Association between out-of-pocket health expenditures and low birth weight in Eastern Ethiopia: a generalized structural equation modeling (GSEM)

Abstract

Background

Globally, approximately 15% to 20% of newborns are born with low birth weight (LBW), with over 90% of these cases occurring in low- and middle-income countries (LMICs). Although previous research on LBW has largely focused on clinical and nutritional factors, economic barriers associated with LBW remain under-researched. This study aimed to assess the association between out-of-pocket (OOP) payment for antenatal care and LBW in Eastern Ethiopia.

Methods

A prospective cohort study followed pregnant women for ten months to examine the incidence of LBW. The cost of ANC and other follow up variables were collected during pregnancy. Direct medical and non-medical costs were summed to calculate total OOP expenditures. Face to face interviews were used to collect baseline and follow-up data. Poisson regression with robust variance was used to assess the independent predictors of LBW. Adjusted risk ratios (aRR) with 95% confidence intervals (CI) were computed. The direct and indirect association between OOP and LBW were estimated using Generalized Structural Equation Modeling (GSEM).

Results

A total of 385 women was followed for 10 months. The study found that 10.9% of women gave birth to LBW neonates. After controlling for confounding factors, OOP expenditure (aRR = 3.21, 95% CI: 1.19, 8.64), prenatal depression (aRR = 2.91, 95% CI: 1.65, 5.13), and lack of birth preparedness and complication readiness (BPCR) (aRR = 4.12, 95% CI: 1.52, 11.20), poor wealth status (aRR = 3.30, 95% CI: 1.16, 9.38), incomplete ANC visits (aRR = 2.37, 95% CI: 1.01, 5.53), unplanned pregnancy (aRR = 1.92, 95% CI: 1.14, 3.22) and long travelling time (1.99, 95% CI: 1.15, 3.44) were significantly associated with LBW. In GSEM, prenatal depression (β = 1.30 (95% CI: 0.21, 2.80) and lack of preparation for birth (β = 1.55 (95% CI: 0.29, 2.80) mediated the association between LBW and OOP expenditures, while ANC visits mediated the association between long travelling time and LBW (β = 1.04, 95% CI: 0.04, 1.05).

Conclusion

There was a significant positive association between OOP payment and LBW which was partly mediated by prenatal depression and lack of BPCR. To reduce the incidence of LBW, an integrated approach should be adopted that combines financial risk protection, psychosocial support and geographical accessibility of services.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-025-26091-9.

Introduction

Globally, around 15%-20% of newborns are affected by low birth weight (LBW). This is more than just a statistic, it shows neonates who are at higher risk of death within the first month of life [1]. Those who survive their first month of life are more likely to experience stunted growth, developmental delays, and a greater risk of chronic diseases in their adulthood period [2, 3]. Over 90% of LBW cases occur in low- and middle-income countries (LMICs), with the highest burden in Southern Asia and sub-Saharan Africa (SSA) [4]. This is supported by research by Okwaraji et. (2024) showing these two regions share three quarters of the global LBW [5].

While appropriate antenatal care (ANC) can reduce the risk of LBW, ANC utilization in SSA remains low [6]. Therefore, improving the accessibility of ANC in terms of its availability, quality and affordability has the potential to reduce the risk of pregnancy complications and improve birth outcomes. The World Health Organization (WHO) emphasizes that pregnant women must be able to access the right care at the right time [7]. However, poor healthcare financing systems significantly affect the maternal service utilization of pregnant women [8]. In many parts of the region, out-of-pocket (OOP) payments for maternal healthcare exceed 50% of total health spending, especially in East Africa [9]. For families in low-income settings and impoverished households, these OOP expenses can significantly hinder access to healthcare services [10, 11], which leads to adverse birth outcomes.

Women often travel more than one hour to reach a health facility [12, 13], and transportation can be costly [14]. Moreover, due to supply shortages in health facilities, women frequently have to pay for medication, laboratory services, consultation and other informal fees to access ANC [15]. When OOP spending is high, many women, particularly those from poor socioeconomic backgrounds, are forced to delay, discontinue or entirely miss ANC [16]. Financial hardship worsened by OOP costs may also lead them to compromise other aspects of essential prenatal care, such as proper nutrition and other key interventions critical for a healthy pregnancy and positive birth outcomes [17]. Due to OOP costs for formal healthcare services, some women seek care from traditional birth attendants for ANC and childbirth. This reliance on informal care can lead to increased risk of complications during pregnancy and childbirth such as such as LBW [18].

Financial hardship can lead to underutilization of ANC visits [19], and insufficient ANC has also been recognized as a risk factor for adverse birth outcomes [20], and it is reasonable that the OOP costs due to ANC might directly or indirectly increase the risk of adverse birth outcomes such as preterm birth, LBW, and stillbirth. However, this association has not been thoroughly explored in empirical studies. This research is among the first to empirically investigate the association between ANC-related expenses and adverse birth outcomes in resource limited settings. By analysing longitudinal data from resource limited settings, it seeks to explore the association and offer evidence to guide strategies and policies for maternal health financing.

