Abstract: The consumer health informatics community is increasingly handicapped by ill interpretation of patient need. One missing key is a behavioral learning mechanism that can sift through user-generated health data to identify outcome-differential behavioral patterns. To address this important problem, this presentation will introduce an interpretable clustering approach that adds behavioral response pattern understanding into patient subgroup analysis. The proposed method is then evaluated by scanning through NHANES data to identify behavioral profiles and prototypical examples related to blood pressure control or depression.
Learning Objective 1: Learn to further inform personalized care plan by accurately and reliably quantifying the behavioral patterns from target user-generated health data
Learning Objective 2: Review existing guidelines and data-driven methods of patient subgroup identification and digital phenotyping and Identifiy gaps for realizing the personalization goal of guidelines
Learning Objective 3: Introduce an interpretable clustering approach that adds behavioral response pattern understanding into patient subgroup analysis and informs guideline deployment.
Learning Objective 4: Provide evaluation by scanning through NHANES data to identify behavioral profiles and prototypical examples related to blood pressure control and depression
Pei-Yun Hsueh (Presenter)
IBM T.J. Watson Research Center
Subhro Das, IBM T.J. Watson Research Center