Maiden klusteroinnista kuolinsyydatan perusteella

Comparative analysis of health data across countries helps understand inequalities in health outcomes and evaluate the effects of health policies. Mortality profiles are defined by cause-specific mortality rates, and they reflect the living conditions within a country. Thus, using mortality profiles to group countries with similar mortality patterns can provide meaningful information about the similarities and differences in health outcomes and living conditions across countries. When multiple causes of death are considered in the analysis, the setting is inherently multivariate, suggesting the use of multivariate data analysis methods. This thesis examines clusters of countries identified from cause-specific mortality data across 26 countries, mainly from Europe, Latin America, and North America. Clustering is performed separately for three years: 2001, 2018, and 2021. The data used for clustering consist of gender-specific age-standardized death rates for five high-level causes of death, including neoplasms, accidents and assaults, diseases of the circulatory system, diseases of the respiratory system, and intentional self-harm. The clusters are derived from the data by applying agglomerative hierarchical clustering using Ward's method. For each considered year, two-, three-, and four-cluster solutions are examined and compared between the years. To assess the socioeconomic similarity within the identified clusters, the Human Development Index (HDI) is used as an external validation measure. The clustering results suggest that interpretable groups of countries in terms of geographical and socioeconomic similarity can be identified from the mortality clusters. Across all examined years, Eastern European and Latin American countries are often clustered into distinct clusters. Countries within clusters also tend to have similar HDI levels, with high-HDI countries clustering together almost always. The changes in the cluster composition from 2018 to 2021 are driven by increased mortality from respiratory diseases, potentially linked to the COVID-19 pandemic. However, this connection is not extensively studied in this thesis. Overall, this thesis demonstrates the usefulness of clustering in comparative health analysis.