Small area analysis of levels and dispersion in cause-specific mortality in Brazil and Sub-Saharan Africa with bayesian models for proportion data

• Grant number: 18/18649-7
• Support type: Research Grants - Visiting Researcher Grant - International
• Duration: March 30, 2019 - May 29, 2020
• Field of knowledge: Applied Social Sciences - Demography - Demographic Data Sources
• Principal Investigator: Everton Emanuel Campos de Lima
• Grantee: Everton Emanuel Campos de Lima
• Visiting researcher: Ezra Gayawan
• Visiting researcher institution: Federal University of Technology, Akure (FUTA), Nigeria
• Home Institution: Instituto de Filosofia e Ciências Humanas (IFCH). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil

Abstract

The planning, implementation and evaluation of health policy and programmes at sub-national, national and international levels require valid, comparable and timely information on the nature and magnitude of health problems. However, these efforts have been hampered by lack of mortality estimates in most developing countries especially at small-area levels. Much effort have been put into understanding the dynamics of child mortality in most developing countries (Gayawan et al., 2016, Almeida et al., 2014) but little is still known about the spatio-temporal variability in the patterns of adult mortality and in mortality by specific cause.Cause-specific mortality is one of the fundamental metrics of population health. Though many developing countries have taken actions to improve the quality of mortality data, there still exist problems of completeness of registration and proportions of deaths classified to non-specific causes of death, with major differentials among geographic regions (França et al., 2008). Even within countries, there are evidence of huge differentials in the records among small-area settings. Demographers and statisticians have therefore developed models for estimation in cases of incomplete or underreported demographic data (Brass, 1971). Statistical smoothing methods that adopt robust procedures are combined with available data to provide local estimates. Modern approaches based on Bayesian methods are extremely useful in this direction due to their ability to "borrow strength" from neighbouring locations to improve local estimates in areas with limited data (Gelfand et al., 2010). Bayesian methods of estimating spatio-temporal patterns of mortality and other demographic rates have largely been focused on quantifying the levels of the rates under consideration based on mean regression models.To this end, this project is aimed at proposing the use of Bayesian distributional spatial regression models for the analysis of cause-specific and ill-defined cause of mortality in developing countries. Of special interest to the project are models for proportional outcomes such as proportion of deaths by specific cause or proportion of ill-defined cause of deaths within small geographical settings and over time where the natural candidate model for such ratios is the beta distribution parameterised such that one parameter represents the expectations while the other relates to a general shape parameter. Though recent efforts have been made within the beta regression models (Ferrari and Cribari-Neto, 2004), we intend to relate the whole distribution conditional on spatial covariates rather than just the mean. Also of interest to this research is the possibility of having censored observations in mortality data leading to excessive number of zeros in the cause-specific mortality data such that the beta distribution would have to be inflated with zeros similar to the zero-inflated count data regression. Another goal is to ensure that our Bayesian regression approach can address the fundamental issues in small-area mortality estimation importantly, the defective registration system common with most developing countries. The study would focus on estimating these mortality data from Brazil and sub-Saharan African countries. (AU)