A medical domain collaborative anomaly detection framework for identifying medical identity theft

Medical identity theft is a serious problem in healthcare systems around the world, especially the US healthcare system. In addition to financial losses to the patients, healthcare providers, and insurance providers, it has devastating effects on the patients healthcare services. Current mechanisms to detect medical identity theft rely on weak detection mechanisms like human operators, network anomalies, or audit. These methods have very low success rates and only identify medical identity theft cases after the fact which does little to actually stop them. In this paper, we present a novel framework that uses sophisticated anomaly detectors using disease ontologies from the medical domain, institutional anomalies, and network anomalies to detect medical identity theft. These various detectors reside with different healthcare entities and collaborate to merge their outputs into a single inference engine to determine suspected cases of medical identity theft with high efficiency. We argue that this method is very effective and cannot be circumvented by medical identity thieves using regular means because the framework leverages inherent relationship between diseases to ensure that a new request for medical care is aligned with the patients medical history. The reasoning engine decides this based on the relationships between the diseases and reasoning whether the new request is an anomaly or not. We develop the system architecture of the framework in a cloud based setting. We implement this framework as a prototype system and evaluate it analytically based on real life use cases and experimentally using performance experiments.