Neighbourhood characteristics and under-five mortality in Nigeria

Despite global decline in childhood mortality, under-5 mortality remains high in Nigeria. While many studies have reported individual level factors as important determinants of under-five mortality in Nigeria, similar studies on the effects of neighbourhood contexts have been minimal. Hence, this study examines the effects of neighbourhood contexts on under-5 mortality in Nigeria. Using 2003 and 2008 Nigeria Demographic and Health Survey (NDHS) data, multilevel Cox regression analysis was performed on a nationally representative sample of 6,028 children (2003 NDHS) and 28,647 children (2008 NDHS). Results indicated neighbourhood context as important factor for child survival. For instance, findings showed that being born or raised in poor neighbourhoods (HR:1.54,p<0.05), rural communities (HR:1.25,p<0.05), and North-eastern region of Nigeria (HR:1.56,p<0.05) was associated with elevated hazards of death before age 5. Findings of this study suggest that achieving improved neighbourhood contexts holds great potentials for acceleration of under-five mortality reduction in Nigeria.


Introduction
Despite the global decline in childhood mortality, demographic and public health literature indicates that child health outcomes are generally poor in the sub-Saharan Africa. In 2011, the variation in underfive mortality rates between the developing and developed nations is more than 78-fold ranging from 180 per 1000 live births in Angola to only 2.31 per 1000 live births in Singapore (World Factbook, 2011). Worse still, evidence showed that only about one-third of all countries in Africa showed a decline of 30% or more in under-five mortality, while many countries sadly showed a considerable increase in the recent time (Becher, 2010). In Nigeria, with about 1 in 6 children dying before the age of five (NPC & ICF Macro, 2009), the country like many other nations in sub-Saharan Africa, is not on track to achieve the Millennium Development Goal 4 (reducing childhood mortality) by 2015 (Lykens et al., 2009).
Nigeria remains a major contributor to global statistics on under-five mortality. The country's rate of under-five mortality -156.9 per 1000 live births (ICF Macro and NPC, 2009) -is among the highest in the world. Besides, there is a substantial geographic variation in the levels of under-five mortality in the country -from 89 per 1000 live births in the South-west to 222 deaths per 1000 live birth in the North-east.
To understand the factors driving high under-five mortality in Nigeria and other parts of sub-Saharan Africa, many studies with diverse findings have been reported. While many studies on childhood mortality in Nigeria have indicated that individual level factors such as maternal education and other socioeconomic factors are important predictors of underfive mortality (Adebowale et al., 2012;Fayeun & Omololu, 2011;Grais et al., 2007;Nwokocha & Awomoyi, 2009), studies on the effects of neighbourhood contexts on child survival have been few.
Meanwhile, literature has established that living in a deprived neighbourhood or community is associated with poor health outcomes of individuals (Harttgen & Misselhorn, 2006;Omariba et al., 2007;Sastry, 1997;Zanini et al., 2009). For instance, Zanini et al (2009) found that about half of the variability in infant mortality rates in Brazil was due largely to community-level characteristics. Sastry (1997) argued that neighbourhood characteristics can aggravate or alleviate mortality risks of individuals depending on the neighbourhood where individuals reside. Recent evidences have continued to underscore the importance of neighbourhood characteristics on health outcomes. Aminzadeh et al (2013) found an association between neighbourhood deprivation and wellbeing of young people. Unger (2013) observed that areas of broad economic and social disadvantage (due to overcrowding, substandard housing, poor water and sanitation) tend to have higher under-five mortality compared to socially and economically advantaged areas. Although a study by Gilbert et al (2013) established a decline in infant mortality across neighbourhood income quintiles over time in Canada, the study indicates the need for infant health promotion policies among the vulnerable populations. A study by Becares et al (2013) established association between area deprivation and poor health outcomes, and concludes that addressing neighbourhood poverty and area deprivation is essential to improving health outcomes of individuals.
Although, the effect of neighbourhood contexts on child survival has been recognised in a number of countries in sub-Saharan Africa (Antai, 2011b;Boco, 2010;Omariba, et al., 2007) and elsewhere (Sastry, 1997), evidence is sparse on the effects of neighbourhood contexts on under-five mortality in Nigeria.
Several studies on under-five mortality in Nigeria seem to emphasize individual-level determinants and clinical-related causes of infant and child deaths, e.g.; (Ekenze et al., 2009;Ifesanya et al., 2009;Melliez et al., 2007;Okafor et al., 2009;Okoro et al., 2009). Many other studies that are not hospital-based, but population-based failed to consider the influence of contextual situation where children are raised on their survival chances. To this end, the present study seeks to advance the existing knowledge beyond the understanding of clinical-related or individual-level factors by examining the influences of broad familial and neighbourhood contexts on under-five mortality in Nigeria.

