Combined determinants of adverse birth outcomes in Ethiopia: an application of ecological model using Demographic and Health Survey data

Abstract

Background

Studies of determinants of adverse birth outcomes (ABOs) were conducted in Ethiopia; however, there is a lack of a single study considering the factors operating at multiple levels (individual, interpersonal, organisational, environmental and policy levels). Therefore, this study identified combined determinants of ABOs at all levels in Ethiopia by analysing the Demographic and Health Survey data guided by the Ecological model, considering that birth outcomes are shaped by the interaction between a mother’s environment and her biological and psychological health.

Objective

This study aims to identify combined determinants of ABOs at all levels in Ethiopia by analysing the Demographic and Health Survey data guided by the Ecological model.

Design

A cross-sectional study design based on interviewer-administered questionnaires was used for the respective Demographic and Health Surveys.

Setting

We used data from the 2016 Ethiopian and Demographic Health Survey, a stochastically national representative study with inclusive information on ABOs, to examine how various levels of influence from individual behaviours to environmental-level factors are affecting birth outcomes.

Participants

An effective number of 11 023 live births within the 5 years preceding the survey.

Main outcome measure

ABOs, including low birth weight and preterm birth. Multivariable multilevel mixed-effects logistic regression was used to identify determinants of ABOs through five hierarchical models in Stata V.14. Model I was the null model; models II, III, IV and V sequentially included intrapersonal, interpersonal, organisational and environmental variables, respectively. Statistical significance was determined using ORs with 95% CIs at p<0.05.

Results

The weighted prevalence of ABOs in Ethiopia is 27.0% (95% CI 25.7% to 28.3%). The final model of the multivariable multilevel mixed-effects logistic regression identified several predictors of ABOs at the intrapersonal or individual level, including maternal age of 15–24 completed years (adjusted OR (AOR)=1.24, 95% CI 1.02 to 1.51); poorest (AOR=1.41, 95% CI 1.01 to 2.00), poorer (AOR=1.42, 95% CI 1.02 to 2.01) and middle wealth quintiles (AOR=1.45, 95% CI 1.02 to 2.06); first-born twin (AOR=2.61, 95% CI 1.31 to 5.21) and second-born twin (AOR=4.05, 95% CI 2.16 to 7.61); and female childbirth (AOR=1.41, 95% CI 1.22 to 1.63). On the other hand, intimate partner physical violence (AOR=1.19, 95% CI 1.07 to 1.34) was the only factor associated with ABOs at the interpersonal level; cluster altitudes of 180–1500 m (AOR=1.28, 95% CI 1.05 to 1.55) and 2501–3455 m (AOR=1.51, 95% CI 1.15 to 1.99) were found to be an exposure of ABOs at the environmental level.

Conclusions

The prevalence of ABOs in Ethiopia is high. Factors associated with ABOs at the individual level include maternal age, wealth quintile, twin birth and female birth. In contrast, exposure variables at the interpersonal level comprise intimate partner violence, and those at the environmental level include cluster altitude. To improve ABOs and consequently reduce neonatal mortality, maternal and child health investment and future studies should act at all levels.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This study has strengths in that it applied the ecological model to analyse determinants of adverse birth outcomes at all levels, from individual to environmental-level factors.

• The study used nationally representative data with relatively inclusive information on adverse birth outcomes, and it applied advanced analysis.

• However, the study was prone to some limitations in that the size of the child for birth weight records was dependent on the maternal recall and classified as very large to very small. Thus, the study needs cautious interpretation, as it could not be free from recall bias.

• The other limitation is data incompleteness, as the Ethiopian Demographic and Health Survey data missed variables at the policy-level analysis and some variables at individual, organisational and environmental levels.

• Changes in childbirth service coverage and socioeconomic conditions since 2016 may limit the direct applicability of this study’s findings to the current context, although environmental factors such as cluster altitude are relatively stable over time.

Background

Adverse birth outcomes (ABOs) are still a major global health concern, especially in developing nations.¹ The leading causes of neonatal and infant mortality globally are ABOs, which include congenital abnormalities, stillbirth, preterm birth (PTB) and low birth weight (LBW).² Due to ABOs, over 75% of neonatal deaths happen during the first few weeks of life.³ Not only is the risk of death increased, but stunting and poor cognitive development are as well higher for children born with ABOs.⁴ There were also higher long-term risk factors for women’s premature mortality associated with all significant adverse pregnancy outcomes.⁵ Cognisant of these facts, identifying determinants of ABOs at all levels, from personal to societal and system-level risk factors, will have paramount importance in reducing those mortalities. For this reason, an ecological model has been used for exploring the aetiology of ABOs as a theoretical basis.

