Turning datasets into patient-centered knowledge utilities

Abstract

This paper describes an approach utilizing both data analysis and visualization to help diabetes patients improve medication compliance. Through information visualization tools, we aim to provide feedback to patients to encourage behavior change. The system has two core building blocks: (1) Data analysis combining several statistical and machine learning models, founded under different principles and assumptions, into a single meta-model for predicting compliance behavior. The aim is to create superior models for behavior prediction - knowledge that can then be translated into patient-centered decision-support tools. (2) Incorporating data analysis and visualization enabling datasets to be turned into knowledge utilities that can intelligently interact with participants by alerting them to any interesting correlations within the data. Such tools could provide feedback indicating, for example, a high-risk to medication non-compliance behavior in which case appropriate resources could be directed to those who need help the most.