Wednesday, July 11, 2012

Computer Assisted Coding

In the past, I've highlighted candidates such as "analytics", "HIE", and "gamification" for the hottest technology concepts of the year, the "Plastics" of 2012.   Recently, I've seen a new strong contender - "Computer Assisted Coding"

With ICD10 looming on the horizon, companies such as m-Modal, Dolbey, 3M, and Optum are offering applications that process the structured and unstructured data associated with an inpatient hospitalization or outpatient encounter into suggested ICD9 or ICD10 codes.

Using linguists, informaticians, natural language processing experts, and proprietary algorithms, each company promises to increase the efficiency of coders, provide a audit trail of the logic used to code each case (very useful if CMS/RAC asks for justification), and more accurately code case complexity.    Better documentation with accurate coding may even lead to reimbursement increases because the severity of illness of the patient and the nature of the treatment rendered is more completely described.

We're speaking with each of the major vendors of computer assisted coding products to understand their interface requirements, the nature of the clinical data they require as inputs, and their integration into workflow.

Workflow is a tricky question.

Suppose that a patient visits an emergency department after a finger injury.   Accurate ICD-10 coding requires laterality (left or right), open or closed fracture, simple or compound etc.     If the provider dictates a note that contains the text "fracture of the index finger at the PIP joint" there may not be enough detail to accurately code the injury.   Some computer assisted coding products intervene at the documentation point instructing the clinician what is needed to minimally specify the patient's condition or procedure.   Others ingest all the inputs from documentation created by caregivers and recommend possible codes.   Getting the data in right to begin with generates more accurate codes, but some clinicians will be fatigued by the alerts that prompt them during the documentation process.

I've studied some of the interoperability required to connect EHRs to Computer Assisted Coding products.   Some ingest print dumps or PDFs of text documents.  Others require HL7 2.x messages (ORU messages for structured data, MDM messages from unstructured data).   None yet accept the CCD or Consolidated CDA, although the Meaningful Use 2014 edition will require that EHRs export clinical summaries using CDA standards, not HL7 2.x

ICD-10 has spooked the industry with tales of 50% loss in coder productivity.   Computer Assisted Coding may just be the silver bullet.

More to come as we pilot it.


Paul G. said...

Hi John, Be sure to check out HealthFidelity's product. ( They're the primary licensor of MedLEE, the highly regarded NLP system from Columbia University.

drscarlat said...

IMHO the reason ICD-10 has so many codes, much more than ICD-9 and more than actually needed is that its data structure is not normalized:
The clinical concept of Ulnar Fracture is the same whether it is the left or right ulna. The laterality of the fracture should NOT be part of the unique identifier of the fracture. But unfortunately in ICD 10 it is: S52.601 is a rt. ulnar fracture while s52.602 is a lt. ulnar fracture.
It should have been the same code and the laterality should be related to it.
The above increases the number of codes, causes redundancy, inconsistencies and other data anomalies.

Anand Shroff said...


Good points. Running "right ulnar fracture" in an H&P through our product, REVEAL, we find the following things:
1. The problem is discretely identified as "fracture".
2. The body location is discretely identified as "ulna".
3. The body side is discretely identified as "right".
4. The ICD-9 codes are identified as 813.82 - "Closed fracture of right ulna" and 813.92 - "Open fracture of right ulna".
5. There are several hits on ICD-10, including S52.291A, S52.201A etc.
6. SNOMED returns the cleanest matches - 54556006 - "Fracture of ulna", 263204007 - "Fracture of shaft of ulna" etc. without reference to the body side.

Perhaps this is one of the reasons for the AMA pushing to go to ICD-11 (which is based on SNOMED) bypassing ICD-10 entirely.