Hi! Tom Beale wrote: > is DCM now trying to be totally model-agnostic?
Andrew McIntyre wrote: > Unless everyone wants to throw away their model and start afresh it has to be. [...] Andrew McIntyre wrote: > The concept of two level modelling perhaps needs to be 3 level, > > 1. Information System > 2. Glue layer > 3. DCM Model I wasn't involved in medical informatics at the age of the (friendly) dinosaurs when Arden Syntax was developed as way to represent detailed knowledge and rules for decision support, but parts of the research group were. If I interpret them correctly, knowledge representation worked pretty good in the Arden based "medical logic modules", the problem was the cost and resource consumption required to make what you call "2. Glue layer" since the layer "1. Information System" was so different in different systems. In the Arden days I believe a lot of the "glue" was called "data dictionary", neat services that you could ask for things like "latest fasting glucose value". As you can imagine this dictionary will be hosting a lot of complexity and development/maintenance costs... A good paper describing parts of this is: Desperately seeking data: knowledge base-database links G Hripcsak, SB Johnson, PD Clayton http://scholar.google.com/scholar?cluster=13853775547974845989 http://www.ncbi.nlm.nih.gov/pubmed/8130552 One way of looking at openEHR is that it is trying to achieve semantic coherence at level "1. Information System", and if the detailed clinical models (3) are done using archetypes, then no "glue layer" is needed (or the glue to e.g. decisions support can be AQL-queries reusable between systems). A model agnostic DCM (in practice probably a DCM having it's own internal model) can be useful in order to gather requirements, but I doubt that anybody will be able to invent general algorithms that can be used to automatically transform it to something that can be used in all information systems. Instead manual reinterpretations will be necessary - and BANG - you are back at the "Desperately seeking data"-problem and associated costs. So yes, mindmap/UML-model-agnostic DCMs can be useful for requirements gathering, somewhat independent of information systems/specifications, if the people paying for this find it to be a useful exercise and are then willing to fund the manual transformation to archetypes, HL7 etc afterwards. If a country/region on the other hand already has decided to use a specific information system model (e.g. openEHR/13606) then they might consider it wiser to spend the money on doing the detailed clinical modeling directly as templates and archetypes and avoid the extra reinterpretation cost/time/ambiguities. I guess the same goes for HL7-based thinking, except that the system owners will need to do one more reinterpretation anyway to map between HL7 and the "1. Information System", but modelling directly using HL7-building blocks instead of a model agnostic DCM will at least save them one reinterpretation. I am curious; who is the driving force behind a model agnostic DCM? Is it the same force that will take the costs for the "2. Glue layer"? Another question: Will it be cheaper to manually reinterpret from a "model agnostic" DCM to both HL7 and openEHR, than it would be to manually reinterpret from a openEHR-based DCM to HL7 or from a HL7-based DCM to openEHR? Personally I am working partly with EHR overviews, and from a practical point of view the restrictions/guidance from openEHRs information models regarding e.g. recording of timepoints are a true blessing. I now know where to find many important timepoints without having to manually adjust to every single archetype/DCM. I see it as another angle of the "Desperately seeking data"-issue. The same goes e.g. for finding the coarse grained state of care processes (due to the built in state machine scaffolding). Best regards, Erik Sundvall erik.sundvall at liu.se http://www.imt.liu.se/~erisu/ Tel: +46-13-286733 (Mail & tel. recently changed, so please update your contact lists.)