42
| 4 | 2011
Klinisk Biokemi i Norden
(
Fortsat fra side 41)
The ANOVA approach can also be used to verify the
performance of many instruments in a laboratory or
in many collaborating laboratories. Each instrument/
laboratory would then be regarded as equivalent to a
run in the ANOVA design. The number of observa-
tions in each group may need to be increased.
Evaluation
If the imprecision (
sd
L
)
is smaller than, or equal, to that
claimed, the precision verification is accepted. Howe-
ver, if the precision is larger it may still be acceptable
provided it passes a statistical test. Different tests may
be used, the EP 15 suggests a Chi-square test.
��������������������������������������������������������(4)
where
n
is the number of observations,
s
the found
standard deviation and
σ
the claimed standard devia-
tion. The latter is assumed to have no uncertainty.
The 95% confidence interval for the acceptance of
sd
L
is thus:
������(5)
Trueness and bias
Trueness is the concept and bias its measure. Bias
represents the systematic error of an uncertainty. In
theory, there would be no bias if a correct, traceable
calibrator were used. In the clinical laboratory cali-
brators and patient samples may react differently in
the measurement procedure and the traceability of
a calibrator is no guarantee for trueness
per se
.
The
property of a calibrator that reacting similar to the
test material in different measurement procedures; is
recognized as “commutability”.
2
There are essentially two procedures to estimate bias
≈ compare measurement results with an assigned value
≈ compare sets of patient samples.
Bias by reference materials
Commutable reference materials with known con-
centration and uncertainty shall be used. Verification
usually requires two different concentrations.
The reference material shall be measured repeatedly
to estimate the average value and a representative
SEM
.
The manufacturer shall provide the concentration of
the reference material with its uncertainty (
u(ref)
)
which is equivalent with the standard error of the
mean. If the difference exceeds the uncertainty of the
difference (
u(d)
)
between the target value and the
found value, then the difference is regarded as signi-
ficant and the verification fails. The uncertainty of the
difference is estimated as
������������������������(6)
where the coverage factor (
k
)
expresses the desired
level of confidence, conventionally
k=2
corresponds
to a 95 % level of confidence.
A common question is how many replicates would
be necessary. The formula (6) indicates that the
uncertainty of the difference and thus the detectable
difference cannot be less than the standard uncerta-
inty of the reference material and will approach that
as the number of measured replicates increases. It is
not justified to increase the number beyond reaching
a level that is comparable to the
u(ref)
.
The formula (6) is based on the general rule for
2
VIM (
defines (5.15)
commutability of a
reference material:
property of a reference material, demon-
strated by the closeness of agreement between the relation
among the measurement results for a stated quantity in this
material, obtained according to two given measurement
procedures, and the relation obtained among the measure-
ment results for other specified materials.
Brandgul skålmurkla (Aleuria aurantia) Foto: Henrik Alfthan
