International Classification of Diseases Hampers the Use of Analytics to Improve Health Care


andy oramThe health care field is in the grip of a standard that drains resources while infusing little back in return. Stuck in a paradigm that was defined in 1893 and never revised with regard for the promise offered by modern information processing, ICD symbolizes many of the fetters that keep the health industries from acting more intelligently and efficiently.

We are not going to escape the morass of ICD any time soon. As the “I” indicates in the title, the standard is an international one and the pace of change moves too slowly to be clocked.

In a period when hospitals are gasping to keep their heads above the surface of the water and need to invest in such improvements as analytics and standardized data exchange, the government has weighed them down with costs reaching hundreds of thousands of dollars, even millions just to upgrade from version 9 to 10 of ICD. An absurd appeal to Congress pushed the deadline back another year, penalizing the many institutions that had faithfully made the investment. But the problems of ICD will not be fixed by version 10, nor by version 11–they are fundamental to the committee’s disregard for the information needs of health institutions.

Disease is a multi-faceted and somewhat subjective topic. Among the aspects the health care providers must consider are these:

  • Disease may take years to pin down. At each visit, a person may be entering the doctor’s office with multiple competing diagnoses. Furthermore, each encounter may shift the balance of probability toward some diagnoses and away from others.
  • Disease evolves, sometimes in predictable ways. For instance, Parkinson’s and multiple sclerosis lead to various motor and speech problems that change over the decades.
  • Diseases are interrelated. For instance, obesity may be a factor in such different complaints as Type 2 diabetes and knee pain.

All these things have subtle impacts on treatment and–in the pay-for-value systems we are trying to institute in health care–should affect reimbursements. For instance, if we could run a program that tracked the shifting and coalescing interpretations that eventually lead to a patient’s definitive diagnosis, we might make the process take place much faster for future patients. But all a doctor can do currently is list conditions in a form such as:

E66.0 – Obesity due to excess calories

E11 – Type 2 diabetes mellitus

M25.562 – Pain in left knee

The tragedy is that today’s data analytics allow so much more sophistication in representing the ins and outs of disease.Take the issues of interrelations, for instance.

These are easy to visualize as graphs, a subject I covered recently.

Figure 1 shows how a patient’s obesity contributes to Type 2 diabetes and knee pain. There are many ways to store this information in ways that a computer program can retrieve and make sense of, including a standard called RDF that is widely used on the Web.

Screen Shot 2014-08-05 at 8.44.07 AM

Figure 1. Illustrating contributory factors

In contrast, ICD obscures relationships. The standards do represent a small subset of such relationships; for instance Type 2 Diabetes has 56 ICD-10 codes that list commonly associated conditions. The paucity and gawkiness of such efforts (for instance, how do you represent multiple complications?) just underlines how desperately the health care field needs a different approach.

Wrong-side errors (such as operating on the left side when the problem is on the right) are shockingly common, but it’s still laughable to define separate medical codes for the left and right side, instead of coding left/right as a separate dimension that can be represented in the graph.

Finally, given that the doctor will select the diagnosis that gets the highest reimbursement rather than the diagnosis that best represents the patient’s condition, one can say good-bye to any analytical benefits that supposedly come from the proliferation of ICD codes. The bias of the system toward billing instead of treatment is revealed by the definition of separate codes for the initial encounter and subsequent encounters.

Researchers and analytics firms, I’m confident, will devise standards for representing disease in all its complexity. Health care institutions, eager to cut costs and find the right treatments faster, will use the new systems to track and analyze disease. It’s sad that we’ll be forcing doctors to use at least two parallel diagnostic systems–one tied to the practice of medicine in 1893 and another appropriate for 21st-century data processing.

3 replies »

  1. Creating ICD10 claims for testing can be easy with i10xCodr.

    i10xCodr uses a complex mapping algorithm on top of GEMs to convert ICD9 to equivalent ICD10 claim. It uses concepts like Laterality, Stage, Episode of Care, Combination codes, External cause of injury reason, External cause of injury place of occurrence, Anatomical sites and more.

  2. Kudos. You have an accurate feel for our classification dilemmas. Billing needs are incongruent with studying-disease needs. If you look at one of the proteomics diagrams that Science mag is famous for, you will find that there is no way to do classification at this stage of our science. The combinations and permutations of co-morbidities are too vast: the unbelievable complexity in medicine can’t be classified…yet.

    Futility, I guess, is the proper conclusion.

  3. “We are not going to escape the morass of ICD any time soon.”

    The morass of the ACA, the morass of MU and EHR, now the morass of
    ICD-10. Things don’t look so good right now.