Theoretical framework
This study has its theoretical underpinning in two theoretical models -(1) Mosley-Chen model and (2) Sastry Framework. Mosley and Chen in 1984 postulated a model which proposes that socio-economic determinants of child mortality operate through a common set of biological and proximate mechanisms to influence child mortality (WHO, 2003). The model takes into account a range of factors grouped into three broad categories: individuallevel variables (i.e. individual productivity -a t mothers or fathers' level as well as traditions/ norms/attitudes); household-level variables (e.g. income and wealth) and community-level variables (i.e. ecological setting; political economy and health system). Building upon the premise established by Mosley-Chen model, Sastry (1997) categorized the proximate determinants into three broad categories: genetic, behavioural and environmental. He argued that these determinants can occur at three different levels of operation: child, family and community; and that the three levels provide a logical organization for the variables that are likely to influence child mortality.
The thrust of the two models is that children belonging to the same household are exposed to the same situation while children in the same neighbourhood are exposed to the same neighbourhood situation. Thus in this paper, the relationship between neighbourhood context and child survival was conceptualized based on the foundation established by these two theoretical models.

Data and method
This study draws on 2003 and 2008 Nigeria Demographic and Health Survey (NDHS) data. Data from 2003 and 2008 NDHS were utilized in this study purposely to understand the influence of neighborhood contexts on child survival within a ten-year period -1998-2008. The two surveys adopted similar methodology and this makes results comparable. The primary sampling unit (PSU) which was regarded as a cluster for both surveys was defined on the basis of Enumeration Areas (EAs). While the 2003 NDHS adopted the Enumeration Areas designed for 1991 population census, the 2008 NDHS utilized the Enumeration Areas designed for Nigeria 2006 population and housing census. Samples for the two surveys were selected using stratified two-stage cluster design consisting of 365 clusters for 2003 NDHS (NPC and ORC Macro, 2004) and 888 clusters for 2008 NDHS (NPC and ICF Macro, 2009). Data were gathered from 7620 women aged 15-49 in 2003 and 33,385 women aged 15-49 women in 2008. The relevant data for this study (women age 15-49 years who had at least one live birth within the five years preceding the survey) were extracted from the whole 2003 and 2008 NDHS datasets.
Out of the survey's complete sample size of 7620 women contained in 2003 dataset, the sample size for this paper comprised 3775 women who had a total of 6028 live births within the five years before the survey. Similarly, from a total of 33,385 women contained in the 2008 dataset, the sample size for this study comprised 18,028 women who had a total of 28,647 live births within the five years preceding the survey. The birth recode datasets of both surveys were utilized in the analysis. The data contained information for each live birth with maternal variables replicated for births to the same mother.
To account for oversampling of some sections of the population, weighting factor provided by Measure DHS was applied in data management and analysis. Data quality assessment indicates that both datasets were of good quality. For instance, data quality assessment test for 2008 NDHS data indicates that the percentage of missing information on births and deaths only varied between around 1% and 3% (NPC & ICF Macro, 2009).

Outcome variable
The outcome variable for this study is the risk of death during the first five years of life. This is defined as the risk of a child dying between birth and the fifth birthday. The variable was measured as the duration of survival since birth in months.

Independent variables
The independent variables in this paper included such important characteristics at the individual level, familial level (household level) and community level. The selection of independent variables in this study was guided by the reviewed literature and the theoretical foundation established from the literature. The variables at the individual level included characteristics at the child and mother levels. These are maternal age, child's sex, birth order, birth interval, child's size at birth (the self-reported size of the child at birth), prenatal care, contraceptive use, place of delivery and maternal education. Important selected household-level variables which could influence child survival include: family structure, children ever born and wealth index. The selected community-level variables in this study are: region of residence, place of residence, ethnic diversity, distance to health facility, community maternal level of education, community infrastructures, community prenatal care, community poverty level and community hospital delivery. Apart from region of residence, place of residence and distance to health facility, all other community level characteristics were created from the individual-level and household-level variables. Using Stata software (version 11.1), individual and household-level variables were aggregated at the level of PSU to create the community-level variables of interest. The generated community-level variables were divided into three tertiles and then categorized as low, medium, and high. The decision to create the community variables considered in this study was based on the understanding obtained from the reviewed literature.

