![]()
26
| 3 | 2004
Klinisk Biokemi i Norden
has a different slope, but if only the two lowest
control points are considered, the dotted line has a
slope comparable with the others, and a geometric
mean close to 10 mg/L, which could be solved by
use of control samples with lower concentrations.
Discussion
Measurements performed according to the ordinal scale
have often been considered as simple yes/no results,
which is not so simple as seen from a quantitative point
of view. In these preliminary presentations based on
relatively few results, it is clear that the change from
negative (0) to positive (1) takes place over a consider-
able concentration range for all these analytes. Further,
within this interval, the result is based on some random
process with a percentage of positive (1) results for each
concentration in control materials.
In external control the results reported from each
participant are independent, and thus makes it pos-
sible to present the fractions of 1 in a rankit-plot.
The dispersion may depend not only on random
events, but also on the nature of the process which
may change just like a buffer and on the skill of the
person performing the test.
This presentation of ordinal data in rankit plots,
allows for characterisation of the different tests in
terms related to ln-Gaussian distributions, such as
geometric mean and coefficient of variation, which
may further, make a basis for description of perfor-
mance of these types of tests. Whether deviations
from straight lines in the plots are due to limitations
for the model or on poor reagents or poor perform-
ers may be disclosed when the final results from
the NORDFOND-project are available in the Spring
2005, where the analytes hCG, haemoglobin, glu-
cose, albumin, esterase and nitrites are investigated
in the co-ordinated Nordic project with a great
number of participants.
This model is excellent for assessment of meth-
ods/kits, whereas assessment of the participants
should be performed with control samples with
concentrations just below the 0.1 %-point and just
above the 99.9 % point. Here this model may help
to find these points for each kit/method. The aim
is, however, to define analytical goals based on the
clinical outcome of the interpretation of the patient
data, as is the case for hCG, where a clear goal is
that the highest reference limit for non-pregnant
women must be measured negative (0).
Reference
1. Hyltoft Petersen P, Sandberg S, Fraser CG,
Goldschmidt H. A model for setting analytical
quality specifications and design of control
for measurements on the ordinal scale. Clin
Chem Med Lab 2000;38:545-51.
??????. Foto: ????????
