Re: [COAS-List] New Model for ObservationValue

[Gunther Schadow <gunther@aurora.rg.iupui.edu>]

Tim, 

as you requested, here are my comments on what I see on the three
pages of COAS Information Model. These are all points of
criticism. Sorry to appear so destructive. I do appreciate any
discussion on that and especially I appreciate your trying to
harmonize. Since I recognize some of the idioms you are using here
from other ongoing discussions I am quick to find flaws in them. You
are invited to comment and criticise any of the HL7 v3 data type stuff
too.

- "negative finding" is not normalized. The attribute is found in
DateTime and ConceptCode, but nowhere else, although, logically
it could apply everywhere. If you must have it, I'd suggest pushing
it into ObservationValue.

- what are DateTime.units? What's the difference to .precision?

- ConceptCode.uncertainty of type QualifierCode? How's that used?
  I believe that "narrative probabilities" (e.g. likely, probable,
  consistent-with) are ill concepts and should not be formalized.

- MachineProbability vs. userProbability: you must have built this
  with some input from Stan Huff did you? We discussed this too,
  and I think it is a possible but not very generalizeable
  distinction. What is a user here? How do I care? Probability is
  in the eye of the beholder, subjective. Thus everyone must
  interpret incoming data and adjust belief into it. Along the
  chain of communication (you telling X, X telling me) all of us
  must adjust probabilities. I don't trust a user more than a
  machine nor a machine more than a user, so for my own
  interpretation, that user/machine distinction is not very useful.

  For more on uncertainty, please see
    http://aurora.rg.iupui.edu/~schadow/v3dt/minutes19981214.html

- Null? What is a Null observation value? Why does Null have a value
  of NULL? This reminds me of the metaphysics you can run into
  when you build you own toy LISP and are about to decide what
  (EVAL NIL) is?

- PlainText how do you handle character codes? We chose to use Unicode
  and be done with it. If you did the same that's fine. If not you
  have to reflect encodings somewhere. Maybe it's in String?

- Blob: Everything can be a BLOB. The problem is, PlainText can be a
  Blob too, and so is the boundary between Audio and Curve blur.
  We do use the MIME idea in our MultimediaEnabledFreeText type too
  and have similar problems. But I would say that PlainText would
  be replaced. I would not hardcode the URL vs. PhysicalLocation in
  the PlainText. It does apply for Blob as well. It is not clean here.

- Measurement precision vs. accuracy, think there are other more
  appropriate ways of expressing uncertainty of measurements. Some
  are suggested by ISO, we use ProbabilityDistributions for that
  so far, and I also find confidence intervals very useful.

- Measurement.relationalOperator, I don't find this a good idea.
  I would approach this a bit more mathematically: there are
  numbers and intervals (continuous sets of numbers). The relational 
  operator implies an interval.

  With Range as subtype of Measurement, you get into the same
  discussion the Stan and I had. I believe that <3 is an interval
  (range) "]unk, 3]" and there shouldn't be two ways of specifying
  intervals depending on whether the interval is bounded on one or
  both sides.

  The fact that your two intervals (relational op.) and
  Range are on two different levels and non-exclusive you could
  build expressions like "< [3,5]" and I don't know what that
  could ever mean ("= [3,5]" is not much better, at lest it
  means something, but not what I suppose you'd like it to mean.)

- Actually, the above boils down to a technical correction: Numeric
  has a second relational operator, now you could build "< > 3"
  what does this mean?

- I believe that MeasuredValue should bear the unit, but that may
  be a matter of style.

Again, I apologize for the criticism and I am looking forward to
discuss the issues with you and other COAS folks for the sake of
correctness and simplicity. I heard you are in Orlando in January
24-29? We could meet there and exchange ideas orally, if you want.

regards
-Gunther

Gunther Schadow ----------------------------------- http://aurora.rg.iupui.edu
Regenstrief Institute for Health Care
1001 W 10th Street RG5, Indianapolis IN 46202, Phone: (317) 630 7960
schadow@aurora.rg.iupui.edu ---------------------- #include <usual/disclaimer>