Letters
State of the Art Instead of Biological
Variation to Set Requirements for
Imprecision
To the Editor:
Various models have been proposed
to set requirements for imprecision
in clinical laboratory sciences. The
idea of establishing a hierarchy for
these models based on their objectiv-
ity has been proposed recently (1, 2 ).
Briefly, in the different models, hier-
archically ordered, the requirement
for imprecision is based on:
1. the effect of imprecision on clini-
cal outcomes in specific clinical
situations;
2. within-subject biological varia-
tion;
3. clinicians’ opinions;
4. experts’ recommendations;
5. interlaboratory comparisons orga-
nizers; and
6. current state of the art.
Models higher up in the hierarchy
are to be preferred to those at lower
levels, but the hierarchy may change
depending on the specific clinical
purpose of the measured quantity
(e.g., screening, diagnostics, or mon-
itoring) (1, 2 ).
Notwithstanding this, we think
there is evidence enough to demon-
strate that the use of the current
state-of-the-art model should be the
second preferred model, replacing
the use of within-subject biological
variation for setting requirements for
imprecision for clinical laboratories.
Let CVM be the measurement pro-
cedure imprecision (metrological co-
efficient of variation) and CVBw be
the coefficient of variation corre-
sponding to within-subject biological
variation. At the moment, the model
for setting the allowable imprecision
based on within-subject biological
variation outlines three situations
(1 ):
• Desirable performance:
CVM ⬍ 0.50 CVBw. Useful when
the allowable imprecision value
generated following this inequality
is viewed as being generally appli-
cable.
• Minimum performance:
CVM ⬍ 0.75 CVBw. Useful for mea-
surement procedures for which the
desirable performance (CVM ⬍
0.50 CVBw) is not attainable with
current technology and methodol-
ogy.
• Optimum performance:
CVM ⬍ 0.25 CVBw. Useful for mea-
surement procedures for which the
desirable performance (CVM ⬍
0.50 CVBw) is easily achieved with
current technology and methodol-
ogy.
To understand the consequences
of applying the three factors that
multiply CVBw, it should be remem-
bered that the total variation of a
measurement (CVT) contains both
within-subject biological variation
and metrological variation:
CVT ⫽ (CVBw2 ⫹ CVM2)1/2
From this equation, when CVM ⫽
0.25 CVBw, then CVM is 24.2% of
CVT; when CVM ⫽ 0.50 CVBw, then
CVM is 44.7% of CVT; and when CVM
⫽ 0.75 CVBw, then CVM is 60.0% of
CVT.
Obviously, the selection of these
three factors instead of any others is
not objective and independent of ex-
ternal situations, but is based on the
feasibility of the allowable impreci-
sion value obtained, that is to say, it
is really based on the state of the art!
The use of this model has addi-
tional weakness: (a) there is great
variation among CVBw values (Table
1); and (b) in some cases the within-
subject biological variation is so
small that requirements based on
such data may not be achieved with
the routine systems of measurement.
The conclusion from all of the
above may be that setting require-
ments for imprecision from within-
subject biological variation data is as
arbitrary as setting these require-
ments from a fractile of participating
laboratories in external quality as-
sessment schemes (daily interlabora-
tory quality-control programs), that
is to say, from the state of the art.
One of the three situations pro-
posed in the model using within-
subject biological variation is the so-
called “optimum performance”.
Because optimum is the superlative
of good, “optimum performance”
Clinical Chemistry 46, No. 10, 2000
means “the best performance”; can
we really accept that when CVM ⫽
0.25 CVBw is the best performance
regarding imprecision? In 1980,
Ralph Gräsbeck wrote “[. . . ] An au-
thoritative Scandinavian meeting
stated this in the following terms:
‘The analytical precision must be so
good that it is possible to decide
whether a normal intraindividual
variation or a pathological deviation
is present’. To this we should today
perhaps add: ‘[. . .] and that also hith-
erto undescribed physiological vari-
ations are detectable in the individ-
ual’ ” (3 ). Keeping in mind this idea,
with which we agree absolutely, im-
precision should ideally be zero; in
other words, the optimum impreci-
sion is really zero. Consequently, any
requirement for imprecision should
be consistent with such an optimum
value.
