ethiopia to improve the health of newborns and infants, it is among the regions with the highest burden of neonatal mortality. Although there are studies done on neonatal mortality in different areas of Ethiopia, as to our search of pieces of literature there is no study in Emerging regions of the country. Therefore, this study aimed to investigate the individual and community-level determinants of neonatal mortality in the Emerging regions of Ethiopia. Methods Using the 2016 Ethiopian Demographic and Health Survey (EDHS) data, secondary data analysis was done. A total weighted sample of 4238 live births in Emerging regions were included for the final analysis. A multilevel binary logistic regression was fitted to identify the significant determinants of neonatal mortality. The Intra-class Correlation Coefficient (ICC), Median Odds Ratio (MOR), Proportional Change in Variance (PCV) were used for assessing the clustering effect, and deviance for model comparison. Variables with a p-value < 0.2 in the bi-variable analysis were considered in the multivariable analysis. In the multivariable multilevel binary logistic regression analysis, Adjusted Odds Ratio (AOR) with 95% Confidence Interval (CI) were reported to declare statistically significant determinants of neonatal mortality. Results The neonatal mortality rate in Emerging regions of Ethiopia was 34.9 per 1000 live births (95% CI: 29.8, 40.9). Being born to a mother who had no formal education (AOR = 1.79, 95% CI: 1.12, 2.88), being born to a mother who did not participate in making health care decisions (AOR = 1.25, 95% CI: 1.14, 1.79), and being twin birth (AOR = 6.85, 95% CI: 3.69, 12.70) were significantly associated with higher odds of neonatal mortality. On the other hand, being female (AOR = 0.67, 95% CI: 0.47, 0.95), having 1–3 Antenatal Care (ANC) visits (AOR = 0.34, 95% CI: 0.15, 0.74), high community media exposure (AOR = 0.64, 95% CI: 0.41, 0.98), and preceding birth interval of two to 4 years (AOR = 0.38, 95% CI: 0.24, 0.58) were significantly associated with lower odds of neonatal mortality. Conclusion Neonatal mortality in Emerging regions of Ethiopia was unacceptably high. Maternal education, women’s autonomy in making decisions for health care, sex of a child, type of birth, preceding birth interval, ANC visit, and community media exposure were found significant determinants of neonatal mortality. Therefore, empowering women in making health care decisions and increasing access to mass media play a major role in reducing the incidence of neonatal mortality in Emerging regions of Ethiopia.


(Continued from previous page)
Conclusion: Neonatal mortality in Emerging regions of Ethiopia was unacceptably high. Maternal education, women's autonomy in making decisions for health care, sex of a child, type of birth, preceding birth interval, ANC visit, and community media exposure were found significant determinants of neonatal mortality. Therefore, empowering women in making health care decisions and increasing access to mass media play a major role in reducing the incidence of neonatal mortality in Emerging regions of Ethiopia.
Keywords: Ethiopia, Neonatal mortality, Emerging regions, Multilevel analysis Background Globally, under-five mortality significantly decreased from 12.7 million in 1990 to 6.3 million in 2015 with 2.6 million died during the neonatal period [1]. It accounting for 40% of under-five mortality [2]. Approximately 98% of neonatal deaths occurred in low and middle-income countries [3,4] and 39% in sub-Saharan African (SSA) countries [5]. The neonatal mortality rate has varied across countries ranged from 20 per 1000 live births in high-income countries to 31 per 1000 live births in SSA [6]. It is far below to achieve the Sustainable Development Goal (SDG) target of reducing the neonatal mortality rate of 12 or less per 1000 live births by 2030 [7,8].
Despite the substantial decrease in child and infant mortality, the decline in neonatal mortality is steady [9]. As in most African countries, Ethiopia is one of the countries with the highest burden of neonatal mortality [10]. In Ethiopia, though child and under-5 mortality has dramatically decreased in the last two decades, neonatal mortality has steadily decreased [11]. According to the Ethiopian Demographic and Health Surveys (EDHSs) report, under-five mortality decreased from 166 per 1000 live births to 67 per 1000 live births, and infant mortality decreased from 97 per 1000 births to 48 per 1000 births, while neonatal mortality decreased from 49 per 1000 live births to 29 per 1000 live births, which is lower than under-five and infant mortality [12][13][14].
Infectious diseases, malnutrition, and birth complications are identified as the leading causes of neonatal mortality [2,4,15]. Previous studies conducted on neonatal mortality showed that residence [16], parity [17], educational status [17], mode of delivery [18], ANC utilization [19], birth interval [20], place of delivery [21], maternal nutritional status [22], and maternal obstetric factors [23] were statistically significant determinants of neonatal mortality. Neonatal mortality has significantly reduced in Ethiopia in the last two decades. However, it is marginally higher in Somalia, Afar, Gambella, and Benishangul-Gumuz regions [13,24]. This highlights Ethiopia should work further to reach the Every Newborn Action Plan (ENAP) set a national target of less than 10 per 1000 live births by 2035 [25].
Though there are studies conducted on neonatal mortality in different areas of Ethiopia [26][27][28][29], there is limited evidence on the individual-and community-level determinants of neonatal mortality in Emerging regions. Therefore, this study aimed to investigate the individual and community level determinants of neonatal mortality in Emerging regions of Ethiopia using multilevel logistic regression analysis. Identifying significant individual and community-level determinants of neonatal mortality is crucial to reduce the incidence of neonatal mortality in Emerging regions of Ethiopia.