In low- and middle-income countries, the burden of ABOs remains high.⁶ LBW made up 5.7%, and PTB made up 7.5% in Southwestern Ontario, Canada.⁷ About 29.7% of births in sub-Saharan Africa (SSA) result in adverse outcomes.³ In Ghana, the prevalence of PTB was 24.0%, LBW was present in 27.6% of newborns and both PTB and LBW occurred in 17.4% of pregnancies.⁸ The prevalence of ABOs in Northwest Ethiopia is 26.8%,⁹ 28.3% in Debre Berhan⁶ and 31.8% in Northeast Ethiopia,¹⁰ and a lifetime prevalence of ABOs of 14.5% was reported from Gondar, Northwest Ethiopia.¹¹

Many pregnancies continue to be complicated with disastrous outcomes, despite significant efforts to reduce ABOs.¹¹ Significant efforts were made by the government and stakeholders of the Federal Democratic Republic of Ethiopia, including expanding the accessibility and coverage of health services and offering free maternal health services.¹²

Previous studies on the risk factors for ABOs produced incomprehensive results, which might have been caused by a failure to apply an all-inclusive theoretical model. However, evidence from cross-sectional, cohort and case–control studies, as well as those analysed Demographic and Health Survey (DHS) data conducted in community and health facility settings in Ethiopia, Ghana, SSA, Taiwan and Indonesia involving samples of pregnant and recently delivered mothers, has identified determinants of ABOs. These include a history of PTB or LBW, having hypertension,¹³ antepartum haemorrhage (APH), no antenatal care (ANC)310 13,16 or lack of eight contacts,^{2 4 17} multiple pregnancies,^{2 3 16} maternal anaemia (haemoglobin <110 g/L),^{2 17} mid-upper arm circumference <23 cm,¹⁰ early rupture of the membranes in the current pregnancy,^{2 16} being an employee of the government,¹⁵ no use of contraceptives, less than 2 years’ birth interval,¹⁷ rural residence,^{10 13 16 18} ages 23–34, fever,¹⁶ age ≥34, multigravidas,¹⁰ extreme maternal age,¹⁹ poverty, lack of safe water,¹⁸ parity, passive smoking,²⁰ access to a health facility, support from husband,¹¹ khat chewing,¹³ female child, secondary school education, middle-class and wealthy socioeconomic status, physical abuse (beating) by intimate partners, long-distance travel and a lack of women’s autonomy.³ In general, studies emphasised interpersonal and individual-level factors of ABOs more often than institutional, community or policy-level factors.²¹ Our review shows that these factors were not analysed at multiple levels by a single study based on an ecological model.

Determinants of ABOs have also been studied in Ethiopia; however, we lack a single study that considered the factors operating at multiple levels: individual, interpersonal, community and societal. We, therefore, analysed the Ethiopian Demographic and Health Survey (EDHS) data using the multilevel mixed-effects logistic regression analysis. The analysis was guided by the ecological model based on the assumption that birth outcomes are shaped by the interaction between a mother’s environment and her biological and psychological health. The data were used to examine how various levels of influence, from individual behaviours to societal-level factors, affect birth outcomes.

Brief overview of the ecological model

An ecological model is a multilevel, interactive approach that highlights how factors interact and are dependent on one another at every level of a health issue.²² It has long been advised that public health practice be guided by ecological models that depict the interdependent features of people and environments that underpin health outcomes.²¹ According to the ecological model, the larger community and society have a significant influence on maternal and family characteristics, which in turn affect birth outcomes.²³ It is McLeroy et al’s socioecological model that assumes people will change if the social environment is changed appropriately, and enacting environmental changes requires the support of the overall population.²⁴ The classification of factors that impact health behaviours on various levels (intrapersonal, interpersonal, organisational and environmental, as well as policy and legislative) has been made easier by ecological models.²⁵

Intrapersonal-level factors such as maternal age, female child, hypertension, woman autonomy, no ANC, APH, multiple pregnancies, haemoglobin level <110 g/L and maternal Mid Upper Arm Circumference (MUAC) <23 cm23 10 13,17; interpersonal factors such as husband support and intimate partner violence^{3 11}; organisational-level factors such as quality of ANC services, health workers’ competency, inaccessible health facilities and work stress^{3 11}; environmental factors such as passive smoking, radiation, dust and lack of safe water supply^{8 20}; and policy-level factors such as maternal leave and poverty reduction policies¹⁸ can be conceptualised using the ecological model. Their complex interplay can also be shown by applying the model. Previous studies mainly focused on intrapersonal and interpersonal-level factors; however, the public health importance of ABOs and their countereffects on neonatal and maternal mortality need comprehensive and holistic analysis of its factors at all levels. Therefore, based on the data accessibility, the conceptual framework of the ecological model to be tested as a theoretical basis is adapted for this study (online supplemental figure).

