Menu
Menu
Menu
Menu
Background: Health facility delivery has been described as one of the major contributors to improved maternal and child health outcomes. In sub-Saharan Africa where 66% of the global maternal mortality occurred, only 56% of all births take place in health facility. This study examined the individual and contextual predictors of non-use of health service for delivery in Nigeria where less than 40% births occur in health facility. Methods: Data from 2013 Nigeria Demographic and Health Survey (DHS) involving 20,192 women who had delivery within 5 years of the survey were used in the study. Multilevel multivariable logistics regression models which had the structure of non-use of health service for delivery defined at individual, community and state levels were applied in the analysis. Spatial analysis was also used to capture the locations where the phenomenon is prevalent in the country. Results: About 62% of the women did not utilize health service during delivery. More than three-quarter of those with no education and 92% of those who did not attend antenatal clinic during pregnancy never utilized health service for delivery. The odds of non-use of health service during delivery increased for women who had no education, from poor households, aged 25-34 years, unmarried, never attended antenatal clinic, experienced difficulty getting to health facility and lived in the most socioeconomically disadvantaged communities and states. Conclusions: This study has demonstrated that non-utilization of health service for delivery is influenced by individual, community and state level factors, with substantial proportions of women not utilizing such service residing in the northern region of Nigeria. Each level should be adequately considered in the design of the appropriate interventions.
Analyses in this study were done using the 2013 Nigeria Demographic and Health Survey (NDHS) data set. The survey is cross-sectional, population-based and provides information on population and health characteristics. A multi-stage cluster sampling method was used in the 2013 NDHS. The country was categorised into 37 units which included all the 36 states and the Federal Capital Territory (FCT), Abuja. A total of 896 communities (clusters) were selected from these states using the primary sampling unit (PSU) of the 2006 population and census enumeration areas. The chosen communities were further disaggregated into enumeration areas in which 532 were created in the rural areas while the urban areas had 372. Households were then randomly selected from the enumeration areas. A total of 40,680 households were finally chosen with 23,940 and 16,740 in the rural and urban areas respectively. Details of data collection have been published elsewhere [16]. Questionnaires were used to obtain information from women aged 15–49 years through household interviews. Such women were asked to provide information about their socioeconomic characteristics, reproduction, breastfeeding practice, domestic violence, child care practice and health service use during pregnancy, delivery and postnatal period. The study focused on women aged 15–49 years who gave birth to children within five years of the survey. Women who delivered at health facility, either private or public, were defined as utilizing health service for delivery while those who delivered elsewhere were defined as not utilizing health service for delivery. The former was subsequently defined as a binary variable assuming the value of 1 while the latter assumed the value of 0. The variables that constituted individual level factors include: age, education, household wealth index, occupation, marital status, mass media exposure and antenatal care attendance. Age was defined as 15–24 years, 25–34 years and 35+ years. Education was expressed as no education, primary, secondary or higher education. Since participants’ response on income in developing countries is often characterised with inaccuracy, household wealth index was used as a measure for wealth status. This wealth index was obtained by considering the ownership of household commodities such as television, radio, type of roofing/floor, water source and dwelling features. This approach, which is based on principal component analysis, has been used by the World Bank to define household poverty level [17, 18]. Although DHS presented the wealth index in five quintiles, we regrouped these quintiles into three tertiles (poor, middle and rich). Occupation was grouped into working and not working. Marital status has two categories: ever married and never married. Mass media exposure was defined as ever exposed for those who have access to at least one of newspaper, radio or television, and never exposed for those who have access to none. Antenatal care attendance was grouped into women who never attended, those who had less than 4 visits and those who had 4 or more visits. The following factors were considered at community level: place of residence (rural or urban), getting to health facility (being a problem or not a problem), ethnicity diversity index and socioeconomic status. Socioeconomic status was derived from the proportions of individuals who are unemployed, illiterate and poor. This was then categorised into tertile 1 (least disadvantaged), tertile 2 and tertile 3 (most disadvantaged). Ethnicity diversity index was a variable obtained using the formula: Where: xi = population of ethnic group i of the area, y = total population of the area, n = number of ethnic groups in the area. It reflects the spread of ethnic groups by calculating values from 0 to 1. This is then multiplied by 100 to arrive at the diversity [19]. The higher the value the more widespread the community. While an index of 0 indicates a community is mono-ethnic in nature, an index of 1 shows that such a community is multi-ethnic in nature. The state-level factor was derived from the proportions of individuals in the state who are unemployed, illiterate and poor. This was then categorised into tertile 1 (least disadvantaged), tertile 2 and tertile 3 (most disadvantaged). In the descriptive analysis which involved the use of Chi-Square test, the independent variables at each level were presented using numbers and percentages. A three-level binomial regression model consisting of individual, community and state was constructed due to the hierarchical nature of the data set. Four models were thereafter specified. In the first model which was specified in order to decompose the amount of variance found between the community and state levels, no explanatory variables were included. Individual and community level variables were included in the second and third models respectively. The last model contained the state level variables in addition to the variables from individual and community levels. The results of fixed effects were presented in terms of odds ratios (OR) together with their 95% credible intervals (CrI). Results of random effects were presented using three measures: the intra-cluster correlation (ICC), variance partition coefficient (VPC) and median odds ratio (MOR). MOR measures cluster heterogeneity that remains unexplained. Information on the procedure for computing MOR has been published elsewhere [20, 21]. While goodness of fit of the model was checked using Bayesian Deviance Information Criterion (DIC), multicollinearity was assessed by applying Variance Inflation Factor (VIF). MLwiN 2.35 [22] calling Stata Statistical Software version 14 (Stata, 2015) was used to carry out all the multilevel modelling operations. Also, the operation involved Markov Chain Monte Carlo (MCMC) estimation [23]. Results of the spatial analysis were presented using percentile map, excess risk map, global spatial autocorrelation (Moran’s I) map and funnel plot. Percentile map showed the prevalence of non-use of health service for delivery in four categories: low prevalence (3–10%); moderate prevalence (10–25%); high prevalence (25–45%) and; very high prevalence (45–70%). The excess risk map revealed the expected number of women versus the observed number of women who did not utilize health service for delivery. States with value greater than 2 are considered to have excess risk above the expected while states with value less than 2 are considered to have excess risk less than the expected. Global spatial analysis (Moran’s I) presented the distribution of non-use of health facility for delivery in four groups: High-high: this indicates high rate of non-use of health service for delivery in a particular state with the adjoining states experiencing high rates of non-facility delivery. Low-low: low rate of non-use of health facility for delivery in a state with the adjoining states having low rates as well. High-low: high rate of non-use of health service for delivery in a state with the adjoining states experiencing low rates of non-facility delivery. Not significant: this group involves states with values that are not statistically significant. The spatial analysis was performed by applying the exploratory spatial data analysis (ESDA) method using GeoDa software [24].
N/A