Study setting and design
The study used the 2016 Ethiopian Demographic and Health Survey (EDHS) data, which was obtained using a community-based cross-sectional study design. The 2016 EDHS is the fourth survey conducted in every fiveyear interval in Ethiopia. There are nine regional states (Afar, Amhara, Benishangul-Gumuz, Gambella, Harari, Oromia, Somali, Southern Nations Nationalities and People's Region (SNNPR), and Tigray) and two city administrations (Addis Ababa and Dire Dawa) in Ethiopia. These regions are classified into three regions; emerging regions (Afar, Somali, Benishangul-Gumuz, and Gambella), developed regions (Amhara, Oromia, Harari, Southern Nations Nationalities and People's Region (SNNPR) and Tigray) and two city administrations (Addis Ababa and Dire Dawa) [30]. A total of 13.1 million people of Ethiopia are living in Somali, Afar, Benishangul-Gumuz, and Gambella regions [31]. These regions are characterized by scattered pastoralists and semi-pastoral populations with extreme poverty and limited access to health care.

Data source and sampling procedure
The data used for this study were retrieved from the most recent Ethiopian DHS survey (EDHS 2016). The EDHS is conducted every five-year using structured methodology and pretested validated standard tools to generate updated health and health-related indicators. The EDHS employs a multi-stage stratified cluster sampling technique to select the study subjects. In the first stage, a total of 645 Enumeration Areas (EAs) that represent the country were selected. In the second stage, systematic random sampling was employed and on average 28 households per EAs were selected. For this study, neonates born in Emerging regions of Ethiopia within 5 years preceding the survey were included. A total of 4238 neonates were used for analysis. The overall data collection and the sampling procedure was presented in the full EDHS 2016 report [13].

Study variables Outcome variable
The outcome variable for this study was neonatal death as reported by the mother, and it was defined as the death of a neonate within the first months of birth. This takes a binary outcome; such that neonatal death will be regarded as death (1 = if death occurs in the first month of life) or alive (0 = if the newborn alive in the first month of life).

