Background Childhood mortality is a well-known public health issue, particularly in the low and middle income countries. The overarching aim of this study was to examine whether neighbourhood socioeconomic disadvantage is associated with childhood mortality beyond individual-level measures of socioeconomic status in Nigeria.
Methods Multilevel logistic regression models were applied to data on 31 482 under-five children whether alive or dead (level 1) nested within 896 neighbourhoods (level 2) from the 37 states in Nigeria (level 3) using the most recent 2013 Nigeria Demographic and Health Survey (DHS).
Results More than 1 of every 10 children studied had died before reaching the age of 5 years (130/1000 live births). The following factors independently increased the odds of childhood mortality: male sex, mother's age at 15–24 years, uneducated mother or low maternal education attainment, decreasing household wealth index at individual level (level 1), residing in rural area and neighbourhoods with high poverty rate at level 2. There were significant neighbourhoods and states clustering in childhood mortality in Nigeria.
Conclusions The study provides evidence that individual-level and neighbourhood-level socioeconomic conditions are important correlates of childhood mortality in Nigeria. The findings of this study also highlight the need to implement public health prevention strategies at the individual level, as well as at the area/neighbourhood level. These strategies include the establishment of an effective publicly funded healthcare system, as well as health education and poverty alleviation programmes.
- MULTILEVEL MODELLING
- DEVELOPING COUNTR
- SOCIAL EPIDEMIOLOGY
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Childhood mortality is a well-known public health issue. Globally, childhood mortality is about 9 million deaths per year and 70% are preventable, which makes this an important public health problem deserving further investigation. In one of the series on child survival published on Lancet, childhood mortality was described as a public health disaster.1 Secondary to the burden of childhood mortality, the United Nation (UN) has incorporated reduction of childhood mortality by two-third by the year 2015 as one of its Millennium Development Goals (MDGs) which were set in 1990. Furthermore, it is essential to identify the determinants of childhood mortality because this will assist in the formulation and implementation of appropriate health programmes and policies to reduce childhood mortality in Nigeria.
In Nigeria, it has been estimated in a study conducted jointly by the World Bank/WHO in 2005 that the under-five mortality rate (UFMR) is about 159 deaths/1000 live births. There was a slight reduction in this rate in a later study conducted in 2013 where it was estimated at 120 deaths/1000 live births.2 This reduction is still grossly above the global average of 60 deaths/1000 live births for developing countries.2 Another study carried out between 1996 and 2003 indicated that UFMR in Nigeria was 157 deaths/1000 live births.3 Currently, under-five mortality (UFM) in Nigeria stands at 124 deaths/1000 live births.4 If one looks at the above statistics, one will see that childhood mortality is still a major public health problem in Nigeria that needs to be solved.
Several studies conducted in Nigeria have attempted to identify the determinants of childhood mortality. However, the majority of these studies had concentrated on individual-level (maternal and child) factors, neglecting contextual (area/neighbourhood and state) level factors.5–7 The association between contextual factors and health outcomes have also been documented in the literature.8–11 Beyond individual (child and maternal) level factors, survival of a child can be influenced by area and societal factors at large. In addition, concentrating only at one level, either at the micro level (individual) or contextual level (area), can lead to methodological and practical problems. For example, studies that concentrate on individual levels are prone to atomistic fallacy, 12 ,13 because association at the individual level may differ at the area level. Similarly, studies conducted at the area level alone are prone to ecological fallacy.12–14 Thus, the overarching aim of this study was to examine whether neighbourhood socioeconomic status may correlate with childhood mortality beyond individual-level measures of socioeconomic status.
This study was based on secondary analyses of cross-sectional population-based data from the 2013 Nigeria Demographic and Health Survey (DHS).
Methods used in the DHS have been published elsewhere.15 Briefly, the survey used a three-stage cluster sampling technique. The country was stratified into 36 States and the Federal Capital Territory (FCT), Abuja, making 37 districts. The primary sample unit (PSU) was based on the 2006 Nigeria population census enumeration areas. The first stage involved selecting 896 clusters (PSU) with a probability proportional to the size, the size being the number of households in the cluster. The second stage involved the systematic sampling of households from the selected clusters. A total of 40 320 households were selected for the 2013 DHS survey. The third stage involved the distribution of the households in each state proportionately among its urban and rural areas and with this, a total of 40 680 households was finally sampled out of which 16 740 were from urban areas and 23 940 from rural areas.
