Reducing the Risk of Upcoding in DRG Grouping Through a Two-Stage DRG Grouper Based on Machine Learning.

Area of Science:

• Health Economics
• Medical Informatics
• Machine Learning

Background:

• Diagnosis-Related Groups (DRG) implementation incentivizes hospitals to upcode for increased profitability.
• Existing DRG grouping methods are susceptible to manipulation, leading to inaccurate patient classification and financial gain.

Purpose of the Study:

• To propose and evaluate a novel two-stage DRG grouper (ML-DRG) designed to mitigate the risk of upcoding.
• To develop a machine learning model that predicts clinical resource consumption for a more robust patient classification.

Main Methods:

• Utilized a two-stage machine learning approach (ML-DRG) to predict patient resource consumption.
• Developed a resource consumption index based on comprehensive patient characteristics, making it difficult to alter.
• Compared ML-DRG performance against three mainstream DRG grouping methods using Chinese healthcare data (2011-2018).

Main Results:

• ML-DRG demonstrated superior performance compared to existing methods.
• Specific disease groups (intracranial hemorrhagic disease - BR1, respiratory infection/inflammation disease - ES2) were effectively divided into 4 DRGs with low coefficients of variation (<0.8).

Conclusions:

• The ML-DRG grouper shows significant potential in reducing healthcare upcoding by providing a more accurate and less manipulable patient classification system.
• Implementing ML-DRG can help hospitals avoid selecting incorrect DRG codes for inflated reimbursement rates, promoting fairer healthcare billing.