Hello everyone It has been very interesting to read the post and follow the subsequent discussion about [the quest for] "one information model" in e-health and especially the "patterns" part.
I am not sure if there can be "one information model" in e-health but what i think that can be done is a ranking of available models according to how expressive they are and patterns (patterns of data structures and types) would be key to this. For example, at the low end of the spectrum, we could have some really simple models which are able to describe hierarchical information up to a certain depth. For example, you can describe the data about a document as a "sequence of element" where "element" can be "simple" (only one data entry of some supported type) XOR "complex" (an entry composed of many simple supported types). This would be less expressive than "a list of element" where element can be "simple" OR "complex". In the latter case complex elements can also be nested. Something more expressive than this would be a "[structure] of element" where [structure] could be a list, a tree, a graph or other of "complex" OR "simple" elements. The table PATTERN / EXPLANATION in Thomas' blog post is a good start but i think that it is mixing types with some familiar class names. These can be broken down even further: The "data/state/protocol/reasoning" is a Dictionary. The "History of events" is essentially the same but the type of the "index" is now different. An "order state machine" is a directed graph (with loops). A composition document is a Set and participations could also be sets or dictionaries. Could we have defined the COMPOSITION as a special case of a "set archetype" with specific constraints? Could we have done the same for a state machine? Simply counting the supported patterns would not provide a comprehensive picture though. I trust that terms are used consistently. So, a mathematical model provides constraints over a domain. But there is nothing stopping someone to create an overly complex model to fit complex data with minimal error (reminds me of feature creep). You could have two models, one with 5 parameters and one with 50 parameters that seem to be fitting experimental data (observations from the domain) perfectly, which one do you choose? Obviously, the one with the 5 parameters because it is _simpler_. Mathematics does have tools to estimate model complexity versus how well the model describes the observations and i am sure that some parallels can be drawn here. In computer science, there are probably additional markers that need to be taken into account...It's not only a matter of how expressive a model is but also how "easy" it is to be used or how "cheap" it is to implement...These are probably more difficult to quantify but key factors for adopting a particular model. In this way, the model(s) that rank first would be the easiest and cheapest model(s) that can describe a domain most effectively. We don't have a golden standard for a domain of course, but something like this would be irrelevant. Even if we could pull all the workflows and documents from all the healthcare systems in the world, they would be a snapshot of what is required today** and with the exponential pace of progress would quickly become surpassed. This means that the ranking (without taking into account the "shape" of the domain) would only be useful amongst two or more models. All the best Athanasios Anastasiou **: But a very good indication of the patterns that are actually used out there...maybe Google is already doing it. On 05/05/2011 17:20, Thomas Beale wrote: > > this is an often debated question, and after coming across (for the > 100th time) just such a debate recently online, I thought it might be > interesting to try to get to the bottom of the question in some way. The > basic idea posted here > <http://wolandscat.net/2011/05/05/no-single-information-model/>. It is > of course not scientific work, but I would be interested in the views of > others on this concept. > > - thomas beale > **