Although previous research on LBW has largely focused on maternal, clinical and nutritional factors, economic barriers remain under-researched. In assessing the association between dependent and independent variables, a usual and conventional method manages all independent variables conceptually as confounders. However, some variables lie on the causal pathway between cost and adverse birth outcomes and conceptually are mediators rather than confounders. Treating these mediators as confounders under-estimates the total causal effect of cost on adverse birth outcomes. Given that ANC completion, prenatal depression, and birth preparedness are influenced by OOP expenditure [21–23] and are also established risk factors for LBW [22, 24] these variables should be treated as mediators in the analysis. Our study aims to assess the direct and indirect association between OOP and LBW; hence mediation analysis was employed to assess the extent to which OOP expenditure associated with LBW was transmitted through these pathways, using Generalized Structural Equation Modeling (GSEM). The findings will serve as an input for assessing the effectiveness of existing policies and resource allocation to reduce the financial burden on families, ultimately contributing to improved maternal and neonatal health outcomes.

Hypothesis of the study

H0: OOP health expenditures have no direct or indirect association with LBW.

Ha: OOP health expenditures are directly or indirectly associated with LBW.

Methods

Study settings and design

The primary data was collected from the Kersa Health and Demographic Surveillance System (KHDSS). The KHDSS is located in the Kersa District of the Eastern Hararge Zone, Oromia Region, in eastern Ethiopia. KHDSS was Established by Haramaya University in 2007 to monitor demographic and health changes within the community and to record key demographic events. KHDSS uses a population-based, open-cohort longitudinal Study in which individuals residing within the catchment area are followed and surveyed every six months. A total of 19 healthcare facilities; including 4 health centres, 10 health posts and 5 clinics are found in KHDSS [25]. Hiwot Fana Specialized University Hospital is the only specialized and referral center for the population in this area. The hospital has over 230 beds and a maternity unit that attends approximately 5,808 deliveries annually [26]. Additional information about the KHDSS project, including its establishment and the nature and scope of the project has been published previously [27].

Population

The source population was all pregnant women residing in KHDSS and the study population was all pregnant women participating in the ongoing KHDSS study during the study period. We excluded women residing in the study area for less than six months as the KHDSS only collects data from permanent residents.

Sample size determination

The double population proportion formula was used to determine the sample size, based on variables associated with LBW identified in a study conducted in Eastern Ethiopia [28]. The following assumptions were applied: 80% power, 95% confidence level, a 1:1 ratio of exposed to unexposed, and the proportions of LBW among the non-exposed group (P1) and exposed group (P2). Using these assumptions, the sample size was calculated using Epi Info version 7, resulting in a total of 232 participants (the largest sample size was selected). Because the study was conducted at the community level, a design effect of 1.5 was applied. Additionally, considering the prospective cohort design, a 15% non-response rate was added. The final sample size for the study was therefore 410 (S1 file).

Study design and recruitment

This specific study followed participants over a 10-month period, from September 15, 2023, to June 30, 2024. During this time, participants were interviewed three times: the first two interviews took place before delivery, and the women were contacted for the third time to collect birthweight of the neonates within 72 h of delivery.

From September 15 to December 15, 2023, 394 pregnant women with a gestational age (GA) of less than six months (to minimize recall bias in reporting ANC-related costs) were recruited and interviewed. The final sample size of 394 was slightly lower than the calculated sample size of 410 because the total number of pregnant women available during the study period was limited. All eligible pregnant women present during the study period were included in the study.

Baseline information and ANC attendance, timing, and costs were collected. Participants’ addresses and estimated delivery dates were also documented to trace the women for the next round of data collection. Then, women were contacted for the second time from April 1 to June 30, 2024. This was during the last month of pregnancy and collected cost for ANC visits experienced after the first round of data collection. Data on BPCR and prenatal depression were also collected at this second interview. Women were contacted for the third time after delivery to collect birth outcomes, including LBW.

Variables

Dependent variables

Incidence of LBW was the dependent variable of this study. Birth weights under 2500 grams were considered as “LBW” and those ≥ 2500 grams were considered as “normal birth weight” [29, 30]. Data collectors used two methods to assess birth weight. For women who gave birth at a health facility, weight measured after delivery was recorded. For home births attended by trained traditional birth attendants (TTBA), birth weight was measured at home since TTBA were initially trained on birthweight measurement by the project. For births attended by untrained traditional birth attendants (TBAs), data collectors encouraged women to visit a health facility and record birth weight. Birth weight was not recorded for women who moved to other areas after delivery or were not found at home; these were recorded as lost to follow-up (LTFU).

Exogenous and mediator variables

Cost (OOP) was considered as an exogenous variable since it is conceptualised as directly affecting the outcome variables and indirectly affecting the outcome through mediator variables. OOP health expenditure was the summation of direct medical and non-medical costs [31]. OOP cost includes payments for medical services like medications, ultrasounds, and laboratory tests, as well as non-medical expenses such as transportation or accommodation across all ANC visits. For analysis, OOP was categorised into two categories. Women who did not make any payments for direct medical or non-medical costs were classified as “No OOP” and women who paid for direct and nonmedical costs were categorised as “OOP”. The costs in Ethiopia’s local currency (birr) were converted to US dollars based on the 2023/2024 exchange rate [32].

