- From: Peter Ansell <ansell.peter@gmail.com>
- Date: Tue, 2 Feb 2010 09:39:46 +1000
- To: Danny Ayers <danny.ayers@gmail.com>
- Cc: Jim McCusker <mccusker@gmail.com>, John Madden <john.madden@duke.edu>, w3c semweb HCLS <public-semweb-lifesci@w3.org>, Oliver Ruebenacker <curoli@gmail.com>
On 2 February 2010 09:08, Danny Ayers <danny.ayers@gmail.com> wrote: > Peter, I agree with 99% of what you said but this bit bothers me a bit: > > >> People regularly misinterpret medical documents currently by examining >> them without the proper medical training. Adding superclasses etc or >> deleting elements as they feel necessary is just formalising the >> process where normal people interpret advice given by medically >> trained people. > > Surely the point of what we do (or maybe just should do) with online > data is to minimise the risk of misinterpretation? > Classic medic stuff says the doctor isn't always right, but the > patient usually doesn't have a clue so it's based on trust. Trust is > good, but really (for the person on the receiving end) I'd rather see > stuff based on facts. If the focus was on interactive medical documents, and not just marked up versions of what is currently given, then it would provide a bigger benefit, IMO. The main bits I am wary of is using automated complex reasoning outside of the facts that are given in the document because that still requires people to understand pretty much every statement in the set of related documents in order to trust the decision. If it was more focused on making the documents that are given to patients more interactive I am all for it as it gives the possibility to explore the related information without being overwhelmed by everything in a complex ontology at once. One area might be enhanced discharge summaries for instance that I saw a talk on a few days ago. If we get experience in being able to easily annotate documents, without focusing on being able to actually do anything reasoning based in the background, the reasoning element might fall into place gradually. Just having links to go with the medical terminology would be a good start. Practical use of patient data automatically with ontologies will most likely need to wait until there are less brittle methods of doing reasoning where single incorrect statements don't throw the whole process out. The success of bioinformatics ontologies on large statistically invariant populations doesn't easily map to individual cases--where the hard decisions that we are trying to make easier will inevitably contain apparently contradictory facts that will throw anyone who hasn't actually worked on modelling the ontology. Cheers, Peter
Received on Monday, 1 February 2010 23:40:13 UTC