Generally, the degree of metrolog-
ical quality that a clinical laboratory
can achieve is limited by its measure-
ment systems, which for financial
reasons should be working through-
out their lifespans (amortization);
thus, the requirement for imprecision
should be possible to achieve and
updateable. Such requirements can
be one of the fractiles of the distribu-
tion of imprecisions from all partici-
pating laboratories in different exter-
nal quality assessment schemes,
provided that in these schemes im-
precision is estimated from all of the
results corresponding to a particular
control material. This control mate-
rial should be measured in each lab-
oratory every day that a run is per-
formed (4, 5 ). Every country or
region can decide which is the most
appropriate fractile for each quan-
tity. Because science and technology
are constantly advancing, the re-
quirements based in the state of the
art should be periodically updated
(perhaps every 5 years), and conse-
quently, the overall trend in require-
ments will be toward zero. This fact
is consistent with the goal stated
above: imprecision should ideally be
zero.
With this model, depending on
quantities and external quality as-
sessment schemes, requirements
1715
1716 Letters
Table 1. Examples of the range of the observed means or medians of within-
subject coefficients of variation (CVBw) observed in different studies with a time
span greater than 1 month, and on the range of CVBw observed in 20 healthy
individuals in one study.
Range of CVBw means or Range of CVBw in
Quantity a
medians in different studies (6) 20 healthy subjects (7)
S–Alanine aminotransferase; cat.c. 15.0–57.9 (n ⫽ 6)b 0.0–68.8
S–Calcium(II); subst.c. 0.0–4.1 (n ⫽ 14) 0.0–4.1
S–Cholesterol; subst.c. 3.7–9.2 (n ⫽ 23) 2.6–13.1
S–Protein; mass c. 2.2–4.2 (n ⫽ 12) 1.6–8.0
S–Sodium ion; subst.c. 0.0–1.4 (n ⫽ 10) 0.0–1.6
S–Thyroxine; subst.c. 4.8–10.1 (n ⫽ 6) 0.0–14.4
a
According to IUPAC and IFCC (8 ): S, serum; cat.c., catalytic concentration; subst.c, amount of substance
concentration; mass c., mass concentration.
b
n, number of studies.
may be established for one, two, or oratory medicine. Stockholm April 24 –26,
1999]. Scand J Clin Lab Invest 1999;59:585.
three values of the quantity. 3. Gräsbeck R. Implications of reference values in
The use of state of the art as a deciding upon the degree of tolerable variation in
given fractile of imprecision has the analytical procedures. In: Hørder M, ed. Assess-
ing quality requirements in clinical chemistry.
strategic advantage for regulation Helsinki: NORDKEM, 1980:25–7.
purposes that we can know directly 4. Fuentes-Arderiu X, Alı́a P. Goals for imprecision
in advance the approximate fraction derived from the “state of the art”. Eur J Clin
Chem Clin Biochem 1993;31:543– 4.
of laboratories in a given region that 5. Sebastián-Gámbaro MA, González-de-la-Presa B,
will achieve the requirements. The Fuentes-Arderiu X. An improvement on the crite-
rion of the state of the art to estimate the
only weakness of this proposal is that maximal allowable imprecision. Eur J Clin Chem
it depends on publication of the frac- Clin Biochem 1996;34:445– 6.
tiles of imprecision by the organizers 6. Sebastián-Gámbaro MA, Lirón-Hernández FJ, Fu-
entes-Arderiu X. Intra- and inter-individual biolog-
of external quality assessment pro- ical variability data bank. J Clin Chem Clin Bio-
grams (generalization of this activity chem 1997;35:845–52.
should be encouraged). 7. Juan Pereira L. Variabilitat biològica intraindi-
vidual de les magnituds bioquı́miques. Aplica-
In spite of all of the above, when a cions clı́niques [Doctoral Thesis]. Barcelona,
measurement system for a “new” Spain: Universitat de Barcelona, Facultat de
Farmàcia, 1989.
biological quantity is developed by 8. International Union of Pure and Applied Chemis-
the in vitro diagnostics industry, the try, International Federation of Clinical Chemis-
requirements for imprecision based try. Compendium of terminology and nomencla-
ture of properties in clinical laboratory sciences.
on biological variation should al- [Prepared by Rigg JC, Brown SS, Dybkaer R,
ways be taken into account as a Olesen H]. Oxford: Blackwell, 1995.
guide because, as a general rule, the
smaller the within-subject biological Xavier Fuentes-Arderiu1*
variation, the smaller the imprecision Jaume Miró-Balagué
must be. However, industry, for rea-
1
sons of competition and indepen- Servei de Bioquı́mica Clı́nica
dently of clinical chemists’ wishes Ciutat Sanitària i
and theories, will be motivated to Universitària de Bellvitge
develop measurement systems in 08907 L’Hospitalet de Lobregat
which imprecision will be progres- Catalonia, Spain
sively closer to zero.
