Menu
Menu
Menu
Menu
Background: Timely and appropriate health care during pregnancy and childbirth are the pillars of better maternal health outcomes. However, factors such as poverty and low education levels, long distances to a health facility, and high costs of health services may present barriers to timely access and utilisation of maternal health services. Despite antenatal care (ANC), delivery and postnatal care being free at the point of use in Burundi, utilisation of these services remains low: between 2011 and 2017, only 49% of pregnant women attended at least four ANC visits. This study explores the socio-economic determinants that affect utilisation of maternal health services in Burundi. Methods: We use data from the 2016–2017 Burundi Demographic and Health Survey (DHS) collected from 8941 women who reported a live birth in the five years that preceded the survey. We use multivariate regression analysis to explore which individual-, household-, and community-level factors determine the likelihood that women will seek ANC services from a trained health professional, the number of ANC visits they make, and the choice of assisted childbirth. Results: Occupation, marital status, and wealth increase the likelihood that women will seek ANC services from a trained health professional. The likelihood that a woman consults a trained health professional for ANC services is 18 times and 16 times more for married women and women living in partnership, respectively. More educated women and those who currently live a union or partnership attend more ANC visits than non-educated women and women not in union. At higher birth orders, women tend to not attend ANC visits. The more ANC visits attended, and the wealthier women are; the more likely they are to have assisted childbirth. Women who complete four or more ANC visits are 14 times more likely to have an assisted childbirth. Conclusions: In Burundi, utilisation of maternal health services is low and is mainly driven by legal union and wealth status. To improve equitable access to maternal health services for vulnerable population groups such as those with lower wealth status and unmarried women, the government should consider certain demand stimulating policy packages targeted at these groups.
We use data from the nationally representative 2016–2017 Burundi Demographic and Health Survey (DHS). The study sample consists of 8941 Burundian women who reported at least one live birth in the five years preceding the survey. For women who had multiple births, we consider the data for the most recent pregnancy with the view of minimising recall bias. The dependent variables used for analyses are presented in Table 2. These are utilisation of ANC services provided by a qualified health provider, the number of ANC visits, and the place of delivery and birth assistance. These variables are key indicators for the monitoring of maternal health care [40–42]. Dependent variables We included individual-, household-, and community-level independent variables identified by a review of literature and an understanding of the country context (Table (Table33). Independent variables In most cases, we did not need to recode or recategorize the independent variables. However, we did need to recategorize some variables for ease of analysis and precision of estimates. The age of women was categorised as follows: between 15 and 19 years (adolescents), between 20 and 34 years (youth) and between 35 and 49 years (adults). This definition is consistent the African Youth Charter (AYC) which is a legal continental framework used in most of African countries [43]. The birth order was categorised as follows: first, second to third, fourth to fifth, and more than sixth birth orders. The original dataset categorised religion into eight categories. These were collapsed into four namely: Catholic, Protestant, Muslim, and traditional practitioners, to account for the demographic composition of the population and to avoid an enlargement of the error term in the regression due to large differences in values between categories. Marital status was also coded into four categories: women who never lived in union, those who currently live in legal union, those currently living in partnership, and women who ever lived in union but are currently living alone. In the fourth category of women, we combined widows, divorced women, and those who were separated because they have one civil status in common – the transition out of union. For the family size variable, we based our categories on Burundi’s family and reproductive health advice. According to this, couples should have a maximum of three children. Therefore, the family size took a value of 1 for families whose household did not exceed five individuals (three children and two parents), and 2 for households of more than five individuals. All the 18 provinces of Burundi were included in the analysis. We use logistic regression to estimate the following empirical model to understand women’s likelihood of utilising ANC services provided by a trained health professional, controlling for individual-, household-, and community-level characteristics: Here, the dependent variable is the log odds that a woman i will choose alterative j relative to alternative 0, where 0 = non-use of ANC services from a trained provider; and 1 = consultation with a medical doctor, nurse or midwife. Independent variables are grouped into three categories; namely individual-level factors represented by a standard vector of covariates X, household-level determinants corresponding to the standard vector of covariates Y, and community-level determinants represented by the standard vector of covariates Z. The model includes a dummy variable that captures provincial effects. β0 captures fixed effects and β1,2,3 detect random effects on the probabilities of using ANC services from a trained provider. We then estimate the effect of individual-, household-, and community-level factors on the number of ANC visits using linear regression. The empirical model is specified as: Where, the outcome anc _ visitsi is continuous and represents the number of ANC visits that a woman i attends during her pregnancy. X,Z, and Y are the same standard vectors used in the logistic model. This model concerns women who reported attending one or more ANC visits. For the linear regression model, we used 95% p-value to ascertain significance of coefficients. We use a multinomial logistic model to explore women’s likelihood of seeking delivery services from a trained birth attendant. The empirical model is given by: Where the response variable is the log odds that a woman i will choose delivery alterative j (j = 1, 2) relative to 0, where 0 = home delivery without assistance by a skilled birth attendant; 1 = home delivery with assistance by a skilled birth attendant; and 2 = health facility delivery. Independent variables are represented by standard vectors of covariates X,Y, and Z used in previous models. In addition to standard covariates, this model includes the number of ANC visits as this has been found to positively predict assisted delivery [44]. All models assume that community-level effects are invariant for women living in the same setting. With an aim to attempt validate the assumption, we use two variables to account for province and residence. Residence is binary; rural versus urban; and there are 18 provinces each having a rural and an urban component. As such, community-level factors are assumed to be constant for women living in residence i within province j. All discrete models used 95% confidence intervals to ascertain significance of coefficients, which the literature claims to be more reliable [30, 45]. This study used DHS datasets. No patients or members of the public were involved in the design, analysis or reporting of this study.
N/A
