Trend and decomposition analysis of risk factors of childbirths with no one present in Nigeria, 1990-2018

Objectives To assess the trend and decompose the determinants of delivery with no one present (NOP) at birth with an in-depth subnational analysis in Nigeria. Design Cross-sectional. Setting Nigeria, with five waves of nationally representative data in 1990, 2003, 2008, 2013 and 2018. Participants Women with at least one childbirth within 5 years preceding each wave of data collection. Primary and secondary outcome measures The outcome of interest is giving birth with NOP at delivery defined as childbirth assisted by no one. Data were analysed using Χ 2 and multivariate decomposition analyses at a 5% significance level. Results The prevalence of having NOP at delivery was 15% over the studied period, ranges from 27% in 1990 to 11% in 2018. Overall, the prevalence of having NOP at delivery reduced significantly by 35% and 61% within 2003-2018 and 1990-2018, respectively (p<0.001). We found wide variations in NOP across the states in Nigeria. The highest NOP practice was in Zamfara (44%), Kano (40%) and Katsina (35%); while the practice was 0.1% in Bayelsa, 0.8% in Enugu, 0.9% in Osun and 1.1% in Imo state. The decomposition analysis of the changes in having NOP at delivery showed that 85.4% and 14.6% were due to differences in women's characteristics (endowment) and effects (coefficient), respectively. The most significant contributions to the changes were the decision-maker of healthcare utilisation (49%) and women educational status (24%). Only Gombe experienced a significant increase (p<0.05) in the level of having NOP between 2003 and 2018. Conclusion A long-term decreasing secular trend of NOP at delivery was found in Nigeria. NOP is more prevalent in the northern states than in the south. Achieving zero prevalence of NOP at delivery in Nigeria would require a special focus on healthcare utilisation, enhancing maternal education and healthcare utilisation decision-making power. We used secondary data extracted from five successive NDHS conducted in 1990, 2003, 2008, 2013 and 2018.18–22 The NDHS is cross-sectional, population-based and nationally representative in design. The respondents were women aged 15–49 years. However, our analysis was restricted to respondents who reported at least one birth delivery within 5 years preceding each of the surveys. Geographically, Nigeria is divided into six geopolitical zones (regions), and these regions are further subdivided into states and Federal Capital Territory (FCT) for administrative purposes. As of 1990, Nigeria has 21 states. These were then divided and grouped into 30 states and the FCT in 1991. Additional six states were created in 1996 which resulted in the present number of 36 states (figure 1). Map of Nigeria showing the 36 states and the Federal Capital Territory, by the geopolitical zones. A multistage cluster sampling technique was used where the clusters are the primary sampling unit. Local government areas (LGAs) were selected from each state and FCT in the first stage. Enumeration areas were then extracted from each LGA at the second stage, and households and household representatives were randomly selected for questioning in the last stage. For further details on the sampling methodology, please visit wwwdhsprogramcom. In all, 8781, 7620, 33 385, 38 984 and 41 821 women participated in 1990, 2003, 2008, 2013 and 2018, respectively.18–22 We used the data on the delivery of the last pregnant by any of these respondents within 5 years preceding the surveys. A total of 4874, 3761, 17 920, 20 100 and 21 792 eligible deliveries for 1990, 2003, 2008, 2013 and 2018 NDHS, respectively, were included in this study. The outcome variable was whether a birth delivery was assisted or not irrespective of who offered the assistance. The reported birth delivery assistants by the respondents are skilled (doctors, nurses and midwives), unskilled (traditional, community health worker, auxiliary nurses, family, friends) and having NOP at delivery.16–22 The outcome was categorised as NOP at delivery versus anyone present. The explanatory variables used in this study consist of individual, household, community and societal factors. They were identified based on extensive literature search and review.16–19 21 22 Andersen behavioural model and healthcare utilisation30 was also used. In addition, we adopted and enlarged the behavioural–ecological framework of healthcare access and navigation to select the explanatory variables in this study.31 The variables are the following: We used descriptive statistics to report the frequency distribution and prevalence of NOP at delivery as well as its percentage changes by the explanatory characteristics and state of residence. We examined trends in NOP at delivery for 1990–2003, 2003–2008, 2008–2013, 2013–2018, 2003–2018 and 1990–2018. The Χ2 analysis for trend was used to identify the significant changes across multiple time points. Multivariate decomposition analysis (MDA) was employed to decompose changes in NOP at delivery between 2003 and 2018. Data management and analysis were conducted using Stata V.16.0, R statistical software and Power BI were used for the visualisations. Samples were weighted using weighting factors included in the NDHS data to account for unequal group sizes, and all significance tests were at 5%. The MDA allows the quantification of the contributions of different factors to changes in outcome measurements over two time points or among two groups of people with different outcomes. Unlike the logistic regressions that identify the odds of an event occurring, the MDA uses different models including the logistic regression to identify the contributions of explanatory variables to the differentials in the probability of events occurring in different groups. In which case, the groups are mutually exclusive. In the decomposition analysis, we excluded 1990 data and considered only 2003–2018, as there were only 19 states in Nigeria as of 1990 and thereby would disallow full comparison across the current 36 states in Nigeria. The difference in respondents’ NOP at delivery is the response variable, 2003 constituted a ‘group’ while 2018 is another ‘group’, while predictor effects were partitioned into differences in characteristics (endowment) and differences in the effects (coefficient) in the regression decomposition.35 This enables the identification of the root of changes in NOP between 2003 and 2018 and evaluates how changes in NOP at delivery were affected by the explanatory characteristics. The MDA technique is an improvement of the Oaxaca-Blinder decomposition,36 37 which has been extended to handle non-linear models including logit and probit models.35 38 In this study, the decomposition of the difference in the factors influencing NOP at delivery is a function of a linear combination of the predictors and regression coefficients and can be in general, additively decomposed into: where Y is the n by 1 vector of the dependent variable, 0≤p≤1, X is the n by k matrices of the independent variables and β is the k by 1 vector of the regression coefficients in equation (1). The difference in the proportion of respondents with NOP was decomposed in equation (2) into two parts. In equation (3), the component {F(XPβP)–F(X1-PβP)} is the differential attributable to differences in endowment (otherwise called the explained component), while {F(X1-PβP)–F(X1-Pβ1-P)} is the differential attributable to differences in coefficients (unexplained component). Also, YP denotes the proportion of respondents with NOP at delivery (comparison group), while Y1-P denotes the proportion of respondents with someone present at delivery (reference group). The method has been used elsewhere.39