Clinical Semantics - Computational Intelligence in Healthcare: Advanced Methodologies - page 8

learn-while-use paradigm. In relation to this, further research has been carried out by
authors in the domain of language models for clinical medicine. Practical work in this
direction has been reported in [8] and is further elucidated in [9], [10] and [11].
The chapter builds on this foundation and discusses a flexible, evolving clinical
practice supported by open clinical agents for both clinical professionals and patients
capable of learning at the human abstraction level. Clinical semantics is not an add-on
but rather natively integrated to, and an operational principle behind, the functionality
of these agents. A key technical topic is natural language models specifically in the
context of clinical applications.
2 The Challenge of Clinical Informatics
The application of information technology to the computerization of medical records
dates back to 1960s when the first attempts were made at developing comprehensive,
structured hospital information systems. A prominent example of such technology is
MUMPS (Massachusetts General Hospital Utility Multi-Programming System) origi-
nally developed by Dr. G. Octo Barnett's lab at the Massachusetts General Hospital.
The authors of a research paper [12] reporting early experiences on the use of
MUMPS technology at a hypertension clinic state that:
”It is estimated that the data structure developed for hypertension and its complica-
tions represents between 1 and 5 per cent of the content of clinical medicine”.
This estimate turned out to be a wildly positive one considering all the effort put
into developing structured electronic medical records (EMRs) during the last 30+
years. It often seems that the evolution of clinical information outpaces the efforts of
system developers. The authors furthermore state that:
“In any system that is developed, mechanisms must be provided for adapting the
structure to changing needs, so that is remains relevant and appropriate at all times.
It is anticipated that this task of maintaining a structure representation of current
knowledge will be a much simpler task than its de novo development. ”
In fact, the maintenance problem of a conventionally structured medical record has
turned out to be an exceedingly complex one. In the typical clinical setting, different
information systems are tightly interconnected so that changes in one system can
result in a cascade changes in the entire health information system infrastructure of an
institution.
Still, the use of highly structured data remains a key technological goal as the pre-
valence of free-text, natural language composition is seen as one of the major hin-
drances in the work towards increased utility of an electronic medical record [9].
System designers often seek to limit the use of free-text compositions as the ambigu-
ity and unbounded nature of natural language results in difficulty in [9]
- validating the correctness and completeness of human-submitted information to be
recorded;
- generating descriptive and aggregative reporting and visualizations out of recorded
information; and
- developing decision support tools utilizing the recorded information.