Data Source
The study adopted a cross-sectional research design that utilised secondary data extracted from the 2018 Nigeria Demographic and Health Survey (NDHS). The NDHS was implemented in the country by the National Population Commission (NPC) in collaboration with the National Malaria Elimination Programme (NMEP) [40]. The Demographic and Health Survey (DHS) Programme provided technical support for the implementation of the survey. Major international organisations such as Bill and Melinda Gates Foundation, The Global Fund, World Health Organisation (WHO), United States Agency for International Development (USAID), and United Nations Population Fund (UNFPA) provided funding support for the survey. The 2018 NDHS provided reliable information on basic demographic and health characteristics of the Nigerian populace including national information about malaria infection, treatment and control [40]. The data were analysed with authorisation from MEASURE DHS, and its available online via https://dhsprogram.com/data/.
Procedure
The 2018 NDHS sampled survey participants using a multi-stage sampling technique that ensured a nationally representative sample. The whole country was stratified into urban and rural areas after which some urban and rural areas were selected randomly. The selection was based on localities used as Enumeration Areas (EAs) in the penultimate national housing and population census. The EAs served as the primary sampling unit (cluster) in the survey. In the selected EAs, households were randomly selected for the survey following appropriate household listing. From the selected households, eligible men aged 15-59 and women aged 15-49 were randomly selected for the survey. Comprehensive details of the methodology adopted for the 2018 NDHS have been widely published [40].
Participants
Though, the 2018 NDHS covered 41,821 women of reproductive age. Women who had no live birth in the past year preceding the survey (34,549 women), women who had less than three antenatal care visits (2,359 women), and women who do not know their number of antenatal care visits (178 women) were excluded from the analysis. The inclusion criteria were having had at least a live birth in the past year preceding the survey, and having had at least three antenatal care visits during the last pregnancy. These criteria resulted in a weighted sample size of 4,772 women.
Measures
The outcome variable in the study was usage of IPTp through the use of Sulfadoxine-pyrimethamine (SP) under the brand name Fansidar. This was measured in the study by the number of IPTp dose a woman received during her last pregnancy. We categorised the responses into two, namely optimal (woman receives 3 or more doses) or otherwise (woman receives less than 3 doses). This categorisation was not only based on the WHO recommendation that pregnant women should take at least 3 doses of IPTp before delivery [6]. It was also consistent with categories adopted in many existing studies [23, 22, 14, 13]. The explanatory variables examined in the study cut across individual and community levels, and were selected based on existing literature [21-22, 13-14, 18-19]. The variables were however divided into predisposing, enabling and need factors in line with the theoretical constructs of the Andersen model. The predisposing factors are age group (15-24, 25-34, and 35+), parity (primiparity, multiparity, and grand multiparity), formal education (none, primary, secondary, and higher), employment status (unemployed or employed), autonomy on own health care (not autonomous or autonomous), household wealth (poorest, poorer, middle, richer, and richest), place of residence (urban or rural), geo-political zone (north-central, north-east, north-west, south-east, south-south, and south-west), proportion in community who perceived that malaria can cause death (low, middle, and high), and proportion in community who perceived that malaria is easy to treat (low, middle, and high).
The enabling factors are health insurance enrolment (not enrolled or enrolled), source of antenatal care (government or private), partner education (none, primary, secondary, and higher), community literacy level (low, middle, and high). The need factors are experience of death of a child (ever or never experienced), timing of first antenatal visit (first trimester, second trimester, and third trimester), possession of mosquito bed net for sleeping (no or yes), and actual sleeping under mosquito bed net (no or yes). A number of these variables were re-coded in the study. The community variables were generated from individual responses through aggregation at the cluster level, and then divided into three equal proportions (low, middle and high), using tertile value as cut off reference. This method is generally used for the generation of community variables using DHS data sets [41-45].
Data analysis
Two types of analyses were carried out. Firstly, frequency distribution was used to describe the socio-demographic characteristics of the respondents as well as the realised access to IPTp. Secondly, two multilevel mixed-effects logistic regression models were fitted to examine the predictors of optimal usage of IPTp. Prior to fitting the model, three mini analyses were carried out. One, variables were selected into the two models based on results of a bivariate analysis. Any variable not showing significance at p<0.025 was not selected. Two, a Variance Inflation Factor (VIF) was performed to ensure no multicollinear variable was selected for modelling. The bench mark for this test was that no variable with VIF score of five or higher values should be selected into the models [46]. Three, a ‘null model’ was fitted. This model did not include any explanatory variables. The essence of the null model was to ascertain whether significant variation exists in the optimal usage of IPTp across the communities. This is determined by the significance of the intercept of the model. Model 1 used the predisposing and enabling factors to explain inequitable use of IPTp in the country. Model 2 used all the predictor variables to examine equitable use of IPTp in Nigeria. The analytical tool adopted not only aligned with the theoretical position of the Andersen model but is also suitable for examining predictors of an outcome with hierarchical influences such as individual and community levels. This tool is widely applied in multilevel studies [41-45]. The multilevel mixed-effects logistic model partitions influences on an outcome into fixed and random effects [47]. The fixed effects in the current study were examined using the adjusted Odds Ratio (aOR) while the random effects were examined using the Intra-Cluster Correlation Coefficient (ICC). The ICC which ranges from zero to one indicate the importance of the community factors in the overall variance observed in the outcome variable. The models were check for adequacy using the Akaike Information Criterion (AIC). The model with the lowest AIC value is dim to have the best goodness-of-fit. All analyses were performed using Stata 14 [48]. Statistical significance was set at 5%.