Learning to Extract Actionable Evidence from Medical Insurance Claims Data

Abstract

This chapter demonstrates a few examples of applying this methodology to enable purposive analysis of medical insurance claim data. It motivates a wider adoption of advanced machine learning–based analytics to this and other similar types of readily available and currently underutilized data. The first case demonstrates how anomaly detection can be used to detect break points in billing patterns that may reflect systematic inefficiencies. The second case develops metrics to measure emergency room utilization by the chronically ill and exposes differences of efficiency of managing such patients between public and private insurers. The third case uses adaptive supervised classification to predict undocumented conditions from the history of a patient's medical claims. Finally, the last case illustrates how sequential rule learning algorithm can help identify patients suffering from rare and hard-to-diagnose diseases such as Gaucher disease by revealing characteristic patterns in their medical claims histories.