Subject: health care, waste, abuse, fraud, machine learning, unsupervised learning
Year: 2020
Type: Proceeding article
Title: How Simple Predictive Analysis of Health Care Claims Data can Detect Fraud, Waste and Abuse Threats in Health Care Insurance - The Case Study of United Arab Emirates
Author: Kristijan Jankoski
Author: Kiril Milev
Author: Gjorgji Madjarov
Abstract: The usage of unethical practices which does not follow prescribed clinical standards and leads to the unnecessarily high expenditure for health care (waste, abuse and fraud) is increasing day by day in the Middle East countries. Reports show that about 30% of health care companies expenditures are based on a fraudulent medical claim. The rule-based approaches and expert systems that are used traditionally for tackling the health care waste, abuse and fraud (WAF) are very limited and require experts with extensive knowledge of medicine and expertise in the domain itself. The predictive analysis can be more flexible and less susceptible to some of the problems encountered with rules-based systems by focusing on the outcomes rather than the entire decision making process. In this paper, we present how simple predictive analysis and unsupervised learning on health care claims data can be used for detecting waste, abuse, and fraud threats in health care insurance in UAE. Our focus is to detect abnormal behavior of the clinicians from different specialties from different medical providers using the patterns made on the diagnosis and activity level prescription. The results obtained from the experiments performed on over 370K medical claims showed that only 0.007% of the clinicians caused potentially over 10% of the WAF marked claims. 27 clinicians marked with the analysis and scored as being most suspicious by the auditors made total of 4.929 claims.
Publisher: Ss. Cyril and Methodius University in Skopje, Faculty of Computer Science and Engineering, Republic of North Macedonia
Relation: CIIT 2020 full papers;42
Identifier: oai:repository.ukim.mk:20.500.12188/8265
Identifier: http://hdl.handle.net/20.500.12188/8265
