Predictors of antenatal care utilization among the Rohingya population in the refugee camps of Cox’s Bazar, Bangladesh

Abstract

Background

While studies in both high- and low-income settings have demonstrated the importance of early and regular antenatal care (ANC) utilization, evidence from crisis-affected populations remains limited. Rohingya refugees in Cox’s Bazar, Bangladesh, face unique barriers to healthcare; previous studies have found low rates of ANC utilization. This study examines the predictors of number of visits and timing of initiation of ANC among pregnant Rohingya women in Cox’s Bazar.

Methods

Data came from a prenatal birth cohort study of “Intergenerational Risk and Resilience of Rohingya in Displacement” (iRRRd). Women (N = 2,322) were recruited during pregnancy between 2023 and 2024. ANC utilization was assessed via surveys at the birth follow-up visit with 2,065 new mothers. Primary outcomes included the number of ANC visits and the timing of the first visit. Predictors were categorized using the Andersen-Newman model (predisposing, enabling, and need factors). We used Poisson regression for number of visits, and logistic regression to analyze timing of first visit.

Results

The average number of ANC visits was 6.9 (SD = 2.6), with only 19.2% of women delaying care until after the first trimester. Women with one to three children reported fewer visits (IRR = 0.94, p = 0.01), as did those with more than three (IRR = 0.91, p = 0.02). Women with more years of education also showed higher number of visits (IRR = 1.015, p = 0.02) and higher likelihood of first trimester initiation (aOR = 0.891, p = 0.02) of ANC. More freedom of movement was associated with lower likelihood of delaying care (aOR = 0.871, p = 0.03), whereas higher self-rated health predicted delayed initiation (aOR = 1.187, p = 0.01). Socioeconomic indicators and education of relatives were not associated with ANC use. Receiving information via radio was unexpectedly associated with delayed initiation (aOR = 1.534, p = 0.03). Location of residence explained minimal variance (2%).

Conclusions

ANC coverage in the Rohingya camps appears high, yet some groups remain at risk for delayed care. Feeling healthy, high parity, and restricted mobility contributed to late initiation. Conventional predictors like household resources showed limited relevance in this context. Further work is needed to better understand the quality of services and effective communication strategies to address remaining underutilization of ANC.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13031-025-00733-6.

Introduction

The UN Sustainable Development Goal 3.2. aims to reduce neonatal and under-5 mortality worldwide [1]. Deaths of young children have been linked to adverse birth outcomes, like preterm delivery or low birth weight [2, 3]. Studies in high-income countries show that many of those adverse outcomes could be prevented with early and regular antenatal care (ANC) [4, 5]. ANC, as stipulated by the World Health Organization (WHO), refers to care provided by skilled professionals to pregnant women to ensure the best outcomes for both mother and baby. ANC includes regular check-ups, screening for risk indicators, health education, and interventions to prevent and manage pregnancy and birth complications. Similarly, literature in low- and middle-income countries shows the importance of adequate ANC utilization to avoid low birth weight [6, 7]. In such countries, nutritional and medical advice during pregnancy is particularly critical, and even more so in contexts of conflict and displacement. ANC serves as a vital factor in the effort to decrease maternal and infant mortality and to enhance the overall health of both mothers and infants [8]. Enhancing ANC utilization has the potential to benefit individuals and families by reducing suffering and improving livelihoods through decreased medical costs and increased productivity and learning [9].

Predictors of ANC utilization

The selection of an appropriate conceptual framework is essential for organizing and interpreting the diverse factors that shape maternal health service utilization. Frameworks help move beyond documenting associations to clarifying the pathways through which individual, household, and contextual factors influence care-seeking. For this study, we drew on the Andersen-Newman Behavioral Model, one of the most widely applied models in research on health services utilization in low- and middle-income countries (LMICs) to look at individual care-seeking behaviors. According to the Andersen-Newman model, three types of factors influence healthcare utilization: predisposing, enabling, and need factors [10]. Predisposing factors include socio-cultural characteristics that exist before a person’s need for health services arose. In low-income countries and fragile and conflict areas, pregnant women’s formal education [6, 11–15] and their partner’s education [13, 14, 16] significantly predicted adequate utilization of ANC. However, literacy showed mixed effects on ANC utilization in the literature [17]. Additionally, while some studies have shown that the younger a pregnant woman is, the less adequate [18] or later into her pregnancy her ANC utilization is [16], others reported no effect of a women’s age on ANC [19, 20]. Interestingly, one study found that being over the age of 30 was a risk factor for late ANC initiation [6].

Enabling factors concern the logistics of receiving care [10]. Socioeconomic status (SES) conceptualizes access to resources and is often measured using the possession of assets and in low-income countries, access to basic sanitary facilities [21]. Studies show that SES or wealth enable people to utilize ANC [11, 12, 14]. Additionally, the quality of available ANC has been shown to predict its utilization [6, 22].

The need factor is a person’s perceived need to visit a healthcare professional [10]. A study in Bangladesh showed that with each additional child, pregnant women utilized ANC significantly less often [14], and Duodu and colleagues [12] found in one of their cohorts that multiparity was associated with fewer ANC visits compared to women who were pregnant for the first time; however, this finding was not replicated in recent data [12]. The study’s authors hypothesized that pregnant women who have experienced pregnancy before may feel able to handle issues in their current pregnancy independently. Overall, some predictors of ANC are well-established in different contexts, most importantly, formal education and SES. In contrast, others vary depending on the definitions of the variables or the specific population investigated.

Adequate ANC utilization

The differences in definitions of adequate ANC utilization complicate the comparison of many studies and may contribute to the discrepancies in their results. To simplify and make their studies comparable, many authors focus solely on the number of visits [14, 23–25]. However, the adequacy of ANC utilization depends not only on the number of visits and the time of initiation, but also type and quality of services [26].

Until 2016, the World Health Organization recommended that pregnant women worldwide have at least 4 ANC visits. However, in specific resource-limited settings with strained services, it proves challenging to accommodate all elements of care within the confined 4 appointments. Hence, since 2016, the WHO started to recommend 8 visits while keeping up the recommendation of ANC initiation within the first trimester [8].

