Mechanics of Autonomous Medical Coding

Parse trees, ontologies, and autonomous coding, oh-my!?Medical coding is very complex, and the advent of autonomous medical coding technology is hard for some to grasp.?There are 250,000 medical coders in the United States dedicated to reviewing medical charts and assigning the appropriate codes.?In becoming a medical coder, one must get the proper education, e.g., a certificate program, an associate’s degree, or a bachelor’s degree.?Those studying must learn the basics of body systems, treatment procedures, disease processes, and medical terminology, to name a few.?They also learn about billing procedures and regulatory compliance.??There is so much more medical coders must learn to have a successful career as their production and quality are always under a microscope. Medical coding is complicated!

An improperly coded medical record will impact a health organization’s revenue and could pose severe challenges if over coding is found.?It’s essential to remain up to date on annual coding changes and to follow standard coding guidelines. It is understandable to code patient records cautiously to mitigate the risk of over coding, but in doing so, you are possibly leaving compliant revenue on the table.?What if technology could offer inter-rater reliability to your medical coding?

Understandably, someone could have a hard time grasping what was thought to be practically impossible, autonomous medical coding.?To calm those reservations, we talked about the history of the technology and the challenges of using NLP for understanding language, specifically clinical language, in the previous two articles.?Now let's put it all together and show how technology can provide autonomous coding with inter-rater reliability at blisteringly fast speeds of <5 seconds per chart.??

Mechanics of the Nym Technology

The Nym CLU technology’s linguistic rule-based approach uses structured maps of human knowledge for identifying key elements in the text of a medical record. The engine reconstructs the clinical narrative of the patient chart by analyzing each textual component to understand linguistic intent and resolve context and word ambiguity. There are four main mechanics Nym works through to fully understand the patient chart and apply the appropriate ICD10, CPT, E&M levels, and all appropriate modifiers. Let's look at each step carefully.

Step 1: Analyze Sentence Structure?

The Nym engine analyzes each sentence in the patient chart, breaking it down into individual syntactic parts. A parse tree and a semantic representation are created, showing hierarchical relations to understand the structure and logic of the sentence. Nym identifies and flags any incomplete or ambiguous sentences, describing the specific reasons for its inability to linguistically understand the sentence. These charts are sent back to the healthcare facility for manual coding and review to determine areas for improved clinical documentation.??

Step 2: Reconstruct Medical Narrative?

The Nym engine combines meanings and medical logic with knowledge graphs (or ‘ontologies’) to create a logical explanation of the structure and the relationships between the sentences of the medical narrative.

References such as pronouns (e.g., I, my, he, his) and definite articles (e.g., the), as well as objective or subjective speech, tenses, or negated issues are identified and marked.

Step 3: Extract Insights

To understand the context of the narrative, the Nym engine searches through the medical narrative and extracts meaningful insights from the text about active speakers, locations, and anatomy by correlating the information to knowledge graphs. Derived insights are converted into a list; parts-of-speech are tagged as subjective or objective and marked as negated if needed; and assessed and classified according to the confidence level. For example, in the sentence: “The patient denies any previous trauma to the leg,” the Nym engine would detect the negated subjective speech.?

?Step 4: Assign Medical Codes

The engine reviews the insights collected about the narrative and maps them to the knowledge graph to identify relevant diagnoses and medical procedures in the patient chart. It then automatically applies the applicable codes based on coding guidelines.?

Outcomes and Coding Quality

What about coding quality? I am glad you asked, Nym will only code records it fully understands. It will offer inter-rater reliability for your coding. It will pass unclearly documented records to your coding staff for review. Moreover, all steps described above are documented in a transparent audit trail, tracing back the logic for why each medical code was selected. As mentioned earlier, the coding, validation, and audit results of the chart take place at an average of 2.5 seconds per chart. Autonomous medical coding is here, are you ready for a test drive?