Andrew McIntyre wrote:
> Hello Tim,
> 
> 
> Tuesday, February 21, 2006, 8:36:58 PM, you wrote:
> 
> TC> Andrew McIntyre wrote:
>>> Hello Tim,
>>>
>>> OK, you can arrow down and select the appropriate concept. Really you
>>> need 2 words or a very complex word or its to vague for a search. Its
>>> also essential to restrict the top level concept to a defined parent,
>>> such as procedure or clinical finding, as a minimum.
> 
> TC> Um, the attached screenshot shows a tree list of terms containing
> TC> duodenal. Am I right in thinking that one needs to chose various
> TC> restrictions as you describe before typing some words to look up? Not
> TC> quite what I had in mind...
> 
> No, that's in code, behind the scenes. You need to know what sort of
> values you want in the field, which you would, in the example its a
> "Clinical finding" rather than a procedure. In general you want to
> know what type of code you want in a field, or you may make a
> nonsensical statement. In this case I typed "Duodenal Ulcer" and the
> list popped up.

OK, thanks. As an aside, "screencasts" - short movies of software in
action, usually with a voice-over, seem to be the tool of choice for
illustrating such things. We just bought a copy of Camtasia for that
purpose - several hundred dollars only - and it's the ant's pants. Now,
where's my Director's Chair. Lights, cameras, action!

>>> Restricting the search is vital as there are 2 many concepts
>>> otherwise.
> 
> TC> The trick, Andrew, is to sort the choices by likelihood of being the
> TC> correct choice.
> 
> Well if I type "duodenal ulcer" as the free text the choices look
> pretty correct to me.
> 
>>> We use some complexity testing before we do a background
>>> search. Archetypes are excellent for restricting the query.
> 
> TC> Complexity testing? You've lost me there? Something to do with teh
> TC> entropy of the words (as in password complexity testing)?
> 
> Yes, if you type "Ulcer" as a search term then there are to many
> possible choices, if you type "Thrombocytopenia" or
> "Choledocholithiasis" then a search is worth a go. Finding a reliable
> rule for this is something that needs some more R&D. Typing "Duodenal
> ulcer" seems good enough to get a reasonable pick list in this case.
> Its also what people are likely to call it. Exploring some natural
> language Processing here is likely to help also. There are 2 snomed CT
> coding schools however - The "choice restrainers"  and the "Natural
> Language Processing" camp. I probably lean to the former, while
> wanting to leverage the later.

OK, I understand what you mean now - basically "inverse term frequency"
as researchers working in information retrieval tend to call it (but
yes, it can be more complex than that).

> Its a bit like google, typing "and" is not going to get you very far
> (NASA interestingly!)
> 
> The aim of archetypes is to restrict choices for you (You don't have
> to type them in, they are in the ADL) That way if you choose a report
> about eg Cholecystectomy, you are not shown concepts about colectomy
> or cystoscopy or rabies.
> 
> If people are presented with a reasonable list of choices they might
> use it. Browsing raw snomed for a term is not practical, you also need
> to be sure that you are choosing a concept from the correct branch of
> snomed-CT. We are looking for Name=Value Pairs and unless you choose
> from the correct branch you may end up with Value=name, name=name or
> value=value.

Yes, we are in the process of creating a set of extensions to the
current Australian standards for path reporting to allow path labs to
unambiguously report communicable disease notifications (to health
departments) in a canonical form. That process includes the selection of
all relevant SNOMED-CT codes from which path labs can chose for any
particular notification message. Such pre-selection of SCT codes in
itself is not a trivial task, we are finding.

> Snomed-CT has a lot of ip behind it, its not just a "Big List" and you
> can do some really complex things with it, but it does take some
> investment in time to understand it, I am certainly still learning,
> but its impressive, if a little flawed in areas. Snomed-CT are aware
> of some of the flaws and its evolving and improving with time. It
> seems its going to come to Australia so we may as well use it to its
> fullest advantage. An Australian Medication extension would be nice.

Agreed on all points.

Tim C

>>> Tuesday, February 21, 2006, 4:22:38 PM, you wrote:
>>>
>>> TC> Andrew McIntyre <[EMAIL PROTECTED]> wrote:
>>>>> Hello Tim,
>>>>>
>>>>> TC> Having seen your demo, I still think that the dream of a mechanism
>>>>> TC> in which the GP types the first few characters of each term into a
>>>>> TC> text box and the system provides an intelligent set of guesses
>>>>> TC> about whcih SCT concept (or English language word) was intended
>>>>> TC> and auto-completes it for him/her is feasible.
>>>>>
>>>>> You can stop dreaming, its done ;-)
>>> TC> Andrew,
>>>
>>> TC> Jolly good, show us! 
>>>
>>> TC> Tim C
>>>
>>>
>>>
>>>
>>>
>>> ------------------------------------------------------------------------
>>>
>>>
>>> ------------------------------------------------------------------------
>>>
>>> _______________________________________________
>>> Gpcg_talk mailing list
>>> [email protected]
>>> http://ozdocit.org/cgi-bin/mailman/listinfo/gpcg_talk
> 
> 
> 
> 

_______________________________________________
Gpcg_talk mailing list
[email protected]
http://ozdocit.org/cgi-bin/mailman/listinfo/gpcg_talk

Reply via email to