Menu
Menu
Menu
Menu
Objective: Inadequate overall progress has been made towards the 4th Millennium Development Goal of reducing under-five mortality rates by two-thirds between 1990 and 2015. Progress has been variable across African countries. We examined health, economic and social factors potentially associated with reductions in under-five mortality (U5M) from 2000 to 2013. Setting: Ecological analysis using publicly available data from the 46 nations within the WHO African Region. Outcome measures: We assessed the annual rate of change (ARC) of 70 different factors and their association with the annual rate of reduction (ARR) of U5M rates using robust linear regression models. Results: Most factors improved over the study period for most countries, with the largest increases seen for economic or technological development and external financing factors. The median (IQR) U5M ARR was 3.6% (2.8 to 5.1%). Only 4 of 70 factors demonstrated a strong and significant association with U5M ARRs, adjusting for potential confounders. Higher ARRs were associated with more rapidly increasing coverage of seeking treatment for acute respiratory infection (ß=0.22 (ie, a 1% increase in the ARC was associated with a 0.22% increase in ARR); 90% CI 0.09 to 0.35; p=0.01), increasing health expenditure relative to gross domestic product (ß=0.26; 95% CI 0.11 to 0.41; p=0.02), increasing fertility rate (ß=0.54; 95% CI 0.07 to 1.02; p=0.07) and decreasing maternal mortality ratio (p=-0.47; 95% CI -0.69 to -0.24; p≤0.01). The majority of factors showed no association or raised validity concerns due to missing data from a large number of countries. Conclusions: Improvements in sociodemographic, maternal health and governance and financing factors were more likely associated with U5M ARR. These underscore the essential role of contextual factors facilitating child health interventions and services. Surveillance of these factors could help monitor which countries need additional support in reducing U5M.
The study was approved by the Institutional Review Boards at Vanderbilt University. Mortality data were obtained for the period 2000–2013, while data on all factors of interest were obtained for the period 1998–2011. These are described in detail below. We obtained country-specific U5M rates from annual estimates provided by http://www.childmortality.org, the data used in the United Nations Children’s Fund (UNICEF) Report on Levels and Trends in Child Mortality. For each of the 46 WHO African Region countries, the estimated U5M rates for 2000 and 2013 (the latest estimates available at the time of the analysis (accessed June 2015)) were obtained and the ARR from 2000 to 2013 was calculated. Use of the ARR as an outcome facilitates interpretation of results in the context of MDG4 progress metrics. The U5M ARR reflects a constant rate of change in the U5M rate between two time periods and is calculated using the following equation (equation 1): where yt is the mortality rate for a given year (eg, 2000) and n is the number of years between the two rates (eg, 13 years when calculating ARR from 2000 and 2013 rates). Consistent with how it is calculated and reported by UNICEF1 and others,12 the ARR is expressed as a per cent and will have a positive value when a country is reducing its mortality rate. As an example, an ARR of 4.4% or greater is needed for a country to meet MDG4 of reducing U5M by two-thirds between 1990 and 2015. In the years leading up to 2015, expressing the reduction in U5M as an ARR made it possible to monitor progress across countries and over different time periods. Since we wanted to assess a broad range of factors potentially associated with U5M ARR, factors to be used in the analysis included those monitored by Countdown to 2015 as well as others identified through a comprehensive literature search of the PubMed database. We searched for studies on under-five, infant or neonatal mortality within any of the 46 countries in the WHO African Region that were published between 2002 and 2012. Abstracts were reviewed to identify factors that were (1) associated with under-five, infant or neonatal mortality, (2) not already reported by Countdown to 2015 and (3) had aggregate country-level data available for the analysis. Thirty-four factors met these criteria. These were combined with 20 Countdown 2015 intervention indicators (or closely related) and 16 non-intervention indicators reported in the Countdown 2015 country profiles (or were closely related). The final list used in the analysis included 70 factors from the following categories: sociodemographics (18 factors), access to healthcare (16), governance and financing (11), maternal health (6), child survival interventions (7), clinical and health conditions (7), and other country infrastructure (5) (see online supplemental table S1). Data for 58 (83%) of the 70 indicators were obtained from the World Bank Data Catalogue.13 The World Bank Data Catalogue is a repository of national, regional and global indicator data that have been compiled from officially recognised international sources. In many instances, a single indicator may be derived from multiple data sources using modelling or aggregation techniques. Data for the remaining 12 factors not available through the World Bank Data Catalogue were obtained directly from each country’s Demographic and Health Surveys14 (10 factors) or Countdown 2015 country profiles (2 factors).8 10 In total, 26 (37%) of 70 factors were obtained entirely or in part from country Demographic and Health Surveys or other household survey data, including the majority of the maternal health, access to