2
Laboratori Clı́nic
References Hospital de Viladecans
1. Fraser CG, Hyltoft Petersen P. Analytical perfor- 08840 Viladecans
mance characteristics should be judged against
objective quality specifications. Clin Chem
Catalonia, Spain
1999;45:321–3.
2. Kenny D, Fraser CG, Hyltoft Petersen P, Kallner
A. Consensus agreement [Conference on strat- *Author for correspondence. Fax 34-93-
egies to set global quality specifications in lab- 260-7546; e-mail xfa@csub.scs.es.
Following is a reply to the above letter
by X. Fuentes-Arderiu and J. Miró-
Balagué
To the Editor:
Fuentes-Arderiu and Miró-Balagué
(1 ) put forward ideas for a change in
the hierarchy of models that should
be applied to set analytical quality
specifications for imprecision. This
hierarchical model was based on our
proposal (2 ) and, with minor
changes, was adopted as the basis for
the Technical Report (type 2) cur-
rently being generated by ISO/
TC212/WG3 Report 15196 (3 ) and
was the consensus achieved at the
Conference sponsored by IFCC, IU-
PAC, and WHO in Stockholm in 1999
(4 ).
The change proposed by Fuentes-
Arderiu and Miró-Balagué (1 ) is to
place quality specifications based on
“the state of the art” higher in the
hierarchical structure than those de-
rived from data on “biological varia-
tion”.
It is important for us to be recep-
tive to other ideas and proposals and
assess their relevance to accepted
ideas, so here we restate the objective
basis for the hierarchical structure.
The core concept of the hierarchy is
to relate analytical quality specifica-
tions first and foremost to the use of
laboratory data. Consequently, the
highest hierarchical level is to evalu-
ate specifications directly from the
clinical settings (e.g., clinical strat-
egy, diagnosis, and follow-up). These
calculations are, however, cumber-
some, and clinical approaches and
algorithms may vary over time and
geography, so the next step in the
hierarchical structure is the more
general approach, using data on bio-
logical variation, to set analytical
quality specifications. This is directly
related to the clinical use of reference
intervals (for bias) and serial sam-
pling and measurement, that is, clin-
ical monitoring (for imprecision).
The telling point is that, in contrast,
the state of the art, which is sug-
gested to rank above the use of bio-
logical variation (1 ), is in no way
related to the use of laboratory data
and is, therefore, placed far below
other strategies in the agreed hierar-

chical structure (2– 4 ). Therefore, we
find no reason for changing the rank-
ing of the state of the art in the
hierarchical structure of models to
define analytical quality specifica-
tions.
Fuentes-Arderiu and Miró-
Balagué (1 ) have some critical re-
marks on the use of data on biologi-
cal variation and argue that the state-
of-the-art proposal has only one
drawback. These postulates deserve
refutation.
The authors state that the use of
the published factors for the three
levels of quality, 0.25 for optimum,
0.50 for desirable, and 0.75 for mini-
mum quality, are not objective, and
these values are based on the state of
the art (1 ). These values are easy to
remember and use and are firmly
based on serious investigation of the
influence of analytical imprecision
on the within-subject variation in
monitoring of patients using clinical
laboratory results as has been clearly
demonstrated in the literature and in
tables (5 ) and figures (6 ). Further-
more, these values are not state-of-
the-art values but are used to define
a method for ranking of the current
analytical imprecision according to
its influence on such clinical situa-
tions. The increases in CVtotal com-
pared with the inherent biological
variation, CVwithin-subject, are approx-
imately 3%, 12%, and 25% for the
three rankings, i.e., optimum, desir-
able, and minimum quality, respec-
tively.