To estimate the direct and indirect associations between OOP payments and LBW, two variables were used as mediators: prenatal depression (yes/no), and BPCR (prepared/not prepared) (Table 1). These mediators are considered endogenous, as they may be influenced by ANC-related costs [21–23], and they are also risk factors for LBW [22, 24]. Review of the existing literature was used to ensure that mediators met key considerations that there must be sufficient evidence that mediators are affected by the exogenous variable and are also associated with the outcome [21–24]. The timing of variable measurement must ensure that mediators follow the exogenous variable. The overall OOP (across all ANC visits) was used as the exogenous variable, allowing BPCR and prenatal depression as mediators, all measured in late pregnancy. For other confounding factors like wealth status and travelling time, ANC completion was used as the mediator. This process ensured chronological precedence of confounding before mediator development. ANC completion was not included as a mediator with overall OOP, as total cost was collected at the same time as ANC completion, making causal interpretation invalid. Prenatal depression was assessed using the Edinburgh postnatal depression scale (EPDS) assessment tool. The tool was originally developed to assess postnatal depression but is now widely used for screening depressive symptoms during pregnancy [33]. The EPDS has 10 items, and each item measured from 0 to 3 (never, rarely, sometimes, most of the time), with a maximum score of 30 and a minimum score of zero. Women with EPDS ≥ 12 were considered having clinically significant prenatal depression [34, 35]. The tool has been standardised, validated and used in many LMICs such as Ethiopia (with specificity and sensitivity of 78.9% and 75.3%, respectively) [36]. The tool was administered during the third trimester of pregnancy, consistent with standard practice for assessing antenatal depressive symptoms and aligned with the study’s conceptual framework. The EPDS was administered by the project’s regular data collectors, all of whom received specific training on the use of the EPDS tool before data collection began.

Table 1.

Definition of exogenous, mediators and confounding variables, KHDSS 2023/24

Variables	Definition/measurement	Coding	Phase of data collection	Types of variables
Age of the women	• Adolescent for women aged ≤ 19 years • Adult for women aged 20–49 years	• 0: adolescent • 1: Adult	Phase 1	Confounding
OOP	• The summation of direct medical and nonmedical costs across all ANC visits. Women who did not make any payments were classified as “No OOP” and women who paid for direct and nonmedical costs were categorised as “OOP”.	• 0: No • 1: Yes	Phase 1 and 2	Exogenous
BPCR	• Measured by asking six questions on the last ANC visit (prepared/not prepared)	• 0: Prepared • 1: Not prepared	Phase 2	Mediator
Prenatal depression	• Measured using EPDS tool, EPDS ≥ 12 indicates the presence of prenatal depression and EPDS < 12 indicate absence of depressive symptoms.	• 0: No • 1: Yes	Phase 2	Mediator
ANC completion	• ≥ 4 ANC visits “complete’ • < 4 ANC visits “incomplete”	• 0: Complete • 1: Incomplete	Phase 1 and 2	Mediator and Confounding
Marital status	• Married • Divorced	• 0: Divorced • 1: Married	Phase 1	Confounding
Residence	• Urban • Rural	• 0: Urban • 1: Rural	Phase 1	Confounding
Wealth Index	• Poor • Medium • Rich	• 0: Rich • 1: Medium • 2: Poor	Phase 1	Confounding
Occupation	• Employed • Unemployed	• 0: Employed • 1: Unemployed	Phase 1	Confounding
Educational status	• Literate- who can read and write • Illiterate- who cannot read and write	• 0: Literate • 1: Illiterate	Phase 1	Confounding
Plan for pregnancy	• Planned • Unplanned	• 0: Planned • 1: Unplanned	Phase 1	Confounding
Parity	• Parity was measured by asking women “number previous livebirths”	• 0: One • 1: Two-four • 2: > Four	Phase 1	Confounding
Travelling time	• Travel time from home to the health facility	• 0: ≤ 2 h • 1: > 2 h	Phase 1 and 2	Confounding
Length of health facility stay	• The time spent in health facility (summation of waiting and service time)	• 0: ≤ 2 h • 1: > 2 h	Phase 1 and 2	Confounding
LBW	• Birth weights < 2500 grams considered as “LBW” and those ≥ 2500 grams as “normal birth weight”	• 0: Normal • 1: LBW	Phase 3	Outcome

ANC Antenatal care, BPCR Birth preparedness and complication readiness, EPDS Edinburgh Postnatal Depression Scale, OOP Out of pocket