ANC in contexts of humanitarian crisis and forced displacement

The benefits of adequate ANC are of particular relevance in contexts of humanitarian crises and forced displacement, and understanding how to ensure timely initiation, sufficient frequency, and quality of ANC services has become paramount. The unique barriers faced in crisis settings—ranging from disrupted healthcare infrastructure to cultural and logistical constraints—demand targeted interventions [27]. The past decade has seen a push toward understanding ANC utilization behavior and the effects of targeted community-based programs, including the training of community health workers, home visits, and peer-led educational campaigns on ANC utilization and facility-based deliveries in crisis settings [18, 28–30]. The unique communal specificities of a crisis setting need to be understood to evaluate and adapt such approaches. ANC in humanitarian settings is usually provided free of cost, yet the proximity to services depends on the response plan, and availability alone is not always sufficient.

Significant gaps remain in our understanding of ANC use in the Rohingya communities in Bangladesh, home to the largest and most densely populated refugee settlement in the world. While some progress has been made toward building an evidence-base on sexual and reproductive health among Rohingya communities in Bangladesh, not that much is known about ANC specifically. To our knowledge there is to date only one study looking quantitatively at predictors of ANC usage [31]. This study relies on data collected in 2023 using retrospective self-report on pregnancies in the two years prior to data collection. The authors find that almost half the women had not received any ANC. The other half received at least one; how many exactly, or what the mean and range was, is not mentioned in the study, nor when in their pregnancy they initiated ANC. Given the timing of the pregnancies reported on in this study it is important to note that there was a marked decline in ANC utilization during the COVID-19 pandemic in Bangladesh more generally [32], and in refugee settings specifically [33]. In addition, in light of likely changes in norms, attitudes, and practices of refugee populations as they are exposed to and interact with host country communities and humanitarian aid workers [34], a more current understanding of ANC utilization is needed in order to understand and start to address persistently high rates of poor pregnancy and birth outcomes.

Objective of the study

This study aims to fill these gaps by comprehensively examining how socio-cultural and demographic factors predict ANC utilization more recently within the Rohingya population residing in refugee camps in Cox’s Bazar, Bangladesh, utilizing data gathered from early 2023 through August 2024. The goal is to contribute essential insights into the barriers and facilitators influencing ANC, ultimately providing the groundwork for developing targeted, integrated interventions and policies designed to enhance maternal and infant health within this vulnerable population.

Methods

Study design and setting

Data stems from the “intergenerational risk and resilience of Rohingya in displacement” (iRRRd) study, a longitudinal prenatal birth cohort study investigating how being gestated, born, and raised in a context of conflict and displacement affects child development [35]. In 2017/2018 almost 700,000 Rohingya were violently displaced from Rakhine, Myanmar, bordering the southern tip of Bangladesh. As of 2024 almost 1 million Rohingya live in camps near Cox’s Bazar, one of which now constitutes the largest and most densely populated refugee settlement in the world [36, 37]. It has been shown that Rohingya in Bangladesh have an exceptionally high stillbirth rate [38], and Rohingya displaced to Malaysia have on average higher rates of low birth weight babies than their host community [39]. Moreover, a third of all deaths of Rohingya women are reported to be related to pregnancy complications [40]. In a 2018 survey, more than half of the Rohingya women in Cox’s Bazar who had given birth in the previous year reported that they had not received any ANC [41].

The situation of the Rohingya living in Cox’s Bazar is unique since they have been systematically excluded from (reproductive) health services in their country of origin and find themselves in a refugee camp setting [42]. Specific guidance on the number of ANC visits for Cox’s Bazar could not be found. However, the Bangladeshi government recommends a minimum of 4 ANC visits [14]. Similarly, the Inter-Agency Working Group on Reproductive Health in Crisis recommends at least 4 visits in humanitarian response settings [27].

Reports relying on health sector data show that ANC coverage and facility-based births have gradually increased since 2017 and rates of receiving at least 4 ANC visits are now around 77%⁴³, while a study using self-report data finds over 47% of women did not receive ANC during their most recent pregnancy [31]. Understanding the considerations and barriers to take-up and timely initiation of adequate ANC can lead to targeted interventions to optimize ANC utilization. This could reduce maternal, neonatal and under-5 mortality rates, and address developmental delays observed in the Rohingya community [43, 44].

Participants and sampling

Participants came from 11 purposively selected camps (accounting for differences in terrain reflecting more or less challenging locations) located near Ukhia, Cox’s Bazar, Bangladesh. A total of 2,322 pregnant Rohingya women were recruited between February 2023 and March 2024. All women under the age of 19 identified in our listing of households with pregnant women were recruited into the study. The remainder of the sample, drawn from the listing, was recruited from the area surrounding households already recruited. This was done for feasibility reasons as movement in the camps, to and between households, is only possible on foot, and thus proximity of households was deemed necessary. However, we had practically no refusals at recruitment of households listed (total of 5 households initially refused to participate due to time constraints, 3 agreed to participate when followed up with at a more opportune time), and are thus not concerned about systematic bias related to participation. Additional details on study design and recruitment are described in the cohort profile paper [35]. For the purpose of the present study, only participants who were followed up with at the 1-month post-partum follow-up were considered (N = 2070). Reasons for not completing the 1-month follow-up were predominantly due to the loss of the pregnancy or child (n = 118); only 17 women refused, 29 had migrated and could not be found, and the remaining 18 had other reasons why the follow-up was not conducted. One participant’s education data was uninterpretable (reported 16 years of schooling but also could not read in any language), and 4 participants were lost due to merging issues. The remaining 2,065 participants were included for the purposes of this study.

Data collection procedures

The data for this study stems from structured surveys conducted in the homes of participants. Enumerators conducted surveys on tablets using SurveyCTO after answering any questions and concerns about participation in the study and completing formal consent procedures. Female enumerators, who spoke Rohingya, many with experience working in the camps, were trained extensively on fielding sensitive questions, including ensuring privacy from other household members. As the survey was on average 2–3 h long, enumerators had to pause to allow for the respondent to tend to children or other household chores.

As women were recruited across trimesters, the recruitment process took place over 12 months, whereas follow-up waves spanned approximately 18 months (March 2023-September 2024). Participants were provided roughly USD1 per visit to compensate for their time. The study was approved by the ethical review committee of icddr,b (PR-22064) and from New York University (IRB-FY2021-4875).

