Case-Based Reasoning in the Health Sciences: Foundations and Research Directions - Computational Intelligence in Healthcare: Advanced Methodologies - page 130

from the onset, to ensure that the data collected will be able to support or refute
the stated research hypothesis given the laws of probability. Researchers gather
data in very strict contexts such as randomized clinical trials in medicine. The
subsequent statistical data analysis of collected data represents only a small part
of the statistician's work. The descriptive approach deals with summarizing or
representing a set of data in a meaningful way through primarily quantitative
features, although qualitative variables are also considered. Statistical data anal-
ysis stems from the descriptive approach, but deals only with the data analysis
part. This aspect of statistics, and in particular inferential statistics, is related
to data mining, which builds data models to produce inferences from data. Here,
data analysis has freed itself from the constraints of probability theory to analyze
data a posteriori .
5.4 The Role of CBR in the Health Sciences
CBR brings to the life sciences a method for processing and reusing data without
generalizing them, which statistics clearly considers outside of its scope. CBR,
like statistics, promotes “the scientific study of data describing natural varia-
tion.” CBR deals with natural variation in a novel manner, through analogical
inference and similarity reasoning. Current computer capacity has made it fea-
sible to study individual cases, because large numbers of cases can be eciently
processed without having to be summarized. Therefore, CBR can be seen as an
advance in the scientific study of data made possible by progress in computer
science. This study of how CBR can complement statistics has been a main focus
of CBR in the Health Sciences research. This is also one of the most salient con-
tributions CBR in the Health Sciences can make to CBR in general. Advances
in this area will be useful in any application of CBR to experimental sciences.
Many of the tasks performed by CBR in the Health Sciences systems compete
with corresponding statistical methods, particularly those of inferential statis-
tics. For example, Schmidt et al. present a CBR system for the prediction of in-
fluenza waves for influenza surveillance [60]. The authors compare their method
with classical prediction methods, which are statistical, and argue that CBR is
more appropriate in this domain due to the irregular cyclicality of the spread of
influenza. The rationale behind this is that statistical methods rely on laws of
probability theory which are not always met in practice. In these circumstances,
methods like CBR can be advantageous because they do not rely on these laws.
Another interesting example demonstrates how CBR can be used to explain
exceptions to statistical analysis [61] and particularly data summaries [62].
Some of the most interesting research in this area focuses on the role of CBR
as an evidence gathering mechanism for medicine [34]. CBR can detect and
represent how cases can illustrate contextual applications of guidelines, and spark
the generation of new research hypotheses, such as how repeated exceptions to
clinical guidelines can lead to modifications of the clinical guidelines [34,54].
More generally, one of the main motivations for the development of case-based
reasoning systems in biomedicine is that cases, as traces of the clinical experience
of the experts, play a unique and irreplaceable role for representing knowledge in