On Mon, 2003-06-30 at 14:57, Tim Cook wrote:
> For those healthcare providers not on the TORCH list I wanted to point
> out this effort by Dr. Maynard. 
>> I want to ask about a barrier to initial impression of TORCH that
>> I recall as a newcomer to TORCH and which is also a barrier to
>> thorough testing of the modules. There are not enough realistic
>> patient EMR entrees in the demo, nor enough with multiple
>> encounters and completed encounters.
> 
> If this effort is successful I will develop a method to export these
> records into a file that can be used by other projects that need this
> kind of data.

1) In the Febrl probabilistic record linkage suite (see
http://febrl.sf.net) we have a little utility to synthesise names and
addresses (with or without duplicates with random errors). It could
readily be adapted to furnish alrge numbers of plausible sounding but
fictitious patient identities, and teh technique could be extended to
clinical as well as demographic data. I have just used it to generate
100,000 cases to do some initial stress-testing of the public health
application on which we are working: user interfaces which were fine
with 10 or 20 test cases suddenly don't seem so great with 100,000
cases...for things like TORCH or GnuMed, testing with 10,000 or 20,000
patients would not be unrealistic - and better to discover scalability
problems (with the user interface as well as the underlying database and
application technology) with dummy data than with real life data.
2) There are two ways to get such data into the application: import it
directly into the underlying database, or automatically injecting it
through the user interface, possibly using multiple workstations at once
to simulate real-life load and to check for multi-user issues. For
Web-based applications, the following looks like it may be useful -
agian, we intend to investigate it for testing our public health app:
http://wwwsearch.sourceforge.net/ClientForm/
3) Shareable EHR/EMR test data requires a shareable format which
captures the relationships between dta elements. XML is teh obviousl
choice, but it would be nice if the format was more readily
humna-readable and -writable. YAML (see http://yaml.org) might be a good
choice.
4) Which raises the issue of the appropriate (as generic as possible)
schema/data model for such test data. Are there candidate schema
available - for example, what schema would be required by TORCH?
-- 

Tim C

PGP/GnuPG Key 1024D/EAF993D0 available from keyservers everywhere
or at http://members.optushome.com.au/tchur/pubkey.asc
Key fingerprint = 8C22 BF76 33BA B3B5 1D5B  EB37 7891 46A9 EAF9 93D0


Attachment: signature.asc
Description: This is a digitally signed message part

Reply via email to