Small area clustering of Under-five children's mortality and - PowerPoint PPT Presentation

Small area clustering of Under-five children's mortality and associated factor using Geo-additive Bayesian Discrete-time Survival Model in Kersa HDSS, Ethiopia By Melkamu Dedefo The 28 th International Population Conference of the International Union for the Scientific Study of Population (IUSSP) in Cape Town, South Africa 29 October to 4 November 2017

Introduction Under-five Child Mortality Rate is one of the most important sensitive indicators of the socio- • economic and health status of a community. This is because more than any other age group of a population, child’s survival depends on • the socio-economic conditions of their environment. Millennium Development Goal 4 (MDG 4) calls for reducing the under-five mortality rate by • two-thirds between 1990 and 2015. To achieve MDG 4 on time, the global annual rate of reduction in under-five mortality rate • would need to rise to 15.6 percent for 2012–2015, much faster than the 3.9 percent achieved over 2005–2012. Many countries still have very high under-five mortality rates particularly those in Sub- • Saharan Africa, home to all 16 countries with an under-five mortality rate above 100 deaths per 1,000 live births. Reducing these inequities across countries and saving more children’s lives by ending • preventable child deaths are important priorities.

In 2012 the governments of Ethiopia, India and the United States, in close collaboration with • UNICEF, the Child Survival Call to Action Forum to mobilize political leadership to end preventable child deaths. More than 170 governments have signed a pledge to redouble their efforts to end • preventable child deaths so that more countries achieve MDG 4 and sustain momentum beyond 2015 Ethiopia achieved a remarkable decline in all levels of childhood mortality during the years • 2000 to 2011. Under-five mortality dropped from 166 in 2000 to 88 in 2000 per 1000 children. However, the reduction in Under-five mortality showed marked regional variations. Historically, variations in prevalence of childhood diseases have been related to household • socio-economic factors (such as food, good sanitation, and health care) surprisingly, geographical associations with prevalence have been neglected. The links between health, geographic location, environment and economic development • need to be better understood if the problems associated with these issues that affect developing countries are to be overcome.

In Ethiopia, limited studies were conducted on Under-five mortality and almost none of them • tried to identify the spatial effect on mortality by incorporating the coordinate axis. This study explored the small area clustering of under-five mortality and associated factors in • Kersa HDSS of the Eastern Harage Zone of Oromia Regional Sate, Eastern Ethiopia using Geo- additive Bayesian Discrete-time Survival Model

Methodology Data Five years follow up secondary data from Kersa Demographic Surveillance and Health • Research Center (KDS-HRC) dataset was conducted. According to the first census (2007) there were 10,256 households and 53,462 people in the • study site with an average household size of 5.2 . The study population included all deaths of Under-five children registered in the surveillance • site from September, 2008 to august 31, 2012. Statistical analysis Spatial variation in under-five mortality is modeled with a flexible Bayesian geo-additive • discrete-time survival model This models mortality events as person-specific Cox processes while controlling spatial • dependence and possibly nonlinear effects of covariates within a simultaneous and coherent regression framework. And a full Bayesian approach based on Markov priors using Markov Chain Monte Carlo • (MCMC) techniques applied for inference and model.

Data were analyzed with relevant variables to carry out survival analyses for under-five • mortality. The analysis was carried out using version 2.1 of the BayesX software package, which permits • Bayesian inference based on Markov chain Monte Carlo (MCMC) simulation techniques. Model specification For model choice, the Deviance Information Criterion (DIC),developed as a measure of fit • and model complexity. A number of models were explored. Model comparison was based on the DIC . • This is given by DIC=D+P_D , where D is the deviance of the model evaluated at the • posterior mean of the parameters and represents the fit of the model to the data and P_D is the effective number of parameters, which assesses the complexity of the model Small values of D indicate a good fit and small values of P_D, indicate a parsimonious model, • small values of DIC indicate a better model.

.

The Model uses mother’s age at first birth (m_age) as non-linear accounts for the • unobserved heterogeneity that might exist in the data. The effects of f 0 (t) and f 1 are estimated using second-order random walk prior, and Markov • random walk priors for both structured(f str (s)) and unstructured spatial effect(f unstr (s)) The sensitivity of the effects to choice of different priors for the non-linear effects (p -splines) and suitable and proper choice of the hyper parameter values a and b are investigated. Penalized splines . Approximate f(x) by a weighted sum of B-spline basis functions by employing a large number • of basic functions to enable flexibility And Penalizes differences between parameters of adjacent basis functions to ensure • smoothness, assuming that the expected value of S spat (s) is the average of the function evaluations of adjacent sites.

Results

Unadjusted and fully adjusted odds ratios and 95% confidence intervals for the risk of under-five mortality by background characteristics (Kersa_HDSS, 2008-2013)

 The Table displays both marginal and posterior odds ratios of Under-five mortality risks across the selected study characteristics.  Results from both standard logistic regression and multivariate Bayesian geo- additive survival analyses provided evidence of the factors that are significantly associated with higher Under-five mortality with posterior odds ratio and 95% credible region.  Maternal educational status , Area of residence , Place of Delivery 1.016, No of live birth at a delivery, Household wealth index, pre-term Duration of Pregnancy, post-term Duration of Pregnancy and Antenatal visit were the selected factors from the model which are significantly associated with Under-five children’s mortality.

 The estimated nonlinear effect of child’s age (baseline time) obtained from the Bayesian p-splines are shown in the follwing Figure.  The posterior means of log-odds are presented within 80% & 95% credible regions, and show that there is a noticeable effect of the baseline time on child survival during the first months of life

 The posterior means of log-odds are presented within 80% & 95% credible regions, and show that there is a noticeable effect of the baseline time on child survival during the first months of life.

 The follwing figure shows the estimated non parametric effect of mother age at first birth.  The nonlinear effect (U-shape) and association between mother’s age at child’s birth and mortality rates clearly depicted as shown in the Figure.  Higher mortality rates are observed among mothers giving birth at younger ages.

 The map below shows the Posterior means of the estimated residual spatial effects on under-five mortality.  This map shows a strong spatial pattern, which suggests that survival chances of Under-five children are highest within Kersa town and water town compared to the other regions.

 On the other hand, the survival chances of children under-five years are lowest among children from Yabeta-lecha and Ifa-jalala kebeles compared to the children from the rest of the regions.  Tthese with result from the estimated odds ratio from the previous Table reveals the emergence of a clear spatial pattern of under-five mortality risk.  These spatial effects could therefore be interpreted as representing the cumulative effect of unidentified or unmeasured additional covariates that may reflect impacts of environmental and socio-cultural factors.  Thus, failure to take into consideration the posterior uncertainty in the spatial location (kebeles or districts) would invariably lead to an overestimation of the precision in predicting childhood mortality risks in among sub-districts.