Models for Decision Support in Healthcare

Abstract

One of the many challenges in the field of medicine is to make the best decisions about optimal treatment plans for patients. Medical practitioners often have differing opinions about the best treatment among multiple available options. While standard protocols are in place for the first and second lines of treatment for most diseases, a lot of variation exists in the treatment plans subsequently chosen. We propose to extensively study recommended treatment guidelines and plans for selected rare and chronic diseases. As representative diseases we study Glioblastoma Multiforme (brain cancer) which is classified as a rare disease, and Diabetes Mellitus, which is a nationally and globally widespread chronic disease. A graph model is designed to capture the data pertaining to the treatment options and historical evidence and further analyzed to discover sequential treatment patterns based on different outcome classes based on longevity, complications etc. The notion of `Patient Similarity' would be explored to form cohorts of clinically similar patients. The treatment patterns would be ranked, and highly ranked patterns would be ordered depending on expected outcomes before being assigned to cohorts of patients. A prototype decision support system is planned for recommending treatment options based on a patients clinical profile. Evaluation of the models involves using historical data with various evaluation metrics and also by a qualitative assessment by expert physicians.