Independent variables
The independent variable considered for this study were from two levels (at individual and community levels). At the individual, maternal age, marital status, religion, maternal education, paternal education, wealth index, maternal occupation, media exposure, maternal Body Mass Index (BMI)), number of ANC visit, the timing of first ANC visit, mode of delivery, preceding birth interval, place of delivery, women health care decision autonomy, size at birth, type of birth and birth order were included. At the community level, region, residence, community women education, community poverty, community media exposure, and distance to a health facility were considered. Community-level variables used in the analysis were from two sources; direct community-level variables that were used without any manipulation and aggregated community-level variables that were generated by aggregating individual-level variables at the cluster level.
Media exposure was measured from three variables such as reading the newspaper, listening to the radio, and watching television. These variables were merged and categorized as no "when there was no exposure to either of the three" and yes "when there was exposure to either of reading newspaper, listening radio and watching television". Women's health care decision making autonomy was assessed in EDHS, as a person decides on the respondent's own health care. Which was categorized as women participating in making their own health care decisions and didn't participate in making health care decisions (decides by their husband/partner). Birth weight was categorized as small, average, and large size at birth. Small size at birth is defined as birth weight less than 2500 g while birth weight greater than 4000 g is considered as large size at birth.

Data collection procedure
The research was performed based on the 2016 EDHS data by accessing the data from the official database of the MEASURE DHS program www.measuredhs.com. For the study, we used the Birth Record (BR) data set.

Data management and analysis
The variables were extracted from the BR dataset using STATA version 14 statistical software. The weighted data were used for analysis to adjust for unequal probability of selection and non-response. In EDHS, multistage stratified cluster sampling techniques were employed and the data were not flat. So, to draw valid inference and conclusion advanced statistical models such as hierarchical modelling, which consider independent variables measured at individual and community levels should be employed to consider the clustering effect/dependency. A two-level binary logistic regression model was employed to estimate the effect size of independent variables on neonatal mortality. Four models were fitted. The first model was the null model (a model without the explanatory variable), which was a model fitted to calculate the extent of cluster variability on neonatal mortality. It was assessed using the Likelihood Ratio test (LR), Intra-class Correlation Coefficient (ICC), Median Odds Ratio (MOR), and Proportional Change in Variance (PCV). The ICC is the proportion of total variance explained by the cluster variation [32]. ICC = σ 2 /(σ 2 + π 2 /3) Where ∂ 2 indicates that cluster variance. MOR is the median values of the odds ratio of the cluster at high risk and cluster at lower risk of neonatal mortality when randomly picking two neonates from two clusters (EAs) [33].
PCV is defined as the total variation of neonatal mortality explained by the final model (a model with individual-level factors and community-level variables) relative to the null model (a model without independent variables). Both bivariable and multivariable analyses were done. In the bivariable two-level binary logistic regression analysis, variables with a p-value ≤0.2 were considered in the multivariable analysis. The Adjusted Odds Ratio (AOR) with a 95% Confidence Interval (CI) in the multivariable multilevel analysis were reported to declare the statistical significance and strength of association between neonatal mortality and independent variables. By doing the pseudo linear regression analysis, the multi-collinearity was checked and the mean VIF was less than 5.

Socio-demographic and economic characteristics of the mothers
A total weighted sample of 4238 neonates were used for this study. The majority (74.8%) of neonates were born to mothers with no formal education and 3997 (94.3%) of neonate's mother were married. Of the total neonates, 3397 (80.2%) of their mothers did not have media exposure and 3082 (72.7%) of neonates were born to mothers aged 20-34 years. The majority (72.4%) of neonate's mother was Muslim religion followers (Table 1).

Child and maternal obstetric related characteristics
From a total of 4238 neonates, 52.5% were males and 68.2% were born at home. About 13.1% of the mothers had 1-3 ANC visits during their pregnancy and 2.8% were delivered through caesarean section. About 2.4% were twin births and 40.6% were large size at birth. Nearly three-fourths (70.2%) of the mothers were participated in making their own health care decisions ( Table 2).

Community-level characteristics of the mothers
About 85.1% of neonate's mothers were from rural residents and 36.0% were in the Somali region. The majority (55.7%) of their mother was from a community with high poverty and 58.3% of the mothers reported perceived distance to visit health facilities as a big problem (Table 3).

