Menu
Menu
Menu
Menu
Introduction Despite the decline in infant and under-five mortality rates since the last decade, Ghana did not meet the millennium development goal (MDG) 4 target. To implement effective interventions that could fast-track progress towards achieving the sustainable development goal 3 in 2030, factors contributing to the decline in child mortality throughout the MDG period and which factor(s) has/have been consistent in affecting child survival in the last decade need to be understood. Methods This study used Demographic and Health Surveys (DHS) from 2003, 2008 and 2014 and data from World Bank Development Indicators (2000-2018). We employed modified Poisson with robust SE and multivariate decomposition approach to assess risk factors of child mortality using DHS data from 2003, 2008 and 2014. Penalised regression was used assess the effect of 25 country-level contextual factors on child survival. Results The risk of infant mortality is approximately five times higher among mothers who had multiple births compared with mothers who had single birth over the last decade (adjusted relative risk 4.6, 95% CI 3.2 to 6.6, p<0.001). An increase in the annual percentage of female labour force participation (FLFP) is associated with the reduction of approximately 10 and 18 infant and under-five annual deaths per 1000 live births, respectively. Conclusions This study found that multiple births and shorter birth spacing are associated with increased risk of infant and under-five deaths over the last decade. Increased in FLFP, and the proportion of children sleeping under bed-net are associated with reduced risk of both infants and under-five deaths.
Patients were not involved in this study. The study used data from the three Ghana Demographic and Health Surveys (GDHS) conducted in Ghana, in 2003, 2008 and 2014. Data were downloaded from the DHS website (http://dhsprogram.com) after been granted permission. The DHS include the full birth history of all women within the reproductive age (15–49 years). The birth history includes all children born alive and their survival status to women of reproductive age (15–49 years). The children data file were merged with household data to obtain a complete dataset required for the analysis. The unit of analysis in this study is the children of women born in the last 5 years (0–59 months) preceding the survey. Online supplementary table S1 presents information on the three surveys included in this analysis. bmjgh-2019-001658supp001.pdf Each specific survey data sample was obtained using multistage stratified cluster sampling. Ghana has 10 administrative regions. Each region was stratified into urban and rural areas, yielding 20 sampling strata. Samples of enumeration areas (EAs) were selected independently in each stratum in two stages. The EA size is basically the number of residential households residing in a particular EA. In the first stage, stratification and proportional allocation are conducted at each of the lower administrative levels by sorting the sampling frame within each sampling stratum before sample selection using a probability proportional to size selection at the first stage of sampling. The second stage involves the selection of a fixed number of approximately 20–30 households per cluster selected with an equal probability systematic selection from the newly created household listing. Details of the sampling design for the GDHS can be obtained from the DHS programme website (www.dhsprogramme.com) or from the Ghana Statistical Service.15 This study investigated two primary outcome measures: infant and U5MRs in Ghana. The IMR is the number of deaths in the first year of life (per 1000 live births) and measures the probability of dying before a child’s first birthday. The U5MR measures the probability of death before a child’s fifth birthday. Both IMR and U5MR were estimated within 5 years preceding the survey, including those born exactly 5 years before the survey. This study used information about the year of child's birth, whether each child was alive at the time of the survey, and how old a child was if s/he died to define the primary outcome measures. Specifically, a child who was born in the 5 years preceding the survey but unfortunately died within the first year of life was classified as infant death and coded as 1 or 0 otherwise. All deaths that occurred in the 5 years preceding the survey were classified as under-five death and they were respectfully coded as 1 or 0 otherwise. The unit of analysis in this study is all children born in the 5 years preceding each survey. The choice of the selected covariates in determining infant and under-five mortality were adapted from the analytical framework for the study of child survival in developing countries by Mosley and Chen.16 This study examined the determinants of IMR and U5MR at four different levels of indicators: characteristics of the household (sex of the household head, age of the household head, household size, place of residence, region, household wealth, household access to improved water and sanitation), maternal characteristics (mothers age childbirth, marital status, highest educational level and body mass index), child indicators (age of the child, sex of the child, multiple birth, birth order and preceding birth interval) and finally, the maternal and delivery care received along with coverage of other interventions that could affect both IMR and U5MR (place of delivery, tetanus injection, antenatal care (ANC) attendance and valid national health insurance card). Twenty-five indicators at the country level were also assessed to examine their effect on infant and under-five mortality. These indicators have been assessed previously.17 Our analysis was restricted to 25 indicators out of the 70 indicators previously studied because of lack of data on some covariates. Data were obtained from the World Bank Data Catalogue2 (online supplementary table S8 shows the country-level indicators that were studied). The analysis adjusted for the complex survey design structure (clustering, stratification and weighting) to reduce bias and improve the precision of our estimates. Since the study pooled complex survey data from different surveys at different time points, the women/children standard weight were de-normalised. This was achieved by dividing the women standard weight by the women survey sampling fraction, that is, the ratio of a total number of women aged 15–49 years interviewed in the survey year over the total number of women aged 15–49 years in the country at the time of the survey. The total number of women aged 15–49 interviewed in the survey year was obtained from the DHS datasets, while the total number of women aged 15–49 years in the country at the time of the survey were obtained from OurWorldinData.18 The de-normalised women sampling weight is given by: where Ψ is the women sampling weight as estimated in the DHS, ϕ15-49 is the total females aged 15–49 in the country at the time of the survey and ϕ15-49S is the number of women age 15–49 interviewed in the survey. Four different levels of statistical analysis were conducted to address the aforementioned research questions. First, the Rao-Scott χ2 test and the log-rank test that follows the Kaplan Meier procedure were used to test association and differences in mortality, test homogeneity of these groups and test for equality of survivor functions. Second, modified Poisson with the robust SE was used to assess the relationship between IMR, U5MR and all explanatory variables specifying the time at risk in every Poisson model that was fitted. Poisson was used in the analysis since the primary estimate of interest was a relative risk instead of HR from the Cox-proportional Model. That notwithstanding, a sensitivity analysis using was Cox-proportional Model was conducted to determine whether the results obtained were robust to the model specification. Finally, we applied weighted modified Poisson based multivariate decomposition technique19 which is comparable to the Oaxaca-Binder Method20 but provides flexibility to use non-linear models to assess factors contributing to the decline in child mortality. Multivariate decomposition provides a way to analyse factors that contribute to the differences in mortality rates between two points of time: 2003–2008, 2008–2014 and 2003–2014.3 To determine country-level contextual factors associated with infant and under-five mortality, this study employed theory-driven rigorous penalisation of the ordinary least square estimate with least absolute shrinkage and selection operator (LASSO) and square-root LASSO as proposed by Belloni et al.21 Rigorous penalisation is a modified version of the well-known LASSO.22 The use of rigorous penalisation became necessary because of the high dimensionality of the data set (25 covariates investigated compared with sample size of 18) and the ability of the theory-driven penalisation for lasso and square-root to allow for heteroskedasticity, cluster-dependent and non-Gaussian errors. Sensitivity analysis was conducted using two other different approaches for selecting the penalisation parameters: information criteria (implemented in lasso2), K-fold cross-validation for cross-section, panel and time-series data. The sensitivity analysis was to determine whether our results were robust to the different ways of selecting the penalisation parameters and to identify which indicator(s) will be selected by all the different penalisation procedure. Post-estimation ordinary least square was performed to address the shrinkage bias associated with estimates from rigorous penalisation. The LASSO procedure is implemented in Stata via lassopack; a collection of programme for regularised regression in Stata developed by Ahrens et al.23 All analyses were performed using Stata V.15 and a p value <0.05 was considered statistically significant.
N/A
