NLP Overview

" Documentation is not just about describing medical signs and findings, it also has to do with describing and coding measures and actions, as part of the payment model from the payers. These days, the need for better coding is universal and Capio Denmark is no exception. "

That’s why Capio Denmark decided to contact the Innovation hub for co-operation. One of the hub’s tasks is to monitor what’s going on in the innovation area and try to cross-fertilize ideas. A possible solution worth exploring, in this case, was the technology of Natural Language Processing (NLP), the recognition and handling of voice and text. Machines can translate spoken language into text, and then analyze the text and learn to see patterns, make suggestions according to them, and then learn again from the feedback they get.



From what we have learned so far, we saw a possibility to test this technology together with our Danish operations.


As we have seen from other use cases NLP can free time for the healthcare professionals, transferring it from administrative tasks to patient value time.

But there are at least two other benefits of using structured text as well!

1.      If the data entering the systems are of higher quality, the administrative quality will be higher. Diagnosis codes or measure codes affecting reimbursement and other measurements for follow-ups will be easier to handle, lowering the need for manual handling or extra work. The process of dictation gives a better outcome.


2.      Also, the medical quality can be improved. When data is structured, the use of machine learning gives conditions to find connections between data previously hidden among too vast and unstructured piles of information. There are examples of NLP finding overlooked diagnoses. As patients, we eagerly expect healthcare to jump into these possibilities.


By streamlining the dictation process, you can find time for physicians to meet patients and for medical secretaries to have time for a second opinion looks at the documented codes. If we could find a way to make 80 percent of the documentation flawless enough to read, code, and reimburse, it would definitely make an impact.


To get to that point, the NLP needs teaching and training: It does not know what to do from the start but learns from large amounts of data. Receiving feedback on its suggested codes, it gets a grip on how we want it to work. The more data is analyzed over time, the more solid and accurate models can be designed.