Characterizing Subjects with Chronic Obstructive Pulmonary Disease in GOLD Stage 2 (303218)

Keywords: bayesian hierarchical clustering, COPD, GOLD criteria

Traditional clustering methods have limitations in predetermining the number and height of clusters (where to “cut the tree”), as well as evaluating goodness of fit. Bayesian Hierarchical Clustering (BHC) has advantages over traditional methods as it uses a probabilistic model to determine the ideal number and height of clusters using Dirichlet Distribution by Markov Chain Monte Carlo (MCMC). Chronic Obstructive Pulmonary disease (COPD) is a lung disease that limits airflow. The GOLD classifications, used to describe the severity of COPD, can be calculated through pulmonary function tests adjusted by demographic with symptoms of shortness of breath. GOLD criteria stage 2 is especially known to exhibit heterogeneous subgroups. We apply BHC to classify patients into clusters, or subpopulations, in order to characterize types of airflow limitation such as small airflow disease and parenchymal destruction. We observed 5 top-level dendrograms with hyperparameter of 0.51 based on high mutual information shared between COPD Gold stage 2 patients.