4 Reasons NLP Should Work With Human Coders – Not Replace Them

Written By
Erik Simonsen
Chief Operating Officer

Natural Language Processing (NLP) promises to reduce coding to a fast and effortless process. Running a coding solution powered by artificial intelligence and machine learning, companies can now code charts at an unprecedented speed and with a greater accuracy – or at least that’s what many vendors would have you believe.


Unfortunately, the reality is that if NLP isn’t properly implemented and backed up by human coders, it can actually make the coding process slower and more error-prone. A decrease in output and increase in errors happens because NLP has four major drawbacks:

1. Limited assessment of M.E.A.T.

NLP tools generally have a limited ability to evaluate the M.E.A.T. of a code. Even those that promise to assess M.E.A.T. are often looking for a limited number of medicines or procedures that are associated with a disease as opposed to definitively assessing if a condition can be coded.  As a result, the number of false positive codes often exceeds the number of codeable ones, requiring human coders to review and correct the NLP results.

2. Little ability to adhere to in-house coding guidelines

NLP engines lack the flexibility to be tuned to capture all the nuances of most disease-specific coding guidelines. If customization is required or coding needs to adhere standards that aren’t already programmed in to the NLP solution, codes will be missed or incorrectly added, and charts can fail to meet internal regulations.

3. Only works with high quality images

Over 20% of charts have poor fax quality or have handwritten notes, both of which are extremely difficult for NLP to process accurately. While NLP solutions will pick up some diagnoses from these charts, they will always require a manual audit to make sure that illegible codes are accounted for.

4. Potential compliance risk

NLP-based coding solutions often audit for new codes, but will not check to see if there are any incorrectly coded diagnoses in claims. That presents a huge regulatory issue if you’re not blind coding each chart to look for non-compliant provider coded diagnoses.


How to properly implement NLP

NLP does have several major drawbacks, but it can make the coding process significantly easier when integrated with human coders. At Episource, we believe that NLP is a powerful tool to augment the coders’ work – not to replace it. Only a skilled coder can meet in-house guidelines and CMS compliance requirements, interpret poor quality images, and confidently hunt down all the M.E.A.T. in a chart. However, NLP can speed up that manual process by finding codes faster within charts, speeding up one of the slowest parts of manual coding and reducing coder fatigue.

Will NLP result in 4x faster coding and greater accuracy? Highly doubtful, unless you’re willing to deal with costly audits and pay for a team to review the results to ensure compliance. Can it be integrated into coding workflows to make already skilled coders more efficient? Absolutely.

To learn more about how Episource takes the best of NLP and combines it with the industry’s best coders, contact us today.


We simplify the management of member programs by making it radically more efficient, and increasing value to healthcare organizations and their members

Presentation @ RISE VBC: Effective Collaboration Strategies Between Payer & Provider