B1111
Title: Multilevel factor clustering on matrix time series EHR data
Authors: Hulin Wu - University of Texas Health Science Center at Houston (United States) [presenting]
Abstract: In electronic health records (EHR), the patient information can be abstracted as a three-dimensional tensor or matrix time series: the number of patients N, the observation time T, and the number of clinical variables d. To cluster patients based on the tensor representation from EHR, the multilevel factor clustering (MFC) model is proposed, which consists of global factors and group-specific factors. The global factors represent the commonly driven latent force for each subject while the group-specific factors reflect the fact that each subject is influenced by a certain number of latent cluster factors. The theoretical properties of MFC are studied for the cases where N and T go to infinity. The simulation studies show that the model performs well in terms of clustering and parameter estimation. A real data application for EHR data demonstrates that the proposed method can be used to cluster patients based on their longitudinal EHR data.