Intrapersonal risk factors are mainly biological (eg, maternal age, parity and maternal anaemia), behavioural (eg, drug use in pregnancy, substance use) and psychosocial (eg, history of LBW or PTB, maternal stress), which showed the effect of maternal factors can affect the offspring physiologically. These factors can be exacerbated by the interpersonal-level factors (eg, intimate partner violence), making the problem serious. Organisational-level factors (eg, inaccessible health facilities, health insurance) may interplay both at intrapersonal and interpersonal-level factors, which can be influenced by the environmental and policy-level factors (eg, cluster altitude, maternal leave policy). Therefore, this study has identified the interplay of combined determinants of ABOs at all levels in Ethiopia by applying an ecological model.

Methods

To model combined determinants of ABOs from individual-level to environmental-level factors in Ethiopia, we used the Births Recode file (ETBR71FL.dta) dataset from the 2016 EDHS (dataset),²⁶ a population-based cross-sectional study having national coverage conducted from 18 January 2016 to 27 June 2016. Despite the time lag between data collection and analysis, we have used these data because the most recent mini EDHS 2019 lacks ABO parameters such as gestational age and birth weight to enable robust examination of their determinants. The survey employed a two-stage cluster sampling design to select participants. The data thus inherently create a hierarchy of households within a cluster, household members within each household, interviewed women as a subset of household members and each interviewed woman’s pregnancies and children as study populations. The details of the selection process of study participants and data collection are covered in the 2016 EDHS report’s methodology section.²⁷ A total of 10 641 live births in the 5 years before the interview had a reported birth weight as study mothers reported. ABOs such as birth weights and gestational ages were recorded along with maternal sociodemographic variables and access to healthcare facilities, family planning and ANC services. Environmental and policy-level factors, such as exposure to radiation and maternal leave policy, were missed in the dataset. In this analysis, the weighted samples of 11 023 effective observations were considered. This is larger than the actual number of observations because the EDHS weights are not normalised to the sample size; weights greater than one were assigned to larger regions, such as Oromia, Amhara and the Southern Nations, Nationalities, and Peoples’ Region, to compensate for under-represented groups (online supplemental table 1).

Creating analytical dataset from 2016 EDHS data

The analytical dataset was created from the 2016 EDHS Births Recode file (ETBR71FL.SAV) (dataset)²⁶ based on inclusion and exclusion criteria. We included all live births containing outcome variables such as birth weight and gestational age at birth within the last 5 years of the survey which determined the sample size in this study, although weighted. We excluded births with missing information on key confounder maternal variables such as maternal anaemia, and excluded some irrelevant variables in the dataset for this analysis such as child vaccination in the postpartum period.

The main outcome variable, ABO, comprises LBW and PTB in 2016 EDHS data. A birth is considered an ABO if one of the mentioned outcomes is present. Based on this assumption, pregnancy duration based on pregnancy history and LBW—categorised as very big, bigger than average, average, smaller than average or very small based on mothers’ recall interview from the EDHS data (dataset)²⁶—were used. Therefore, in our analysis, smaller than average or very small were taken as LBW, and a duration of pregnancy that ends before 37 weeks of gestation was taken as preterm. Thus, a composite variable of ABOs was computed from these two parameters.

The exposure variables considered in this analysis were as follows.

Intrapersonal-level variables

Maternal age, marital status, age at first birth, role as household head, multiple pregnancies (twin child birth) or birth order within a multiple pregnancy, anaemia, socioeconomic status (wealth index), government employee, educational level, use of family planning method, khat chewing, cigarette smoking, alcohol use, gravidity and parity, birth order, sex of child, deworming drugs use in pregnancy, caesarean section as the last birth, ever had a terminated pregnancy and female child are intrapersonal-level variables analysed. However, history of LBW or PTB, hypertension, APH, maternal nutritional status, premature rupture of membrane and birth interval were the missing data in the EDHS 2016 ‘Births Recode file (ETBR71FL.dta) dataset’ (dataset).²⁶

Birth order within a multiple pregnancy refers to the sequence in which infants from the same gestation are delivered (eg, first-born or second-born twin). This is considered in the analysis that a second-born twin may be more prone to ABOs.

Interpersonal-level variables

Support from husband, intimate partner violence, women’s autonomy status, husband’s educational level and number of under-5 children.

Organisational-level variables

Inaccessible health facilities (distance) and health insurance. However, data about work-related stress, quality of ANC and health workers’ competency were missing.

Environmental-level factors

Passive smoking, stove type, lack of safe water supply, floor type, community cluster altitude, type of toilet, region and type of place of residence.

