Coders and Computers, People and Machines

September 30th, 2013 / By Rhonda Butler

The bi-annual public meeting at CMS headquarters of the Coordination and Maintenance Committee is where interested parties come to propose changes to the U.S. version of the International Classification of Disease. Until the most recent meeting on September 18 and 19 this meant ICD-9, —now all proposals are exclusively for future updates of ICD-10-CM/PCS.

Discussion during these meetings centers on whether a proposed change would be useful to coders, whether additional instruction needs to be added for coders, and whether a distinction would be clear or confusing to coders. The reason for this is obvious: If new codes or coding instruction (which includes index entries, instructional notes in the codebook, and the official coding guidelines themselves) are incorrect or inconsistent, the resulting coded data is useless.

As I listened I was struck by how different the results could be, depending on the answer to this fundamental question: What is a coder? Is a coder a human being who uses his or her own eyes to review the medical record, look up the correct codes, and assign them directly? Or is a “coder” sophisticated software that sifts through all the information in a provider’s EHR and suggests correct codes to a human gatekeeper at the tail end of the process? Both answers are correct: they represent two ends of a fully populated spectrum. A coder can be a person aided by software in varying degrees, or a coder can be software aided by a person in varying degrees.

If computerized coding systems can be said to have a cognitive style, it would be the polar opposite of the cognitive style of human coders. That is why it is so difficult to create classification systems and coding instructions that satisfy the cognitive styles of both humans and software. A classic example of this dilemma surfaced in the September 18 meeting. ICD-10-PCS has a “device key” which cross-references brand name devices with correct device values in ICD-10-PCS codes. Vascular stent manufacturers are apparently eager to have all their stents listed by brand name, even though the vast majority of the brands are assigned to the same PCS device value. So a list of previously unlisted vascular stent brand names was proposed for the device key.

Some people at the meeting felt we would be making the device key less useful for coders by adding more and more entries, and others wanted to make the device key as complete as possible. For human coders, including all stents by brand name can be overwhelming, especially if a person is looking up information in a paper book, or doesn’t have the habit of using the search function in a PDF file or e-book. For software, the more the merrier –the machine can do its job better. A human coder is easily overwhelmed by too much information, but these days computers can’t really relate to the concept of “too much information.” What is that?

The transition of a job function from person-centered to machine-centered is not new. In fact, the very word “computer” originally referred to a person and not a machine—a woman, typically, in a room full of women, trained to take information stored in one form (WWII encrypted messages) and turn it into another form (readable messages).

The word “coder” in the context of health information still refers to a person, not a machine—172,000 of them by some measure. For some of them, the job is painfully similar to WWII military intelligence operations—paper-based, manual, tedious, prone to error. Others, who are the beneficiaries of the latest and best that machine learning, natural language processing, and a soup to nuts electronic health record have to offer, see themselves as coding auditors, leaving the exhaustive (and exhausted) review of documents to the machine.

Someday using the word “coder” to refer to a person instead of a machine may seem as quaint as “computer”—just a story that we tell about our past. But for now we struggle with the dilemma of trying to satisfy the information processing needs of both human and machine coders.

Rhonda Butler is a Senior Clinical Research Analyst with 3M Health Information Systems.