The Apixio Blog

A Data Science Defense of ICD-10

This fall, healthcare organizations are going to have to start using a new naming convention for conditions patients have. The current ontology being used in the U.S. is the ninth revision of the International Classification of Diseases (ICD-9), and it’s 40 years old. There is a newer version of ICD codes which is around 20 years old (ICD-10) that we have yet to adopt, and even 20 years after being created it strikes fear in the hearts of the healthcare ecosystem. The important question is “why?”and the answer is only partly because the new code set is more extensive and complex than the old one.

The real source of provider anxiety is that ICD codes are tied to healthcare billing, and if physician offices don’t supply the correct codes, they might not get paid for the healthcare services they provide. Indeed, the angst about the new coding system is primarily a concern about breaking the billing system, which in a fee-for-service healthcare economy is a bad thing.

An even deeper understanding of this truth, is that if the coding system that describes what conditions patients have is being held hostage by billing concerns, what does that say about the healthcare software infrastructure at large? How beholden is it to “billing concerns”? The answer is “a lot.” In, The Digital Doctor, author Robert Wachter reflects on the impact of wholesale adoption of Electronic Medical Records (EMR) and makes the observation at one point that EMRs are in large part better billing mechanisms. This doesn’t make EMRs intrinsically bad, but when the bias for creating a system is to assure that it can safeguard billing information (which is not the most clinically relevant data), other ideals like creating better models of  the patient and improving quality outcomes are likely not being attended to.

The fact is, the new ICD-10 coding system will help us better achieve these goals. ICD-10’s complexity enables more supporting information about patients’ conditions, and in the current billing-centric healthcare data environment, this is a big deal. Every step taken to supply more information that is less noisy helps. The healthcare ecosystem should wholeheartedly support the adoption of richer coding ontologies, because each bit of better data helps make our understanding of patients that much richer.

Unfortunately, while ICD-10 is a small step in the right direction, it isn’t going to yield enough data to enable truly useful patient models. Plus, ICD-10 has been stalled for years, so we’re not likely to see another federal regulation to increase healthcare data anytime soon. Given these constraints, one might ask, how is progress every going to be made in analytics to add value back into the healthcare ecosystem?

Well, it turns out there is a valuable source of data right in front of us that has not been taken full advantage of: doctor’s clinical notes about the patients they see. As Wachter explains in, The Digital Doctor, doctors write text notes about their patients that provide a lot of rich information. These notes can be stitched to condition codes for billing and patient metrics/measures (labs) to produce a useful picture of patients and the care they receive. Creating these patient models involves a lot of heavy lifting and it takes time to develop infrastructures and models to acquire, manage and analyze raw patient records.

My team and I at Apixio have been on this mission to create useful patient models from available clinical documentation for the past four years. Our initial product text mines patient records to enable a clearer picture of the risk status of Medicare Advantage patients. It’s a starting point from which we hope to enable more personalized treatment and better quality outcomes, and a mission in which detailed data capture systems such as ICD-10 will play a valuable part.

© Apixio 2015