Data analysis methods

An ecological model for analysing determinants of ABOs was used to analyse the dataset at all levels except at policy level due to a lack of appropriate data. The remaining four levels, such as intrapersonal, interpersonal, organisational and environmental-level determinants and their interplay, have been analysed. To restore the survey’s representativeness and adjust for disproportionate sampling design when calculating robust SEs and accurate estimates, we weighted the data using women’s sample weight as the sampling unit and unit of analysis before any statistical analysis. Sampling weights (women’s individual sample weight/1 000 000) using ‘gen wt=V005/1000000’ command, clustering (cluster number) and stratification (ultimate area unit) were applied using the ‘svyset V001 [pw=wt], strata (V004)’ command in Stata to account for the complex survey design, hence the sample became representative of Ethiopia for generalisation of the study outcomes. Descriptive and multivariable multilevel mixed-effects logistic regression analysis was employed to determine the determinant factors of ABOs fitting five models using Stata V.14. Model I was the null model, having no independent variables, and it was used to justify whether multilevel analysis is required based on the intraclass correlation coefficient (ICC) information. As the nature of the EDHS data of birth records is nested in cluster households with 645 groups and the ICC value of 14%, a multilevel mixed-effects logistic regression analysis was justified rather than standard logistic regression to account for the effect of cluster variability. Model II contained the intrapersonal-level variables; this model was the extension for the rest of the models by adding more variables at other levels. Model III had interpersonal-level variables added to model II, model IV comprised organisational-level variables added to model III and the final model V contained environmental-level variables added to model IV. Detailed results of each model were presented in online supplemental table 2. Similarly, further analyses were conducted separately for LBW and PTB as distinct outcomes fitting models I–V for each of them (table 4). Statistical significance was assessed using ORs and 95% CIs and declared at a p value <0.05. The postestimation Akaike information criterion (AIC) and Bayesian information criterion were used to evaluate the model’s fitness and to select the final parsimonious model, and the variance inflation factor (VIF) was employed to look for multicollinearity. As all the variables in the final model exhibited a VIF of less than 5, our model was free from multicollinearity (online supplemental table 3).

Patient and public involvement

No patients or public were involved in the design of this study.

We used the Strengthening the Reporting of Observational Studies in Epidemiology cross-sectional checklist when writing our report.²⁸

Results

The results of this analysis were based on the 2016 EDHS data. Specifically, a ‘Births Recode file (ETBR71FL.dta) dataset’ (dataset)²⁶ was analysed after a composite variable (ABO) was created from birth weights and PTBs. An effective sample of 11 023 weighted observations or births in the 5 years preceding the survey was analysed.

Intrapersonal (individual-level) factors

The median age of the respondent women who gave birth in the last 5 years preceding the survey was 28.0 years (IQR 25–34), with 69.1% in the 25–39 years’ category. The majority of women (93.8%) were married at the time of data collection, and 66.1% had no education. One-seventh of women (13.8%) had the role of head of household, and half (51.9%) of them were 15–18 years old at their first birth. Regarding their substance use habit, only 0.8% smoke cigarettes, 16.4% ever chew khat and 29.5% ever take drinks that contain alcohol. About half (50.8%) of births were higher order (≥4), 2.6% were multiple births and 48.1% were female births (table 1).

Table 1. Individual-level characteristics of the DHS study population in Ethiopia, 2016.

Variables	Category	Weighted frequency	%
Maternal age in completed years	15–24	2446	22.2
	25–39	7615	69.1
	40–49	962	8.7
Current marital status	Married	10 339	93.8
	Not married*	684	6.2
Highest educational level	No education	7284	66.1
	Primary	2951	26.8
	Secondary	514	4.6
	Higher	274	2.5
Sex of household head	Male	9494	86.1
	Female	1529	13.9
Wealth index combined	Poorest	2636	24.0
	Poorer	2520	22.8
	Middle	2280	20.7
	Richer	1999	18.1
	Richest	1588	14.4
Age at first birth	15–18	5715	51.9
	19–30	5218	47.3
	31–40	90	0.8
Ever had a terminated pregnancy	No	10 056	91.2
	Yes	967	8.8
Current contraceptive use by method type	No method	7526	68.3
	Traditional method	48	0.4
	Modern method	3449	31.3
The last birth was a caesarean section	No	10 786	97.8
	Yes	237	2.2
Anaemia level (n=10 641)	Severe	158	1.5
	Moderate	745	7.0
	Mild	2322	21.8
	Not anaemic	7416	69.7
Smokes cigarettes	No	10 938	99.2
	Yes	85	0.8
Birth order	1–3	5417	49.2
	Higher order births (≥4)	5606	50.8
The child is twin	Single birth	10 731	97.4
	1st of multiple	146	1.3
	2nd of multiple	146	1.3
Sex of child	Male	5725	51.9
	Female	5298	48.1
Drugs for deworming during pregnancy (n=7589)	No	7060	93.0
	Yes	432	5.7
	Don’t know	97	1.3
Ever chewed khat	No	9216	83.6
	Yes	1807	16.4
Ever taken a drink that contains alcohol	No	7774	70.5
	Yes	3249	29.5

^*Not married includes those living with a partner without formal marriage, or never in union, widowed, divorced or separated.

