Socioeconomic Inequalities in Adult Mortality across Small Areas in Brazil: Exploration Analysis Using Machine Learning Models

Abstract

BACKGROUND Future progress in life expectancy in all countries depends, to some degree, on a reduction in adult mortality. Significant regional differences are found in Brazil, with high adult mortality rates in some areas. OBJECTIVE The aim of the present study was to investigate associations between the probability of adult deaths in microregions of Brazil in 2010 and socioeconomic, structural, contextual and health-related factors. METHODS The analyses were based on data from the 2010 demographic census and the Mortality Information System. The machine learning model was used to establish determinants of the probability of adult deaths. Machine learning methods have considerable potential for this type of analysis, enabling a better understanding of the interactions among different factors. The algorithms employed (Random Forest, Extreme Gradient Boosted Trees and Support Vector Machine) obtained a good performance and proved to be effective at analyzing the variables and correlations with the outcome (probability of adult death). RESULTS The results showed the mortality due to external causes, the employment rate, sex ratio, vaccine coverage, proportion of blacks, poverty rate, aging index and proportion of whites had the best predictive power regarding the probability of adult deaths using the algorithm with the best performance (Extreme Gradient Boosted Trees). CONTRIBUTION We use a methodological approach directed at understanding patterns in data and focusing on small areas within a highly unequal and diverse country, that is, use a data-driven approach in place a hypothesis-driven approach.

Estimation and projection of probabilistic age- and sex-specific mortality rates across Brazilian municipalities between 2010 and 2030

“…Age-specific mortality rates, and their accurate estimation and projection in small areas, are essential for public health, economic, and social planning [1,2]. They may also be applied to monitor the Sustainable Development Goals [3,4] and support the development of other small area measures [5].…”