Multivariate decomposition of trends, inequalities and predictors of skilled birth attendants utilisation in Nigeria (1990-2018): a cross-sectional analysis of change drivers

Objectives Literature has assessed skilled birth attendants (SBAs) utilisation, but little is known about what contributes to the changes in SBA use. Multivariate decomposition analysis was thus applied in this study to examine; levels, trends, inequalities and drivers of changes in SBA utilisation. Design and setting A cross-sectional analysis of five-waves of NDHS-data (1990, 2003, 2008, 2013, and 2018), collected through similar multistage sampling across the 36 states and the federal-capital-territory of Nigeria. Participants Women of reproductive age (15-49 years), and with at least one birth in the last 5 years preceding each of the surveys. Main outcome measure SBA use is the response variable while explanatory variables were classified into; Demographics, Health, Economic and Corporal factors. Methods Chi-square test for trends of proportions across the ordered survey years assessed trends in SBA use. MDA that quantifies and partition predictors effect into endowment and coefficient components evaluated contributors to changes in SBA use. Statistical analysis was carried out at a 95% confidence interval in Stata 16. Results SBA use increased with significant (p<0.05) linear trends by 12% between 2003 and 2018. The decomposition analysis showed that differences in characteristics (endowment) accounted for 11.5% of the changes while the remaining 88.5% were due to differences in effects (coefficient). SBA utilisation rises by 61% when respondents decided on her health compared to when such decisions were made by the spouse. Utilisation of SBA, however, fell by 88% among women who reside in the states with high rural populations percentage. Conclusions SBA use remained low in Nigeria, and slowly increase at the rate of <1% yearly. Women health decision-making power contributed most to positive changes. Residing in states with high rural populations has a negative impact on SBA use. Maternal health programmes that strengthen women's health autonomy and capacity building in rural communities should be encouraged. The study is a secondary analysis of data extracted from the five successive NDHS conducted in 1990, 2003, 2008, 2013, and 2018. The NDHS is a cross-sectional population-based nationally representative survey, routinely collected across all states and the Federal Capital Territory (FCT) of Nigeria. The sampling design is similar across the surveys. The survey usually uses stratified and multistage sampling techniques that accommodate household clusters (primary sampling unit) of respondents providing information on their demographic status and reproductive health behaviours of women aged 15–49 years. Nigeria is divided into six geopolitical zones called regions and each region is subdivided into states and FCT. As of 1990, Nigeria has 21 states. These were then divided and grouped into 30 states and the FCT in 1991. Additional 6 states were created in 1996, which resulted in the present number of 36 states as shown in figure 1. Map of Nigeria showing the 36 states and the federal capital territory, by the geopolitical zones. Similar two-stage cluster sampling was used in each of the five-waves of the survey. The 36 states and FCT were subdivided into local government areas (LGAs) whereby rural and urban LGAs were separated. Enumeration areas were selected from the LGAs at the first stage and households were then selected at the second stage where all women aged 15–49 years in the selected households were interviewed. In-depth information on the NDHS sampling methodology where 8781, 7620, 33385, 38 984 and 41 821 women participated in 1990, 2003, 2008, 2013 and 2018 surveys respectively have been documented.10 11 28 No patient was involved. We applied the strobe reporting guidelines.29 Utilisation of SBA during the last childbirth within 5 years preceding each survey was the outcome variable and was measured as whether birth was assisted by skilled provider or not.4 14 30 Skilled delivery services are rendered by doctors, nurses, midwives and auxiliary nurses/midwives. Independent/explanatory variables that includes; maternal age, education, ANC visit, parity, socioeconomic status and place of residence that are consistently associated with SBA use in Nigeria and SSA were studied.4 7 19–22 31–33 Other set of independent factors associated with the use of SBA and captured in each survey year of the DHS were included.10 11 28 To ensure uniformity in all the survey data used, independent variables were classified in this study as; demographic/societal, women health, economic and corporal factors based on extensive literature search and review.5 16 17 31 34 We further adopted the extended behavioural-ecological framework for healthcare access and navigation in selecting and classifying independents variables.35 The independent variables and the respective classification are as follows. Simple descriptive statistics reporting