Data analysis
Descriptive and inferential statistics were employed in data analysis. The background characteristics of the study sample were presented using percentage distribution. Multilevel Cox proportional hazards model was employed to examine association between the outcome and the explanatory variables; and to examine measures of variation in under-five mortality across contexts. All analysis was done using Stata (version 11.1). The multilevel Cox regression analysis was performed using the generalized linear latent and mixed models (GLLAMM) procedure downloadable and implementable in Stata (Rabe-Hesketh et al., 2004).

Multilevel Cox proportional hazards model: methodological procedure
A number of researchers had employed multilevel analysis to identify correlates of under-five mortality.
The assumption is that "individuals (level 1) are nested within households (level 2), and households are nested within communities (level 3)", (Harttgen and Misselhorn, 2006:6). To date, with the exception of one or two studies, Nigerian studies on under-five mortality have rarely employed multilevel model approach. The multilevel approach is good at identifying the broad social, economic and environmental contexts in which a child lives and experiences a particular health outcome (Griffiths et al., 2004). Individuals with similar household characteristics can have different health outcomes when residing in different communities with contrasting characteristics. Griffiths and colleagues (2004) opined that it would be methodologically wrong to fit a single-level standard regression model in the analysis of child survival. This is because standard regression models cannot handle hierarchical structure in the datasets due to its assumption of independence. In this paper, therefore, multilevel Cox proportional hazards regression analysis was undertaken to determine the extent to which the contextual determinants explain underfive mortality in Nigeria. Further, Cox proportional hazards model (i.e. survival analysis) is appropriate in analysing censored observations. This means that, using Cox proportional hazards regression analysis, both the children's survival status and the time when the child died (or censored) were combined to generate the outcome variable. The Cox regression procedure is a useful technique for analysis of survival data and it takes care of censoring problem in mortality data. This is because some children are not fully exposed to the mortality risk. In social science research, an observation is said to be censored when the outcome of interest has not occurred. Cox regression analysis allows for the inclusion of censored observation and it models censored time-until-event data as a dependent variable where it can be assumed that the covariates have a multiplying effect on the baseline hazard.
The outcome variable in this study was treated as the time between birth and death of a child under age five years; or until the observation is censored. Children known to have died (i.e. non-censored) were regarded as the cases, while children who were still alive at the time of the survey were treated as right-censored observations. The probability of under-five mortality is called the hazard. The hazard was modelled using the following equation: Where ... are a collection of explanatory variables and is the baseline hazard at time t, representing the hazard for a person with the value 0 for all the explanatory variables (Fox, 2002). By dividing both sides of equation 1 by H 0 (t) and taking logarithms, the equation 1 becomes: Where H(t) / H 0 (t) is regarded as the hazard ratio. The coefficients b 1 ...b k are estimated by Cox regression. To estimate both the fixed and random effects in the multilevel survival analysis, it could be assumed that the hazards of any two units are proportional (Rabe-Hesketh et al., 2004) and this is modelled as: In the above equation, there are two levels -(i.e. the two subscripts) -where i represents the level 1units and j stands for the level two units, denotes the linear predictor of GLLAMM.
Further, the multilevel analysis involved fitting seven models each using 2003 and 2008 DHS datasets. These models provide understanding on the influences of contextual factors on children's survival chances. Model 0 is the empty or null model and contains no explanatory variables, but focused mainly on decomposing total variance into both individual and community levels components. Model 1 considered only the region of residence in order to examine the independent influence of region where children were born or raised on their survival chances. Model 2 incorporated the child-level variables into the multilevel analysis. Model 3 incorporated the mother-level (familial) variables into the multilevel analysis. Model 4 considered only the contextual factors in order to examine the effect of community-level factors on child survival; independent of other factors. Model 5 is the full model that incorporated all variables into the multilevel analysis. Model 6 is the final model. Fitting this model involved two steps. First, stepwise survival analysis was done to determine the key variables associated with under-five mortality. Second, all the variables selected using stepwise Cox regression analysis were incorporated into the multilevel model.
Measures of association (i.e. fixed effects) were expressed in this paper as hazard ratios (HR) and pvalue (α= 0.05). The random effects which were regarded as measures of variations in under-five mortality across communities were expressed in this paper as intra-class correlation (ICC) (or variance partition coefficient (VPC), and proportional change in variance (PCV). The precision of random effects was determined by the standard error (SE) of the covariates. To determine the goodness of fit of the consecutive models, regression diagnostic was done using Akaike Information Criteria (AIC). Introduced by Hitrotugu Akaike in 1971, AIC measures the relative goodness of fit of statistical model (Reviews, 1988). Lower value of AIC indicates a better fit (Boco, 2010). Table 1 revealed that proportion of the children was almost the same for males and females in 2003 (50.8% vs. 49.2%) and 2008 (50.9% vs. 49.1%). For both surveys, Table 1 also shows that percentage of children was highest for children of birth order 2-4, children with preceding birth interval of 2 years or more, children with large size at birth, children of uneducated women, children of women aged 25-34 years, children of currently married women, children delivered at home, children of Muslim mothers, children from poorest households and children whose mothers had never used contraceptive.  Results from analysis of both surveys indicated that inclusion of child-level variables into the multilevel models (Model 2, Tables 2 and 3) did not significantly alter the results obtained in Model 1. This suggests that the effect of region where children were born or raised (on their survival chances) was independent of child-level characteristics.

