[Gnumed-devel] graphing is_alpha as an approximated pseudo-number ?

[Busser, Jim]

A colleague elsewhere wrote the following:

****************
I have been spending so much time on EMR requirements and functionality I am 
not interested in working on interfaces except where they impact on the EMR. An 
interesting little thing is on how to deal with a lab result that comes in like 
"<0.01" or ">10". The EMRs are set up to receive a numeric result but as soon 
as the "<" or other symbols are included things get screwed up. When looking at 
a text base report this is not a problem as the human reader understands what 
this means. The computer does not. When graphing out these results some EMRs do 
not display that result. The labs insist that that is how they should report 
results that are out of the accurately measurable range. I get stuck with 
looking at a graph and not seeing these significant high or low results. Many 
doctors say they do not bother with graphs. My take is that they have not seen 
good graphs and they have been taught to "look at the numbers". My feeling is 
that when they were taught it was not easy to produce graphs and all that was 
available were raw results or flowsheets. Graphs can be a wonderful way of 
displaying a lot of results in a way that it makes it easier to understand 
things. It is also a way of combining various things like lab results, vital 
signs and medication use. I think that there should be some teaching done on 
these issues.
****************

so I was wondering in the case of (say) TSH values which were

        2011-09-30      0.7 uU/mL
        2012-07-20      1.2 uU/mL
        2013-05-28      <0.01 uU/mL

would it be computationally manageable, and clinically reasonable, to pass over 
to gnuplot

        0.7     (as is)
        1.2     (as is)
        a derived value

where in the case of an is_alpha which contains < to parse what is beyond the 
symbol and to pass to gnuplot either one half, or perhaps one third, of a

        CAST (numeral text as a number)

and, in the case of an is_alpha which contains >, to pass say double or triple 
the numeral text?

I am just thinking that despite that such values are known precisely, they 
confer no wrong or even misleading semantic and, besides, these do not alter 
the original value stored in GNUmed's columns. 

Thoughts?

-- Jim