Advanced AI optimizes risk coding and slashes excess  spend

Project Summary

Client: One of the largest integrated health care delivery and financing systems in America.

Challenge: Manual review of medical records was cumbersome and drove up health plan administrative costs.

Solution: Deploy natural language processing to reduce manual review and coding.

Results: 35 more records reviewed each day, 52,000 records reviewed in first year, $1.7M in savings and revenue

Challenge

The client’s Revenue Program Management department validates and submits treatment reports to the Centers for Medicare and Medicaid Services (CMS) for provider reimbursement. To do this, RPM employees compare patients’ billing records with their medical histories in a process called retrospective risk coding. Done manually, this process was lengthy, complicated, and costly.

Revenue employees reviewed nearly 600,000 medical records – each with hundreds of pages – a year. This quickly became monotonous, leading to low employee engagement and high risk of error due to fatigue.

Further, medical histories and billing records are riddled with complex medical jargon and highly technical language. Even the most qualified revenue employees made mistakes navigating the intricate documents.

To guard against this high chance for human error, the client used two vendors to independently verify the reimbursement reports – this duplicative spending, of course, was expensive for the client.

The revenue team needed a new system of retrospective risk coding that increased efficiency, reduced risk of human error, and cut excess spend.

Solution

We worked with the department to develop a natural language processing program to read the records and search for high-risk conditions.   

Natural language processing combines linguistics and computer science to help machines understand, interact with, and communicate in human language.

Applied to retrospective risk coding, the client’s new natural language processing machines read through medical records and search for high-risk conditions.

Once the machines read the documents, they flag the high-risk conditions that need human attention for further validation.

Then, medical coders use a graphical user interface to efficiently validate the results. This interface uses highlighting, page numbering, and jump navigation for each suggestion to quickly direct the coders to only the conditions that need their attention.

More efficient and more user-friendly, the new natural language processing system optimized the retrospective risk procedure and reduced the likelihood of human fatigue and error. Now, human coders could confidently, quickly, and accurately submit treatment reports for revenue capture.

Takeaway

This innovative collaboration between the client’s revenue team and Lumevity highlights one of our fundamental beliefs: transformation is self-funding and self-propelling. As independent pieces of technology, the natural language processing machines are certainly exciting, but their real value lies in how they kickstarted the revenue team’s holistic evolution. The automated machines optimized risk coding, which freed employees from mundane labor. The freed employees, in turn, analyzed records with more enthusiasm and efficiency – and the cycle of activation and optimization spins on.

35 more records read daily

With the natural language processing machines handling the manual search for high-risk conditions, human coders analyzed twice as many charts per day than with the old system.

52,000 records reviewed

In total, this optimized system reviewed 52,000 records in its first year of use.

$1.7M in savings and revenue

With increased efficiency and accuracy, these new solutions led to prompt reimbursements from payors, eliminated the need for external vendors, and generated $1.7 million in savings and revenue in its first year of use.