Title: Sequential Mining Based Data Analytics For Personalized and Precision Medicine

Kunal Malhotra

School of Computer Science

College of Computing

Georgia Institute of Technology

Date: Tuesday, Dec 8th, 2015

Time: 9:00 AM - 11:00 AM

Location: Klaus Room 3402

Committee:

----------

Dr.Sham Navathe, (Advisor, School of Computer Science, Georgia

Tech)

Dr.Jimeng Sun, (Co-advisor, School of Computer Science & Engineering, Georgia Tech)

Dr.Polo Chau, (School of Computer Science & Engineering, Georgia Tech)

Dr.Edward Omiecinski , (School of Computer Science, Georgia Tech)

Dr.Arvind Ramanathan, (Computational Science and Engineering Division, Health Data Sciences Institute, Oak Ridge National Laboratory)

Abstract:

---------

Healthcare analytics as an area of study is having an increasing impact as the clinicians are getting valuable insights on the efficacy of treatments based on historical data. Personalized treatment recommendation involves studying patient profiles and recommending appropriate treatment to them. Healthcare providers across the United States have a difference of opinion in treating patients. The challenge is to find out the significant sequences of treatment events that have shown a strong correlation with a positive outcome in available historical datasets to be recommended to clinicians.

This dissertation examines sequential mining approaches to study treatment patterns for a variety of diseases ranging from the rarest of rare cancers such as Glioblastoma to the most prevalent disease world-wide such as heart disease and epilepsy. We propose a non- conventional graph based approach to mine sequential patterns from medical data and come up with medically relevant constraints to be applied on the graph. A known challenge in healthcare is data sharing across multiple vendors and healthcare organizations, which results in analysis biased to an identical population. Another challenge is the size of the data, which is growing speedily. We have leveraged the semantic web technologies to address these problems at a large scale and designed a sequential mining approach called Sequential Mining on Shared Memory (SM2) using SPARQL Protocol and RDF Query Language on data represented as a Resource Description Framework (RDF) graph in a shared memory setting. We have also conducted an analysis of nation wide claims data to identify variations in treatment patterns for Autism, heart disease and Breast Cancer. 

As part of the future work we propose to design a parallel computing approach to mine sequential patterns for a distributed memory setting and do a comparative performance evaluation of this algorithm in shared and distributed memory framework. We also plan to incorporate our sequential mining approaches in a treatment recommendation model, which we are in the process of developing to give personalized treatment recommendations for clinicians to choose from for a particular patient.