Data collection procedures have been published elsewhere.15 Briefly, data were collected by visiting households and conducting face-to-face interviews to obtain information on demographic characteristics, wealth, anthropometry and female genital cutting, and awareness of HIV/AIDS, knowledge of HIV prevention, sexual behaviour and domestic violence.
Each woman interviewed in the survey was asked to provide a detailed history of all her live births in a chronological order, including whether a birth was single or multiple, sex of the child, date of birth, survival status, age of the child on the date of interview if alive and if not alive, age at death of each live birth. These data from the birth histories were used to calculate UFMR, defined as the probability of dying before reaching the age of 5 years, using a synthetic cohort life table. The rate was expressed as deaths per 1000 live births.
In this study, we considered three measures of individual socioeconomic status: wealth status, educational attainment and occupation of mother. Wealth index is measured in the DHS surveys in terms of assets, rather than income. Household assets and utility services such as a radio or car as well as characteristics of the dwelling such as floor or roof type, toilet facilities and water source were the items that measured the concept of poverty in these settings. This concept has been used by the World Bank to allocate households and their members into poverty quintiles, using principal components analysis.18 ,19 For analytical purposes, we divided the weighted scores into quintiles. The level of education attained by the mother (the respondent) was defined as never been to school, primary and secondary or higher education. The respondent’s current occupation was categorised into unemployed, white collar (professional, technical and managerial), manual (services, agricultural, skilled and unskilled manual) and other jobs. Gender of the child and age of the mothers were included as control variables.
At the neighbourhood level, we included place of residence and neighbourhood poverty, unemployment and illiteracy rates. Place of residence was categorised into urban and rural residence. We categorised neighbourhood-level poverty, unemployment and adult illiteracy rates into two categories (low and high) to allow for non-linear effects and to provide results that were more readily interpretable in the policy domain.20 Median values served as the reference group for comparison.
At the state level, we included adult illiteracy rate, unemployment rate and poverty rate. We categorised state-level poverty, unemployment and adult illiteracy rates into two categories (low and high) to allow for non-linear effects and to provide results that were more readily interpretable in the policy domain.20 Median values served as the reference group for comparison.
In the descriptive statistics, the distributions of respondents by the key variables were expressed as numbers and percentages as well as deaths per 1000 live births.
We specified a three-level multilevel model for whether a child was alive or dead (level 1), living in a neighbourhood (level 2) and within a state (level 3). We constructed four models. The first model, an empty or null model without any determinant variables, was specified to decompose the amount of variance that existed between the neighbourhood and state levels. The second model contained individual-level variables. The third and fourth models were extended to include neighbourhood-level and state-level measures of socioeconomic status variables, respectively.
Fixed effects (measures of association)
The results of fixed effects (measures of association) were shown as ORs with their 95% credible intervals (CrI).
Random effects (measures of variation)
Measures of random effects included an intracluster correlation and a variance partition coefficient and median OR (MOR). MOR is a measure of unexplained cluster heterogeneity; it measures the area variance as OR. The method used for calculating MOR has been published and described elsewhere.21 ,22
Model fit and specifications
Bayesian deviance information criterion (DIC) was used to judge the goodness-of-fit of the model while variance inflation factor (VIF) was used to check for multicollinearity. All multilevel modelling were performed using MLwiN 2.3223 on the platform of Stata statistical software for windows V.1324 using the runmlwin routine. For the multilevel logistic regression models, the Markov Chain Monte Carlo (MCMC) estimation was used.25
Characteristics of the sample
Table 1 presents descriptive statistics which show that the prevalence of childhood mortality was 130/1000 live births as estimated from the data set across the 37 states. The prevalence of UFM was 154.1/1000 live births in mothers in the age group 15–24 years, while it was 131.5/1000 live births in the age group 35 years and above. There was a significantly higher prevalence of childhood mortality among under-five male children. Likewise, childhood mortality was highest in children born to mothers who were not working. The prevalence was highest in the under-fives whose parents were the poorer with respect to wealth index status and those that had no formal education. Furthermore, childhood mortality was more prevalent with under-five children who resided in rural areas and neighbourhoods with high poverty, illiteracy and unemployment rates.
Measures of association
Table 2 shows results of fitting the model including individual-level, neighbourhood-level and state-level factors. With individual-level, neighbourhood-level and state-level measures of socioeconomic status controlled for, male children had higher odds of dying compared with the female children adjusted OR (aOR) 1.20; 95% CrI 1.11–1.30. Children of mothers aged 25–34 years had odds of dying reduced by 10% compared with mothers aged 15–24 years. The probability of a child dying decreases with increasing level of education of the child's mother. Children whose mothers had secondary or higher education had higher odds of survival compared with those whose mothers were without any formal education by about 32%. The probability of a child dying decreased with increasing household wealth status, such that children from poorest households were more likely to die compared with those from richest households with odds of about 41%.
