Study setting
The republic of Benin is located in West Sub-Saharan Africa and reports a maternal mortality ratio of 400 per 100,000 live births [11]. Benin ranks in the bottom third of countries recently assessed for ANC coverage and in 2009, the Ministry of family and National Solidarity reported 69% of Beninese women suffered abuse at least once in their lifetime [9, 12,13,14]. More recently, in 2019, the UN Women’s Global Database on Violence against Women reported a 24% prevalence of lifetime physical and/or sexual IPV in Benin, with no report of emotional of psychological IPV [15]. Cross-sectional studies from previous Demographic and Health Surveys (DHS) found close to 40% of mothers received less than 4 ANC visits or never attended services, suggesting household wealth and female education as key factors [16, 17]. With the existing issue of high maternal mortality rates, low utilization of necessary antenatal care services and exposure to IPV in Benin, the aim of our study was to examine the impact of intimate partner violence on utilization of ANC services by women in Benin, where utilization is suboptimal and rates of violence against women are high.
Data source
The data used for this study were collected from the 2017–2018 Benin Demographic Health Survey (BDHS) [18]. This nationally representative population and health survey was conducted in Benin by the Ministry of Planning and Development and the National Institute of Statistics and Economic Analysis (INSAE) in collaboration with the Ministry of Health, Centre for Parasitology & Cardiology Laboratory of the National Hospital and the Permanent Secretariat of the Food and Nutrition Council. Technical support was provided by ICF International and funding support by UNAIDS, UNICEF, UFPA and the World Bank [18].
Sampling
The sampling process is described in detail in the 2017–2018 BDHS report; results are representative of the country as a whole and of urban and rural areas separately for 12 administrative departments in Benin [18]. A total of 14,435 households were selected, 16,233 females aged 15–49 were eligible for interviews and 15, 928 females were actually interviewed (98.1% response rate). A total of 3317 women were eligible for our study (i.e. answered the ANC question and the domestic violence module). After completing listwise deletion, the final sample size was 3084 Beninese women aged 15–49.
Variable selection and measurement
The outcome variable was ANC utilization by ever-partnered women aged 15–49 in Benin in 2017–2018. Although the WHO revised the recommendation from at least a four-visit to at least an eight-visit model in 2016, only 54% of the women in sub-Saharan Africa reached four ANC visits, and for the few countries which reached the recommended eight ANC visits, the quality of health services and distribution across the region was an issue [1, 19]. In order to adequately assess the utilization of ANC in Benin, we used both the basic antenatal care coverage model of at least four-visits outlined by the WHO Global health Observatory [20] and the recently recommended at least 8 visit model. The ratio variable ANC-4 was recoded into a binary level measurement; 0 for “0 to 3” visits and 1 for “4 or more” ANC visits during last pregnancy for the basic model. For the updated at least 8 visit model, the variable (ANC-8) was recoded as 0 for “0 to 7” visits and 1 for “8 or more’ ANC visits during the last pregnancy.
The main independent variable was intimate partner violence (IPV), which was recoded to a dichotomous form. IPV included 10 questions on emotional, sexual and physical violence experienced by the women (i.e. ever: pushed, slapped, twist arm/pull hair, punch/hit with something, kick/drag/beat, choke/burn, threaten with knife/gun/weapon, physically force sexual intercourse, physically force sexual acts, force sexual acts with threats) [18]. Each question was recoded into a dummy variable; if the answer was “often”, “sometimes” (which pertain to violence in the last 12 months) and “yes, but not in the last 12 months” for any of the questions on emotional, sexual or physical violence, it was coded yes = 1 and if the answer was never then it was no = 0. The question dummy variables were summed to generate a summed dummy variable that provided a ratio level variable accounting for women who never experienced IPV and women who ever experienced at least one form of violence (answered yes to one or more of the questions). These were then recoded into a binary variable of ever experienced IPV where never = 0 and ever = 1 or more.
The exploratory variables were used as controls to inform the predictive models and outcome of utilization of ANC services in Benin. Decision making was created using four questions on participation in household decision making (i.e. person who usually decides on: respondent’s health, large household purchases, visits to family/relatives, what to do with money husband earns) [18]. Decisions made by “respondent alone” or “respondent and husband/partner” were recoded as 1, decision making. The decisions made by ‘husband/partner alone”, “someone else”, or “other” were recoded as 0, no decision making. A summed dummy variable provided a count which was recoded to generate a dichotomous form of no decision making (participated in no decisions = 0) and some decision making (participate in one or more decisions = 1).
Age of woman was divided into three sub-groups (15–24, 25–34, and 35–49); the youngest subgroup was the reference for regression analysis. Two categories existed for place of residence: rurality (recoded as rural = 1, urban = 0) and urbanicity (recoded as urban = 1, rural = 0). The DHS wealth index (poorest, poor, middle, rich, richest) was used and the poorest quintile was the reference group in regression analysis. In DHS, the wealth index is a composite measure of cumulative living standard of a household based on ownership of select assets (i.e. televisions, bicycles, cars, dwelling characteristics, type of drinking water source, toilet and sanitation facilities) [18].
The level of education for respondent and partner were divided into: (i) no education (reference group) and included women or partners with no education and incomplete primary education; (ii) primary education which included women or partners who had primary and incomplete secondary education; and (iii) secondary or higher which included individuals with secondary or higher education. The DHS asks the female participants if they are ‘working’ and this was recoded as 1 for “yes” and 0 for “no”. For the husband or partner employment, if they “did not work” it was recoded as 0, not employed, and if they had employment (professional, sales, agricultural, services, skilled manual, other unclassified or don’t know), it was recoded as 1, employed. The questionnaire also provided insight into husband or partner’s alcohol drinking and answers were recoded as 1 for “yes” (drinks alcohol) and “no” (does not drink alcohol).
Religion was subdivided into Muslim, Christian, Catholic, Traditional, Other or no religion; Muslim was the reference group in the regression analysis based on previous studies looking at religion and ANC [21]. The number of children was divided into one child (reference group), two to four children and five or more children. Finally, attitude toward wife beating was included as an exploratory variable based on empirical evidence in Benin [14]. The five questions regarding ‘beating being justified if the wife goes out without telling husband, wife neglects the children, wife argues with husband, wife refuses to have sex with husband, wife burns food [18] were recoded 1 if the answer was “yes” to any of the questions. If participants answered “no” or “I don’t know”, it was recoded as 0. A summed dummy variable provided a count which was recoded to generate a dichotomous form of justified = 1 and not justified = 0. Detailed syntax for coding structure and running the analysis is available (see Additional file 1).
Statistical analysis
The complex sample design of the DHS was accounted for in significance testing with sample weights using STATA’s svyset and svy commands. Statistical significance between groups were estimated using bivariate regression tests for women who ever experienced IPV and those who never experienced IPV. To investigate the relationship between IPV and ANC utilization, multiple logistic regression was carried out with all of the covariates included in the model. The data were analyzed using the software package STATA version 16 [22].