Professional Documents
Culture Documents
State of the Art Instead of Biological surement procedures for which the means “the best performance”; can
Variation to Set Requirements for desirable performance (CVM ⬍ we really accept that when CVM ⫽
Imprecision 0.50 CVBw) is not attainable with 0.25 CVBw is the best performance
current technology and methodol- regarding imprecision? In 1980,
To the Editor: ogy. Ralph Gräsbeck wrote “[. . . ] An au-
Various models have been proposed • Optimum performance: thoritative Scandinavian meeting
to set requirements for imprecision CVM ⬍ 0.25 CVBw. Useful for mea- stated this in the following terms:
in clinical laboratory sciences. The surement procedures for which the ‘The analytical precision must be so
idea of establishing a hierarchy for desirable performance (CVM ⬍ good that it is possible to decide
these models based on their objectiv- 0.50 CVBw) is easily achieved with whether a normal intraindividual
ity has been proposed recently (1, 2 ). current technology and methodol- variation or a pathological deviation
Briefly, in the different models, hier- ogy. is present’. To this we should today
archically ordered, the requirement To understand the consequences perhaps add: ‘[. . .] and that also hith-
for imprecision is based on: of applying the three factors that erto undescribed physiological vari-
multiply CVBw, it should be remem- ations are detectable in the individ-
1. the effect of imprecision on clini- ual’ ” (3 ). Keeping in mind this idea,
cal outcomes in specific clinical bered that the total variation of a
measurement (CVT) contains both with which we agree absolutely, im-
situations; precision should ideally be zero; in
2. within-subject biological varia- within-subject biological variation
and metrological variation: other words, the optimum impreci-
tion;
sion is really zero. Consequently, any
3. clinicians’ opinions;
CVT ⫽ (CVBw2 ⫹ CVM2)1/2 requirement for imprecision should
4. experts’ recommendations;
be consistent with such an optimum
5. interlaboratory comparisons orga- From this equation, when CVM ⫽ value.
nizers; and 0.25 CVBw, then CVM is 24.2% of Generally, the degree of metrolog-
6. current state of the art. CVT; when CVM ⫽ 0.50 CVBw, then ical quality that a clinical laboratory
Models higher up in the hierarchy CVM is 44.7% of CVT; and when CVM can achieve is limited by its measure-
are to be preferred to those at lower ⫽ 0.75 CVBw, then CVM is 60.0% of ment systems, which for financial
levels, but the hierarchy may change CVT. reasons should be working through-
depending on the specific clinical Obviously, the selection of these out their lifespans (amortization);
purpose of the measured quantity three factors instead of any others is thus, the requirement for imprecision
(e.g., screening, diagnostics, or mon- not objective and independent of ex-
should be possible to achieve and
itoring) (1, 2 ). ternal situations, but is based on the
updateable. Such requirements can
Notwithstanding this, we think feasibility of the allowable impreci-
be one of the fractiles of the distribu-
there is evidence enough to demon- sion value obtained, that is to say, it
tion of imprecisions from all partici-
strate that the use of the current is really based on the state of the art!
pating laboratories in different exter-
state-of-the-art model should be the The use of this model has addi-
nal quality assessment schemes,
second preferred model, replacing tional weakness: (a) there is great
provided that in these schemes im-
the use of within-subject biological variation among CVBw values (Table
1); and (b) in some cases the within- precision is estimated from all of the
variation for setting requirements for results corresponding to a particular
imprecision for clinical laboratories. subject biological variation is so
small that requirements based on control material. This control mate-
Let CVM be the measurement pro- rial should be measured in each lab-
cedure imprecision (metrological co- such data may not be achieved with
the routine systems of measurement. oratory every day that a run is per-
efficient of variation) and CVBw be formed (4, 5 ). Every country or
the coefficient of variation corre- The conclusion from all of the
above may be that setting require- region can decide which is the most
sponding to within-subject biological appropriate fractile for each quan-
variation. At the moment, the model ments for imprecision from within-
subject biological variation data is as tity. Because science and technology
for setting the allowable imprecision are constantly advancing, the re-
based on within-subject biological arbitrary as setting these require-
ments from a fractile of participating quirements based in the state of the
variation outlines three situations art should be periodically updated
(1 ): laboratories in external quality as-
sessment schemes (daily interlabora- (perhaps every 5 years), and conse-
• Desirable performance: tory quality-control programs), that quently, the overall trend in require-
CVM ⬍ 0.50 CVBw. Useful when is to say, from the state of the art. ments will be toward zero. This fact
the allowable imprecision value One of the three situations pro- is consistent with the goal stated
generated following this inequality posed in the model using within- above: imprecision should ideally be
is viewed as being generally appli- subject biological variation is the so- zero.
cable. called “optimum performance”. With this model, depending on
• Minimum performance: Because optimum is the superlative quantities and external quality as-
CVM ⬍ 0.75 CVBw. Useful for mea- of good, “optimum performance” sessment schemes, requirements
chical structure (2– 4 ). Therefore, we CVwithin-subject values and so generate have put forward a proposal for
find no reason for changing the rank- appropriate quality specifications. ranking the state of the art higher
ing of the state of the art in the Fuentes-Arderiu and Miró- than biological variation in the hier-
hierarchical structure of models to Balagué (1 ) then make the proposal archical structure, but we find no
define analytical quality specifica- for the state of the art (to be placed convincing arguments for such a
tions. higher in the hierarchy than biologi- change.
