Information Technology Reference
In-Depth Information
the final decision based on their own clinical judgment. Furthermore, the users of
CDDS expect the main component of CDSS, KB (1) to be transparent, i.e., human-
readable; (2) to be updatable; (3) to be adaptable to local sites; and (4) to be able to
learn from experience, i.e., from existing “solved” cases [3] .
14.3
Knowledge Base in CDDS
The two main tasks of KE are building and maintaining knowledge bases. These
tasks require knowledge acquisition, verification, and validation. Knowledge acqui-
sition is one of the well-known “bottlenecks” in the development of CDSS. How-
ever, medical knowledge acquisition is not the only problem of CDSS. Once KB
has been created, it requires continuous maintenance. As pointed out by Spooner
[14] : “Maintaining the knowledge base in such systems is the most significant bot-
tleneck in the maintenance of such systems, since the knowledge base needs to be
expanded and updated as medical knowledge grows.” KB requires regular updates
and verification that the newly introduced rules do not interfere with the existing set
of rules.
In this paper, we concentrate on the verification and validation (V&V) of KB
in CDDS. We argue that the V&V process should be explicitly incorporated in the
design and implementation of the CDSS. Moreover, the knowledge representation
used for KB and the V&V process should meet the four requirements of the CDDS
users: transparency, updatability, adaptability, and learnability.
Transparency of KB:
The users of CDSS expect to be able to review the KB and
follow the reasoning apparatus. Thus, the KB representation should be human-
readable. To fulfill the transparency requirement, we have used a rule-based knowl-
edge representation. We have chosen this specific representation for two reasons.
First, rules are common in medicine, and medical practitioners use rules such as
Clinical Prediction Rules (CPR) in their daily practice. The prediction rules sim-
plify the assessment process, expedite diagnosis and treatment for serious cases,
and limit unnecessary tests for low-probability cases. Second, rule-based systems
have been the prevailing representation for medical expert system since the 1970's,
when Shortliffe created a well-known rule-based expert system, MYCIN [13].
Updateability of KB:
The users of CDSS expect to be able to update the existing
KB. To provide for updateability of the KB, we have utilized the KE tools such as
knowledge editors. The updates of the KB rules must be verified, so the new rules
or modified rules are not changing the rule-based inference system in an unintended
way. We discuss the verification of the updates in the next section. Also, updateabil-
ity is interrelated with the learning-from-experience process, which may necessitate
modification of the existing rules.
Adaptability of KB:
The users of CDSS expect to be able to adapt particular KB
to their own clinical settings. This requirement is essential since (1) clinics must