Children living in rural areas had a higher probability of dying than their counterparts from urban areas. In addition, the odds of a child dying was associated with neighbourhood socioeconomic disadvantages, such that those from neighbourhoods with a high poverty rate were more likely to die compared with their counterparts from neighbourhoods where there was a low poverty rate (aOR 1.22; 95% CrI 1.05–1.43). However, there is no statistically significant association between the odds of dying and state-level socioeconomic factors.
Measures of variation
Table 2 shows the random effect (measures of variation) results from multilevel analysis. In model 1 (the empty model), there was a significant variation in the log odds of childhood mortality across the neighbourhoods (τ=0.165, 95% CrI 0.114–0.218) and across the states (τ=0.116, 95% CrI 0.062–0.204). According to the intrastate and intraneighbourhood correlation coefficient implied by the estimated intercept component variance, 3% and 8% of the variance of childhood mortality could be linked to the state-level and neighbourhood-level factors, respectively. Variations across neighbourhoods and states remained statistically significant, even after controlling for individual-level, neighbourhood-level and state-level socioeconomic factors in the final model 4, thereby giving credence to the use of multilevel modelling to account for neighbourhood and state variations. According to the proportional change in variance, approximately about 60% and 23% of the variance in the log odds of childhood mortality across the states and neighbourhoods, respectively, were explained by the individual-level socioeconomic factors (model 2). The final model (model 4) accounted for 66% and 27% of the variance in the log odds childhood mortality across the states and neighbourhoods.
The MOR results also confirmed the evidence of neighbourhood and state contextual phenomena modifying the likelihood of a child dying before the age of 5 years. The MOR (1.5) in model 1 between children with a higher and lower propensity of dying in a neighbourhood suggests that the neighbourhood heterogeneity is moderate. Controlling for individual-level socioeconomic factors reduces the unexplained heterogeneity between neighbourhoods to an MOR of 1.4. In model 2, for two children with the same individual-level socioeconomic status, the MOR between the two children living in a state with a higher compared with a lower disposition to having childhood mortality was 1.3, which also suggests that the clustering effect was moderate. The unexplained state heterogeneity decreased further when all the factors were controlled for in the final model 4 to 1.1. Thus, there were little variations between states in the likelihood of having childhood mortality.
This study confirms the importance of neighbourhood socioeconomic variations with respect to childhood mortality in Nigeria. In particular, the study showed that the neighbourhood context in which people reside is linked to childhood mortality even after taking into consideration the individual-level and state-level socioeconomic status.
We found that male children were more likely to die compared with their female counterparts. This has been reported in previous studies that examined the association of children's sex and the likelihood of dying before the age of 5 years.26–29 This outcome could be attributed to the fact that male children tend to have a larger body size particularly during pregnancy and the intrapartum period, which leads to more difficult births, asphyxia and birth trauma, leading to higher neonatal mortality.30
The reduced odds of childhood mortality seen among women aged 25–34 years compared with adolescent/young mothers (aged 15–24 years) is consistent with previous studies.31–33 Teenage mothers often experience financial and social difficulties, which leads to inadequate provision of child's care.32 Some researchers have also said that adolescent mothers have less matured body organs like the uterus and bony pelvis that are narrow causing neonatal deaths.
The findings of this study support those of other similar studies, which indicates that maternal education has a positive effect on childhood survival.34–37 Mothers with formal education would have knowledge of what to do to prevent childhood mortality from occurring. Moreover, their health-seeking behaviour index is expected to be higher than the those who are uneducated as they are better able to access information important for child survival.
Furthermore, the findings of this study revealed that under-five children from less wealthy households have greater odds of dying than under-fives from wealthy households. Similar results were documented in previous studies carried out in different developing countries, which gives further proof that poverty is a significant predictor of childhood survival.38–40 Household's wealth index status is very important as this most times predicts if there will be availability of good and nutritious foods essential for the growth and development of children or not. In addition, good and nutritious foods would also enhance the ability of the body to fight off infectious diseases. Moreover, children from poorer households in most developing countries like Nigeria, where a publicly funded healthcare system is not practised, would lack access to good and basic healthcare services when they fall ill.
