How big a chunk of information do we want to exchange?

John Halamka and Wes Rishel have a very interesting approach to the recent debates on just what it is we might be exchanging in Health Information Exchanges. The problem arises from a tension among the need to care for the individual patient in front of you right now, to put that patient's care in the context of their history over time, and to take a broader view of a population of patients.

The provider with an individual case in the here and now traditionally generates a progress note - a document with whatever he or she thinks is required for the purposes of care. Much of professional education is transmitting the law and lore of what must/may/mustn't go into these documents and how to organize them for their various audiences. In general, they are free text, minimally organized with perhaps a few headings (Chief Complaint, Past Medical History, Assessment, etc.) and highly variable both across and within practitioners. (Some ten years ago, when Medicare mandated the presence of a Review of Systems for most office and hospital visits, the push-back from providers was huge and some actually left the profession rather than comply with this minimal nod towards standardization.)

Caring for patients in context requires the assembly of data across time and space and is the first place that electronic health records really shine. Knowing that the patient has an abnormal cardiogram now is valuable. Knowing that it is unchanged from one taken a year ago by another physician is gold. Sharing  documents across settings and providers accomplishes much of this, although it can certainly be improved upon with a little more granular organization.

At the population level, however, unstructured documents are not all that useful. Searching for the incidence of diabetic foot examination in our institution required natural language processing of many thousands of documents and failed to produce a reliable estimate [Kost, Chen, Littenberg et al. AMIA 2009]. Rather, we need standardized components of documents that can be extracted, transmitted, understood both as stand-alone entities and in the context they arose from, and analyzed for population as well as individual purposes.

Clinical laboratory data have long had these characteristics. A more-or-less complete documentation of a clinical chemistry test will consist of patient identifiers, specimen source (blood, urine, etc.), date and time of acquisition, analytic method, result, and units. However,  even these simple concepts are represented by a bewildering array of formats and definitions. To make matters worse, clinical labs routinely change their analytic methods, units, test names and any of these details, sometimes without even warning the users! A huge amount of effort goes into maintaining the clinical laboratory interfaces at Vermedx to ensure that these data continue to be useful over time.

More complex and less structured data such as pathology reports, operative notes, and discharge summaries are going to require a very large amount of work to make them transmissible and useful at this level. The first steps seem to be to agree upon what the smaller, irreducible units of information should be. This will undoubtedly be the battlefield of yet another IS standards war, but the outcome will have very big implications for how, and even whether, we can exchange health information into the future.