Thursday, October 11, 2012

CDISC Discovers "Concept Maps"

My attention was drawn to interesting developments on the CDISC front by the following, from Kerstin Forsberg of AstraZeneca:

Checkout "mind maps" slide 13-15, 27 for#CDISC SHARE  < Then have a look at

The CDISC slide deck she refers to does indeed contain interesting representations of, for example, deep brain simulation as a treatment of Parkinson's Disease (slide 14):

But the question arises: how are the nodes and edges in such a graph to be interpreted. On slide 14 we are told that:  
[the b]asic building block is a “concept” which is a piece of clinical information. Examples include:
–systolic blood pressure observation
–systolic blood pressure result
–sodium concentration in plasma observation
–subject's birth weight result
–study subject
Each of these concepts has component parts (including what we would conventionally call variables)
The 'treatment' and 'Parkinson's Disease' nodes in the above, therefore, represent pieces of clinical information. Given the context in which slide 14 appears, I assume that these are pieces of clinical information about some given patient. But how, on this basis, can we provide a coherent interpretation of the edges on the graph. How, for example, can 'implanted in' be understand as linking a piece of information about this patient's brain with a piece of information about some lead?

Questions such as this were addressed (and I had optimistically assumed) put to rest already in 2006. The answer to such questions is that there is no coherent interpretation of the edges in a graph of the sort displayed if the graph is taken to be about relations between concepts. The Ontology for General Medical Science (OGMS) shows, I believe, how to create and interpret such graphs in a coherent fashion -- paying careful attention to the distinction between portions of reality on the side of the patient -- for example actions of treating, brains, neurostimulators -- and the types of which these portions of reality are instances.

OGMS is built on the basis of the assumption that each term and each relation used in the graph representation of a clinical encounter needs to be defined in a logical way. Only thus can the information contained in the graph serve computational inference. It is sad that, after so many years, important groups investing considerable efforts in healthcare informatics have still not apprehended the need for such definitions.

Update: October 18, 2012

XML4Pharma submitted the following query:

Not being a specialist in ontologies, I need some more explanation.
Do you mean that for each node, the edges (predicates I presume in the RDF context) cannot be defined in a unique way, i.e. there is an infinite number of possibilities for each edge to be named? E.g.
1. What one person defines as "is part of" can be defined by another as "is component of". Is that the problem?
2. Or should there for each pair of nodes be only a distinct set of predicates available?
3. Or do you mean that to assign a name to an edge, some systematic rules must be followed?
Ad 1. Currently, the standards defined for RDF, as for OWL (as for XML), place very few restrictions on what relations and what sorts of relations can be used to link nodes in an ontology graph, and they place no restrictions at all on what such relations should be called. This leads to the same sort of chaos as would be created if diferent airlines used different standards for representing time in publishing their schedules. In "Relations in Biomedical Ontology" we suggested a solution to this problem, and this solution has been applied and refined within the framework of the OBO Foundry.

Ad 2. I believe that, for each pair of nodes, only a small set of relations will be meaningfully applicable.

Ad 3. As stated under 1., we need standards which will ensure that the same relations receive the same names on all occasions of use. OGMS is an attempt to set forth the needed standards for annotating data relating to clinical encounters.