Additionally, since the latest WHO recommendation of eight or more ANC visits [37] has not fully been implemented in Ethiopia [38], ANC was categorized as complete (≥ 4 visits) and incomplete (< 4 visits). BPCR was measured by six standard questions on the last ANC visit. (I) identified place of delivery, (II) saved money, (III) skilled birth attendant identified, (IV) mode of transport arranged, (V) essential items prepared, (VI) identified blood donors [39]. Women who responded “yes” to at least four of these questions were categorized as “well prepared” and all others were classified as “not well prepared” (Table 1). Although there is no globally validated tool that measures “birth preparedness” and “complication readiness” as two separate independent constructs, the BPCR framework conceptualizes these components as interlinked and mutually reinforcing. Therefore, the indicators used in our study capture both preparation for birth and readiness for potential complications. Items such as saving money, arranging transport, and identifying blood donors specifically reflect complication readiness, as they address preparedness for emergency response. Meanwhile, identifying a place of delivery, preparing essential items, and identifying a skilled attendant reflect birth preparedness. Collectively, these items provide a comprehensive assessment of BPCR, including readiness for obstetric complications.

Length of time spent at the health facility was measured based on the total duration women spent waiting for and receiving ANC services. For each ANC visit, data collectors recorded the time spent in the facility. The durations from all ANC visits were then summed to generate a total time spent across the pregnancy. Finally, this total was categorized into two groups: >2 h and ≤ 2 h, based on prior literature. Different confounding variables were also considered in the study and summarized in Table 1.

Data collection tool and procedure

The data collection instrument consisted of two components. The first component included variables that were originally part of the KHDSS data collection tool, which has been previously published [40]. The second component comprised variables that were not included in the original project tool and was developed by the principal investigator and integrated into the main data collection instrument. These additional components were costing tools, ANC depression, birth preparedness and complication readiness, and catastrophic health expenditure. The tool for OOP was adapted from the OOP health expenditure tool used by the DHS and SSA countries Living Standard Survey [41, 42] (S3 file). The data was collected through face-to-face interviews by using a tablet computer with Open Data Kit (ODK) application at the community level.

Statistical analysis

The data were analysed using STATA version 18. Descriptive and summary statistical analysis was performed to provide an overview of the data. The Mann-Whitney U test was employed to assess the median difference in the cost of ANC utilization between women who gave birth to a LBW baby and women who had a baby of normal birth weight.

Since a prospective cohort study design was used to assess the incidence of LBW, a Poisson regression with robust variance estimation was fitted to identify predictors of LBW. Initially, we attempted to use log-binomial regression, which is the preferred model for binary outcomes in cohort study, as it directly estimates risk ratios (RRs). However, one of the main limitations of log-binomial regression is its tendency to experience convergence issues, particularly when the model is not well-fitted to the data [43, 44]. Due to the convergence problem of log-binomial regression in the analysis, Poisson regression with robust variance estimation was employed as an alternative model [43]. This model is widely recommended when dealing with binary outcomes in cohort studies, especially when the outcome is not rare (i.e., incidence > 10%) [45]. Adjusted RR with 95% confidence intervals was computed, and statistical significance was declared when the p value < 0.05 in the final model.

Model Building for generalized structural equation model

To examine the mediation effects between covariates and LBW, GSEM was employed using STATA version 18 “gsem” package. To run a GSEM, two steps were undertaken [46]. First, to include exogenous and mediator variables in GSEM, there must be a statistically significant association with the outcome variables in the multivariable Poisson regression with robust variance analysis. Second, to run GSEM, all endogenous and outcome variables should be binary outcomes and were fitted using the appropriate link function and family. Different functions were used, and the best model was selected based on their AIC values [47]. Logit function with the binomial family were selected as the best model.

The analysis started with the proposed hypothetical model. The graphical representation of hypothetical paths illustrates the interactions between exposure, mediators and outcome variable. Only variables that showed a significant association with the outcome in the multivariable analysis were included in the path analysis. This hypothetical graphical model was primarily used to examine the direct effects of each exposure variable on both the mediators and the outcome, as well as the effects of the mediators on the outcome (Fig. 1). Results are presented as coefficients with 95% CI. Then, exposure and mediators were run using “nlcom” command to see the indirect and total effect on outcome variables and proportion of mediation. Mediation is classified into two types: complete and partial mediators. In complete mediation, the entire effect of exposure variables on outcome variables is transmitted through the proposed mediators [48]. The OOP has no direct effect on LBW in this case (entirely indirect effect on outcome variable). In partial mediation, the OOP has both direct and indirect effects on the outcome variables, and the direct effect has not been fully mediated by mediators [49]. The hypothesis was (Null hypothesis (H₀): The selected variables do not mediate the association between OOP and LBW. Alternative hypothesis (Hₐ): the selected variables mediate the association between OOP and LBW. In mediation testing, H_a is accepted if the OOP is associated with the mediators (effect a) and if the mediators are associated with the outcome variable (effect b). If the mediator variables have no association with the outcome variable, OOP has direct association (effect c) with outcome variable (in this case direct effect = total effect). Finally, the proportion of mediation for each mediator was calculated using ab/ (c + ab) (Fig. 1).

Fig. 1.

