40
| 4 | 2011
Klinisk Biokemi i Norden
On the comparison and verification of
measurement results
Anders Kallner
Department of Clinical Chemistry
Karolinska University Hospital, Stockholm, Sweden
Introduction
We would expect, as axiomatic,
that a value obtained for a given
quantity that is measured in one
and the same sample should
be the same, irrespective of the
procedure/instrument/laboratory.
Unfortunately, reality is different.
Instead we recognise the need to define “compara-
bility” of results and talk about harmonization of
measurement procedures as a proactive strategy.
Comparability is understood as the closeness between
results obtained in different laboratories/instruments
and has become more important and interesting as
mobility – or volatility – of patients and health care
workers increase and laboratories grow larger. The
difference between results is often called bias even
if bias formally is estimated in relation to results
obtained by a reference procedure and represents the
systematic error – i.e. the trueness. In estimating the
comparability we need to consider also the precision
of the measurement.
Ironically, as a consequence of an improved preci-
sion in routine measurements, the demand on trueness
has increased; the better precision, the smaller diffe-
rences between results can be observed. This can for
instance be illustrated by the formula for calculating
the Student’s t-value for independent datasets:
������������������������������������������������������(1)
Clearly, the
t
-
value will be proportional to the diffe-
rence between the averages and inversely proportional
to the square root of the sum of the squared standard
error of the means (
SEM
);
estimated as the impreci-
sion of the results (variance =
sd
2
)
and the number of
observations (
n
).
Consequently, if the imprecision (
sd
)
is small and the number of results is large the denomi-
nator will be small and even small differences between
means will indicate a statistical significant difference
(
a large
t
-
value). Therefore high precision will allow
identification of small differences in the means.
Verification is the demonstration that a measure-
ment procedure, or measuring system, fulfils specified
requirements whereas validation is verification that the
given requirements are fit for the intended purpose.
When manufacturers verify specifications of mea-
suring systems they often follow the CLSI EP5 [1]
and EP9 [2]documents. It is of some merit that simi-
lar strategies and statistical procedures are used by
the laboratories in their verification. Verifications in
the routine laboratory maybe simplified as described
in the CLSI EP15 [3]which addresses precision and
trueness. Other properties that are important are for
instance detectability, linearity, interferences, sensiti-
vity and specificity but these are, with few exceptions,
not usually verified in the routine laboratory and not
within the scope of this presentation.
Uncertainty
The concept of uncertainty was introduced in 1993
and has gained wide acceptance in all disciplines
dealing with measurements. The principle is simple;
all sources of uncertainty is identified, quantified and
then combined. The uncertainty can also be estima-
ted from the performance of the complete procedure.
This concept not only allows, but requires, that any
known and significant bias is eliminated. The success
of an elimination of bias is associated with an uncer-
tainty and this needs to be added when estimating the
combined uncertainty. Results are reported as a “best