Results presented in
In contrast, incorporating the mother-level variables into the multilevel analysis in Model 3 (Tables 2 and 3) indicated mother's attributes as important characteristics, as the risks of dying before age five became statistically insignificant in many regions. This suggests that mother-level variables included in the multilevel analysis; such as maternal education, wealth index, contraceptive use, number of children ever born and maternal age are important factors that influence variations in the risks of under-five mortality in various regions of the country. For instance, Table 2 showed that being a child of an educated woman with at least secondary education (HR: 0.70, p<0.05); being a child of a woman from households in the richest wealth quintile (HR: 0.64, p<0.05); and being a child of woman using modern contraceptives (HR: 0.67, p<0.05) was associated with lower risks of under-five mortality, irrespective of the region of residence where a child was born. Further, Model 4 considered only community-level factors in the multilevel models. Results in Model 4 Table 2 indicated that the risks of dying before age five were statistically insignificant across all regions, while results in Model 4 Table 3 indicated further reductions in the risks of dying during childhood across regions. The results in Table 3 (Model 4) suggests that the characteristics of the neighbourhood contexts tend to mitigate the risks of under-five death in some regions, while neighbourhood characteristics tend to increase risks of under-five death in the North-east (HR:1.42, p<0.05) and South-south (HR: 1.30, p<0.05).
Results from Model 5 (full model) which incorporated all covariates as presented in Tables 2 and 3 indicated some elevated risks of under-five mortality during the period under study in the North-central (HR increased from 0.92 to 1.39, p<0.05); Northeast (HR increased from 1.24 to 1.45, p<0.05) and North-west (HR increased from 1.22 to 1.44, p<0.05), compared to the South-west.
Further results from the full model (Model 5 Table 2) showed that child-level variables (such as child's sex, birth interval, and child's size at birth), mother-level variables (including number of CEB and contraceptive use) were significantly associated with under-five mortality. Results from 2008 dataset (Model 5 Table 3) indicate individual level factors (such as child's sex, birth interval, contraceptive use, family structure); and neighbourhood factors (such as region of residence, ethnic diversity, and community hospital delivery) as important determinants of under-five mortality in Nigeria.
Results from final model (Model 6 Table 3) showed that the risks of dying before age five were significantly higher for children in the North-central (HR: 1.43, p<0.05), North-east (HR: 1.56, p<0.05), North-west (HR: 1.55, p<0.05) and rural areas (HR: 1.25, p<0.05). Model 6 also shows that residence in communities where high proportion had access to electricity (HR: 0.64, p<0.05 - Table 2); residence in ethnically heterogeneous communities (HR: 0.82, p<0.05 - Table 3) was significantly associated with lower risks of under-five mortality. In addition, Model 6 ( was associated with significantly higher risks of under-five mortality. Finally, the decreasing values of AIC (Tables 2  and 3) with each successive model indicated a good fit of the multilevel models with each successive model signifying a significant improvement of the previous model.

