43
| 4 | 2011
Klinisk Biokemi i Norden
(
Fortsætter side 44)
propagation of the uncertainty in a sum or a diffe-
rence. Thus, if:
C=A±B
and the uncertainty of
A
and
B
are
u(A)
and
u(B)
,
respectively, then the uncertainty of the sum or
difference is
���������������������������������������������������(7)
Bias by comparing patient samples
At least 20 patient samples are recommended for a
comparison between a “comparative” or reference
procedure and test procedure. Representative patient
samples which concentrations cover the entire measur-
ing interval should be chosen. It is often recommended
to measure the samples in duplicates, at least those of
the comparative procedure.
The preferred comparative procedure is a reference
or definitive procedure but these are few and those
which are available are usually too laborious for the
routine laboratory. The laboratory therefore needs
to select a procedure that can be regarded as con-
ventionally ‘true’. It is thus different from a reference
procedure and may conveniently be called a “mentor”
procedure. The laboratory shall define its performance
and how that is monitored.
The significance of the difference between the
results is estimated using the Student’s
t
-
test for depen-
dent variables.
��������������������������������������������������������������������(8)
where is the average of the differences between the
sample pairs. The degrees of freedom is
n-1
where
n
is the number of pairs.
Student’s
t
-
test is a parametric method and assumes
that the distribution of the differences between the
quantities is Gaussian distributed. If not, a non-para-
metric method should be used, e.g. the Wilcoxon sign
rank test. The Student’s
t-
test is regarded as a relatively
robust test and thus allows some deviation from a true
Gaussian distribution.
The relation between the results of the comparative
and test procedures should be illustrated in a scatter-
gram (Figure 1). If the same quantity is measured by
both procedures, there must be a linear relationship i.e.
a direct first order proportionality between the results.
Any deviation from a perceived linear relation should
be further investigated. Partitioning of the results may
indicate limitations in the measuring interval.
The scatter of the results can be expressed by the cor-
relation coefficient (
r
)
or the coefficient of determina-
tion (
r
2
).
A value of (
r
)
above 0.975 is usually regarded
as reasonable for a method comparison.
Figure 1.
The ordinary linear regression
and the DRA are displayed in this diagram,
together with the equal line, the mean and
the median of the measurement results.
