Janani Venugopalan

Ph.D. Proposal Presentation

Tuesday, September 8th 2015, 9am

Whittaker 2nd Floor Conference Room (2110)

Advisor: May D. Wang, Ph.D. (Georgia Institute of Technology, Winship Cancer Institute, Emory University)

Committee:

Robert Butera, Ph.D. (Georgia Institute of Technology, Emory University)

Peng Qiu, Ph.D. (Georgia Institute of Technology, Emory University)

Mark Braunstein, MD. (Georgia Institute of Technology)

Nikhil Chanani, MD. (Children’s Healthcare of Atlanta)

Kevin Maher, MD. (Children’s Healthcare of Atlanta, Emory University)

“Clinical Decision Support Using Time Series Data Analysis for Electronic Health Records”

US healthcare is undergoing major reforms towards evidence based and precision medicine with an emphasis on data driven models. There is a strong impetus for improving the quality of healthcare while decreasing the cost incurred. The objective of this research is to develop temporal data mining models for clinical decision support using retrospective analysis of electronic health records. These models will target the prevention of readmission while reducing adverse events such as mortality, cardiac arrest and long stay in critical care units. More specifically the work aims to perform the following, 1) Quality control practices in electronic health record data to develop robust clinical markers indicative of increased length of stay and associated adverse events in the ICU such as mortality, 2) To develop advanced data-driven analytics by integrating static and temporal predictive markers extracted from electronic health record data to facilitate clinical decision-making and better clinical care and 3) To build an intelligent visualization system to provide assistive information for care providers.  

In the first aim we focus on the data issues and is aimed at solving challenges such as missing data. We divide the missing data into multiple types on the basis of the statistical properties of the data and develop novel methodologies to impute each missing data type. In the second aim, we perform temporal analysis of quality-controlled data and compare with conventional non-temporal analysis. We also develop models which combine non-temporal and temporal methods to make improved prediction on mortality and 30 day ICU readmission. In the third aim we develop a system to visualize the prediction probabilities of adverse events (output from our models) and propose to provide a framework for self-learning. We will demonstrate our results on adult and pediatric electronic health record data.