DHS, Demographic and Health Survey.

Interpersonal-level factors

Close to two-thirds (68.1%) of respondent women experienced intimate partner physical violence for reasons like going outside without telling, neglecting children, arguing with their husbands, refusing sex and burning food. About 11.2% of women in this sample were autonomous for decisions related to healthcare, large household purchases, visits to family or friends or money. One-third (36.2%) of women had support from their husbands on household chores, and 47.8% of husbands had no schooling (table 2).

Table 2. Interpersonal-level variables of the DHS households in Ethiopia, 2016.

Variables	Category	Weighted frequency	%
Intimate partner physical violence	No	3512	31.9
	Yes	7511	68.1
Women autonomy (n=10 270)	Unautonomous	9123	88.8
	Autonomous	1147	11.2
Husband support on household chores (n=10 462)	No	6178	59.1
	Yes	3788	36.2
	Not living with a husband/partner	497	4.7
Husband/partner’s educational level (n=10 462)	No education	5003	47.8
	Primary	4116	39.4
	Secondary	798	7.6
	Higher	471	4.5
	Don’t know	74	0.7
Number of children under 5 in households	0	315	2.9
	1–3	10 436	94.7
	4–6	272	2.4

DHS, Demographic and Health Survey.

Organisational-level (health facilities) factors

Distance to a health facility is reported by mothers in the DHS survey as a big problem or not to them than by distance in hours or kilometres. More than half (60.6%) of women respondents experienced a big problem with distance to health facilities, while only 3.5% had health insurance coverage.

Environmental-level factors

89% of births and women respondents’ residences are rural, whereas more than half (68.1%) reside at an altitude of 1501–2500 m. The major sources of drinking water were unprotected wells or springs (30.0%), whereas 38.4% of households had no toilet facility or used unimproved type; 86.7% of houses had natural earth, dung or rudimentary floor types; and 85.0% of women respondents used wood, straw or crop residues as cooking fuel (table 3).

Table 3. Environmental-level factors of adverse birth outcomes from DHS survey in Ethiopia, 2016.

Variables	Category	Weighted frequency	%
Type of place of residence	Urban	1216	11.0
	Rural	9807	89.0
Cluster altitude in metres	180–1500	1946	17.7
	1501–2500	7506	68.1
	2501–3455	1571	14.2
Region	Tigray	716	6.5
	Afar	114	1.0
	Amhara	2072	18.8
	Oromia	4851	44.0
	Somali	508	4.6
	Benishangul	122	1.1
	SNNPR	2296	20.8
	Gambela	27	0.3
	Harari	26	0.2
	Addis Ababa	244	2.2
	Dire Dawa	47	0.5
Source of drinking water	Piped water	995	9.0
	Public standpipe	1936	17.6
	Protected well/spring	3097	28.1
	Unprotected well/spring	3307	30.0
	Surface water	1363	12.4
	Rainwater	88	0.8
	Truck/cart tanker	66	0.6
	Bottled water	10	0.1
	Other sources	161	1.4
Type of toilet facility	Flush type	212	1.9
	Pit latrine with slab	754	6.9
	Pit latrines without a slab	5703	51.7
	ventilated improved pit latrines or compost latrines	120	1.1
	No toilet facility or unimproved type	4234	38.4
Main floor material	Natural, earth or dung, or rudimentary	9557	86.7
	Wood or bamboo	205	1.8
	Asphalt, cement or ceramic tiles	715	6.5
	Carpet or other type	546	5.0
Type of cooking fuel	Electricity	305	2.8
	LPG, natural or biogas	38	0.3
	Kerosene	7	0.1
	Charcoal	498	4.5
	Wood, straw, crop residues	9372	85.0
	Animal dung	639	5.8
	No cooking in a house or other	164	1.5

DHS, Demographic and Health Survey; LPG, liquefied petroleum gas, others include not a dejure household; SNNPR, Southern Nations, Nationalities, and Peoples’ Region.

Adverse birth outcomes

Based on the 2016 EDHS data, the weighted prevalence of ABOs in Ethiopia was 27.0% (95% CI 25.7% to 28.3%), and regional variations ranged from 18.7% (95% CI 16.2% to 21.6%) least in Benishangul to 53.1% (95% CI 49.6% to 56.5%) highest prevalence in Afar (figure 1). Specifically, the weighted prevalence of LBW in Ethiopia was 26.0%, while that of PTB was 1.6%.

