DRG grouping by machine learning: from expert-oriented to data-based method.

Area of Science:

• Health economics
• Machine learning in healthcare
• Healthcare policy

Background:

• Diagnosis-related groups (DRGs) are crucial for healthcare reform in China, aiming to control rising costs.
• Expert-oriented DRG grouping presents challenges, including opacity ('black box' problem), potential for upcoding, and high implementation costs.

Purpose of the Study:

• To propose and evaluate a data-based DRG grouping method utilizing machine learning algorithms.
• To enhance the transparency and reduce the cost associated with DRG system design and updates.

Main Methods:

• Developed a data-based grouping approach trained on real or simulated patient cases.
• Employed machine learning algorithms, assessing five typical classification methods.
• Utilized kappa coefficients to evaluate grouping performance via tenfold cross-validation.

Main Results:

• Data-based grouping demonstrated comparable classification performance to expert-oriented methods with appropriate algorithm selection.
• Models trained on simulated cases showed lower accuracy on real cases but achieved kappa coefficients > 0.6.
• Updating the DRG system with data-based grouping significantly improved average kappa coefficients from 0.1534 to 0.6435.

Conclusions:

• Data-based grouping offers a transparent and cost-effective alternative for DRG systems.
• Machine learning-driven DRG grouping meets system requirements and facilitates efficient updates.
• This approach presents a viable new option for optimizing healthcare payment systems.