The Utility of Shapelets for Analyzing Physical Activity of COPD Patients and non-COPD controls

摘要

Physical activity is an attractive endpoint for novel therapies in Chronic Obstructive Pulmonary Disease (COPD). However, a deep understanding about COPD physical activity patterns and disease severity is lacking. In this research, we study the physical activity patterns for 184 individuals with and without COPD from a single center in the COPDGene cohort. These subjects participated in a 3-week observational study wearing wrist-worn accelerometers for collecting physical activity data. Our exploratory data analysis finds using the whole range of activity data is insufficient for patient clustering. Alternatively, we use shapelets, small and local sub-sequences, to better capture patients' behaviors in different groups. We develop a length-bound heuristic algorithm for choosing the subset that has the best clustering result. The study shows the potentials of using shapelets for helping providers in assessing COPD patients' status.