Neonatal mortality rate by respondent characteristics
The neonatal mortality rate in Emerging regions of Ethiopia was 34.9 (95% CI: 29.8, 40.9) per 1000 live births, which was highest in the Somali region (41 per 1000 live births) and lowest in the Benishangul region (35 per 1000 live births) (Fig. 1). The neonatal mortality rate among rural residents was 38.6 per 1000 live births ( Table 4).

Determinants of neonatal mortality Model comparison
The final model was the best-fitted model since it had the lowest deviance value. The ICC was 13.5% in the null model indicated that 13.5% of the total variability of neonatal mortality was due to differences between clusters/EA, with the remaining unexplained 86.5% was attributable to individual differences. Moreover, the MOR was 1.98 in the null model which indicates that there was variation between clusters, if we randomly select neonate from two different clusters, neonate at the cluster with a higher risk of neonatal mortality had 1.98 times higher odds of neonatal mortality as compared with neonate at cluster with a lower risk of neonatal mortality. PCV for the final model was 37.3%, indicated that 37.3% of the variability in neonatal mortality was explained by the full model (Table 5).
In the multivariable multilevel analysis; maternal education, women who didn't participate in making their own health care decisions, twin births, preceding birth interval, number of ANC visits, community media exposure, and sex of child were significantly associated with neonatal mortality. The odds of neonatal mortality among live births born to mothers who didn't attend formal education had 1.79 (AOR = 1.79, 95% CI: 1.12, 2.88) times higher than live births born to mothers who attained secondary education and above. The odds of neonatal mortality among female births were decreased by 33% (AOR = 0.67, 95% CI: 0.47, 0.95) compared to male births. Being born to mothers who didn't participate in making their own health care decisions were 1.25 (AOR = 1.25, 95% CI: 1.14, 1.79) times higher odds of neonatal mortality than births whose mother who participated in making health care decisions. The odds of neonatal mortality among twin births were 6.85 (AOR = 6.85, 95% CI: 3.69, 12.70) times higher compared to singletons. Furthermore, the odds of neonatal mortality among neonates in the community that had high media exposure were decreased by 36% (AOR = 0.64, 95% CI: 0.41, 0.98) compared to neonates in the community with low media exposure. The odds of neonatal death for neonates with preceding birth interval 2 to 4 years were decreased by 62% (AOR = 0.38, 95% CI: 0.24, 0.58) compared to neonates with preceding birth interval less than 2 years. The odds of neonatal mortality among children born to mothers who had 1-3 ANC visits during pregnancy were decreased by 66% (AOR = 0.34, 95% CI: 0.15, 0.74) than a child born to a mother who didn't have ANC visit during pregnancy (Table 5).

