The interesting discussion about database persistence layers for EHRs minds me to point the group to my PhD student Tony Austin’s research, which underpinned the current CHIME clinical cardiology record and decision support system demonstrators, based on our precursor work to openEHR, in the EU GEHR and Synapses projects.
The thesis can be accessed at:
http://www.ehr.chime.ucl.ac.uk/download/attachments/71/Austin%2C+Tony+%28PhD+2004%29.pdf?version=1
The main quantitative results begin on page 191, although that whole chapter needs to be read to understand how the tests were set up and performed.
Here is the abstract, to give an idea of the scope. It was a huge piece of experimental work, with live records, and probably could have made two PhDs; it took a long time!
The new version of the server, in beta test, is archetype based, but is not a native implementation of the openEHR Reference Model specifications, as that is a step too far for us at our demanding clinical application coal face, at the moment.
Abstract of Tony Austin’s PhD thesis, April 2004.
Records grow and multiply for a patient throughout their life. Healthcare enterprises are presently
unable to integrate the clinical information continuously accumulating from patient care or to
communicate it to a point of need. Their medical records are expected to be legally valid, complete,
accurate, and available in a timely fashion. As a consequence they must be adequately expressive and
maintained in such a way as to facilitate secure physical retrieval.
This thesis begins by considering the history of structured and electronic medical records, along with
the histories of distributed systems and databases. It proposes key infrastructure technologies that
could be used for the development of an electronic healthcare record server. Two models are then
established, to represent transferable clinical records in general, and to represent the structure that
those records are required to take. With these there is a means to ratify healthcare data between
different providers and ensure that received healthcare data of any type can be meaningfully
processed.
The performance strengths and weaknesses of seven different physical databases are assessed. These
represent a cross-section of relational, object-oriented and XML technologies. Another
implementation is created for the purposes of federation. A large-scale deployment involving
anticoagulant therapy management within a hospital Cardiology Department provides clinically valid
evaluation data.
An object-oriented model is shown to be highly suited to the task of representing medical data.
Middleware code converting object model data to any of the database architecture formats is never
demonstrably the limiting factor in the performance of the overall server. Consequently relational
databases are equally suitable as persistence engines for medical records. The new class of XML
databases perform well but incur significant space overhead.
Hope this may be of interest to those considering these issues, now. Obviously the world has moved on but the way we tackled the issues may still have some relevance.