Comparison of PAM and CLARA clustering algorithms for assessing the causes of mortality in the population covered by Mashhad University of Medical Sciences in 2017

Analysis of mortality information helps in making better health policies. The present study discusses causes of mortality observed in Mashhad, Iran in 2017 using partitioning around medoids (PAM) and Clustering Large Applications (CLARA).

Data from 21838 subjects were collected from the “death registration system” of Mashhad University of Medical Sciences, Mashhad, Iran, and after data cleaning and dealing with incomplete data, individuals were clustered based on 4 variables: age, sex, leading cause of death and ​​place of residence using PAM and CLARA. The optimal number of clusters was determined before clustering using the mean of silhouette. Finally, the clustering performance of each algorithm was evaluated.

Based on the silhouette criterion, the highest difference among clusters was found when  five clusters were considered. In four of these clusters, which mostly included people aged over 65, circulatory, neoplasms, diseases of the respiratory system and endocrine, nutritional and metabolic diseases were the first to fourth leading causes of death, respectively. In the remaining cluster, the most common leading causes of death were congenital malformations, deformation and certain conditions originating in the perinatal period for children under 1 year of age.

Our results indicate that health policy makers should consider the importance of age distribution in clusters of a given population. Studying on death from cancer at the age of under 14 years and high percentage of death from diseases of the nervous system can contribute to health policies.