Measures and variables

Data for this study were from the surveys conducted at recruitment and birth follow-up waves. Survey instruments were carefully translated to Bangla with special attention to Rohingya language particularities and piloted extensively in July and August of 2022. Figure 1 provides a visual representation of the theory of change, based on the Andersen-Newman model, underlying our study.

Fig. 1.

A Behavioral Framework for ANC based on the Andersen-Newman Model. * Years of schooling retained as primary education indicator; moderate VIF (~ 4) with school type. † School type created for conceptual completeness; dropped due to collinearity.** In this context, resources treated as predisposing (ANC free of cost); asset & PREI factors/scores from EFA/CFA

Primary outcomes

Number of ANC Visits was the first variable of interest. It was defined as the total number of ANC visits reported at the post-birth survey, regardless of where (health clinic, home, etc.) ANC services were received. Timing of ANC Initiation was the second variable of interest and was coded as a dummy variable determining whether a pregnant woman who had at least one ANC visit initiated ANC later than the first trimester of her pregnancy.

Predisposing factors

Maternal Age was grouped into under 15, 15–19, 20–24, 25–29, 30–34 and above 34.

Educational background was collected for the pregnant woman, her husband, father, and mother. For all four individuals, a continuous variable for Years of Formal Education Completed was constructed from the reported highest grade level achieved. Respondents could select a grade category (e.g. Class 1–12, SSC/HSC, Graduate) or indicate “religious education.” Graded categories were mapped to the minimum corresponding number of school years, while religious education was coded as zero years, as it could not be placed on the formal schooling scale. Additionally, for the pregnant woman, we created a categorical variable for the Type of Schooling (no schooling, non-governmental organization (NGO), only religious, governmental). Years of schooling captures cumulative exposure to formal education, while school type reflects potential differences in curriculum quality and accessibility across contexts. Preliminary analyses indicated moderate collinearity between these indicators (VIF ≈ 4); therefore, years of schooling was retained as the primary indicator to ensure comparability and reduce redundancy. Self-Reported Literacy was defined by being unable to read a newspaper in any language, being able to read a newspaper in Burmese, or being able to read it in any other (foreign) language. The variable was created from two questions with a 3-point scale asking participants about their Burmese and foreign language reading skills.

In this study, resources are treated as a predisposing rather than an enabling factor, as ANC in the camps is free of cost. The indicators were obtained using factor analysis that utilized items that identified how many of certain Assets the household possesses (chairs, beds, fans, chicken, solar panels, gas cylinders, tables, trunks, plastic drums, grams of gold) and questions from the perceived refugee environment index (PREI), which was developed to assess the quality of the refugee environment and Access to Resources [45]. From the PREI, only 20 out of 30 items were used that covered questions on livelihoods, basic needs, housing, family environment, access to services and community environment using a 4-point Likert scale (almost never; less than half of the time; more than half of the time; almost always). The 10 items on electricity, heat, protection from rain, sun, wind, and water, or access to the internet, texting, or phone calls had to be dropped from the survey per request of the Refugee Response and Repatriation Commissioner (RRRC)—the Bangladesh government entity that regulates activities in the camps.

Enabling factors

Agency in Mobility was defined as how often a woman could visit their family outside their block by themselves: almost never (0), less than half of the time (1), more than half of the time (2), or almost always (3). Camp was a categorical variable defined as the camp where the pregnant woman lived.

Need factors

Number of Children was grouped into 0, 1–3, and more than 3. Previous Adverse Pregnancy/Perinatal Outcomes was defined as a binary variable as having at least one prior miscarriage, abortion, or stillbirth. The Self-Reported Health Rating described how well someone rated their physical health on a 5-point scale ranging from very bad (0) to very good (4).

Informational pathways

We include two variables that relate to where women received information about ANC. We treat both items as a separate category, since these items intersect with all three Andersen-Newman factors: trusted sources and exposure to media reflect predisposing values; access to such channels enables women by providing practical knowledge on where and how to seek ANC; and the content of messages shapes their perceived need for care. A set of non-mutually exclusive dummy variables identified where participants Received Information about ANC from: example, the radio, pamphlet, phone, online, and/or TV. A further set of non-mutually exclusive dummy variables identified whom a pregnant woman had Talked to About the Pregnancy: her husband, mother-in-law, father-in-law, father, mother, sister, brother, friend, and/or sister-in-law.

Other

We control for Gestational Length defined as the total weeks of gestation at delivery, a variable that was triangulated across several data sources. Gestational length would necessarily affect how many ANC check-ups a woman received. We further controlled for when the participant came to Arrived in Bangladesh—born in Bangladesh, came before the 2017 mass displacement, came in 2017 or after—as time spent in the camps and in Bangladesh could affect their perceptions and practices of ANC.

Analysis plans

Stata18 and MPlus were used for the analyses. First, a descriptive analysis was conducted. Most studies similar to this used principal component analysis (PCA) to construct wealth or asset indices, and then divided the population into quintiles or terciles relying on assets and some additional variables measuring basic living conditions [11, 12, 14, 46–50]. This study additionally captures items from the perceived refugee environment index [45]. Exploratory factor analysis was used to examine the shared variance between these items and identify factors/sub-domains, based on which confirmatory factor analysis was used to obtain the factor scores. Factor analysis was preferred over PCA since the aim was to extract the latent common factors between the items and not merely reduce data [51]. The asset and PREI items went through this factor analysis process separately.

In three of the camps, there was only one participant each. These single observations were dropped from the analyses. One case was dropped as the woman reported having been to medical school yet was not able to read. The Number of ANC Visits data was capped at 16 to reduce the impact of outliers on the analysis (this affected 14 observations).

Previous literature has often focused on defining an adequate and inadequate number of visits using a binary logistic regression model with cut-off points of three or four [11, 16, 50], or eight [52] visits. Others used Poisson regression models [53, 54]. This paper used Poisson regression to model ANC utilization in terms of the number of visits. Overdispersion was tested by comparing whether the variance exceeded the mean [55].

