A Bayesian Hierarchical Analysis of Geographical Patterns for Child Mortality in Nigeria

Abstract

Background: In an epidemiological study, disease mapping models are commonly used to estimate the spatial (or temporal) patterns in disease risk and to identify high-risk clusters, allowing for health interventions and allocation of the resources. The present study proposes a hierarchical Bayesian modeling approach to simultaneously capture the over-dispersion due to the effect of varying population sizes across the districts (regions), and the spatial auto-correlation inherent in the childhood mortality at districts (state) level in Nigeria. Methods: This cross-sectional study was based on 31842 children data extracted from the 2013 Nigeria Demographic and Health Survey (DHS). Of these children, 2886 died before reaching the age of five years. A Standardized Mortality Ratio (SMR) was estimated for each district (state) and mapped to highlight the risk patterns and detect an unusual low (high) clusters relative risk of childhood mortality. Generalized Poisson regression models were formulated with random effects to estimate the mortality risk and then explored to investigate the relationship of under-five child mortality and the regional risk factors. The random effects are formulated to reflect the potential tendency of “neighbouring” regions to have similar risk patterns and the spatial heterogeneity effect was used to capture geographical inequalities in the mortality outcomes. The models were implemented using a full Bayesian framework. All model parameters were estimated in WinBUGS via Markov Chain Monte Carlos (MCMC) simulation techniques. Results: The results showed that of the economically deprived households, 2.088: 95% CI (1.088, 3.165) were significantly associated with childhood mortality, while unhygienic sanitation and lack of access to improved water sources were positively associated with child mortality, but not statistically significant at 5% probability level. The geographical variation of the under-five mortality prevalence was found to be attributed to 69% clustering and 31% was due to spatial heterogeneity factors. The predicted probability maps identified clusters of high risk mortality in the northern regions and low prevalence of concentrated mortality in the south-west regions of Nigeria. Conclusion: The results demonstrated the flexibility of the approach that explored the geographical variation in the potential risk factors of child mortality and that it provides a better understanding of the regional variations of mortality risks. Nonetheless, both representations can help to provide information for the initiation of public health interventions.

Spatial regression models to assess variations of composite index for anthropometric failure across the administrative zones in Ethiopia

“…The number of CIAF cases observed in the zone i (i = 1, 2,…, 72), denoted by y i , was assumed to follow a Poisson distribution with mean λ i = E i γ i , where E i (offset term, which is used as a correction factor for the model) [ 40 , 41 ] denotes the number of expected cases (CIAF) in zone i, and γ i is the relative risk for zone i. Moreover, E i was calculated as where n i is the number of under-five children in zone i [ 42 – 44 ]. The spatial matrix can be constructed in many ways depending on the definition of the neighborhood employed ( Fig 2 ).…”

“…where n i is the number of under-five children in zone i [42][43][44]. The spatial matrix can be constructed in many ways depending on the definition of the neighborhood employed (Fig 2).…”

Bayesian disease mapping and the ‘High‐Risk’ oral cancer population in Hong Kong

“…to generate reliable posterior probability estimates, thus informing reliable map construction. [2][3][4][5][6][7][8][9] Oral cancer, primarily squamous cell carcinoma (SCC) arising from the mucosal lining of the lips, oral cavity and oropharynx, is the 14th highest malignancy worldwide in terms of both incidence and mortality. 1,10,11 With substantive geographic variation in disease distribution, proactive preventive measures and targeted screening of disease-prone individuals in "high-risk" regions are pivotal methods to reduce the high five-year mortality rates associated with late cancer detection.…”

“…Cancer mapping has been trialled to determine spatial variations in disease clustering and to document risk factor distribution, case‐related mortality and survival data; "cancer atlases" have been published in Australia, Italy and the USA. Statistical models based on Bayesian conditional probability have proved most effective in generating robust "smoothed" cancer outcomes combining information from data likelihood and prior distributions (an uncertainty measure) to generate reliable posterior probability estimates, thus informing reliable map construction 2‐9 …”