Study Setting and Data Source
The study was retrospective and cross-sectional in design. Data were extracted from the nationally representative 2018 Nigeria Demographic Health Survey (NDHS) conducted by ICF Macro Calverton, Maryland, USA in conjunction with the Nigeria National Population Commission (NPC), Nigeria . Administratively, Nigeria is divided into 36 states and the Federal Capital Territory (FCT). The states are further grouped into 6 regions as shown in Fig. 1. Although the regions have no administrative functions, people within each region are deemed to have similar characteristics, culture, ethnicity, vegetation etc. Each state is subdivided into local government areas (LGAs). Using the most recent 2006 Population Census, each LGA was subdivided into convenient areas called census enumeration areas (EAs).
A two-stage probability sampling was adopted to select the respondents who were women of child-bearing age (15–49). The EAs were the primary sampling unit (PSU). In each state, depending on the population, about 38 EAs were selected with probability proportional to EA size making a total of 1,400 EAs in the first stage. Then a full list of households in all selected EAs in the first stage was drawn to serve as the sampling frame for household selection in the second stage. In the second stage, 30 households each were selected in every EAs using equal probability systematic sampling. All eligible women of reproductive age (15–49 years) in each selected household were interviewed. Due to the non-proportional allocation of sample sizes at the different states, LGAs and EAs, as well as the possible differences in response rates, sampling weights were applied in all the descriptive analysis. A total of 41,821 women aged 15–49 years were interviewed .
All respondents (women aged 15–49 years) were asked if they had any pregnancy or birth within 5 years preceding the survey (2013 to 2018). Those who had at least one birth were thereafter asked questions relating to conception, pregnancy-related care including ANC visits made, time ANC began, ANC care provider, etc of each birth delivery starting from the most recent. Our analysis is based on the information provided on the most recent deliveries and the associated pregnancy information. Among the 21,785 women who provided information about the ANC visits during their most recent pregnancy within the five years that preceded the survey, 16,448 (75.5%) made at least one ANC visit. We used all pregnancy history from 2013 to 2018 because the 2016 WGO guideline only reaffirmed the preceding guidelines in the 1990s, and 2006. Furthermore, of those that had at least one ANC contact, only 11867 (72.1%) provided the ANC components received during all ANC contacts for the most recent pregnancy. Our analysis was based on the components of ANC received by the 11867 women.
The outcome variable in this study is the quality of ANC services using the number of ANC components received as a proxy. The identified components were: receiving tetanus injection, intestinal parasite drugs (IPD), blood pressure, urine test, blood test, iron supplement, three or more doses of malaria intermittent preventive treatment in pregnancy (IPTp), told about danger signs, and counselled on HIV/PMTCT (either talked about HIV transmitted mother to child, things to do to prevent getting HIV, getting tested for HIV). These components have been recommended and recognized in the literature [7, 8, 15, 18, 20, 22].
Having ANC components could be influenced women and their household compositional factors as well as the contextual based on community and state of residence and the health system. These include women and their spouse background characteristics, their employment status, and household wealth status that could determine the type of facilities women patronize. Similarly, the birth order and preceding birth interval, as well as family mobility, could affect health care utilization. Due to the secondary nature of the data used, the health system factors considered in this study are having health insurance, the skill of the ANC provider and whether the ANC was received at a health facility or not. Having health insurance could dictate the type of health facility visited and by extension the skill of the ANC providers which are major influencers of giving ANC components to women. Of importance are the community factors (such as poverty level, illiteracy, rural residence, poor media access) and societal factors such as religion and ethnicity as they could affect the uptake of components.