healthcare and child survival intervention factors. Other sources included data collected and maintained by WHO, various UN divisions, UNAIDS, the Organisation for Economic Co-operation and Development, and the World Bank (see online supplemental tables S2 and S3 for further details). For each of the 46 WHO African Region countries, data on the 70 factors were obtained that corresponded as close as possible to the years 2000 and 2011. To be considered sufficient for inclusion in the analyses, data for each factor had to meet the following three criteria: (1) reported for one of the years between 1998 and 2003, termed 2000 data, (2) reported for one of the years between 2006 and 2011, termed 2011 data and (3) the pair of data points for each factor had to be at least 5 years apart in order to minimise incorrect extrapolation when calculating changes in the indicator. If any of these criteria were not met, the change over time for that particular factor was not calculated and was deemed missing. The annual rate of change (ARC) for each indicator is conceptually similar to the ARR for U5M and was calculated using the same ARR equation shown above in equation 1 but with one difference: the rate of change is multiplied by positive 100 instead of negative 100 so that the ARC has a negative value when the indicator decreases over time (ie, Coverage2011 <Coverage2000) and a positive value when the indicator increases over time (equation 2): This is in contrast to the U5M ARR which has a positive value when mortality is decreasing over time (ie, U5M2013 <U5M2000). The dependent variable of interest was the U5M ARR for 2000–2013. Each indicator ARC was an independent variable of interest. Both indicator ARCs and U5M ARR were analysed as continuous variables and no transformations were performed. The distribution of ARRs was inspected visually and was confirmed to be approximately normal by the Shapiro-Wilk test for normality (p=0.33). Associations were explored using linear regression. Given the sample size (n=46 countries), results from traditional linear regression methods may be overly influenced by outliers. These outliers, however, likely represent true data rather than data errors and exclusion would unnecessarily decrease the sample size. Hence, robust linear regression was used to minimise the influence of outlying observations, without excluding them15–17 using iteratively reweighted least squares (M-estimation with Huber weighting). Multivariable robust linear regression models were fit for each factor of interest, resulting in 70 different regression models (one for each factor). A standard set of factors was identified a priori to be included in each model as covariates to adjust for potential confounding of the specific factor association being analysed. Given the sample size of 46 countries, at most, a decision was also made a priori to include no more than five covariates in the multivariable analyses to avoid overfitting. We selected the following factor ARCs for inclusion as covariates in each model based on previous ecological studies, consideration of what macro-level or system-level factors would influence the majority of the factors, and having nearly complete data: (1) access to improved water source, (2) health expenditure (relative to gross domestic product (GDP)), (3) adult HIV prevalence, (4) urban population prevalence and (5) receipt of antenatal care (when applicable). All regression models included these core factor ARCs as covariates unless expected to be highly correlated with the primary indicator of interest (eg, improved water sources was excluded from the model for the association between improved sanitation facilities; health expenditure relative to GDP was excluded from the model for the association between health expenditure relative to government expenditure). Changes in the receipt of antenatal care was only included when the indicator being analysed would occur following the pregnancy period (eg, maternal mortality ratio, births delivered at a health facility, measles immunisation, wasting prevalence) and was not included for sociodemographic factors, macro-level factors such as health expenditure and system-level factors such as physician density. The estimated robust linear regression β coefficient for each indicator ARC and U5M ARR association can be interpreted as the change in ARR associated with every 1% increase in the indicator ARC. For example, a β coefficient of 0.20 indicates that for every 1% increase in the indicator ARC there is a corresponding 0.2% increase in the ARR. Stated differently, a 5% increase in the indicator ARC (eg, 6% ARC compared to 1% ARC) corresponds to a 1% increase in the ARR (eg, 4% ARR vs 3% ARR). To help avoid type II errors, which can occur with small samples, we reported all associations when the p values were <0.10, acknowledging that some of these associations may be due to chance alone, especially with higher p values. We report an indicator to be strongly associated with ARR when the adjusted β coefficient is ≥0.20 or ≤−0.20. A preliminary analysis showed significantly different ARRs between countries reporting a specific indicator and countries not reporting a specific indicator when the indicator was reported by <50% of the countries. Since these differences suggest selection bias, we only present results for those factors for which at least 23 countries (≥50%) have sufficient data (ie, non-missing indicator data for both time periods). Results for all factors are available in online supplemental tables S4 and S5. All analyses were conducted using R-software V.2.15.2 (http://www.r-project.org).
N/A