Discussion
Thousands of newborns die each year from preventable causes such as infectious diseases, malnutrition, and accidents, despite impressive success in reducing neonatal, infant, and child mortality in Ethiopia [34]. Neonatal mortality is the most sensitive indicator of limited health care access such as institutional delivery, vaccination, medical treatment of diseases, nutrition, and hygiene [35,36].
This study found that the neonatal mortality rate in emerging regions of Ethiopia was 34.9 [95% CI: 29.8, 40.9] per 1000 live births. It was consistent with studies reported in the Jimma zone [26], and Nigeria [37]. However, it was higher than the 2016 EDHS report [13], and  Afghanistan [38]. The possible explanation could be due to the present study was undertaken in Emerging regions (Somali, Afar, Gambella, and Benishangul-Gumuz) of Ethiopia where maternal and child health care services are relatively low and economically disadvantaged in contrast to other regions [39]. Furthermore, lower vaccine coverage reduced access to healthcare, poorly urbanized, and a comparatively high incidence of childhood infectious diseases such as malaria, and acute respiratory tract infections are found in Emerging regions compared to the other regions of the country [40,41]. In the multilevel analysis; maternal education, preceding birth interval, type of birth, sex of neonate, community media exposure, number of ANC visits, and women participation in making health care decisions were significantly associated with neonatal mortality. A neonate born to mothers who do not have formal education had higher odds of neonatal mortality than a neonate born to mothers who attained secondary education and above. It is in line with studies reported in Sudan [42], Bangladesh [22], and Nigeria [16]. This may be because uneducated mothers may not have access to health information and less likely to visit maternal health care such as institutional delivery, ANC, and PNC [43,44]. Another reason is uneducated mothers are reluctant to pursue childhood vaccination [45,46] and more likely to practice prelacteal feeding [47], this could increase the risk of neonatal mortality. Besides, maternal education could result in good childhood feeding practices and have an improved awareness of common childhood disease preventive approaches that play a significant role in increasing newborn survival [48,49].
In this study, being twin birth was a significant predictor of neonatal mortality. Twin births had higher odds of death in the first month of birth than singletons.
It is consistent with the study finding in Ghana [17]. This could be since twin births are at higher risk of preterm delivery and fetal growth restriction and this could increase their risk of hypothermia, sepsis, and hypoglycemia that might increase the risk of neonatal mortality [50]. Neonates born within the preceding birth interval of 2 to 4 years had lower odds of dying within the neonatal period than those having a preceding birth interval less than 2 years. It was consistent with prior studies conducted in India [51], Afghanistan [38], and Indonesia [52]. The possible justification might be due to the reason that optimal birth spacing is vital for the health of the mother and newborn. The interbirth interval of 2 to 4 years could result in good pregnancy outcomes as women restore their nutritional and physiological loss from a previous birth, this could decrease their incidence of neonatal mortality.
Births to women who did not participate in making health care decisions had higher odds of neonatal death. This was in line with a study conducted in Bangladesh [53], it might be due to the reason that women who have participated in making health care decisions are more likely to use antenatal care service, gave birth at the health facility, and have a postnatal checkup in the early neonatal period, this could help to early identify danger signs of pregnancy and neonates and to seek medical treatment [54,55].
The odds of neonatal mortality among female neonates were lower than male neonates. This was consistent with studies reported in Indonesia [52], and Nigeria [37]. This could be due to the sex differences in genetic and biological makeup, with males being biologically weaker and more susceptible to diseases and mortality [56]. Besides, the difference in mortality might be attributed to the different protein and gene expression variation in the placenta [57]. This study found that community media exposure was a significant predictor of neonatal mortality. Newborns from the community with high media exposure had decreased odds of death in the neonatal period than neonates from the community with low media exposure. This is in line with the study done in Bangladesh [58], the possible explanation might be the reason that mothers who have media exposure had better awareness of ANC utilization, institutional delivery, and childhood illness [59].
Newborns born to mothers who had 1-3 ANC visits during pregnancy had lower odds of neonatal mortality than newborns born to mothers who did not have ANC visits during pregnancy. It was consistent with studies in Kenya [60] and India [61]. This might be because a pregnant mother who had antenatal care visits receives health care such as iron, deworming, folic acid, and tetanus immunizations, this could decrease the risk of neonatal mortality. Besides, ANC creates an opportunity for mothers and newborns to receive different interventions such as anti-D, childhood vaccinations, and nutritional supplementation.
The strength of this study was the use of multilevel modelling taking into account the clustering effect in EDHS to draw valid inferences and conclusions. This study had limitations. As this stud was a cross-sectional study, it shares the limitations of cross-sectional study design. Besides, variables such as infectious diseases, sepsis, congenital anomalies, transplacental infections, HIV status, and medication use which are considered as the most common cause of neonatal mortality were not included in this study since it was not collected in EDHS 2016.

Conclusion
Neonatal mortality in emerging regions of Ethiopia remains a major public health concern. Maternal education, women's participation in health care decision making, sex of the child, type of birth, preceding birth interval, number of ANC visits, and community media exposure were significantly associated with neonatal mortality. Therefore, empowering women in education and their autonomy in making health care decisions as well as improving access to media plays a significant role in reducing neonatal mortality in emerging regions of Ethiopia. The government should scale up maternal and child health services in these regions to reduce neonatal mortality at the national level.