Figure 1. Weighted prevalence of ABOs by regions of Ethiopia, 2016. ABOs, adverse birth outcomes; SNNPR, Southern Nations, Nationalities, and Peoples’ Region.

Combined determinants of ABOs

In the multivariable multilevel mixed-effects logistic regression’s final model, maternal age of 15–24 completed years (adjusted OR (AOR)=1.24, 95% CI 1.02 to 1.51); poorest (AOR=1.41, 95% CI 1.01 to 2.00), poorer (AOR=1.42, 95% CI 1.02 to 2.01) and middle wealth quintiles (AOR=1.45, 95% CI 1.02 to 2.06); first-born twin (AOR=2.61, 95% CI 1.31 to 5.21) and second-born twin (AOR=4.05, 95% CI 2.16 to 7.61); and female childbirth (AOR=1.41, 95% CI 1.22 to 1.63) were associated with ABOs at intrapersonal or individual level. Furthermore, intimate partner physical violence (AOR=1.19, 95% CI 1.07 to 1.34) was the only predictor of ABOs at the interpersonal level, and cluster altitudes of 180–1500 m (AOR=1.28, 95% CI 1.05 to 1.55) and 2501–3455 m (AOR=1.51, 95% CI 1.15 to 1.99) are exposure variables to ABOs at the environmental level. In separate analyses of factors associated with LBW and PTB, the determinants of LBW were largely similar to those of ABOs overall. In contrast, the factors associated with PTB differed somewhat, although urban residence (AOR=4.55, 95% CI 1.75 to 11.80) emerged as an additional factor uniquely associated with PTB at environmental level. There are no determinant factors found statistically significant at organisational and policy levels (table 4).

Table 4. Multivariable multilevel mixed-effects logistic regression’s final model to identify the combined determinants of adverse birth outcomes in Ethiopia; analysis of the 2016 EDHS data based on the ecological model.

Variables	Low birth weight		Preterm birth		Adverse birth outcomes
	AOR	95% CI	AOR	95% CI	AOR	95% CI
Fixed effects
Maternal age in completed years*
15–24	1.25	(1.02 to 1.52)†	1.42	(0.71 to 2.83)	1.24	(1.02 to 1.51)†
25–39	1		1		1
40–49	1.06	(0.73 to 1.54)	0.90	(0.26 to 3.18)	1.05	(0.73 to 1.51)
Sex of household head
Male	1		1		1
Female	1.10	(0.86 to 1.42)	0.40	(0.14 to 1.13)	1.06	(0.82 to 1.37)
Wealth index*
Poorest	1.45	(1.03 to 2.04)†	1.07	(0.38 to 3.04)	1.41	(1.01 to 2.00)†
Poorer	1.43	(1.01 to 2.05)†	1.32	(0.45 to 3.83)	1.42	(1.02 to 2.01)†
Middle	1.46	(1.03 to 2.06)†	1.53	(0.48 to 4.83)	1.45	(1.02 to 2.06)†
Richer	1.09	(0.78 to 1.52)	0.86	(0.29 to 2.53)	1.10	(0.77 to 1.51)
Richest	1		1		1
Anaemia level
Severe	1.40	(0.85 to 2.32)	0.18	(0.01 to 1.69)	1.28	(0.78 to 2.10)
Moderate	1.08	(0.77 to 1.52)	0.29	(0.04 to 2.15)	1.02	(0.73 to 1.41)
Mild	1.17	(0.96 to 1.44)	0.72	(0.27 to 1.93)	1.16	(0.96 to 1.41)
Not anaemic	1		1		1
A child is twin birth*
Single birth	1		1		1
First-born twin	2.21	(1.21 to 4.05)†	9.43	(1.99 to 44.63)†	2.61	(1.31 to 5.21)†
Second-born twin	3.37	(1.97 to 5.76)‡	10.53	(2.48 to 44.64)†	4.05	(2.16 to 7.61)‡
Sex of child*
Male	1		1		1
Female	1.48	(1.27 to 1.71)‡	0.92	(0.50 to 1.69)	1.41	(1.22 to 1.63)‡
Ever taken an alcoholic drink
No	1		1		1
Yes	1.17	(0.96 to 1.42)	0.55	(0.27 to 1.11)	1.11	(0.92 to 1.35)
Husband support on HH chores
No	0.93	(0.78 to 1.11)	0.63	(0.30 to 1.35)	0.94	(0.68 to 1.39)
Yes	1		1		1
Not living with husband/partner	0.94	(0.61 to 1.45)	1.44	(0.33 to 6.26)	1.01	(0.80 to 1.27)
Intimate partner physical violence*
No	1		1		1
Yes	1.08	(0.91 to 1.27)	1.01	(0.55 to 1.84)	1.19	(1.07 to 1.34)‡
Distance to a health facility
Big problem	0.99	(0.84 to 1.19)	1.09	(0.53 to 2.23)	1.01	(0.85 to 1.21)
Not a big problem	1		1		1
Type of place of residence
Urban	1		4.55	(1.75 to 11.80)†	1
Rural	1.18	(0.83 to 1.68)	1		1.08	(0.76 to 1.54)
Cluster altitude in metres*
180–1500	1.28	(1.05 to 1.57)†	1.12	(0.53 to 2.38)	1.28	(1.05 to 1.55)†
1501–2500	1		1		1
2501–3455	1.48	(1.12 to 1.96)†	1.21	(0.45 to 3.29)	1.51	(1.15 to 1.99)†