We assessed potential multicollinearity among education-related and information-related predictors using variance inflation factors (VIF). Both years and type of schooling were initially included as education-related predictors to capture different aspects of educational attainment. Most predictors had VIF values between 2 and 3, which is within acceptable limits. An exception to this was the Pregnant Woman’s Years of Formal Education (VIF = 4.24), reflecting conceptual overlap with the Pregnant Woman’s Type of Schooling. The majority of women who reported having zero years of formal education completed also reported religious education as their type of schooling. Given the conceptual overlap and the broader interpretability of years of schooling, total years of formal education was retained as the primary education indicator in the final models. We conducted sensitivity analyses, including the Pregnant Woman’s Type of Schooling in the model.

Because pseudo R-squared in Poisson regression models measures improvements in model fit relative to the null model rather than explained variance, it does not allow for a direct interpretation of the proportion of variance accounted for by the predictors [56]. To provide a descriptive approximation of the model’s explanatory power, we therefore conducted a sensitivity analysis by estimating OLS regression using the same set of predictors. If the OLS and Poisson models showed similar patterns, the R² from the OLS model could offer a rough indication of the amount of variance explained by the included variables. Based on this, we examined the change in R² when omitting the variable Camp to assess the proportion of variance explained by camp-level differences.

The results of the Poisson regression model were presented as the Incidence Rate Ratio (IRR). IRR can be interpreted as the percent change in the event rate for a one-unit increase in the predictor variable. It is calculated by exponentiating the coefficient of the predictor from the Poisson regression model [57].

The Time of ANC Initiation was modelled using a binary logistic regression model. The results were presented as adjusted Odds Ratios (aOR). Both the ANC utilization and the time of initiation models used robust standard errors to overcome the effects of potential clustering [58] and used complete cases only, which excluded approximately 6% of the data points due to missingness (119 and 118 observations respectively).

Results

Model selection for constructs related to SES

Best fit was found for a 1-factor model for assets (Assets), and a 3-factor model for the PREI (Basic Needs, Support and Acceptance, Safety). The asset scores were subsequently broken into quintiles when used in the regression analyses. The latter was done for comparability to other studies. Details on these analyses can be found in supplemental materials.

Sample characteristics and descriptive statistics

Table 1 presents an overview of the participants’ general characteristics. However, individual variables had missing observations (between 0% and 2.08%). Summary statistics were calculated based on the available data for each variable. The sample consisted of 2,065 pregnant women with a mean age of 21.1 years. In our sample, 3.05% were born in Bangladesh, whereas 89.15% arrived in 2017 or later. Almost three-quarters reported not being allowed to visit their family outside their block alone, while only 12% were almost always allowed. Approximately half of our sample rated their health as good or very good.

Table 1.

Sample descriptives and demographic characteristics

	N	missing(%)	n(%)/Mean(SD)	Min	Max
Maternal Age	2,065	0 (0.00%)	21.10 (5.11)	13	40
Under 15			61 (2.95%)
15–19			952 (46.1%)
20–24			574 (27.8%)
25–29			301 (14.58%)
30–34			139 (6.73%)
Over 34			38 (1.84%)
Maternal Years of Schooling	2,040	25 (1.21%)	1.51 (1.89)	0	9
Husband’s Years of Schooling	2,036	29 (1.40%)	2.83 (3.36)	0	16
Father’s Years of Schooling	2,022	43 (2.08%)	1.53 (2.95)	0	16
Mother’s Years of Schooling	2,051	14 (0.68%)	0.45 (1.48)	0	16
Agency in Mobility (travel alone)	2,065	0 (0.00%)		0	3
Almost never			1528 (74%)
Less than half of the time			192 (9.3%)
More than half of the time			91 (4.41%)
Almost always			254 (12.3%)
Self-Reported Health	2,064	1 (0.05%)
Very Bad			74 (3.58%)
Bad			438 (21.21%)
Regular			423 (20.48%)
Good			1006 (48.72%)
Very Good			123 (5.96%)
Able to Read	2065	0 (0.00%)
No reading			1416 (68.57%)
Burmese			477 (23.1%)
Only foreign			172 (8.33%)
Camp	2062	3(0.15%)
Camp 01E			268 (12.98%)
Camp 02 W			232 (11.23%)
Camp 03			394 (19.08%)
Camp 04			221 (10.7%)
Camp 04 Ext			59 (2.86%)
Camp 07			425 (20.58%)
Camp 08 W			256 (12.4%)
Camp 17			207 (10.02%)
Arrived in Bangladesh	2065	0 (0.00%)
Born in Bangladesh			63 (3.05%)
Arrived before 2017			161 (7.8%)
Arrived after 2017			1841 (89.15%)

The mean of maternal years of formal education was extremely low not even reaching 2 years; two thirds reported that they could not read the newspaper in any language. Husbands, on average, had nearly twice as many years of formal education as their pregnant wives. Fathers of pregnant women had similar average years of schooling compared to the pregnant women, but exhibited a greater standard deviation. This means that, on the one hand, fathers were nearly four times more likely to have completed four or more years of formal education than their daughters, while also one and a half times more likely to have had no schooling at all compared to their daughters. Lastly, mothers of pregnant women attained the lowest level of formal education. On average, they received only a third of the years of schooling their daughters received.

Table 2 depicts ANC/pregnancy/fertility-related details. Remarkably, only two participants did not receive any ANC care (< 1%) and only 5.67% of the sample were lagging behind the adequacy mark of four visits, while more than a third received at least eight visits. Moreover, four out of five women attended their first ANC visit within the initial three months of pregnancy. Notably, two-thirds of the sample already had at least one child, and almost one in four had previously experienced miscarriages, abortions, or stillbirths. That left 28% of the recruited pregnant women without any previous pregnancy experience.

Table 2.

