Anomaly Detection with Sub-Extreme Values: Health Provider Billing

Abstract

Anomaly detection within the context of healthcare billing requires a method or algorithm which is flexible to the practicalities and requirements of manual case review, the volumes and associated effort of which can determine whether anomalous output is ultimately actioned or not. In this paper, we apply a modified version of a previously introduced anomaly detection algorithm to address this very issue by enacting refined targeting capability based on the identification of sub-extreme anomalies. By balancing the anomaly identification process with appropriate threshold setting tailored to each sample health provider discipline, it is shown that final candidate volumes are controlled with greater accuracy and sensitivity. A comparison with standard local outlier factor (LOF) scores is included for benchmark purposes.