PhD Student Seminar: Non-Confidential Clinical Prediction with Summarized Data

Presenter: Mahsa Rouzbahman, PhD candidate, Mechanical and Industrial Engineering

Abstract:

This research reports on a series of four studies that were motivated by the need to incorporate information about similar patients into clinical decision making. The studies examined questions relating to the feasibility of a tool that could identify similar patients in health data repositories, and then use the resulting patient types for prediction of unknown values in the health record for a current patient. The two main questions addressed in this research were: 1) How interpretable are patient types developed using k-means cluster analysis, and 2) how well do predictions based on summarized clusters match predictions based on the original (unsummarized) data in terms of accuracy? The motivation for using summarized prediction in this research is that summarization of sufficiently large numbers of cases removes the privacy concerns typically associated with using individual data.

Two different datasets were used in this research. First, intensive care data (MIMIC II database) was utilized. A set of patient types was developed and the accuracy of predictions (length of stay and vital status) made with summarized data based on the patient types was assessed. Then, using cancer data (ICES) from the province of Ontario, a set of patient types was developed and the accuracy of predictions (emergency department visit and vital status) made with summarized data, based on the set of cancer patient types, was evaluated.

The overall findings in this research were that interpretable patient types (at least in situations similar to the two case studies considered here) can be developed relatively easily, and then used to make predictions based on summarized patient data in a way that does not raise privacy concerns. The present results point the way towards improved clinical decision support tools, and, possibly, the development of generalized health data search engines.

Biography

Mahsa Rouzbahman is currently a PhD candidate of Mechanical and Industrial Engineering at the University of Toronto. She received a M.Sc. degree in industrial engineering from University of Tehran. As part of her PhD thesis, she has been working at Institute for Clinical Evaluative Sciences (ICES) at Sunnybrook hospital since Jan 2014, working on creating cancer patient types and making predictions based on those types. Her main fields of interest are data mining of medical records, human factors research, user interface design for health care environments and clinical decision support systems. Contact her at mrouz@mie.utoronto.ca.