Graphical Association Analysis for Identifying Variation in Provider Claims for Joint Replacement Surgery

Area of Science:

• Health economics
• Medical data analysis
• Machine learning applications

Background:

• Detecting fraudulent or wasteful medical insurance claims is challenging due to data volume and manual effort.
• Variations in provider billing behavior can indicate potential issues.

Purpose of the Study:

• To apply unsupervised machine learning for identifying and ranking unusual medical provider claiming behaviors.
• To develop interpretable models for analyzing claims related to unilateral joint replacement surgery.

Main Methods:

• Utilized unsupervised machine learning on Australian Medicare Benefits Schedule data.
• Constructed reference models for surgical procedures and compared individual provider claims.
• Ranked providers based on additional fees beyond typical claims.

Main Results:

• Identified providers with significantly unusual claiming patterns.
• Outlying providers' claims could increase procedure costs by up to 192%.
• The developed method is efficient and generalizable.

Conclusions:

• Unsupervised machine learning provides an effective and interpretable method for detecting unusual medical billing.
• This approach can be integrated into existing workflows to combat healthcare waste.
• The model's generalizability allows application to various medical procedures.