Add support for referents


Work by W Ceusters, B Smith, and others shows that referents to real-world phenomena are not supported properly in EHRs or other health information systems. See the paper "Strategies for referent tracking in Electronic Health Records" by W Ceusters, IMIA 2005.


Koray Atalag
May 25, 2015, 2:04 AM

Hi Tom, I think this clearly covers the PR I raised about adding instance level annotations ()
Though, I must say, this is wider than that - in my case I was after asserting a Particular_to_Universal type relationship here you have to deal with two other types: P_P and U-U. An interesting paper - especially Werner's left testicle is a bit over-emphasised but certainly ontologically very sound (this paper I'm talking about: Ceusters W, Smith B. Strategies for referent tracking in electronic health records. Journal of Biomedical Informatics. 2006 Jun;39(3):362–78.)

From p12: "The ‘‘annotation’’ of an entry in a database by means of a term from a controlled vocabulary such as the Gene Ontology [41] is a typical example"

I must also say the way to go would be to incorporate Linked Data principles - which are underpinned by URI based identification and a simple RDF based (subject, predicate, object) annotation approach where each could have an URI - whether a universal or a particular. I'm not saying we adopt RDF by any means, just the means that would allow for capturing same info.

Thomas Beale
May 25, 2015, 9:09 AM

I would suggest there are two distinct needs -
1. referent tracking - that means using the system of IUIs etc to properly identify real things in the EHR so that repeated mentions of them mention the same one or not, according to what happened in reality

2. annotations of data - these probably need to be set up as some sort of triple store registries that point in to data items in the EHR (using reliable EHR URIs) and relate them to specific ontology or other entities according to research programmes.

Koray Atalag
May 25, 2015, 10:12 AM

Hi Tom, I agree but in the paper mentioned you'll see this is also given as part of referent tracking. You may be right that the sensible way to implement would be like that but I still think there is a place for instance annotations - or as the paper puts particular to universal relations for EHR data. Furthermore the changes required for particular to particular type relations could support this kind of depiction without additional complexity. Two birds in one go....

Thomas Beale
May 25, 2015, 10:27 AM

There are probably two types of 'annotation' needs (both separate from referent tracking, which is a whole different thing):
1. annotations that users/apps committing data want to include at moment of committal - some sort of operational meta-data
2. research and secondary use meta-data. This would typically be added after commit, and could be voluminous, consist of multiple whole siloes of meta-data corresponding to different research projects.

I would say that we might consider changes to the RM to accommodate 1, if there was strong enough evidence of the need. I would say that 2 definitely needs to be outside the EHR data proper, and be done by a system of references pointing into the data. The logical result is still something like 'annotations', but done without weighing down the operational DB with a growing accumulation of research project content.

Koray Atalag
May 25, 2015, 10:54 AM

I'd completely agree with that.


Thomas Beale

Raised By



Affects versions