Friday, May 27, 2011

Chocolate Teapot Not Otherwise Classified

The methodological approach which underlies this blog is based on the position of ontological realism. One standard objection from the HL7 community to this position is that the ontological realist strives for 'perfection', where HL7 and its ilk, more sensibly, strive for mere 'practical utility'.

Examination of the many concrete proposals made under the ontologically realist heading proves, however, that they are both (a) quite modest, and (b) of considerable practical utility.

In many respects these proposals echo the "Desiderata" advanced by Cimino in responding to the poor quality of many of the terminology artifacts created in the field of medical informations. In one central respect, however, they go beyond Cimino who still (in 1998, at least) sees the way forward as lying in a move from the orientation around terms to an orientation around what he calls "concepts". For inspection reveals that the concept orientation is itself one reason for the massive inconsistencies that exist, both between different specialist terminologies, and among the different versions of these terminologies produced at successive points in time. This is because the concepts in the minds of specialists are too etherial to serve as a constraint on the builders of terminologies, especially when the latter are themselves subject to the pressures of third-party payers, gatherers of public health statistics, patient groups and other stakeholder lobbies. The result is confusion; the waste of investment in the creation of poorly maintained mappings; ever more intensified forking leading to a failure of cumulativity of data and to a lack of IT support, for example, in areas such as continuity of care (for example when patients move from one hospital to another).

The solution proposed by ontological realism -- now embraced in part also by Cimino -- rests on the idea that terms in ontologies (or terminologies, or messaging standards) should as far as possible be oriented, not around concepts in people's heads, but rather around the types of entities in the corresponding domains of reality. In this respect, terminology work should mirror the practice of scientists, who as a matter of course use terms such as 'electron' or 'planet' or 'mammal' in describing their data, rather than terms such as 'electron NOS' or 'planet NEC' or 'mammal that is of interest to the human resources department'. Terms in the former group are used in remarkably stable ways, by multiple, ever expanding communities throughout the world, and there is empirical data associated with the corresponding types in reality deriving from large numbers of heterogeneous sources.

The proposal is (1) that representations of types of the mentioned sort should be used in terminology resources wherever possible, not merely as terms in their own right, but also in the definitions of further terms, and (2) that the term 'concept' should be avoided entirely. (A partial exception might be made in areas such as linguistics or psychology, where the scientists in question are able to define what the term 'concept' means.) In this way, terminologies will be anchored -- to some degree at least -- to a stable, empirical benchmark, following a pattern already realized by the OBO Foundry ontologies. Terminology resources will also be anchored to each other, through the use of a common set of feeder terms in their respective definitions, in ways which can support consistency over time and thereby also allow the more effective aggregation of data.

We do not see this solution as a magic bullet. But we do believe that it has the chance to constrain forking in some small areas of clinical research, and that putting it to the test in pilot experiments may provide a generalizable strategy for the formulation of coherent terminologies and associated definitions in broader areas in the future.

Moreover, we are absolutely sure that the negative results which have flowed from not following this proposal are immense. These results are visible, for example, when we examine some of the problems faced by users of SNOMED as a result of the fact that the latter has been subject to such a large number of changes in its successive versions -- changes that can be attributed precisely to the concept orientation, which infects SNOMED from the top down.

I now discover that these results have been documented in a beautiful piece by Malcolm Duncan entitled "Medical terminology version control discussion paper: The chocolate teapot (Version 2.3)" in relation to the Systematized Nomenclature of Kitchen Terminology (SNoKitch), and specifically to its two non-contiguous branches Crockery and Teamaking Related Findings in its n.3 release:

Crockery
---Teapot
----- Brown teapot
------White teapot
----------White china teapot
------Blue teapot
----------Blue china teapot
------China teapot
---------White china teapot
---------Blue china teapot
------Pink teapot
------Aluminium teapot
------Chocolate teapot
------Ornamental teapot
------Industrial teapot
Tea making related findings
---Teapot related findings
------Large teapot
------Full teapot
------Empty teapot
------Teapot with warm water
------Teapot with cold water.
The document in question is cited in a (long) HL7 discussion concerning the problems which arise in connection with the Cimino desideratum of "concept persistence". This discussion, incidentally, reveals one flaw in Cimino's formulation of his desiderata: for it shows that recommending "concept persistence" as a desideratum is a bit like recommending "life persistence" for the patients in your hospital. Life persistence is, certainly, a nice thing to recommend; but it would be even better, surely, if one can formulate a desideratum which identifies some specific strategy for achieving it -- in such a way that the desideratum of persistence will follow as by-product. It is something like this which ontological realism, in its stumbling fashion, is attempting to do. For its counterpart of 'concept persistence' -- to the effect that ontologies should be developed in such a way that versions should evolve gracefully in reflection of scientific advance -- follows immediately from the recommendation to use, as far as possible, the terms employed by scientists.

Saturday, May 14, 2011

HL7 Refresh

Here, continuing the theme of "the fall of the RIM", are some recent contributions on the future of HL7: