Clustering Countries on Development Indicators Reveals Structure Relevant for H5N1 Mortality Analysis

Infectious diseases are often observed to have different epidemiology in different countries, which arises due to various factors including those that are ecological, socioeconomic, and healthcare-related. Such variability can sometimes be best captured through looking at groups of countries that are similar within-group but variable between-group. In this study we use statistical learning methods to generate data-driven disease-centric groupings of countries rather than those developed for administrative or political reasons by e.g. the WHO, World Bank, and the United Nations. In particular, we apply hierarchical clustering to group countries based on shared disease-relevant characteristics for zoonotic H5N1 influenza. Using statistical methods such as classification and regression trees (CART)-based imputation and dynamic tree cutting, the analysis accounts for missing data and identifies epidemiologically (rather than politically or economically) meaningful clusters. Applying health metric relevant indicators, we cluster the countries of the world and using a Bayesian approach compute CFRs of zoonotic H5N1 influenza before comparing across clusters. We find that countries with stronger healthcare systems and lower poverty rates tend to have lower and more stable CFRs, whereas resource-limited settings face higher fatality risks.