'Content Neutrality' and Why It Is Important for EHR

by Daniel Essin | April 05, 2012 2:44 AM EDT

In response to Udaya Padakandia

The short answer is yes, I am most likely referring to the same concept.

One can imagine a spectrum of approaches. One end is represented by a data model that specifies each data "field"and how the fields are related or organized. The other end is represented by free-text narrative such as one might dictate. Between the extremes lie a number of hybrid approaches, some combining pre-defined fields and free-text, others imposing some structure on the narrative for the purpose of segregating the material into discrete elements, each with two personalities: data field and narrative.

The preferred location along the spectrum is somewhere in the middle for two reasons. First, theory tells us that a mixed-strategy is always best. Second, the extremes suffer from virtually insurmountable obstacles. The essentially infinite complexity of medical data means that it is never possible to "complete" a data model. What is available for use at any moment always lacks something important that interferes with achieving the desired goal and which frustrates user. Natural language processing, the holy grail of those who favor free-text also has what I consider to be a fatal flaw. It lacks sufficient sensitivity and specificity for me to be willing to trust the results.

A record based entirely on data fields would represent that maximum in content specificity while a free-text record would represents the maximum in content neutrality. The goal is to choose a mid-point on the spectrum that provides as much content-neutrality as possible without giving up the guideposts necessary to allow computer programs to locate and extract information reliably and efficiently. An obvious solution is to incorporate some sort of tagging into the body of each document.

Since there are many possible tagging schemes. The challenge, and perhaps the art, lies in what choice is made. The main feature which distinguishes different approaches involves whether content-specific choices are made early or late. Earl choices are usually made by a standard-setting group, before actual applications are developed and actual documents are created and used by practitioners. It appears complete and erudite but lacks the benefit of experience.

If the choice of how specific content will be incorporated can be deferred until later, then the only choice that needs to be made early is to select a simple tagging scheme that demarcates the document's internal organization, for example sections that may contain sections or items, and items which can contain narrative and other annotations indicating various qualitative or quantitative modifiers that have been attached to the item to add meaning (metadata). The other decision that can be made early or late is which coding schemes, lexicons and nomenclatures will be attached to an item as metadata and whether there is any restriction on the source of such metadata.

The notion of using tagged documents, sometimes called structured documents, as a mechanism for representing medical record content is now pretty well established. This approach is embodied in the Clinical Document Architecture (CDA) standard, of which I am an original co-author. It has been in existence for almost 15 years now and I find two things interesting. First, it is viewed primarily as a vehicle for exchanging records between facilities. To my knowledge, there has only been one EHR product that used this approach as the primary storage form of its medical records. Second, although this standard is supposed to enable the long sought after interoperability, and it has now gone through 2 major revisions, lack of interoperability is still bemoaned. My conclusion is that the CDA standard is not being used much, or if it is, people must have meant something else by the word "interoperability".

I hope this discussion creates more clarity than confusion for you.

by Udaya Padakandla | March 27, 2012 9:31 PM EDT

Daniel

Very interesting article indeed. I remember seeing this discussion in HMT or HCI or some such IT journal about the contest between designing EMRs on purely pre formatted data fields, versus allowing physicians to create a free text flow based record (from which a search engine would later extract the necessary "standardized" terminology). Structured versus free text-flow, in other words. I am still trying to fathom the depth of your article , and it makes me wonder if you happen to be referring to the same concept. Thanks.