Number of effective observations for the null model=11 023, for the final model=9469; group variable is the cluster number. Number of groups for the null model=643, for the final model=640.

^*Statistically significant variables.

^†P<0.05.

^‡P<0.001.

AOR, adjusted OR; EDHS, Ethiopian Demographic and Health Survey; HH, household.

The initial ICC of the ABO’s null model is 0.14, which is >0.10, suggesting that a significant portion of the variance in ABOs is due to differences between the clusters, indicating that a multilevel approach is justified.²⁹ As the ICC of the ABO’s final model is 0.124, this indicates that 12.4% of the observed total variance in ABOs is occurring at the cluster level (table 5).

Table 5. Estimation methods and model fitness statistics of all the five models in multilevel logistic regression to identify determinants of adverse birth outcomes in Ethiopia to compare them and identify the most parsimonious one.

Model	ICC	AIC	BIC	Log-likelihood	LR χ2 (P value)
1	0.1405	12 222.75	12 237.30	−6109.373	435.3 (<0.001)
2	0.1219	11 429.01	11 580.56	−5693.505	286.8 (<0.001)
3	0.1264	10 712.65	10 870.07	−5334.323	267.19 (<0.001)
4	0.1263	10 714.60	10 879.19	−5334.301	266.77 (<0.001)
5	0.1246	10 706.01	10 885.15	−5327.005	250.43 (<0.001)

AIC, Akaike information criterion; BIC, Bayesian information criterion; ICC, intraclass correlation coefficient; LR, likelihood ratio.

Compared with the other models, model V is the most parsimonious because its AIC value is lower than other models, and it was the final model with all level factors combined.

Discussion

The prevalence of ABOs in Ethiopia is 28.2%, as per this study. This finding was comparable with the study that analysed DHS data of the 10 SSA countries and reported a pooled prevalence of ABOs, which was 29.7%.³ Similarly, comparable findings were reported from different parts of Ethiopia: 26.8% in Northwest Ethiopia,⁹ 28.3% in Debre Berhan⁶ and 31.8% in Northeast Ethiopia.¹⁰ However, the prevalence in Ethiopia is higher than in Southwestern Ontario, Canada, where LBW made up 5.7%, and PTB made up 7.5%.⁷ This difference could be attributed to the differences between the two countries in terms of their level of socioeconomic status and the advancement of their healthcare system.

Several factors at the individual level are associated with ABOs. Women in the age group 15–24 years have 1.31 times higher odds of experiencing ABOs compared with women in the middle age group (24–39 years). However, there is no difference between women in the upper age group (40–49 years) and the middle age group. Similarly, studies from Gondar in Northwest Ethiopia reported that the lower age of 23–34 years is a risk factor,¹⁶ and another study, on the other hand, from the North Wollo zone in Northeast Ethiopia revealed that maternal age ≥34 is the risk factor for ABOs.¹⁰ Additionally, a study from Taiwan reported that ABOs were high at extremes of maternal ages <26 and >30.¹⁹ The other two retrospective cohort studies from China and Italy reported that young maternal age and maternal age over 40 years have been associated with increased risks of adverse pregnancy outcomes.^{30 31} This association is explained by the physiological immaturity among young mothers, which may lead to cervical insufficiency, and older women may experience chronic conditions like hypertension, which may lead to ABOs. Therefore, attention to maternal age extremes should be the concern of the healthcare system.

