systems, with the exception of ECG, have been
accepted for use by health care professionals
(Shortliffe 1987). In our opinion, there are two
main reasons for this: l) Many systems have tried
to cover too large an area ofmedical knowledge.
Thismeans that the system has useful features but
the level of knowledge is not considerably higher
than that ofahealth careprofessional. 2)TheDSSs
require active use of the user. They are built as
stand-alone PC programs for decision support of
problem cases. The aim has been that clinicians
could use theDSS when they feel uncertain about
a case. However, the major problem here is that
clinicians have no objectivemeans to knowwhen
they do not even realise the problem.Most errors
of clinicians are such that it has never even
occurred to them that there might be a problem.
Errors cannot be avoided by a system that requires
active use by the expert. Therefore it is easy to
understand that stand-aloneDSSs have usuallyhad
little impact on the quality of care.
Ifdecision supportcanbebrought automatically
to the users when needed, we could perhaps avoid
certain errors and reduce costs. However,
aurornatic decision support is muchmore difficult
to build than a stand-alone DSS. In a stand-alone
DSS, it is possible to include a great deal of infor–
mation and numerous references that the user can
search and retrieve. Experts using the DSS will
judge the clinical significance of the information.
The expert alsobecomes farniliarwith the software
and knows the logic behind it.
When decision support is provided automati–
cally by a computerized system, the information
is displayed on paper or on a screen together with
other patient information. Therefore, the space is
usually very lirnited. The decision support should
not destroy the clarityofother patient documenta–
tion. Therefore, themessages should be short and,
at the same time, clear and understandable. For
drug effects on laboratory tests this means thatwe
have to report a short summaryofall the available
information of the drug effect on a laboratory test.
To be able to do so, we have to make our own
conclusion of each drug - laboratory test combi–
nation based on the publications found in the li–
terature and code it in a knowledge base.
We also have to take into account the fact that a
patient record is a juridical document. All texts
Kli"isk Kemi
i
Norden3, 1998
displayed in it are also seenbypatients. Therefore,
thewamings have to be formulared sothat they do
not lirnit the freedom of clinicians to make final
judgements. For example, in an application that
warns about drug interactions, a too strictly
formularedwamingmaymake it difficult to choose
optimal treatment for a certain patient.
DUCODE
It was evident that maintaining a knowledge base
ofdrug effects on laboratory test requires a strictly
systematicway todescribe theknowledge. As free
text, rules are difficult to update and use in an
aurornatic system. The essential features have to
be coded in a way suitable for computerized
processing, but at the same time, understandable
to ahuman expert.For this purpose,we developed
theDrugLaboratory Interference (DLI) code.The
first version was published in 1995 (Grönroos et
al. 1995). During the work some changes were
made in the code.
Nine parameters were selected to describe an
effect of a drug on a laboratory test. Each of these
parameters can obtain different values according
to the knowledge we have of the effect. The para–
meters are:
Nature of effect
Route of administration
Direction and strength of effect
Level ofdocumentation
Sex of patient
Age of patient
Onset of effect
Durationofeffect after stopping themedication
Clinical significance of effect
Note that we have not included the dose of the
drug in the code. Because the number of
publications in which the effect has been exami–
ned using different doses is very lirnited. We have
had some problems with acetosalicylic acid and
corticosteroids in which the used dose differs
greatly according to indication. However, this in–
formation is presenred in the field "Additional in–
formation" that is included for eachDLI codewhen
necessary.
Coding the DLI knowledge base is a slow and
difficult task. We cannot code each single
publication to the code because the information in
95