ANC characteristics

	N	missing(%)	n (%)/Mean(SD)	Min	Max
Total ANC Visits	2,061	4 (0.19%)	6.94 (2.61)	0	16
Less than four			117(5.67%)
Eight or More			775(37.55%)
Months Pregnant when 1 st ANC	2,059	6 (0.29%)	2.68 (1.12)	1	10
ANC Initiation post 1 st TM	2,064	1(0.05%)	397(19.2%)
Gestational Length (weeks)	2,052	13 (0.63%)	39.86 (1.88)	32	42
Parity	2,065	0 (0.00%)
Nulliparity			648 (31.38%)
Low multiparity			1170 (56.66%)
Grand multiparity			247 (11.96%)
Previous Adverse Pregnancy/Perinatal Outcomes	2064	1(0.05%)
No			1573 (76.17%)
Yes			491 (23.78%)
Talked About Pregnancy with	2065	0 (0.00%)
No one			125 (6.05%)
Husband			1228 (59.47%)
Mother-in-law			1047 (50.7%)
Father-in-law			124 (6%)
Father			218 (10.56%)
Mother			1313 (63.58%)
Sister			699 (33.85%)
Brother			47 (2.28%)
Friend			241 (11.67%)
Sister-in-law			804 (38.93%)
Received Information about ANC via	2065	0 (0.00%)
None			111 (5.38%)
Radio			607 (29.39%)
Leaflet/Pamphlet/Written Info			1474 (71.38%)
Phone			116 (5.62%)
Online			143 (6.92%)
TV			781 (37.82%)
Speaker Announcements			539 (26.1%)

Communication about pregnancy was common; nearly all pregnant women talked with someone (94%), most often husbands, mothers, or mothers- and sisters-in-law. Communication with male family members (brothers, fathers, father-in-law) was notably rare. Only a small percentage received pregnancy information via their phone or online. Most women received information about their pregnancy from pamphlets.

ANC utilization – Regression results

The dispersion parameter (ϕ = 0.95) indicated no evidence of overdispersion, supporting the adequacy of the Poisson model for the Number of ANC Visits. Table 3 presents the results from the two regression models. Regarding Predisposing Factors, only the pregnant woman’s education was associated with earlier and more ANC visits. For each additional year of formal schooling the odds of delaying ANC until after the first trimester decreased by 10% and was associated with a 1.5% increase in ANC visits. The education of the participants’ husbands, mothers, or fathers did not predict the pregnant women’s ANC utilization behavior. A sensitivity check revealed that including the Pregnant Woman’s Type of Schooling, which was omitted from the model due to potential collinearity, did not predict the timing or amount of ANC visits. None of the predictors derived from factor analysis and associated with household resources were associated with ANC utilization behavior.

Table 3.

Regression results

	Number of ANC Visits				ANC Initiation Delayed Post 1 st TM
	(Poisson)				(logistic)
	Number of observations = 1,946				Number of observations = 1,945
	IRR	P > z	[95% CI]		aOR	P > z	[95% CI]
Age of Pregnant Woman
Under 15	1.000				1.000
15–19	0.967	0.55	0.867, 1.079		1.579	0.27	0.705, 3.537
20–24	0.958	0.47	0.852, 1.077		1.803	0.18	0.762, 4.261
25–29	1.029	0.66	0.907, 1.167		1.297	0.58	0.516, 3.259
30–34	1.011	0.88	0.879, 1.164		2.446	0.08	0.898, 6.666
Over 34	1.078	0.42	0.899, 1.294		1.189	0.79	0.327, 4.319
Pregnant Woman Years of Education	1.015	0.02	1.002, 1.029		0.891	0.02	0.811, 0.979
Husband Years of Education	1.000	0.98	0.994, 1.006		1.008	0.68	0.971, 1.046
Father Years of Education	1.001	0.68	0.995, 1.007		1.035	0.13	0.990, 1.082
Mother Years of Education	0.999	0.85	0.986, 1.012		0.940	0.22	0.853, 1.037
Literacy
No Reading	1.000				1.000
Burmese	0.978	0.44	0.924, 1.035		1.375	0.10	0.938, 2.014
Other Language	1.002	0.96	0.940, 1.067		0.754	0.29	0.449, 1.265
Basic Needs (factor scores)	0.959	0.16	0.905, 1.016		1.032	0.88	0.684, 1.555
Support and Acceptance (factor scores)	0.983	0.49	0.936, 1.032		0.813	0.24	0.576, 1.146
Safety (factor scores)	1.022	0.12	0.994, 1.051		0.948	0.61	0.772, 1.164
Asset Quintiles (factor scores)
Poorest	1.000				1.000
Poorer	1.035	0.21	0.981, 1.092		0.871	0.49	0.591, 1.285
Middle	1.002	0.95	0.950, 1.056		1.198	0.35	0.819, 1.753
Richer	0.979	0.45	0.926, 1.035		1.031	0.88	0.693, 1.535
Richest	0.971	0.36	0.913, 1.034		1.142	0.54	0.749, 1.742
Camp
Camp 01E	1.000				1.000
Camp 02 W	0.930	0.04	0.870, 0.995		0.833	0.46	0.512, 1.356
Camp 03	0.923	0.01	0.870, 0.978		0.518	0.00	0.339, 0.791
Camp 04	1.015	0.65	0.951, 1.084		0.926	0.75	0.574, 1.492
Camp 04 Ext	1.020	0.63	0.941, 1.106		0.167	0.01	0.048, 0.582
Camp 07	0.866	0.00	0.815, 0.921		1.398	0.08	0.960, 2.037
Camp 08 W	0.868	0.00	0.815, 0.925		0.377	0.00	0.226, 0.627
Camp 17	0.975	0.46	0.913, 1.042		0.548	0.02	0.326, 0.921
Agency in Mobility	0.997	0.73	0.981, 1.014		0.871	0.03	0.767, 0.989
Self-Reported Health Rating	0.985	0.08	0.969, 1.002		1.187	0.01	1.046, 1.348
Number of Previous Children
Zero	1.000				1.000
One to Three	0.937	0.01	0.895, 0.981		0.924	0.64	0.664, 1.287
More than Three	0.909	0.02	0.838, 0.985		1.115	0.71	0.626, 1.988
Previous Adverse Pregnancy/Perinatal Outcomes	1.013	0.53	0.973, 1.054		1.316	0.05	1.000, 1.731
Talked About the Pregnancy with
No One	1.002	0.97	0.896, 1.120		1.273	0.53	0.598, 2.711
Husband	1.009	0.69	0.967, 1.052		0.747	0.09	0.535, 1.043
Mother-in-law	0.979	0.36	0.935, 1.025		1.051	0.79	0.731, 1.513
Father-in-law	1.076	0.11	0.984, 1.176		0.517	0.08	0.250, 1.070
Father	0.980	0.54	0.916, 1.047		1.066	0.83	0.603, 1.882
Mother	1.003	0.89	0.963, 1.045		0.974	0.88	0.704, 1.348
Sister	0.993	0.73	0.952, 1.035		0.915	0.58	0.669, 1.252
Brother	0.988	0.82	0.890, 1.097		1.085	0.84	0.489, 2.409
Friend	0.977	0.39	0.926, 1.031		0.997	0.99	0.672, 1.478
Sister-in-law	1.012	0.51	0.976, 1.050		0.825	0.14	0.640, 1.063
Received Information about ANC via
None Received	0.964	0.53	0.858, 1.082		0.770	0.54	0.336, 1.765
Radio	1.018	0.52	0.964, 1.074		1.534	0.03	1.033, 2.278
Leaflet/Pamphlet/Written Info	0.989	0.71	0.934, 1.048		0.885	0.58	0.572, 1.369
Phone	0.979	0.63	0.898, 1.068		1.326	0.43	0.660, 2.663
Online	1.048	0.28	0.962, 1.142		0.998	1.00	0.507, 1.965
TV	0.976	0.28	0.933, 1.021		1.176	0.34	0.840, 1.647
Speaker Announcements	1.020	0.40	0.975, 1.066		1.130	0.48	0.807, 1.581
Arrived in Bangladesh
Born in Bangladesh	1.000				1.000
Arrived before 2017	1.057	0.35	0.941, 1.187		0.646	0.33	0.267, 1.563
Arrived after 2017	1.002	0.97	0.901, 1.114		0.830	0.63	0.390, 1.768
Gestational Length (Weeks)	1.013	0.01	1.003, 1.023		-	-	-
_cons	4.872	0.00	3.174, 7.477		0.183	0.01	0.049, 0.686