Discussion
This paper examines the effects of neighbourhood contexts on under-five mortality in Nigeria. Multilevel Cox proportional hazard model was employed to account for the hierarchical nature of the DHS data. This is because the children were nested within mothers, and mothers were nested within communities.
Findings showed the importance of both individual-level characteristics (i.e. mother and child levels) and community or neighbourhood contexts for child survival in Nigeria. For instance, results from the null model indicated a significant variation in under-five mortality across communities. This result suggests that situations of the community contexts where children are born or raised significantly influence under-five mortality in Nigeria. This underscores the need to take neighbourhood contexts into consideration (Aremu et al., 2011;Boco, 2010) in the efforts to address the high level of under-five mortality in the country.
Findings in this paper established a strong association between under-five mortality and region of residence. As previously found (Antai, 2011b), this result indicates region of residence where children were born or raised as important contextual factors influencing under-five mortality in Nigeria. In particular, the results demonstrated that while characteristics of the neighbourhood contexts appear to alleviate under-five mortality risks in the South-west Nigeria, community characteristics tend to aggravate risks of under-five mortality in other regions, especially North-east, North-west and South-south. Sastry (1996) noted that characteristics of the community contexts can either mitigate or exacerbate mortality risks of individuals depending on the neighbourhood where individuals reside.
Findings further established significant relationship between under-five mortality and other neigh-bourhood characteristics such as place of residence, community poverty, and community infrastructure. For instance, neighbourhood infrastructure such as access to electricity in a community was established as important predictor of under-five mortality in Nigeria. This result points to the importance of good infrastructural contexts for child survival. Constant electric supply could possibly help to prevent some infectious diseases through the use of electrical appliances such as fridge (to preserve foodstuffs) and microwaves (to keep food warm for children). In addition, availability and access to drinkable water within the neighbourhood could prevent children from contacting avoidable infections like waterborne diseases such as diarrheal disease and other forms of infectious diseases. Sastry (1996) had showed that infrastructures such as electricity and water supply are significantly related to child survival.
The socio-economic context of the community where children live was also found as important predictor of under-five mortality in the country. It was shown that poverty concentration within a community appears to significantly increase risks of underfive mortality. Factors that contribute to poor socioeconomic contexts in Nigeria include poor infrastructure (National Planning Commission, 2004). Nigeria is one of the sub-Saharan African countries that face most critical infrastructural challenge (Akinwale, 2010;Foster & Pushak, 2011). The negative effect of poor socio-economic context on child survival is expected to be severe in a country like Nigeria where there is a high level of infrastructural deficiency.
Corrupt practices at various levels of governance in Nigeria have led to non-availability of good infrastructures in the country (National Planning Commission, 2004;Osoba, 1996) provide social services such as good roads, electricity, piped-borne water, health facilities as well as adequate securities among others, the usual practice now is for the relatively better-off communities to make communal arrangements to provide services such as bore-holes, dispensaries, motorable roads and even vigilante groups for their communal benefits. While these activities may be affordable for very rich and relatively better-off communities, communities predominantly occupied by poor people would find it difficult to make provision for such services. Thus, the present study established an elevated risk of under-five mortality for communities that had poor socio-economic contexts, perhaps because young children are likely to be highly vulnerable to unfavourable community contexts arising from lack of essential social services. This establishes one of the reasons why poor neighbourhoods like slums tend to have poor child health outcomes (Unger, 2013). Results of this study provide empirical explanations supporting the adopted theories which posit that community or neighbourhood contexts matter for health outcomes of individuals. Findings of this study suggest that achieving improved neighbourhood contexts holds great potentials for acceleration of under-five mortality reduction in Nigeria.

Conclusion
This study found that substantial variations in underfive mortality exist across neighbourhood and regions in Nigeria, with economically and socially deprived neighbourhoods and regions having elevated risks of under-five mortality compared to better-off communities and regions. Findings of this study suggest that efforts to attain Millennium Development Goal four (childhood mortality reduction) should not only include policies that address individual-level factors. Government and well-meaning individuals are urgently needed to intensify efforts towards undertaking neighbourhood or communitylevel interventions aimed at improving child survival in the socially and economically deprived neighbourhoods and regions. lier version of this paper was presented at the 27 th International Population Conference of the International Union for the Scientific Study of Population (IUSSP) held in Busan, South Korea, from 26-31 September, 2013. Insightful comments from conference participants are gratefully acknowledged. Author is also grateful to ICF Macro for permission to use Nigeria DHS data.