Consistent with results of studies from other developing countries, 41–43 we found that the risk of childhood mortality is associated with place of residency. Children residing in rural areas experience high mortality risks than those from urban areas where there are better healthcare facilities. Children residing in neighbourhoods with high poverty rates in this study had a higher likelihood of dying before getting to the age of 5 years compared with their counterparts from neighbourhoods with low poverty rates after controlling for individual-level socioeconomic status. Our finding is consistent with those of other previous studies, though the participants in these studies were not limited to only under-five children but from age 0 to 14 years,44–46 and the plausible explanations for this association could be that residing in a neighbourhood where there is a high rate of poverty will influence childhood survival.
This study shows that there is some evidence of geographical variation with respect to childhood mortality in Nigeria. Approximately 8% and 3% of the total individual variations regarding childhood mortality were at the neighbourhood and state levels, respectively. It is theoretically expected that individuals from the same area are more similar to one another in relation to health outcomes than others from different areas.47 Individuals with similar attributes may have different health outcomes according to whether they live in one neighbourhood or in another because of differing sociocultural, economic, political, climatic and geographical contexts.47 Hence, individuals residing in the same neighbourhood tend to have similar health outcomes. This may be because individuals in the same neighbourhood are susceptible to the same contextual influences. This contextual effect shows itself as a clustering of individual health outcomes within a neighbourhood; that is, a portion of the health outcomes among people may be linked to the areas in which they live.47 On the basis of this, we might infer that there is some evidence for possible neighbourhood and state contextual influences modifying childhood survival, and that neighbourhoods are very invaluable in understanding individual childhood survival differences.
Study limitations and strengths
Caution should be taken when interpreting the results of this study. The cross-sectional nature of the data limits our ability to draw casual inferences. The study can be criticised for using an indirect measure of household wealth. However, owing to the fact that in developing countries it is hard to obtain reliable income and expenditure data, an asset-based index is generally considered a good proxy for household wealth status. Our study focused on understanding the role of individual, neighbourhood and state socioeconomic status as correlates of childhood mortality. We did not incorporate an assessment of the effect of interactions between such variables in our study design. Another point to consider when interpreting the findings of this study is the fact that what we called neighbourhoods are administrative demarcations used for survey purposes, which may not properly capture the social context vital for understanding the causes of childhood mortality.
Despite these limitations, the study's strengths are significant. It is a large population-based study with national coverage. DHS data sets are often nationally representative, allowing for conclusions that cover the entire country. In addition, data of the DHS are widely perceived to be of high quality, as they were based on sound sampling methodology with a high response rate from participants. Moreover, there are advantages to studying factors associated with childhood mortality using multilevel modelling approach as state-level and neighbourhood-level factors like we used in our study would help identify economic, cultural and social context in which individual lives and experiences health outcomes. Aside from the neighbourhood factors, individuals will be affected by state policies and, by extension, country policies which affect the proximate correlates of childhood mortality.
The study provides evidence that individual-level and neighbourhood-level socioeconomic status are important correlates of childhood mortality in Nigeria. The findings highlight the need to implement public health prevention strategies at the individual level, as well as at the neighbourhood (context) level. These strategies include the establishment of a publicly funded healthcare system, as well as health education and poverty alleviation programmes.
Studies to be conducted in the future should examine other factors that may be responsible for the unexplained neighbourhood and state clustering in childhood mortality in Nigeria and other similar countries.
What is already known about this subject?
Several studies conducted in Nigeria on the determinants of childhood mortality had concentrated on individual-level (maternal and child) factors using regular single level regression analysis, neglecting contextual (area/neighbourhood and state) level factors.
Studies are still lacking on the use of multilevel modelling to understand the independent contribution of individual-level, neighbourhood-level and state-level factors on childhood mortality in Nigeria.
What this study adds?
This is the first study to examine the independent contribution of individual-level, neighbourhood-level and state-level socioeconomic factors associated with childhood mortality in Nigeria in a single analytical framework using the latest nationally representative data set in the country.
This study revealed that individual-level and neighbourhood-level socioeconomic status are important correlates of childhood mortality in Nigeria.
The data used in this study were made available through the MEASURE DHS Archive. The data were originally collected by the ICF Macro, Calverton USA.
Contributors VTA contributed in the conception of the study; carried out data extraction; and conducted statistical analysis and drafting of the manuscript with contributions from N-BK, SS and OAU. All authors read and approved the final manuscript.
Competing interests None declared.
Ethics approval Approval was granted for secondary analysis of existing data after the removal of all identifying information of the respondents by the Institutional Review Board (IRB) of the ICF Macro at Calverton in the USA in conjunction with the National Health Research Ethics Committee (NHREC) of the Federal Ministry of Health in Nigeria.
Provenance and peer review Not commissioned; externally peer reviewed.