Notes: ¹The lower number of participants in the regression models compared to the descriptives is due to listwise deletion of cases with missing data. ²Two participants with 0 ANC visits were excluded from the logistic model, one of whom was already excluded in the Poisson model due to missing data in another variable

Regarding predictors that enable pregnant women to receive care (enabling factors), having a unit increase on the Agency in Mobility scale was linked to 13% lower odds of delaying ANC initiation until after the first trimester. However, the predictor had no association with the number of ANC visits. There was substantial variation between Camps. Of those that are statistically significant, being in a specific camp can decrease the number of visits up to 14%, and/or decrease the odds of delaying initiation by 80%. Interestingly, the direction between number of visits and timely initiation often do not go in the same direction, meaning, while being, for example, in Camp 8W decreases the average number of ANC visits by 13%, but it also decreases the odds of delaying initiation by 62%.

Assessing the perceived need to access ANC (need factors), compared to participants with zero children, those with one to three children reported 6% and those with more than three children reported 10% fewer ANC visits. No association between parity and the timing of ANC was found. For each unit increase of the pregnant women’s self-reported health rating, the odds of delaying ANC until after the first trimester increased by nearly 19%. Similarly, Self-Reported Health was approaching significance for the number of ANC visits. However, every unit increase was only associated with 1.5% fewer visits. While right at the threshold for significance, having had previous adverse pregnancies increased the odds of delaying care until after the first trimester by 32%.

ANC-specific predictors (informational pathways), such as whom they talked to about their pregnancy and where they received information about it, were predominantly unrelated to the number or timing of ANC visits. Only speaking to their husband and father-in-law about their pregnancy was trending significance but with a potentially large decrease, 25 and 50% decrease respectively, in the odds of delaying ANC. Interestingly, having received information about the pregnancy on the radio was associated with 53% higher odds of delaying ANC until after the first trimester.

Overall, the Poisson regression model yielded a low McFadden’s pseudo R-squared of 0.015, indicating only minimal improvement in model fit relative to the null model. Consistent with this, few predictors reached statistical significance, and effect sizes were generally small. The OLS sensitivity analysis produced a low R² of 0.07, supporting the interpretation of limited overall explanatory power of the predictors. Omitting the variable Camp from the OLS model reduced the R² by 0.02, suggesting that camp-level differences accounted for approximately 2%, or almost two thirds of the explained variance. Taken together, these indicators consistently point to a modest amount of variation in ANC visits explained by the included predictors.

Discussion

This study examined predictors of the number and timing of initiation of ANC visits among forcibly displaced Rohingya living in refugee camps in Cox’s Bazar, Bangladesh. The average of 7 ANC visits was unexpected, as was the high rate of ANC initiation in the first trimester, practically meeting WHO’s global recommendation for ANC. However, several of the factors that had been identified in previous studies in other populations and contexts did not predict ANC utilization behavior in our study. We found, however, a decrease of 12.1% in the odds of initiating ANC later than the first trimester and an increase of 1.5% in number of ANC visits for each additional year of formal education of the pregnant woman. No association was found between the husband’s, father’s, or mother’s years of education and the pregnant woman’s ANC utilization behavior. Furthermore, the current study found no significant relation between ANC initiation timing or the number of ANC visits and any of the resource and asset factors. Better self-rated health significantly increased the odds of ANC initiation after the first trimester, leading potentially to women missing the critical early window for ANC. In addition, women who could not move freely without accompaniment had a lower likelihood of initiating ANC within the first trimester. Women for whom this was going to be the first child reported having more frequent ANC check-ups, whereas having had a previous adverse pregnancy, while only marginally significant, oddly increased the odds of late initiation. Receiving information about their pregnancy from the radio significantly increased the odds of delaying ANC. And, while only trending, discussing their pregnancy with their husbands and fathers-in-law was associated with lower odds of delaying initiation until after their first trimester.

With the exception of a couple of the camps, the significant predictors that were identified showed modest or weak association. The Poisson model showed a very low pseudo R² (0.012), with few significant predictors and weak associations. A sensitivity check using OLS regression found that only roughly 7% of the variance of the number of ANC visits could be explained by this study’s predictors. Lastly, variance in both dependent variables was exhibited between different camps. However, no camp was associated with more than a 13% increase in ANC visits. Additionally, a sensitivity check omitting the camps entirely from the model showed that the variance accounted for by the location in the number of ANC visits was only 2%, which nonetheless is almost a third of the total variance explained.

