ICD-10 and the problem of interobserver variability

I first heard about the ICD-10 when I was working at a small start-up, trying to develop an EMR for a string of dialysis clinics. It was always spoken of with a certain gravity, like the ominous visit from an aunt that nobody in the family likes, but feels obligated to see.  Practical (read: business) people hate ICD-10. It’s giant and unwieldy. Doctors think it’ll be an excuse to bilk them out of payments.  They dread the day that they get a “false coding” note for a visit for a broken arm because they didn’t specify the patient fell of their bicycle or down a flight of stairs.  So who’s driving this?

I can only assume that it’s research.  ICD-10 must be an epidemiologist’s dream. Want to prove something inane, like the fact that waterskiing accidents are more common in the summer?  ICD-10 is your tool.  If you can collate all the insurance billing from the entire country, you can begin to pull out these vanishingly rare instances and analyze them.

Admittedly, as this article points out, some of the events the ICD-10 tries to capture are so vanishingly rare that they actually, well, vanish.  They’re literally unheard of or actually impossible. But what about some of the other widely panned codes, like falling off a chicken coop?  Theoretically, we could begin to perform real time monitoring of safety conditions in all kind of industries.  These events are rare, which means that if we see a cluster of them occurring in a particular geographic area, an investigation might be warranted. Maybe building inspectors aren’t performing their inspections. Maybe a certain company isn’t enforcing proper safety standards. Again, theoretically, the giant index of ICD-10 codes could drive meaningful data collect and interventions.

The problem is the observer. Interobserver variability is a problem in all sorts of medical fields, from reading chest x-rays to interpreting physical exam findings. For the ICD-10 to be useful for research, you need to code these rare events correctly.  And, with the endless array of options, the chances of this happening seem, to me, to be vanishingly small.  Maybe there’s a good technical solution to this, where an EMR scans the history of present illness and offers a variety of appropriate billing codes (writing “chicken coop” should be a dead giveaway).

The validation and implementation of this for all 155,000 codes is, however, a monumental task at best.  Such an undertaking can (and should) be done by those who created the codes in the first place. Unfortunately, something tells me they can be less than thorough.

Michael Slade is a medical student who blogs at Inside Looking In: Healthcare and the Transformation of Self.

View 36 Comments >

Most Popular