Hypothesized relationship between OOP and LBW using directed acyclic graphs (DAGs) version 3.1

Results

Sociodemographic characteristics of the participants

A total of 394 women were recruited to the cohort. Four women experienced stillbirth and the birth weight of five newborns were not recorded (lost to follow-up), leaving 385 (97.5%) women with complete birth outcome data included in the analysis. The mean age of women was 26.1 (SD 6.7) years. One quarter (24.9%) of participants were adolescent women, and the majority (92.2%) of the women were from rural areas. Most women were from poor households, only 14.3% were employed, and 56.6% were illiterate (Table 2).

Table 2.

Sociodemographic characteristics of pregnant women in KHDSS, Eastern Ethiopia, KHDSS 2023/24

Variables	Total n = 385	%
Age of mother
Adolescent	96	24.9
Adult	289	75.1
Place where mother live
Rural	359	93.2
Urban	26	6.8
Wealth index of household of mother
Poor	160	41.6
Medium	154	40.0
Rich	71	18.4
Education status of mother
Illiterate	216	56.6
Literate	169	43.9
Sex of children born
Female	189	49.1
Male	196	50.9
Occupation of mother
Employed	55	14.3
Unemployed	330	85.7
Marital status of mother
Married	384	99.7
Divorced	1	0.3

ANC utilization, obstetric characteristics of participants and LBW

Of the total participants, only 141 (36.6%) had complete ANC follow-up (≥ 4 visits). 42 women (10.9%) completed at least one ANC visit, and 3 women (0.8%) completed 8 ANC visits. One quarter (24.2%) of pregnancies were unplanned and 156 (40.5%) of the women were fully prepared for birth and potential complications that could happen during delivery. The majority (76.4%) of mothers prepared essential items of delivery and 77.4% saved money for delivery, but only 26.2% arranged blood donors and 51.9% identified the means of transportation to the health facility (S2 file). Prevalence of prenatal depression was 19.0% (95% CI: 15.3, 23.2) (S2 file). Among the study participants, 10.9% (95% CI: 8.2, 14.4) of women gave birth to a LBW baby (Table 3).

Table 3.

Obstetric characteristics, OOP ANC health expenditure and prevalence of LBW among pregnant women in Eastern Ethiopia, KHDSS 2023/24

Variables	Total N = 385	Percentage
OOP in USD (mean (SD))	32.4 (50.7)
Median OOP in USD (IQR)	11.6 (5.8, 36.9)
Presence of OOP across all ANC visits
No	111	28.8
Yes	274	71.2
Travelling time
≤ 2 h	194	50.4
>2 h	191	49.6
Length of health facility stay (waiting and service time)
≤ 2 h	75	19.5
>2 h	310	80.5
Parity
1	92	23.9
2–4	207	53.8
>4	86	22.3
Plan for pregnancy
Planned	292	75.8
Unplanned	93	24.2
ANC completion
Incomplete	244	63.4
Complete	141	36.6
Prenatal depression
No	312	81.0
Yes	73	19.0
Birth preparedness and complication
Well prepared	156	40.5
Not well prepared	212	59.5
Low birthweight
Yes	42	10.9
No	343	89.1

Out of pocket expenditure and time spent to attend ANC visits

The mean OOP expenditure for ANC utilization was 32.4 USD (SD ± 50.7). About seven in ten pregnant women (71.2%, 95% CI: 66.4, 75.4) incurred OOP health expenses while utilizing ANC services. The mean overall OOP expenditure was 26.4 USD for women who gave birth to a LBW baby, compared to 33.2 USD for those with a normal weight baby. Nearly half (49.6%) of women travelled more than two hours to attend ANC and 80.5% stayed more than 2 h at the health facility to receive ANC service (Table 3).

Determinants of LBW

Bivariable and multivariable Poisson regression with robust variance estimate was fitted to estimate the effect of the specified exogeneous, endogenous and other confounding variables on LBW. In the adjusted model, women who faced any OOP expenditures had significantly associated with increased risk of LBW compared to those who did not experience OOP expenditures (aRR = 3.21, 95% CI: 1.19, 8.64). In addition, prenatal depression significantly associated with LBW (aRR = 2.91, 95% CI: 1.65, 5.13) and lack of preparation for birth and complication increased the risk of LBW by 4.12 (aRR = 4.12, 95% CI: 1.52, 11.20). Variables like poor wealth status (aRR = 3.30, 95%CI: 1.16, 9.38), incomplete ANC visits (aRR = 2.37, 95%CI: 1.01, 5.53), unplanned pregnancy (aRR = 1.92, 95%CI: 1.14, 3.22) and long travelling time (1.99, 95%CI: 1.15, 3.44) were also significantly associated with LBW (Table 4).

Table 4.