Even so, this study addresses important gaps in our understanding of ANC behaviors in the Rohingya camps in Bangladesh, and presents a more current picture of the situation with regards to ANC, one that is reflective of the post-COVID recovery of health service provision, and likely changes in attitudes that shape demand for ANC over time. This study also highlights that demand and uptake of ANC in humanitarian settings may not follow the same patterns as observed in non-humanitarian, low-resources settings. This indicates the need to rethink how one might encourage early uptake and adherence to recommended number of ANC visits in both acute and protracted humanitarian settings. The Rohingya camps in Cox’s Bazar, Bangladesh, represent one of the world’s most challenging humanitarian contexts. Nearly one million refugees live in extremely dense, hazard-prone settlements with limited rights, restricted mobility, and chronic underfunding of essential services. The camps are situated in hilly, flood- and landslide-prone terrain with narrow, often unpaved pathways, making movement challenging for residents and humanitarian workers alike, and making effective service delivery more difficult. The combination of scale, vulnerability, and political constraints creates a uniquely complex environment for health service provision.

Our results stand in marked contrast to Khan and colleagues’ findings that 47% of Rohingya women in Cox’s Bazar had not attended any ANC services [31]. In contrast to our study, they were able to identify predictors in line with previous literature. Some of the differences in findings could be of methodological nature including social desirability, sample composition and characteristics, and recall biases. However, we know from official records and from our qualitative investigations that the ANC utilization rates in this context have changed over time. By 2019, the number of ANC visits and facility-based deliveries had already increased by 40% compared to two years earlier [36]. After assessing the quality and distribution of health services in 2019, over a third of health facilities were relocated or closed to support better distribution and quality of services. This was accompanied by continued capacity building among health workers and community advocacy [59]. However, the COVID-19 pandemic led to a significant contraction of the health sector serving the camps which may be driving some of Khan and colleagues’ findings using data collected in 2023 from women who had been pregnant in the preceding 2 years [31]. Meanwhile, the most recent official health sector bulletin reported that by February 2024, 77% of pregnant women had at least four ANC visits, and 90% of births were facility-based [60]. These numbers align with the findings in our work, establishing confidence in our data.

We expected to find relations between the woman’s years of formal education with ANC behaviors given the robust association in literature [6, 11, 13, 14, 46]. Nonetheless, the weak associations are somewhat surprising. A Bangladesh-wide study reported that primary and secondary education was associated with 27% and 57% more ANC visits, respectively, compared to those without schooling [61]. Similarly we were surprised by the findings on assets and household resources. In previous studies they reliably predicted ANC utilization in African and South Asian countries [11, 12, 14, 46–50]. On the other hand, the few studies on ANC utilization in humanitarian settings have found mixed results regarding education and resources. In an Iraqi refugee camp, researchers found that among forcefully displaced Syrians, maternal education is not associated with ANC utilization, while maternal employment status is.⁶² A study in camps for internally displaced people in Sudan reported that employment status was not related to ANC utilization but identified education as a predictor [15].

In fact, some studies have shown a higher rate of ANC utilization for refugees compared to before forced displacement, and compared to the host population. For instance, 99% of Syrian refugees in Iraqi camps attended ANC at least once, and 65% reported at least four visits [62]. Additionally, 70% of Palestinian refugees in Jordanian camps attended ANC at least four times [25]. This is similar to the current study, which observed high ANC attendance rates among Rohingya when compared to Bangladesh more generally and other South Asian countries where most women receive no more than three ANC visits [14, 22, 46, 54]. Fareed and Ismail [62] suggest that special proximity, free services, and incentives from NGOs to attend ANC impact the higher rates in refugee camps. In other words, the humanitarian aid sector can distort and render meaningless factors that are well established in the literature. For instance, household assets may not be a meaningful indicator in predicting ANC usage where assets and services are provided fairly equitably and free of charge and are thus possibly not reflective of the things that resources typically are measuring. The current study adds to this scarce body of research in refugee camp settings, suggesting that other, context-specific factors in the camps alter the roles of resources as a predictor of ANC. This would explain our findings of some camps seeming to be more supportive of adequate ANC usage.

Our finding that women who reported feeling healthier had later initiation may indicate unmet pregnancy care needs among healthier-feeling women and could be reflective of the need for better sexual and reproductive health education. Based on evidence from Bangladesh more generally, it is not uncommon for women not to reveal their pregnancy to their family until after the first trimester, which would preclude them seeking health services [63]. Given some similarity in culture and norms between Rohingya and host communities, this could be a factor for later initiation if they deem their physical health to be good.

Our finding that primiparous women showed higher rates of ANC aligns with previous research in Bangladesh [14, 64]. Duodu and colleagues [12] have argued that such an effect may reflect the confidence of women being able to handle another pregnancy on their own, especially if previous pregnancies and births occurred without complications. Primigravid women are at a higher risk of obstetric complications [65]. However, women with four or more pregnancies also face higher risk [66, 67]. This pattern has been described as a U-shaped association in the literature. However, in our data, ANC utilization did not follow a U-shaped pattern; women of higher parity did not show increased ANC use compared to those with one to three previous pregnancies. In addition, our data indicates that having had a previous adverse pregnancy or perinatal outcome, if anything, is associated with higher odds of later ANC initiation. We can only speculate for why this might be, though it seems worth exploring further.

Furthermore, maternal age is a well established factor in pregnancy outcomes, with adolescents and women aged 30 and above, facing higher risks of complications compared to those in their twenties [68, 69]. Despite these risks, age did not demonstrate a significant association with ANC utilization behavior. However, first-time pregnancy is highly correlated with being under the age of 18 as age of marriage in this population is very young. While the slightly higher ANC utilization among primiparous women is encouraging, there remains a need to raise awareness about pregnancy complications associated with high parity and maternal age below 20 or above 30 years within the community.

The trending associations between conversations with husbands and fathers-in-law (especially fathers-in-law) and lower odds of late ANC initiation was surprising given the rather conservative beliefs around reproductive health in the Rohingya community, where husbands’ and mothers-in-law’s knowledge and opinions often shape reproductive decisions [7, 42]. The association may reflect specific family structures or the women’s own beliefs and attitudes, leading to discussing these topics with their fathers-in-law, though this is purely speculative at this stage. Links between knowledge, attitudes, and ANC utilization have been observed in other low-income countries [70–72]. These insights highlight the importance of further exploring Rohingya women’s knowledge, attitudes, and beliefs regarding ANC and how these are shaped.