The variability of estimated
CVwithin-subject is also suggested to be
a drawback for the hierarchical struc-
ture. It is correct that, for a number of
quantities, the variances for biologi-
cal variation appear to be not homo-
geneous. Moreover, it is stated that
the values in the literature are not
consistent. Unfortunately, some of
the published investigations on bio-
logical variation are poorly planned
and poorly accomplished, but there
is overwhelming evidence that the
real CVwithin-subject values are rather
alike in different populations. The
art is to select the best or better
estimates of biological variation to
produce more robust estimates of
Clinical Chemistry 46, No. 10, 2000
CVwithin-subject values and so generate
appropriate quality specifications.
Fuentes-Arderiu and Miró-
Balagué (1 ) then make the proposal
for the state of the art (to be placed
higher in the hierarchy than biologi-
cal variation) on the basis that the
analytical quality specifications for
imprecision should be defined by a
fraction of the best performers as
estimated from external quality as-
sessment data, arguing that the
only drawback is the lack of pub-
lished data. This may be a valid
argument if the purpose of the
quality specifications is accredita-
tion or regulation, but not for the
ranking in the hierarchical struc-
ture, which should be based on
clinical considerations of quality. In
addition, there are several severe
drawbacks to their proposal:
(a) These estimates of imprecision
are usually based on artificial ma-
terials, which may not reflect the
imprecision of patient materials.
(b) Usually, these data have already
been used in internal control and
are therefore censored.
(c) The estimates will depend on the
current methods, instrumenta-
tion, equipment, and technology
and will vary over time.
(d) Any decided fractile will be com-
pletely arbitrary.
Actually, Fuentes-Arderiu and
Miró-Balagué (1 ) make three sugges-
tions:
(a) The goal for imprecision is
CVanalytical ⫽ 0 (this is, of course,
the best if it could be obtained
without additional cost).
(b) “. . . when a measurement system
for a ‘new’ quantity is developed
. . . , the requirements for impre-
cision based on biological varia-
tion should always be taken into
account as a guide” (we clearly
agree with this).
(c) The fraction of best performers
(for which the evidence is
flawed).
These three proposals are contra-
dictory, and therefore, the real fun-
damental concept of the Letter to the
Editor by Fuentes-Arderiu and Miró-
Balagué (1 ) is difficult to grasp. They
1717
have put forward a proposal for
ranking the state of the art higher
than biological variation in the hier-
archical structure, but we find no
convincing arguments for such a
change.
References
1. Fuentes-Arderiu X, Miró-Balagué J. State of the
art instead of biological variation to set require-
ments for imprecision. Clin Chem 2000;46:
1715– 6.
2. Fraser CG, Hyltoft Petersen P. Analytical perfor-
mance characteristics should be judged
against objective quality specifications [Editori-
al]. Clin Chem 1999;45:321–3.
3. Kaplan LA. Determination and application of
desirable analytical performance goals: the
ISO/TC 212 approach. Scand J Clin Lab Invest
1999;59:479 – 82.
4. Kenny D, Fraser CG, Hyltoft Petersen P, Kallner
A. Consensus agreement. Scand J Clin Lab
Invest 1999;59:585.
5. Fraser CG, Hyltoft Petersen P. The importance
of imprecision. Ann Clin Biochem 1991;28:
207–11.
6. Fraser CG, Hyltoft Petersen P, Libeer JC, Ricós
C. Proposals for setting generally applicable
quality goals solely based on biology. Ann Clin
Biochem 1997;34:8 –12.
Per Hyltoft Petersen1
Callum G. Fraser2
1
Department of Clinical Biochemistry
Odense University Hospital
DK-5000 Odense C, Denmark
Fax 45-65-41-19-11
E-mail phy@imbmed.ou.dk
2
Biochemical Medicine
Ninewells Hospital and Medical School
Dundee DD1 9SY, Scotland
Fax 44-1382-645333
E-mail callumf@tuht.scot.nhs.uk
Evaluation of a Simple Dot-Blot
Method for the Detection of Anti-
Neutrophil Cytoplasmic Antibodies
Directed against Proteinase 3 and
Myeloperoxidase
To the Editor:
Anti-neutrophil cytoplasmic anti-
bodies (ANCAs) usually are demon-
strated by indirect immunofluores-
cence (IF), followed by examination
of patient samples for the antibodies