Predictors of LBW using Poisson regression with robust variance Estimation among pregnant women in Eastern Ethiopia, KHDSS 2023/24

Variables	Category	LBW		cRR (95%CI)	aRR (95%CI)	P-value
		No	Yes
Age of mother	Adolescent	78	18	Ref	Ref	0.495
	Adult	265	24	0.44 (0.25, 0.78)	0.64 (0.48, 1.41)
Wealth status	Rich	67	4	Ref	Ref	0.550 0.027*
	Medium	144	10	1.15 (0.37, 3.55)	1.38 (0.44, 4.55)
	Poor	132	28	3.10 (1.13, 8.53)	3.30 (1.16, 9.38)
ANC visit	Complete	132	9	Ref	Ref	0.043*
	Incomplete	211	33	2.11 (1.04, 4.30)	2.37 (1.01, 5.53)
OOP	No	102	9	Ref	Ref	0.002*
	Yes	241	33	1.49 (1.01, 3.00)	3.21 (1.19, 8.64)
Prenatal depression	No	289	23	Ref	Ref	< 0.001*
	Yes	54	19	4.92 (2.81, 8.63)	2.91 (1.65, 5.13)
BPCR	Prepared	152	4	Ref	Ref	0.005*
	Not prepared	191	38	6.47 (2.35, 17.79)	4.12 (1.52, 11.20)
Plan for pregnancy	Planned	266	26	Ref	Ref	0.013*
	Not planned	77	16	1.93 (1.08, 3.44)	1.92 (1.14, 3.22)
Travelling time	≤ two hours	182	12	Ref	Ref	0.013*
	> two hours	161	30	2.53 (1.33, 4.81)	1.99 (1.15, 3.44)
Length of stay in health facility	≤ two hours	62	13	Ref	Ref	0.439
	> two hours	281	29	0.53 (0.29, 0.98)	0.79 (0.41, 1.52)

Note-aRR- adjusted Risk Ratio, cRR- crude Risk ratio, BPCRs Birth preparedness and complication readiness, LBW Low birth weight, OOP Out of pocket

*-value < 0.05

Direct and indirect association between OOP and LBW using GSEM

Variables which were significant in the multivariable Poisson regression were selected in the path model of GSEM. GSEM using binomial distribution (logit link function) was selected based on AIC and was fitted to determine whether the effect of OOP and other confounding on LBW was mediated by the proposed mediators (Fig. 2).

Fig. 2.

Unstandardized parameter estimates of pathways from OOP to LBW, GSEM. The values (β) are the direct effects (bolded: statistically significant) of exogenous and endogenous variables on LBW, with their respective 95% CIs in brackets

In the mediation analysis (Table 5; Fig. 2), the presence of OOP was associated with a direct increase in the risk of LBW, with a coefficient of 1.53 (95% CI: 0.37, 2.69). Prenatal depression significantly mediates the association between OOP and LBW 1.30 (95% CI: 0.21, 2.80). The association between OOP expenditures and LBW was also significantly mediated by the lack of preparation for birth 1.55 (95% CI: 0.29, 2.80). Therefore, prenatal depression mediated the association by 45.9% and lack of preparation for birth mediated by 50.3% showing substantial mediating role between OOP and LBW. Similarly, long travel time to health facility directly associated with LBW (β = 1.12, 95% CI: 0.13, 2.14). The path from long travel time to LBW was also mediated by incomplete ANC visit (β = 0.90, 95% CI: 0.10, 1.83). Prenatal depression (β = 1.63, 95%CI: 0.81, 2.44), poor preparation for birth (β = 1.92, 95%CI: 0.77, 3.08), poor wealth status (β = 1.04, 95% CI: 0.43, 1.64), unplanned pregnancy (β = 1.13, 95%CI: 0.29, 1.98), and incomplete ANC visit (β = 1.04, 95% CI: 0.04, 2.05) directly associated with LBW (Table 5).

Table 5.

Direct, indirect and total effect of exogenous, mediators and confounding factors on LBW, KHDSS 2023/24

Variables	Direct effect β (95% CI)	Indirect effect β (95% CI)	Total effect β (95% CI)	Proportion of mediation
OOP (Yes)	1.53 (0.36, 2.69) *	-	-	-
BPCR (poor preparation)	1.92 (0.77, 3.08) *
Prenatal depression (Yes)	1.63 (0.81, 2.44) *
Wealth status (Poor)	1.04 (0.43, 1.64) *
Plan for pregnancy (unplanned)	1.13 (0.29, 1.98) *
ANC (incomplete)	1.11 (0.04, 2.05) *
Travel time (> 2 h)	1.12 (0.13, 2.14) *
OOP → prenatal depression		1.30 (0.21, 2.39) *	2.83 (1.22, 4.44) *	45.9%
OOP → BPCR		1.55 (0.29, 2.80) *	3.08 (1.41, 4.74) *	50.3%
Wealth status → prenatal depression		0.21 (-0.34, 0.78)	1.26 (0.42, 2.10)	not statistically significant
Wealth status → BPCR		-0.21 (-0.79, 0 0.33)	0.81 (-0.001, 1.62)	not statistically significant effect
Plan for pregnancy → prenatal depression		-0.29 (-1.26, 0.67)	0.84 (-0.42, 2.11)	not statistically significant
Plan for pregnancy → BPCR		0.26 (-0.70, 1.23)	1.40 (0.10, 2.70) *	not statistically significant
ANC → BPCR		-0.65 (-1.58, 0.28)	0.49 (-0.87, 1.75)	not statistically significant
Travel time → BPCR		0.24 (-0.58, 1.07)	1.32 (0.13, 2.52) *	not statistically significant
Travel time → ANC		0.90 (0.10, 1.83) *	1.99 (0.83, 3.15) *	45.2%