There does not seem to be a logical explanation as to why pregnant women who received information about ANC on the radio had moderately higher odds of delaying ANC until after the first trimester. This seems contradictory to a recent study that found that listening to the radio contributed to good STI and HIV knowledge among women in the Cox’s Bazar refugee camps [73]. Though knowledge does not always translate into behavior. We speculate that this association is likely spurious rather than reflecting a meaningful underlying relationship, though it might be worth exploring further what radio exactly they are listening to.

There are of course a few limitations that deserve some reflecting upon. First, it is worth pointing out that conducting research in the Rohingya context is incredibly difficult. The power differentials between data collectors from the local Bangladeshi community and the Rohingya community and the linguistic barriers, among other things, can contribute to misunderstandings and biased responses. Especially in light of the generally high number of ANC visits reported, it is not outside the realm of possibility that our participants were reporting on health center visits more generally, and not ANC visits specifically. Nonetheless, if they went to a clinic for something other than their pregnancy, the health centers would likely conduct whatever routine ANC check-up was due at that point.

While the majority of women are aware of being pregnant very early on, there was a small percentage of women whose self-reported date of last menstrual period and weeks pregnant were significantly off. This posed some challenges in determining ANC initiation in the absence of ultrasound determined gestational age. Nonetheless, our gestational age variable was triangulated between self-reported data, birth date of the baby, birthweight, and information taken from the health card if available. While there may still be a not insignificant amount of noise, we nonetheless trust our data to be a reasonably accurate representation of pregnancies in our study population.

Our data on assets, education, and literacy warrants further investigations. We were restricted by the RRRC in asking about literacy in certain languages, and we did not conduct any reading comprehension assessments. We know that formal schooling and literacy rates are very low in this population, but we also know that people use text messaging to communicate. Similarly, we suspect that our assets do not reflect the more nuanced household socioeconomic characteristics that typically drive behaviors such as health care seeking.

While the Andersen-Newman Behavioral Model provides a useful structure for examining determinants of ANC utilization in this context, it also has limitations. The model was originally developed in high-income settings, and although it has been widely applied and adapted for maternal and child health research in low- and middle-income countries, it may not fully capture the cultural, gendered, and structural dynamics that influence health-seeking in humanitarian settings. For example, decision-making authority within households, normative expectations around pregnancy care, and the perceived quality of available services are only partially represented by the sociodemographic and exposure variables included in our dataset. Similarly, the model places stronger emphasis on individual and household characteristics than on system-level barriers (e.g., facility readiness, supply chain interruptions, difficulty in getting to a health center) that may be highly salient in refugee camps.

Nonetheless, given the quantitative nature of our data, which included measures of household sociodemographic and resource characteristics and limited indicators of health information exposure, the Andersen-Newman model offered a parsimonious and well-established framework for organizing variables and interpreting associations. Future work, particularly with more qualitative or system-level data, could complement this approach by incorporating frameworks that foreground cultural norms (e.g., PEN-3) or structural barriers to maternal health (e.g., the Three Delays model).

Our study does not address questions pertaining to the quality of ANC. Aside from perceived quality being a pull-factor in individual decision making, the actual quality is a critical aspect of effective ANC. Thus, while our study seems to present a picture of availability and uptake of ANC at recommended levels, it fails to tell us anything about the effectiveness of the ANC systems in place.

The few findings that this study presented do have some important implications for humanitarian programming and service delivery. While primiparous women are seeking ANC earlier, there may be grounds to emphasize risks associated with higher party. However, according to our results, the radio does not necessarily seem to be the most effective medium to do so. Furthermore, as many women are moderately to severely constrained in autonomous mobility, identifying safe and culturally acceptable and trusted ways of transporting or accompanying pregnant women to health centers could be an effective strategy to encourage early initiation and sufficient number of ANC visits. Ultimately, it will be important to further explore Rohingya women’s and men’s knowledge, attitudes, and beliefs on SRH, and how decisions on SRH and ANC are made within households in order to better target, adapt, and disseminate information.

Conclusions

Our study suggests a fairly well-functioning ANC system in the Rohingya camps in Cox’s Bazar, Bangladesh, in line with the official statistics from the health sector. Particular factors contributing to relatively early and high uptake of ANC services remain somewhat unclear and should be explored more carefully in order to target specific messaging addressing some of the gaps in care. Despite our findings of high engagement with ANC, in reality maternal (295/100,000 live births) and under 5 (25.31/100,000 live births) mortality rates remain rather high in the Rohingya context [60]. Thus, future research should thus investigate the links between ANC frequency, initiation, and quality, and adverse pregnancy and perinatal outcomes. This study provides insights into the use of ANC services in the Rohingya context, which constitutes an important component in ensuring maternal and offspring health in a very vulnerable population, and provides insights into possible improvements in reproductive health care in humanitarian contexts.

Abbreviations

ANC

Antenatal care

CXB

Cox’s Bazar

IRR

Incidence rate ratio

iRRRd

Intergenerational risk and resilience of Rohingya in displacement

LMIC

Low- and middle-income country

aOR

Adjusted odds ratio

OR

Odds ratio

PREI

Perceived refugee environment index

RRRC

Refugee Response and Repatriation Commissioner

SES

Socioeconomic status

SRH

Sexual and reproductive health

WHO

World Health Organization

Author contributions

DS contributed to conceptualization of this paper and led in analyses and first draft of manuscript. DM supported analyses and write up of methods and results. MSR supported analyses and provided details and background on field procedures. MCH and AKR supported survey preparation, and interpretation and contextualization of results. KIS contributed to the conceptualization and design of and provided scientific oversight to iRRRd, and reviewed and provided comments on the manuscript. FT co-led the conceptualization and design of iRRRd and reviewed and provided comments on the manuscript. AJW (senior author) co-led the conceptualization and design of iRRRd, supervised and mentored the lead-author on conceptualizing and executing this paper, provided guidance on analyses and substantial feedback and edits on the manuscript throughout the process, and led the R&R process.

Funding

Generous funding for this study was provided by the LEGO Foundation through the Play to Learn initiative.

Data availability

Upon request.

Declarations

Ethics approval and consent to participate

The study was approved by the ethical review committee of icddr, b (PR-22064) and from New York University (IRB-FY2021-4875).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.