Clinical coding is complex—the ICD-10 classification system contains around 68,000 diagnosis codes, and the CPT coding system contains over 10,000. Medical coders can code around 60 cases per day depending on project parameters with varying accuracy.
How do we enable coders to improve output and accuracy while reducing needless sifting and sorting? The answer is natural language processing (NLP).
NLP is a subgroup of artificial intelligence (AI) that understands language and can make associations between clinical evidence and medical codes.
Using this technology alongside medical coders dramatically improves accuracy and completeness to maintain minimum industry standards of 95%. It also supports an increase in output for medical coders to an additional 20-30% production improvement.
Health plans can set NLP programs with clinical rule sets, business rules, and CMS regulatory rules to align with the organization’s risk adjustment strategy.
Let’s review the process from start to finish—from chart retrieval to final CMS submission.
The Benefits of First Pass NLP
The process begins with the NLP program conducting a ‘first pass’ scan of a member’s chart to pre-identify which charts contain risk-adjustable HCC diagnosis codes. This lays the foundation for the retrospective chart review.
Identify open gaps
HCCs that may have been missed in the past can be discovered through suspecting logic and running the chart through multiple passes of the NLP engine to find historical gaps and accurately reflect the member’s burden of illness.
The value of medical coders
The human element cannot be replaced. NLP programs have limitations in distinguishing between nuances, double meanings, conditions with shared acronyms, and confusing indicators of certain conditions. Specific acronyms representing two different diagnoses can be flagged by NLP but require a human coder for final determination. NLP often errs with instances of cancer without active treatment, telehealth visits that do not contain evidence of both audio and video components, and conflicting documentation indicative of diabetes in a pre-diabetic diagnosis.
As new code sets are continually implemented and updated, human coders are necessary for critical thinking and higher-order functions such as final determinations, selecting the best evidence, and providing constant feedback to refine the NLP system.
It’s time for a Second Pass
In some cases, the chart is run a second time through the NLP technology to identify missed or mismatched codes. Finally, the medical chart undergoes a final manual review with a medical coder.
Using NLP alongside seasoned medical coders significantly enhances the speed, efficiency, and accuracy of the medical chart review process.
Are you looking for a risk adjustment program that integrates the latest technology with expert knowledge and insights?
References
¹ Venkatesh, K. P., Raza, M. M., & Kvedar, J. C. (2023). Automating the overburdened clinical coding system: challenges and next steps. Npj Digital Medicine, 6(1). https://doi.org/10.1038/s41746-023-00768-0
