A few of my previous projects, have engendered for me a love-hate relationship with standardized medical vocabularies, particularly diagnosis codes. I was recently asked to provide some thoughts on diagnosis codes and how their input could be improved. Here were my answers.

Are vocabulary standards useful?

In general, the utility of vocabulary standards depends on your frame. Under prospective payment systems such as IPPS, it is helpful to characterize patient acuity for the purposes of assigning DRG codes which have further implications for a health system’s reimbursement. Without structured, universal codes it would not be possible to define the “product” that a health care system provided to a sick patient.

Similarly, structured codes have utility for non-billing reporting processes such as process improvement and public health. In theory, diagnosis codes can be followed over time or used as important patient strata for outcomes assessment. Unfortunately, their utility for these purposes are often undermined by their billing implications. “Gaming” can occur that may lead to artifacts in data such as a steady incline in the number of patients coded for a certain diagnosis code that happens to be well reimbursed.

From a clinical entry standpoint, structured concepts tend to offer little utility and frequently frustrate clinicians. Depending on the robustness of the chosen ontology, clinicians may have to find “workaround” diagnoses such as “Viral infection, NOS”. These NOS or “unspecified” diagnoses tend to be fairly common among ICD9 for example.

Possible UI improvements for fixed vocabulary

Clinicians generally have been spoiled by data entry user interface elements available in consumer products. These time-saving and error-reducing measures include

• auto-complete
• associations (ie people who chose X, also chose Y)
• “most commonly used” or “most recently used” picklists
• spell-check (ie “did you mean …”)
• automatic labelling (ie NLP / regular expression matching of free text for likely concepts)

I’d love to see EMRs move further away from the “diagnosis calculators” concept that involves multiple clicks and high cognitive burden and more towards these well-enjoyed UI elements.