Based on existing literature [15, 17, 20, 21, 23, 24], the independent variables are maternal age (15–19,20–24, 25–29, 30–39, 40–49 years), education (no education, primary, secondary and higher), spouse education (no education, primary, secondary and higher), employment status (currently employed or not), spouse employment status (currently employed or not), access to media (at least one of radio, television, or newspaper), household wealth tertiles (lowest, middle, and highest), women's autonomy using “who decides respondents health care” (respondent alone, respondent/spouse, and spouse alone) as a proxy. Others are birth interval (firstborn, < 36 months, and > = 36 months), birth order (1, 2, 3, 4 and 5+), children ever born (none, 1–2, 3–4, 4+), current marital status (currently married or living together, divorced/separated/widowed, never married), place of residence (rural/urban), religion (Islam, Christian, others), and ethnicity (Hausa/Fulani, Igbo, Yoruba, and others). Family mobility (had stayed less than five years at residence or not), wanted child when became pregnant (then, later or not more), household headship (male or female), health insurance coverage (yes or no), place of ANC (health institution or not) and ANC caregiver (skilled or unskilled).
We also assessed four community-level factors in the descriptive analysis: poverty rate (high or low), unemployment rate (high or low), illiteracy rate (high or low), and media access rate (high or low). The communities are synonymous with the EAs. We computed the community socioeconomic (SES) disadvantage composite score using principal component analysis of the proportion of respondents within each community with no media access, who are illiterates, who are poor, and who are unemployed. The SES was categorized into three tertiles: lowest, middle, and highest. The only state-level characteristics considered was the proportion of the rural population in the states of residence. It was categorised as low rural proportion (0–33.3%); medium rural proportion (33.4–66.7%) and high rural proportion (66.8–100%).
Data were analysed using descriptive statistics, bivariable, and multivariable logistic regression using STATA version 16 (Stata Corp, Texas, USA). We invoked the “SVY” command in STATA to adjust for the study design and the sampling weights. Frequency tables showing percentages were used to describe the distribution of study respondents’ characteristics and the distribution of outcome variables by the respondents’ characteristics (Tables 1).
We fitted a multilevel Bayesian Markov Chain Monte Carlo (MCMC) Poisson-based Generalized Linear Model (GLMs) to the data, with women nested within EAs and the EAs nested within the states. The models have mixed outcomes consisting of the fixed and random parts as shown in Eq. (1).
The “risk” that pregnant woman i of community j from state k will receive ANC component is denoted by γijk, Uojk is the random effect of mothers community j in state k and Vok is the random effect of state k, eijk is the noise such that , and in a model with t covariates.
We reported the measure of the rates of receiving the components as incidence rate ratios (IRRs) with their 95% credible intervals (CrI). Measures of variations were explored using the intraclass correlation (ICC) and median incidence rate ratios (MIRR)[25, 26]. The ICCs, an equivalent of the variance partition coefficient (VPC), is the percentage of the total variance in the risk of a pregnant woman obtaining the components that is related to the community and state where they live (i.e. a measure of clustering of risk of child mortality in the same community and state). We also estimated the proportion of total variance which are accounted for at the community and the state levels. The MIRR is the estimate of the probability that a pregnant woman will receive additional components attributable to the community and state context. The Bayesian MCMC Multilevel Poisson model was implemented using MLwin v3.03 and implemented in Stata V16 with the following parameters: Burnin = 5000; Chain = 50000, Thinning = 50.
At the bivariable level, we first identified all variables that were significant at p < 0.20. The identified independent variables were then used as candidate variables in the multivariable model from where we identified the adjusted odds ratios of characteristics associated with the quality of ANC services. The “collin” command in Stata was used to identify collinear variables and the associated variance Inflation factor (VIF). Based on the VIF, the less important of a pair of collinear variables were dropped from the multiple regression. We used the principal component (PCA) and factor analysis procedure to explore clustering among the components. The PCA method maintains all theoretically relevant variables and avoids the negative influence of high inter-correlation among the variables [27, 28]. The “cluster single-linkage” and “loadingplot” commands in Stata were used to analyze the data and visualize the clustering respectively.
This study was based on the analysis of existing survey data. The Institutional Review Board (IRB) of ICF Macro at Fairfax, Virginia in the USA reviewed and approved the MEASURE Demographic and Health Surveys Project Phase III. The 2010–2018 DHS’s are categorized under that approval. The IRB of ICF Macro complied with the United States Department of Health and Human Services requirements for the “Protection of Human Subjects” (45 CFR 46). Written informed consent was obtained from every study participant before participation and all information was collected without identifiers and kept confidentially. The full details of the ethical approvals can be found at http://dhsprogram.com.