Information Technology Reference
In-Depth Information
Since differences between two cases are sometimes very complex, especially in
medical domains, many case-based systems are so called retrieval-only systems. They
just perform the retrieval task, visualise current and similar cases, and sometimes
additionally point out the important differences between
them [16]. Our early warning
system for the kidney function [12, 13] just performs retrieval too.
2.2
Temporal Abstraction
Temporal Abstraction has become a hot topic in Medical Informatics in the recent
years. For example, in the diabetes domain measured parameters can be abstracted
into states (e.g. low, normal, high) and afterwards aggregated into intervals called
episodes to generate a so-called modal day [17]. The main principles of Temporal
Abstraction have been outlined by Shahar [7]. The idea is to describe a temporal
sequence of values, actions or interactions in a more abstract form, which provides a
tendency about the status of a patient. For example, for monitoring the kidney function
it is fine to provide a daily report of multiple kidney function parameters. However,
information about the development of the kidney function on time and, if appropriate,
an early warning against a forthcoming kidney failure means a huge improvement
[13].
To describe tendencies, an often-realised idea is to use different trend descriptions
for different periods of time, e.g. short-term or long-term trend descriptions etc. [e.g.
16]. The lengths of each trend description can be fixed or they may depend on
concrete values (e.g. successive equivalent states may be concatenated).
However, concrete definitions of the trend descriptions depend on characteristics
of the application domain:
(1)
On the number of states and on their hierarchy,
(2)
On the lengths of the considered courses, and
(3)
On what has to be detected, e.g. long-term developments or short-term
changes.
3
Prognostic Model for TeCoMed
Since we believe that warnings can be appropriate in about four weeks in advance, we
consider courses that consist of four weekly incidences. However, so far this is just an
assumption that might be changed in the future. Figure 3. shows the prognostic model
for TeCoMed. It consists of four steps (the grey boxes on the right side).
3.1
Temporal Abstraction
We have defined three trends concerning the changes on time from last week to this
week, from last but one week to this week, and from the last but two weeks to this
week. The assessments for these three trends are "enormous decrease", "sharp
decrease", "decrease", "steady", "increase", "sharp increase", and "enormous
increase". They are based on the percentage of change. For example, the third, the
long-term trend is assessed as "enormous increase" if the incidences are at least 50%