Women exhibiting the poorest, poorer and middle categories of the wealth quintile when compared with the richest category have higher odds of ABOs. Analogously, a study analysing DHS in Ghana for contextual risk factors of LBW reported that living in poverty is the risk factor.¹⁸ Another study analysed recent DHS data from 10 SSA countries revealed that the odds of ABOs decreased among women with middle-class and wealthy socioeconomic status.³ Low socioeconomic status can affect women’s access to healthcare, living conditions and stress levels. As a result, a couple of effects of lack of access to quality healthcare and stress could lead to preterm delivery by affecting the pregnancy physiologically through increased cortisol levels.³²

Twin birth is the risk factor in this study as the first-born and second-born twins had higher odds of ABOs compared with singleton births. Likewise, twin birth was reported as being associated with ABOs by a study that analysed the DHS of 10 SSA countries.³ In addition, two other studies from the Bale zone of the Oromia region in Southeast Ethiopia and Gondar in Northwest Ethiopia reported multiple pregnancies as a risk factor for ABOs.^{2 3 16} The effect of twin pregnancies can be justified by their physiological demands as they are two in a limited uterine space, and this could lead to their limited growth. A comprehensive ANC follow-up would solve the problem.

In this study, female birth was 1.33 times more likely to have ABOs compared with male birth. Being a female child was also a determining factor of ABOs in a study of DHS in SSA countries.³ Although ABOs can happen among both fetal sexes, the observed association of female birth with ABOs may be due to maternal factors and/or methodological issues rather than inherent biological vulnerability. Therefore, this finding should be interpreted as associative rather than causal. In the settings where fetal sex is identified during the prenatal period, this association may also be influenced by sociocultural preferences, as maternal experiences and behaviours during pregnancy may vary according to fetal sex.

Women who experienced intimate partner physical violence for different reasons had higher odds of ABOs compared with their counterparts. Similarly, physical abuse by intimate partners was reported by a study in SSA countries.³ Being beaten during pregnancy can harm the fetus in a beaten womb, and the membrane may rupture prematurely, so PTB is likely to occur. Legal grounds for intimate partner violence control and husband counselling may solve the problem.

Lastly, women residing at low and high cluster altitudes of 180–1500 m and 2501–3455 m, respectively, when compared with altitudes of 1501–2500 m, had higher odds of experiencing ABOs. According to meta-analyses, for every 1000 m of elevated maternal altitude above sea level, the average birth weight decreased by 96.98 g.³³ As per a study from Peru, LBW was more likely to occur in children whose mothers lived between 2500 and 3499 m and above 3500 m above sea level.³⁴ High altitude has less oxygen and can lead to maternal and fetal hypoxia.³⁵ Therefore, this finding suggests that there should be altitude-sensitive maternal and child healthcare services designed.

Although all associations coincide between stratified analyses of factors associated with LBW, PTB and composite ABOs, urban residence emerged as an additional factor uniquely associated with PTB at environmental level. This finding is supported by a study that analysed health and demographic surveillance system data from Bangladesh, which reported that a substantial proportion of PTBs contributed to neonatal deaths in urban slum settings.³⁶ Conversely, this finding is inconsistent with another study conducted in Ethiopia, which reported that urban residence was a protective factor associated with PTB.³⁷ The observed association between urban residence and PTB may reflect increased exposure to environmental pollution, psychosocial stressors, health facilities overcrowding and socioeconomic inequalities within urban settings. In densely populated or resource-constrained urban areas, women may be particularly exposed to air pollution, psychosocial stress and lifestyle-related risks during pregnancy.

This study has strengths in that it applied the ecological model to analyse determinants of ABOs at all levels, from individual to environmental-level factors. It used nationally representative data with relatively inclusive information on ABOs, and it applied advanced analysis; however, it was prone to some limitations. The size of the child for birth weight records was dependent on the maternal recall and classified as very large to very small. Thus, the study needs cautious interpretation, as it could not be free from recall bias. The other limitation is the data incompleteness as the DHS data missed variables at policy-level analysis and some variables at the individual level like hypertension, APH and Premature Rupture of Membrane (PROM); at an organisational level like service quality and work stress; and at environmental level like passive smoking. Changes in childbirth service coverage and socioeconomic conditions since 2016 may limit the direct applicability of this study’s findings to the current context, although environmental factors such as cluster altitude are relatively stable over time. Also, certain exposures such as wealth quintile, intimate partner violence (IPV) status, media exposure and contraceptive use were measured at the time of the survey rather than at the time of index pregnancy. This may lead to temporal misclassification if household socioeconomic status or living conditions changed during the survey period. Additionally, as the DHS is cross-sectional and retrospective, temporal ordering cannot be fully guaranteed and residual confounding or reverse causality cannot be fully excluded.

Conclusion

Several factors have been identified as determinants of the high prevalence of ABOs in Ethiopia. These factors include maternal age, wealth index, twin births and female birth at the individual level; intimate partner violence at the interpersonal level; and cluster altitude at the environmental level. This study did not identify any organisational or system/policy-level factors. Maternal and child health investments and future research should be altitude sensitive and take place at all levels to improve ABOs and thereby lower neonatal mortality.

Data availability statement

Data are available in a public, open access repository.