The client is a large multinational that provides healthcare solutions in the United States. Healthcare claims are processed in highly complex 837 and 835 EDI Files. The client wanted to build a solution that would aid them in reducing the costs associated with healthcare claim processing. Each claim that is sent is vetted preliminarily and is assessed whether it would require rework/resubmission. The client wanted to reduce this substantial cost associated with the preliminary vetting of claim files. Our idea was to utilize machine learning to develop a predictive model to identify the denial likelihood of healthcare claims.

• Parse highly complex hierarchical 837/835 EDI Files.
• Develop an ML Model to predict the denial probability of healthcare claims.
• Create an end-to-end API for production use.
• Formulate a workflow to identify denials in cases where one claim is associated with more than one remittance.

• With the use of Spark SQL’s fast and optimized functions, the complex hierarchical data was rolled up at the claim line level.
• Based on historical claims, a suitable flow chart was devised to differentiate between multiple responses in the case of COB (Coordination of Benefits).
• Several historical features associated with the Patient’s ICD Codes, CPT Codes and denial patterns of payors were engineered to capture the context of the whole denial management process.
• A machine learning model was developed to accurately predict the denial probability of 837 files.
• A complex neural network architecture was created to accurately predict the high-level denial reasons as well.
• The whole end to end process was packaged into a single API for automated production use.

The client was successfully able to implement the API and reduce the mechanical manual work in the claim management process.