[Gnumed-devel] Re: [wxPython-users] Re: Medical Expert System GUI | Anchoring widgets

[Karsten Hilbert]

On Tue, May 11, 2010 at 01:45:32AM -0700, Dr.James wrote:

> The expert system engine is PyKE (http://pyke.sourceforge.net).Pyke
> has both forward chaining and backward chaining rules unlike PyCLIPS
> which has only forward chaining.

I see.

> Bruce Fredericksen, the author of
> Pyke is guiding us in the project, we are currently creating
> prototypes and refining the methods.

Now, that's good.

> We are using controlled vocabularies i.e each and every symptom/sign
> is already present in the system. These symptoms/signs/investigation
> results are all facts to be fed to the expert system, the expert
> system reasons using the inference engine and makes a list of
> hypothesis (each with a certainty factor) and list of questions to
> test the hypothesis (these too have weights, the question with strong
> weight has to be answered first).

What if there's no answer available to that ? I suppose the
doctor is free to answer another question ?

> Once a diagnosis is made, the management plan is drawn.

This is a problem with much of General Practice care.
There's rarely a "definite" diagnosis to be head. But I
suppose you don't mean scientifically definite but rather
definite in a rules-related computational sense.

> There is also
> provision for empirical and emergency treatments, for which we are
> using forward chaining rules.

That'll probably be what applies to most of GP care.

> Right now the system works in a command line interface (cli) which
> Bruce has build, and we are testing on it.

Is there any API GNUmed can talk to ?

What GNUmed would like to do (at which point we'd want to
move further discussion to a more appropriate list such as
gnumed-devel which I CCed):

When a patient is activated the expert system is notified
about that fact. It then retrieves known data about the
patient from the database and - based on its rules - form
several hypothesis with certainties. It would then notify
GNUmed about its results. GNUmed would display the results
similar to this:

- consider pankreatitis (77%)
- consider acute gastritis (71%)
- consider aortic dissection (15%)

where the user will have the possibility to point at an
entry and be told:

- why the expert system thought this should be considered
- what else should be investigated to increase diagnostic certainty

Another important area for an expert system would be
checking of adherence of patient state (as documented in the
database) with known guidelines for the health conditions
known to exist in the patient. IOW, monitor management of
care.

It is rather unlikely doctors will want to enter data in
order to get trivial suggestions (such as "- consider
pneumonia"). They will welcome "trivial" hints if they
*don't* have to enter anything *additionally*, however.
OTOH, they would be willing to answer a few questions from
the expert system if it helps them detect cases of rare
diseases ("- consider Winiwarther-Burger syndrome").

> I couldn't study the GNU MED GUI as it gives me some errors at
> startup. I am reinstalling the dependency packages and am going to try
> again.

Feel free to post problems on our mailing list. Or else
download a live CD for easy access:

http://sourceforge.net/projects/gnumedlive/files/

> A formalized vocabulary in an EMR can work wonders - basically in
> usual EMRs the data is stored as chunks of information. What if i want
> to tap into the system and gain info about a particular disease-
> symptom, perhaps list all cases with a disease and list the
> unexplained symptoms? Integrating Naimath in an EMR can thus help
> researchers a lot. Also this standardizes the case sheet writing.

There's basically two schools of thought here. The one you
put forth and the other one that presents the argument that
controlled vocabs *alone* cannot ever capture the whole (or
even true) story. Which I tend to agree with from the last
10 years of community medical care.

Karsten
-- 
GPG key ID E4071346 @ wwwkeys.pgp.net
E167 67FD A291 2BEA 73BD  4537 78B9 A9F9 E407 1346