Resent-Date: Mon, 01 Mar 2004 22:07:06 +0000
From: Thomas Beale <[EMAIL PROTECTED]>
Andrew's initial response to this (this is care2x's data model; why should we favour it as a global data model over any other) is about right. Care2x has built a single-level model to make their system work, just like every other system builder - and that's fine. You can see that it's 'single-level' by perusing the class hierarchy - notice for example that the Encounter class (http://care2x.org/phpDocumentor/care_api/care_api/Encounter.html) is a kind of "Note" in their system; notice also the number of hospital specific attributes in the interface. I can guarantee that if you put 50 people from this domain (e.g. from this list) into a room to discuss this modelling, you would get a 2 week unresolved argument. Some would say that 'encounter' is indeed a kind of 'note'; others would say it is a kind of 'act'; others would say it is a kind of patient/health system interaction, and so on. And that's before the debate on detailed attributes and functions. Now - this is not at all to criticise the Care2x people for their product - just to make the point that the model is one designed to meet the requirements set for a particular product, and is (going by the last 15 years of experience in health informatics) unlikely to be directly usable across the entire domain.
Others here and elsewhere of course have thought about the problem (for many years) of making data models that really do work for all of the domain, and it is the main goal of openEHR (with which I am involved) to do this for health records (EHRs); Andrew's OIO product has a similar goal; the HL7v3 RIM is another such model; there are other attempts out there as well.
The point is that models which are initially conceived to satisfy many users across the spectrum (rather than a single product or kind of use) have to be materially different from models of particular products or domains. First, as Andrew points out below:
Has it a good conceptual model?
J, I think it has an excellent data model. That is not the criticism. Instead, the issue is how can a single data model be excellent for every hospital and every living, breathing group of healthcare workers?
Well, here we diverge. It certainly does not have the most accurate
medical data representation model.
Accuracy is never the issue. The biggest issue, when it comes to living, breathing, symbiotic software (from my biased perspective), is change management / adaptability.
Attempts to produce single-level models of whole of domain will invariably fail on these criteria - comprehensive suitability across the domain, and adapability / change management. The key thing to understand in this approach is that any domain concept such as 'Encounter' (in its usual medical sense) most likely cannot afford to appear in the data model - since it will never be the correct model of that idea for all possible users of the model; instead it has to be built using some other formalism, which makes use of a far more generic data model - what we would call a 'reference model'. The openEHR reference model for example only has 3 generic Entry types - Observation, Evaluation, and Instruction; and about a dozen other major types. Encounter is not one of them. To see the openEHR data model see http://www.openehr.org/documentation.htm (NOTE: the openEHR site is under complete reconstruction as I speak, so this page is not up to date, but it doesn't matter for this discussion).
Instead, you have to adopt a different strategy. One such strategy is 'two-level modelling', which I described in some detail (sorry to the rest of you here!) in http://www.deepthought.com.au/it/archetypes/Output/front.html. THis approach is used in openEHR (www.openEHR.org); you can see examples of how this is done from the archetype page currently hosted at http://www.oceaninformatics.biz/adl.html, including a page of examples at http://www.oceaninformatics.biz/archetype_examples.html. There is a tool with which you can experiment with such archetypes at http://www.oceaninformatics.biz/adl/adl_workbench_win_feb_2004.zip.
Now...of course this is not the only attempt to solve this problem, but generically speaking, it seems to be the right kind of approach. Computerised practice guidelines adopt the same approach (generic underlying reference model; then each guideline is a configuration of model objects); and others here will be able to give details on other systems which adopt the approach.
hope this helps,
- thomas beale
------------------------------------------------------- SF.Net is sponsored by: Speed Start Your Linux Apps Now. Build and deploy apps & Web services for Linux with a free DVD software kit from IBM. Click Now! http://ads.osdn.com/?ad_id=1356&alloc_id=3438&op=click _______________________________________________ Care2002-developers mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/care2002-developers