Proportion of mediation was calculated as indirect effect/total effect

ANC Antenatal care, BPCR Birth preparedness and complication readiness, OOP Out of pocket

*Statistically significant

Discussion

OOP health expenditure associated with ANC-related costs may lead to delays, under- or non-utilization of services, potentially resulting in adverse maternal and neonatal outcomes. The Free Maternity Service (FMS) program was introduced to lessen the financial burden on pregnant women and family, aiming to enhance service coverage and improve maternal and neonatal health outcomes [50, 51]. However, its implementation has fallen short of expectations, with many women either avoiding ANC services or not attending fully due to persistent OOP expenses [52], with the potential consequence of increased adverse birth outcomes such as LBW. These costs may happen as direct non-medical expenses or direct medical costs for medication and treatment.

This study found that approximately one in ten pregnant women gave birth to LBW babies. This was consistent with previous studies conducted in Ethiopia [53–55], Kenya [56], Burkina Faso [57], SSA [29] and Indonesia [58]. However, it was lower than findings from other studies in Ethiopia [59, 60], India [61], and South Africa [62]. These variations may be attributed to differences in study design and sample size, as many of the previous studies used retrospective cross-sectional designs in which women were asked to recall their newborn’s weight, or relied on secondary data extracted from patient records. Conversely, the incidence of LBW in the present study was higher than that reported in China [63], Brazil [64, 65] and Malaysia [66], likely reflecting differences in socioeconomic status and overall health system performance across countries.

The study’s conceptual framework proposed that prenatal depression, ANC service utilization, and BPCR would mediate the association between OOP and LBW, while sociodemographic and obstetric characteristics were considered within the pathways. Our findings partially confirmed this framework, demonstrating significant mediating role of prenatal depression and BPCR in the association between OOP expenditure and LBW. Moreover, incomplete ANC visits mediated the association between long travel time to health facilities and LBW.

The study found that OOP was directly associated with increased risk of LBW. The finding is consistent with previous studies assessing the association between OOP and LBW [67, 68]. While this study found that higher OOP expenditures were associated with an increased risk of LBW in a low-resource agro-pastoral setting in Ethiopia, evidence from high-income countries often suggests the opposite, where greater health spending is associated with improved health outcomes [69]. In higher-income settings, OOP spending may reflect improved access to advanced care and greater capacity to pay for health services, often made without compromising basic needs [70]. This difference may reflect underlying differences in socioeconomic status and health system infrastructure. In Ethiopia, women in agro-pastoralist regions often incur OOP expenses not by choice, but due to gaps in service availability, forcing them to seek care elsewhere at a financial cost that can strain already fragile household budgets. These expenses can lead to delayed or reduced ANC visits, compromised nutrition, or skipped medications, all of which increase the risk of LBW [71].

Thus, while OOP expenditures in high-income countries may indicate better access and autonomy in care choices, in low-resource settings they are more likely to signify systemic inequities and financial burden. These differences highlight the importance of contextualizing the impact of OOP spending within the socioeconomic, health system realities and demographic dynamics of the population.

This study extends the findings in the literature by examining mediators in the association between OOP and LBW, offering a more comprehensive understanding of the causal pathway. The study found that OOP not only had a direct association with LBW but was also indirectly associated through inadequate BPCR. Previous studies found that women who are financially poor have low preparation for birth [72, 73]. Women who faced high OOP may have limited resources to save for emergencies, arrange transportation, prepare essential birth supplies, or plan for skilled delivery care, factors that lead to inadequate BPCR and increase the risk of adverse birth outcomes. Strengthening free access to prenatal care during pregnancy improves compliance to prenatal care which in turn improves BPCR, reducing risk of adverse birth outcomes such as LBW [74]. This study was conducted in a agro-pastoral region of Ethiopia, where health facilities are scarce, infrastructure is underdeveloped, and women face significant challenges in accessing services due geographical barriers [75]. Therefore, the findings highlight the need to improve service accessibility to enhance BPCR, which is essential for reducing adverse birth outcomes.

This study also found that prenatal depression mediates the association between OOP and LBW. This aligns with evidence from high-income country which found women’s depressive symptoms mediates the association between financial strain and LBW [76]. Evidences from India and Pakistan also supports this findings in which poor financial status is a strong predictor of perinatal depression [77, 78], while study in the US found that depressive symptoms due to difficulties in affording healthcare services during pregnancy [79]. Although, prenatal depression is influenced by hormonal and lifestyle changes during pregnancy, socioeconomic factors such as poverty and financial hardship are key contributors [52, 78]. High OOP expenses due to transportation, medication, or essential supplies create financial stress that can trigger or worsen depressive symptoms, particularly in resource-limited settings [80, 81]. Prenatal depression has been identified as a significant psychosocial risk factor for adverse birth outcomes, including LBW [82, 83], by reducing sleep quality, appetite, and care-seeking behaviour [84]. These findings highlight the importance of reducing financial stress during pregnancy, empowering women economically, and implementing routine screening and support for prenatal depression to improve maternal and neonatal outcomes.

