Using Symbolic Data Analysis to Detect Fraud, Waste, and Abuse in Healthcare Insurance Claims Data

Abstract

As the nation’s cost of healthcare continues to escalate, so does the exposure to instances of fraud, waste, and abuse. Multiple approaches to detecting these occurrences are in practice today and there are multiple organizations, public and private, which are focused on their identification and reduction. This dissertation investigates symbolic data analysis (SDA) and its applicability to detecting anomalistic behavior. SDA is a growing field of study and has implications far beyond what this dissertation will cover. The core concepts of SDA provide a foundation from which to develop an alternative approach to analyzing healthcare insurance claims data for the presence of anomalistic events. The research introduces a symbolic method that investigates data at a higher concept level as opposed to the traditional line level at which most analyses are performed. Simulated datasets and real-world inspired datasets are studied and results between symbolic and centroidal approaches are compared. Results suggest that symbolic anomaly detection techniques perform equally as well as their classical centroidal counterparts when only changes in mean distinguish one set of data from another. When changes are more subtle, particularly when means are equal but the underlying shapes of the distributions are different, the symbolic approach excels. Using the foundational principles of SDA, this dissertation introduces a novel technique to anomaly detection and provides an alternative way of analyzing healthcare insurance claims data for fraud, waste, and abuse.