frequency and percentages of women utilising SBA viz-a-viz independent characteristics were presented in tables 1 and 2. Bivariate association was examined for each category of nominal/ordinal independent variables and SBA use was examined across the periods between 1990 and 2018 ‘(1990–2003, 2003–2008, 2008–2013 and 2013–2018) and longer periods of 2003–2018 and 1990–2018’ using the χ2 test for trends of proportions, with the survey years being an ordinal exposure variable.36 The χ2 analysis of trend and Rao-Scot χ2,37 38 were used to determine if there are any significant changes or not at alpha (α)=0.05. We found no difference between the conclusions from the χ2 for trend and the Rao-Scot χ2. MDA was employed to decompose changes in SBA use between 2003 and 2018. The MDA presents an opportunity to decompose what contributes to changes over two time points or among two mutually exclusive groups. We excluded 1990 from the MDA and considered 2003–2018, to allow full comparison across the current 36 states in Nigeria. In the MDA, respondents’ SBA use is the response variable with outcomes in 2003 as one ‘group’ and 2018 as another ‘group’ while predictor effects were partitioned into differences in characteristics or endowment and differences in the effects or coefficients in the regression decomposition.39 This is to identify the root of the changes in the utilisation of SBA in the last one and half decades (2003–2018) and evaluate how SBA use responds to changes in women characteristics. Data management and analysis were conducted using Stata V.16.0. Survey design was considered in the analysis due to sample disproportionality and was managed by probability weights. Hence, we applied the sample weight (SW) using the weighting factors included in the NDHS data and adjusted for the complex survey design (that incorporate the sample weighting, clustering, and stratification) through the ‘svy’ analysis on Stata to account for unequal population sizes. Test of statistical significance was carried out at 5% level of significance (95% confidence level) in all the statistical analysis. We computed and applied the year-women weight (YWW) to the analysis to reflect the differences in population sizes of the women in each survey year. The YWW is the product of SW (provided in the NDHS data) and year-specific weight (YSW). We computed the YSW as the number of sampled women aged 15–49 years divided by the population of women aged 15–49 years for each year as earlier reported.40 We controlled for multicollinearity using the ‘colin’ command in Stata and the variance inflation factor (VIF) was evaluated. The mean VIF was 1.97. Distribution of mothers’ background characteristics na, not available. Trends and prevalence of SBA use by background characteristics of mothers All percentages calculated as [(b-a)/a * 100]. *Trend test of proportions across the survey year. na, not available; SBA, skilled birth attendants. MDA technique is useful in decomposing changes or group differences in statistics such as; mean, proportion, and count in linear, logit and count multivariate models into characteristic and coefficient functions, respectively.41 The approach is an improvement of the Oaxaca-Blinder decomposition.42 43 It has been extended to non-linear models including logit and probit models.44 45 The main purpose of MDA is to determine explanatory variables attributed to changing composition or effects, especially in trends spanning overtime to explain the root cause of those changes.46–48 In this study, the options offered in multivariate decomposition were applied to construct a normalised decomposition towards limiting the bias associated with the choice of reference categories (the identification problem). MDA automatically determine the high-outcome group (SBA used) and reference the low-outcome group (SBA not used) in the analysis of group variables (dummy) which was set for ANOVA normalisation, such that the coefficients of the multivariate (logistic) regression for all the level of the categories approximately sum to zero.39 The decomposition or standardisation of the difference in the first moment (1. e proportion using SBA) was based on logit model and is thus a function of a linear combination of the predictors and the regression (logistic) coefficients and can be in general, additively decomposed into: Where Y is the n x 1 vector of the dependent variable 0≤p≤1, X is the n x k matrices of the independent variables and β is the k x 1 vector of the regression coefficients in (1). The difference in the proportion of respondents using and not using SBA was decomposed in (2). In (3) the component {F (XPβP) – F (X1-PβP)} refers to the differential attributable to differences in characteristics or endowment (explained component) while {F (X1-PβP) – F (X1-Pβ1-P)} refers to the differential attributable to differences in coefficients or effects (unexplained component). YP denotes the proportion of mothers who used SBA (comparison group) while subscript denotes the proportion of mothers who did not utilise SBA (reference group).