Fuentes-Arderiu and Miró- cal variation) on the basis that the
Balagué (1 ) have some critical re- analytical quality specifications for References
marks on the use of data on biologi- imprecision should be defined by a 1. Fuentes-Arderiu X, Miró-Balagué J. State of the
art instead of biological variation to set require-
cal variation and argue that the state- fraction of the best performers as ments for imprecision. Clin Chem 2000;46:
of-the-art proposal has only one estimated from external quality as- 1715– 6.
drawback. These postulates deserve sessment data, arguing that the 2. Fraser CG, Hyltoft Petersen P. Analytical perfor-
mance characteristics should be judged
refutation. only drawback is the lack of pub- against objective quality specifications [Editori-
The authors state that the use of lished data. This may be a valid al]. Clin Chem 1999;45:321–3.
the published factors for the three argument if the purpose of the 3. Kaplan LA. Determination and application of
levels of quality, 0.25 for optimum, quality specifications is accredita- desirable analytical performance goals: the
ISO/TC 212 approach. Scand J Clin Lab Invest
0.50 for desirable, and 0.75 for mini- tion or regulation, but not for the 1999;59:479 – 82.
mum quality, are not objective, and ranking in the hierarchical struc- 4. Kenny D, Fraser CG, Hyltoft Petersen P, Kallner
these values are based on the state of ture, which should be based on A. Consensus agreement. Scand J Clin Lab
clinical considerations of quality. In Invest 1999;59:585.
the art (1 ). These values are easy to 5. Fraser CG, Hyltoft Petersen P. The importance
addition, there are several severe
remember and use and are firmly of imprecision. Ann Clin Biochem 1991;28:
drawbacks to their proposal: 207–11.
based on serious investigation of the
influence of analytical imprecision 6. Fraser CG, Hyltoft Petersen P, Libeer JC, Ricós
(a) These estimates of imprecision C. Proposals for setting generally applicable
on the within-subject variation in are usually based on artificial ma- quality goals solely based on biology. Ann Clin
monitoring of patients using clinical terials, which may not reflect the Biochem 1997;34:8 –12.
laboratory results as has been clearly imprecision of patient materials.
demonstrated in the literature and in (b) Usually, these data have already Per Hyltoft Petersen1
tables (5 ) and figures (6 ). Further- been used in internal control and Callum G. Fraser2
more, these values are not state-of- are therefore censored.
the-art values but are used to define (c) The estimates will depend on the 1
Department of Clinical Biochemistry
a method for ranking of the current current methods, instrumenta- Odense University Hospital
analytical imprecision according to tion, equipment, and technology DK-5000 Odense C, Denmark
its influence on such clinical situa- and will vary over time. Fax 45-65-41-19-11
tions. The increases in CVtotal com- (d) Any decided fractile will be com- E-mail phy@imbmed.ou.dk
pared with the inherent biological pletely arbitrary.
variation, CVwithin-subject, are approx- Actually, Fuentes-Arderiu and
2
Biochemical Medicine
imately 3%, 12%, and 25% for the Miró-Balagué (1 ) make three sugges- Ninewells Hospital and Medical School
three rankings, i.e., optimum, desir- tions: Dundee DD1 9SY, Scotland
able, and minimum quality, respec- Fax 44-1382-645333
tively. (a) The goal for imprecision is E-mail callumf@tuht.scot.nhs.uk
The variability of estimated CVanalytical ⫽ 0 (this is, of course,
CVwithin-subject is also suggested to be the best if it could be obtained
a drawback for the hierarchical struc- without additional cost).
ture. It is correct that, for a number of (b) “. . . when a measurement system
quantities, the variances for biologi- for a ‘new’ quantity is developed
cal variation appear to be not homo- . . . , the requirements for impre-
geneous. Moreover, it is stated that cision based on biological varia- Evaluation of a Simple Dot-Blot
the values in the literature are not tion should always be taken into Method for the Detection of Anti-
consistent. Unfortunately, some of account as a guide” (we clearly Neutrophil Cytoplasmic Antibodies
the published investigations on bio- agree with this). Directed against Proteinase 3 and
(c) The fraction of best performers Myeloperoxidase
logical variation are poorly planned
(for which the evidence is
and poorly accomplished, but there
flawed). To the Editor:
is overwhelming evidence that the
real CVwithin-subject values are rather These three proposals are contra- Anti-neutrophil cytoplasmic anti-
alike in different populations. The dictory, and therefore, the real fun- bodies (ANCAs) usually are demon-
art is to select the best or better damental concept of the Letter to the strated by indirect immunofluores-
estimates of biological variation to Editor by Fuentes-Arderiu and Miró- cence (IF), followed by examination
produce more robust estimates of Balagué (1 ) is difficult to grasp. They of patient samples for the antibodies