ANC visits also mediate the association between long travelling time and LBW. This suggests that long travelling time to access ANC increased risk of LBW through reduced frequency of ANC utilization. This shows geographic barriers discourage ANC utilization, which in turn compromises fetal health and lead to adverse birth outcomes. This aligns with previous studies conducted in different setting where longer travelling time affect access to adequate ANC [85] and increase the risk of adverse birth outcomes [86, 87]. In Ethiopia’s agro-pastoralist regions, women are not utilizing service due to lack of physical accessibility and high transportation costs which pose major challenges to accessing maternal health services [75]. These barriers highlight the importance of improving access to ANC in order to promote healthier birth outcomes. Therefore, strategies to reduce LBW in these areas should address not only socioeconomic constraints but also physical accessibility, through interventions such as mobile ANC clinics, outreach services, and infrastructure development.

This study has several strengths and limitations. A key strength is the use of a prospective cohort design, which enabled us to follow women throughout pregnancy and collect ANC cost data prospectively, thereby minimizing recall bias. Moreover, we employed GSEM to examine the mediating role of selected variables in the association between OOP and LBW, ensuring that their inclusion shows the correct temporal ordering of causality. However, the study also has limitations. It focused only on LBW as the outcome variable, incorporating other adverse birth outcomes could have provided a more comprehensive assessment of the association between OOP and maternal-neonatal health outcomes. Second, although cost data were collected at each follow-up visit and verified with receipts when available, most expenses were based on women’s self-report. This reliance on self-reported data may introduce recall bias and social desirability bias, potentially affecting the accuracy of cost estimates. In this study, only the presence of OOP expenditure was included in the model as a binary variable (yes/no). The actual amount of OOP spending was not included in the model, which may underestimate the association between OOP expenditure and LBW. Furthermore, as the objective of this study was to examine the direct and indirect effects of OOP on LBW, some important covariates such as environmental, maternal nutrition, intimate partner violence, gestational age, preterm birth, congenital abnormalities, comorbidities and family-related factors were not collected and therefore were excluded from the path analysis. Future research could benefit from using large dataset to explore the potential moderating effect of OOP and ANC completion and incorporate these unmeasured variables into the analytical model. Given the observational nature of study design, causal relationships between OOP expenditure and LBW cannot be established. Future intervention-based or experimental studies are recommended to strengthen causal inference in this area.

Conclusion and recommendations

This study found that the incidence of LBW was 10.9%. Poor birth preparedness and prenatal depression mediated the association between OOP expenditure and LBW, while incomplete ANC partly mediated the association between long travel time and LBW. These findings highlight the interconnected impact of geographic, psychosocial, financial, and service accessibility barriers on birth outcomes. Therefore, in short run, focusing on counselling, screening, and financial support during pregnancy is recommended, while the long-term recommendations to reduce LBW should emphasize on free maternity service policy strengthening, mental health integration into prenatal care, and improving geographic access for pregnant women.

Abbreviations

ANC

Antenatal care

CI

Confidence Interval

ETB

Ethiopian birr

FMS

Free Maternity Services

GSEM

Generalized Structural Equation Modelling

LBW

Low birth weight

KHDDS

Kersa Health and Demographic Surveillance Site

LMIC

Lower middle-income countries

OOP

Out of pocket

SSA

Sab-Saharan Africa

WHO

World Health Organization

VIF

Variance inflation factor

USD

United States Dollar

Authors’ contributions

TT, LG, and JAO contributed to the proposal preparation, study design, and manuscript write up. TT. MD and ELH contributed to the data analysis and writing the initial draft of the manuscript. TT, JAO and LG involved in writing the final draft of the manuscript. All authors read and approved the final version of the manuscript.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Data availability

Data is available from corresponding authors up on reasonable request.

Declarations

Ethics approval and consent to participate

All methods of this research were conducted in accordance with the Declaration of Helsinki. Ethics approval for this study was granted from Deakin University (No- 2023 − 342), and the copy was submitted to Haramaya University. The broader KHDSS project has ethical approval from the Ethiopian Public Health Association (EPHA), the Centres for Disease Control and Prevention (CDC), the Ethiopian Science and Technology Agency, and the Haramaya University Health Research Ethical Review Committee (HRERC). Informed verbal consent was obtained from all participants after explaining the objectives and purpose of the study prior to data collection. Verbal consent was used for several reasons. First, a substantial proportion of the population in this area is illiterate and unable to read or write. Second, in the Ethiopian context, written consent is typically required for studies involving laboratory procedures, blood sample collection, or clinical trials, which carry greater risk compared to conventional interview-based data collection. All information obtained from each study participant is kept confidential throughout the process of study and kept in approved secure data storage.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.