[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[COAS-List] 80-20 rule
I think I am seeing a clear 80-20 (or better)
scenario here. That is, there is probably 90+ % of observations we can
handle with an admittedly slightly-simplistic solution that takes <10% of the
effort needed to handle everything.
The pure analyst in us likes to exhaustively
model the world and believe that "we see it all perfectly - and nobody will
come up with a structure that breaks our model". The realist always
accomplishes more in the long haul. Of course, however, we must guard
against an overly simplistic model that turns out to be useful for only
10%! "I reject the occasional doctor that says unless its ALL
available to me electronically I don't want any of it". We have found
on several occasions that if you can deliver near-real-time lab results and med
record reports into a CDR these docs will get right in line for a user
ID.
I also think we have to keep in mind that while
"ANYs" and "meta data" (okay, "metaknowledge" if
that feels better) allow intact COMMUNICATION of non-prespecified complex
structures, the ability of the requestor to DO anything with it requires that
the object has the ability to produce its own viewer. Even if you do this
(say, with XML, a pointer to an applet, DII/DSI), you have not enabled the
special format to be usable by MACHINE (unattended) inferencing.
Therefore to make observations both available and usable by
the human client requires standardized format, standardized metadata, or a
bundled viewer. For any observation type that clearly meets this rule,
let's put it in ELSE leave it out.
I am glad Henry Ford didn't withold the model T until the
Mustang was also ready. We could drive to the grocery store just fine in
the T.
Jon