Tuesday, February 21, 2006

Is there a difference between a person and a door?

In the paper "HL7 RIM: An Incoherent Standard" the RIM is criticized for failing to draw a clear distinction between entities in reality (patients, diseases, treatment acts, ...) and the records of such entities, for example in paper documents.

Mead Walker reacts to this criticism in an email as follows:
The paper puts a lot of effort into discussing HL7's mixing of what you call its "information model" with its "reference ontology". I agree that these are distinct concepts, and that HL7 tends to switch from one to another without warning. However, I do not believe the paper substantiates why this is a problem. I think that is its major flaw. It is almost as if you think the seriousness of the defect is self evident, but it is not. Perhaps, this is an omission that only a philosopher would make. Let me use an analogy. You can look at me to see if I have a mustache; you can also look at a mirror image of my face or at a photograph. In either case, the fact that I do have a mustache is evident, and the question of whether you know that by looking at me, or by looking at the picture is normally trivial. Why do you think it is important to know if the knowledge that is being transmitted comes from the thing itself, or from its representation? ... Isn't this simply a fundamental philosophic conundrum that really is not practically important?

My response was as follows:

Dear Mead,
You are right that this is, for me, a self-evident defect. You are right, too, that it sometimes does not matter where you get your knowledge of mustaches fr0m (though it might, e.g. for forensic purposes; and I would imagine that if you are building a messaging standard then these are precisely the sorts of purposes you should be bearing in mind). The main issue, though, relates to the question not of how we know but of what sorts of things our knowledge is about.

If I told you that your messaging standard cannot distinguish reliably between, say, a human being and a door, then you would rightly be troubled.

Why are you not similarly troubled if I tell you that your messaging standard cannot distinguish reliably between a human being and a social security number?

Or between an actual real-world case of cardiac arrest and the information content of a cardiac unit resuscitation note?

Or between John's elevated blood pressure (which endured for several hours) and the information recorded during an action of measuring this blood pressure at 4.07 pm?

Surely it is a serious problem when one does not know what basic categories of entity one is dealing with, so that one confuses real-world phenomena (which might make you sick) with information captured during the observation of such phenomena which sits inside computers. This latter distinction is, for me, as self-evident as the distinction between a real-life performance of Tosca and the musical score you might buy in a music shop.

But, you might still say, is the distinction of genuinely practical importance? Well, consider blood pressure:

On the one hand there is your blood pressure itself, the real-world phenomenon which obeys the laws described in a medical textbook (which will tell you about systemic arterial pressure, about systolic and diastolic phases, about fluid dynamics, etc., etc. complicated physics and physiology that will be of practical importance e.g. when designing an instrument that can accurately measure blood pressure or when dealing with a patient who has atrial fibrillation). On the other hand there is a blood pressure observation, another real-world phenomenon, but of an entirely different sort, involving factors such as:
  • the position of the patient at the time of measuring (sitting, lying, etc.),
  • the tilt of the surface on which the person is lying,
  • the variation in measured blood pressure with respiration,
  • the instrument used to measure the blood pressure,
  • the size of the cuff if a sphygmomanometer is used,

and so forth, as well as the units in which measurements are taken. (One detailed representation of these factors is here.)

Note that blood pressure itself (including the sort of elevated blood pressure that can kill you) exists entirely independently of instruments or units of measure or acts of observation. Blood pressure itself exists as an endurant feature of the organism from one hour or day to the next (constantly rising and falling even while preserving its identity). A blood pressure observation, in contrast, is a process, which does not endure through time but rather unfolds through time in successive temporal phases. (See here.) Thus the features of the former are quite different from the features of the latter, as different as, say, a potato and the act of eating a potato.

Consider, now, for further evidence of the practical significance of this distinction, that there are some words (e.g. 'systolic') which appear both in textbook descriptions of blood pressure itself and in records of acts of observation. Sadly, even leading experts in medical terminology are often misled by this fact. For the two sets of occurrences of the same term refer to entities in reality which are of an entirely different sort. Descriptions of the first kind would be appropriate to ontologies of pathology (of the what it is on the side of the patient) such as we might hope to find in SNOMED-CT; descriptions of the second kind would be appropriate to message descriptions such as we might hope to find in HL7.

1 comment:

Gunther Schadow said...

I still think that you don't demonstrate the practical importance of this issue. Of course we distinguish between a patient and a social-security-number, how don't we? I'd say you should show real issues, not analogies. Don't say "it's like ..." make a real scenario and show what goes wrong.

I agree to much of your blood pressure analysis and I think it's nice to think through these things with a calm mind (!) However, I find a problem in your analysis. You say:

On the one hand there is the blood pressure itself [...] on the other hand there is a blood pressure observation [...] involving factors such as [1] the position of the patient [...] [3] the variation in measured blood pressure with respiration.

The two that I left in the quote are factors which influence the actual blood pressure, not just the observation. So, we can say that to find out what we need to know about a blood pressure observation, we need to know about the blood pressure, and how the observation works. In the end then, some components of the blood pressure Observation are also factor which influence the blood pressure.

These other factors are factors which influence the measurement. We can account for that if we consider that the Act of measuring is sampling in nature. There is the entire blood pressure, then there is the blood pressure at the brachial artery on which we end up measuring, those are different. Then there is the occlusion pressure exerted by the cuff. So we measure a cuff-pressure which by this apparatus is a sample of the blood pressure in that brachial artery at that time.

Thus, the blood pressure Observation is indeed a recorded sample of the whole blood pressure phenomenon.

And so, yes, in the HL7 RIM we are talking about Observation as the only place for talking about blood pressure, because that is all that had ever been required by medical record systems. (And of course T.B.'s thing doesn't do this any different). So, it is true we are stopping at the Observation Act without modeling the phenomenon observed. But this works for what it is meant to do.

One problem with going beyond the Act is that it's not clear how the phenomenon should be modeled and what good it would do. Because people do not need to exchange their theories about physiology but they need to exchange the blood pressure measurements that they make, and they think they know what it is. No EHR system needs to contain blood pressures, only observations about it.

If you built a simulator, then you would have an endurant process somehow that simulates the blood pressure phenomenon. [As a process of regulation and homeostasis, blood pressure is still nicely represented as an Act, but of a very different sort.] As soon as we sample the simulation process, however, we have Observations again. And observations + theories is all that we have when we make medical decisions.

The only known use case for HL7 is to represent the Observations, and it does that well and in fact noone does it much different. Noone had asked HL7 to represent blood pressure simulations or theories of blood pressure phenomena. So, the RIM doesn't need to talk about that at this time.

Now we can do a whole lot of decision support -- practically relevant -- with Observations by using Observations as the basis for the decisions. You have Observations in, decision made based on theory, and concequent Acts out. Again, what we have works. Any decision algorithms that people actually deal with, also work based on blood pressure Observations, not on the endurant phenomenon of blood pressure.

Even simulation-based approaches, may be some models with differential equations in which the equations are a model of the blood pressure phenomenon, would still be fed with blood-pressure observations only, and its recommendations are expressible as observations which we should be able to make (if the simulation is correct.)

So, in the end, noone enters blood pressures into a system, only blood pressure Observations. Noone asks an EHR system for blood pressures, only for blood pressure Observations. That is why what the HL7 RIM does is fit for the purpose. And nothing is in fact broken about this. Certainly not in practice, and it seems not even in theory.