AI-Driven DRG Validation in Healthcare RCM: Challenges, Solutions, and the Path Forward
Loading...
Files
Authors
Reeves, Tiffany
Issue Date
2026-05
Type
Dissertation
Language
en
Keywords
Healthcare Innovation & Delivery , Healthcare Revenue Cycle , DRG Validation , Artificial Intelligence
Alternative Title
Abstract
Artificial intelligence-enabled natural language processing tools are increasingly used in diagnosis-related group validation workflows to improve coding accuracy and reduce revenue leakage. However, there is no clear evidence of the types of errors these systems generate and the organizational factors that influence their performance. The purpose of this applied doctoral project was to understand the nature, frequency, and operational impact of errors generated by artificial intelligence-driven diagnosis-related group validation tools and how human and organizational factors influence their identification and management. The project was guided by the Human Artificial Intelligence Integration Framework. A qualitative descriptive design was employed, combining structured artifact analysis with thematic analysis of stakeholder interviews. A structured audit of 40 synthetic inpatient cases was conducted to assess differences between artificial intelligence-generated recommendations and manual coding validation. Root cause analysis was performed for different types of errors, and the operational and financial impacts were assessed. Ten semi-structured interviews were conducted with stakeholders in coding, clinical documentation improvement, auditing, and revenue cycle leadership. The audit results showed that discrepancies occurred most often in clinically complex cases, particularly respiratory and sepsis-related cases. False positives were detected in 50% of cases and were the most common discrepancy types. Thematic analysis highlighted common elements: the need for human validation, limitations in contextual interpretation, documentation inconsistencies, workflow burden, and the importance of feedback loops. Artificial intelligence tools can help with case identification, but human review is needed to ensure correctness and compliance. We suggest enhancing human validation processes, standardizing the documentation, making the AI logic transparent, and developing feedback loops. As such, we provide practical guidance for integrating artificial intelligence into revenue cycle workflows whilst maintaining coding accuracy, operational efficiency, and compliance.