There is a movement afoot in the health IT world to standardize as much as possible. I recently heard a complaint from a research IT developer about the ongoing use of free text entry of data like patient history and pathology annotations, which is still very common in medicine. The use of standardized encodings makes possible machine-machine communication with semantic accuracy, and also is the key to AI-like applications that could allow clinical informatics to achieve the kind of quantum leap in computability that has recently occurred in the so-called bench sicences like genomics and proteomics.
This is a compelling argument. It is also wrong. Worse than that, it's dangerous.
I say this as a programmer who studied expert systems in the 1980's, helped build them in the early '90's, and has worked to embed intelligence in every system I could ever since. I am currently actively working on the information architecture aspects of a system that depends heavily on SNOMED and LOINC, have actually contributed dontent to LOINC, and have proposed a poster for AMIA 2006 that addresses the topic of how to convert complex natural-language expressions into HL7v3 Observations elements using LOINC and SNOMED as well as the semantics embodied in the HL7 information model.
I am not against standard terminologies, taxonomies and ontologies. What I believe to be wrong and dangerous about the argument is the idea that free text per se is bad.
I wrote most of this on my flight back to Detroit from San Francisco where I attended the Second Annual Velos eResearch User Conference. eResearch is the clinical research data management system we have deployed, but we have an IRB workflow management system from Click Commerce which was given the name eResearch by our IRB folks, so we call the product Velos. For a variety of implausible reasons, given the wealth of SF and its proximity to Silicon Valley, I had very little access to the Internet during my stay. Hilton charges for Internet access (not that it's my money, it's the principle of the thing), as do Starbucks' T-Mobile access points (objecting on the basis that this is both my money and the principle of free access is being violated). I'm not sure I totally agree with Richard Stallman's idea that "software should be free, like air", but I believe the Web is in the category of things that should be "free, like air". Apologies to Stallman if I am misquoting him, but - I had no Internet access up there, so I can't use Google to find the original!
The diatribe against free text came from a very intelligent person whose identity is unknown to me, because due to seating angles I couldn't see who he was, but his argument was cogent and well thought out. There is no question about his being right in the realm of IT, standard encodings make a difficult job easier, but to eliminate free text in the areas he mentioned would be wrong and dangerous from both cultural and medical perspectives.
Consider an entry in a patient chart like the following:
Patient believes has "the sugar" because she gets light-headed when she stands up and is "wiped out" during a lot during most of her day, especially after meals. Spouse reported confidentially that he suspects she is self-administering "Robitussin" by which he means a cough suppressant containing codeine. Spouse was unsure of actual brand or amount being taken; will observe more carefully and is to report by phone before her next visit. (Possible projection- spouse appearance consistent with CNS depressant abuse.) Dx of depression equivocal. DM2 possible but insulin levels tested normal during examination. Patient refuses to self-test due to cost of supplies (self-insured). Similarly refused to schedule next visit; spouse to encourage.
There is a vast amount of information here even to we who are outside the imaginary situation, but in real life the clinic staff may derive even more due to their knowledge of context, including prior medical history of patient and family, cultural norms, fear of skin puncture previously expressed by the patient, family financial and emotiuonal stability, and so on. Encoding many of these data may eventually be possible, but not necessarily in our lifetimes, but meaning is not static or atomic. In human communication, meaning is negotiated and high degrees of ambiguity are tolerated. The right thing is often done for the wrong reasons (i.e. in the absence of sufficient evidence) and vice versa. Human negotiation of meaning is fluid and continuous in a way that may never be possible to duplicate in machine "intelligence".
Where I would argue we need to devote our energies would be to machine understanding and encoding of what factual knowledge can be gleaned from the free text, but without discarding the free text, and with the encoded facts clearly marked as machine interpretation. Computer interpretation of EKGs is routine, and more coverage both in breadth and depth is both possible and desirable. But please, speaking as a patient and child- and eldercare giver, don't lose the free text. Interpersonal communication is the best and cheapest form of intelligence we've got right now.
Comments