10/21/2016

Post date: Oct 23, 2016 9:49:07 PM

Title: Physical Activity as Deep Phenotyping Markers: Association or Prediction

Speaker: Haochang Shou, Department of Biostatistics, University of Pennsylvania

[Abstract]

Physical activity measures collected via wearable computing sensors have been increasingly used in various clinical domains and may provide new insights to the disease pathology by identifying novel markers. However, little has been done to explore how activity patterns can differentiate disease-related human behaviors. We specifically focus on the data from National Institute of Mental Health (NIMH) family study of affective spectrum disorders and aim to find the link between the real-time activity patterns and mental disorders. In particular, we propose a set of methods to test the differential variations across two groups of functions. For this purpose, we first conduct dimension reduction using multilevel Functional Principle Component Analysis (MFPCA) that accounts for both subject- and day-level variations. Based on the PC results, we then suggest two ways to conduct hypothesis testings on the eigen-functions and eigen-scores that examine whether there are differences between Bipolar I (BPI) and Major Depressive Disorder (MDD) patients. Results from both approaches suggest that BPI patients have more variable activity profiles than MDD especially during night period. Such observations could potentially classifying future patients with available activity measures.