7
Biological Variation
ABSTRACT
Background
There are many sources of variation in numerical results
generated by examinations performed in laboratory medi-
cine. Although some measurands have biological variations
over the span of life and others have predictable cyclical varia-
tion, most measurands have random variation around
homeostatic setting points, which differ between individuals.
Knowledge of the generation and application of data on
within-subject and between-subject variation is essential for
the correct interpretation of results.
Content
In this chapter, we explain that numerical estimates of ana-
lytical, within-subject, and between-subject biological varia-
tion are generated by examination of a series of specimens
taken from a cohort of individuals, which is then followed
by statistical analysis of the sources of variation. Databases
of estimates are available that facilitate applications. The
NATURE OF BIOLOGICAL VARIATION
There are many sources of variation that contribute to the
uncertainty of any result generated in laboratory medicine.
Biological variation is one of the most important and should
be taken into account in any interpretation made. In the
previous edition of this textbook, a footnote to the section on
Biological Variability stated: the author has based much of
the discussion on a monograph (and) this source should be
consulted for further details.1 This still holds true, although
there has been considerable progress in this field, which will
be particularly highlighted in this chapter.
There are various types of biological variation. The con-
centration or activity of some measurands changes over the
span of life, some slowly and some more quickly, particu-
larly at times of rapid physiologic development, such as the
neonatal period, childhood, puberty, menopause, and old
age. The concentration or activity of measurands can also
differ between men and women. This variation is taken care
of by the creation of age- and/or sex-stratified (partitioned)
reference intervals when these are needed, although a dis-
advantage is that age-stratified reference intervals are based
on chronological age rather than biological age. A number
of measurands have predictable cyclical rhythms in their
Callum G. Fraser and Sverre Sandberg
individuality of a measurand and the use of conventional
population-based reference intervals are determined by
comparison of the within-subject and between-subject
biological variations. Within-subject biological variation
and analytical imprecision can be used to create reference
change values (RCVs) to assess the statistical significance of
changes in serial results from an individual or the probability
that any change seen is significant. Analytical performance
specifications for imprecision, bias, total error allowable,
measurement uncertainty, and other characteristics can be
created using within-subject and between-subject variation.
The data have many other uses. Currently, there are concerns
regarding the robustness of some data and estimates for
some measurands that show considerable heterogeneity. We
support recent recommendations on the evaluation, genera-
tion, and application of biological variation data. If followed,
more evidence-based data on biological variation will be
published.
concentrations. These can be daily, monthly, or seasonal
in nature. The major ramifications for interpretation are
that reference intervals cannot be generated for every point
during cycles, knowledge of the expected values throughout
the cycle is vital for clinical interpretation, specimen collec-
tion must be at appropriate times, and absence of rhythm
may indicate disease. These types of biological variations
have been described in detail in Chapter 5, and the strati-
fication (or partitioning) of reference intervals is explained
in Chapter 8.
The most important type of biological variation is ran-
dom biological variation. As an example, four specimens
were taken from four individuals at daily intervals, and
serum sodium activity was examined (reference interval:
135–147 mmol/L). The results are provided in Table 7.1. It
is evident that the results for each individual vary from day
to day; this is due to three sources of variation, namely, pre-
analytical, analytical, and within-subject biological variations.
The mean value is termed the homeostatic setting point. In
addition, each individual has a different average serum so-
dium activity; the variation among the homeostatic setting
points of individuals is the between-subject variation that can
translate into reference intervals, whereas the average varia-
tion within each individual is the within-subject variation.
157
158 SECTION I Basics of Laboratory Medicine
TABLE 7.1 Serum Sodium Activity
in Four Specimens Collected at Daily
Intervals From Each of a Cohort of
Four Individuals
Sodium Day 1 Day 2 Day 3 Day 4
Individual 1 137 139 136 138
Individual 2 144 146 145 144
Individual 3 141 143 142 140
Individual 4 139 138 141 140
Values are measured in millimoles per liter.
Generation and subsequent application of numerical data
on the components of biological variation are crucial facets
of laboratory medicine, and both of these are described in
detail in this chapter.
TERMINOLOGY
There is little doubt that global harmonization in laboratory
medicine needs to go beyond the examination phase and
must include all steps within the total process, including ter-
minology, symbols, and units. Unfortunately, the range of
terms and variety of symbols used to define the components
of biological variation has grown with the increasing body of
literature, which undoubtedly causes confusion. A recent
study that investigated papers on biological variation in the
13 most highly cited journals of laboratory medicine found
that, from 2009 to 2013, 62 papers contained terms and
symbols for components of biological variation. There were
68 terms and 25 symbols for the components applicable to
individuals and 47 terms and 18 symbols for the component
applicable to groups of individuals.2 It was proposed that the
following terms and symbols should be used, because they
were mostly applied and were also suggested by Fraser and
Harris in their highly cited review of 19893:
• CVI: within-subject biological variation (variation
within a single individual estimated as a pooled varia-
tion from a [homogenous] group of individuals)
• CVG: between-subject biological variation (varia-
tion between the central tendencies of a group of
individuals)
• CVA: analytical variation (analytical imprecision).
These terms and symbols are used throughout this chapter,
and the other recommendations in the publication by Fraser
and Harris3 are followed. This work has been supported by a
prestigious, high-impact journal of laboratory medicine: the
journal now recommends that both authors and referees
comply with the use of these terms and symbols.4
GENERATION OF DATA ON COMPONENTS
OF BIOLOGICAL VARIATION
Production of data on biological variation is very similar
to derivation of population-based reference intervals (see
Chapter 8) except that, instead of one specimen being taken
from a large number of reference individuals, a number of
specimens are taken from a smaller cohort of reference indi-
viduals. In general terms, the traditional approach recom-
mends that numerical estimates of CVA and both CVI and
CVG components of biological variation should be generated
using the following experimental approach: select a group
of reference individuals (usually apparently healthy volun-
teers); take a set of specimens from each of the individu-
als at regular time intervals while minimizing all sources of
pre-examination variation in preparation of the subjects for
collection, and the collection and handling and transport of
specimens; store the samples derived from the specimens
until ready for examination; and undertake the examination
in duplicate while minimizing examination sources of varia-
tion. Then, after removal of statistical outliers and confirma-
tion that all results from the individuals are homogenous,
dissect out the CVA, CVI, and CVG components using nested
analysis of variance. This approach is described in detail in
the review by Fraser and Harris,3 where it is stated that “the
components of variation can be obtained from a relatively
small number of specimens collected from a small group of
subjects over a reasonably short period of time,” although
good evidence for this subjective statement was not provided
and has been lacking until recently.5
Design of Studies
The preceding general design has been widely used and is
very suitable for those measurands that have low CVI and
tight homeostatic control, but it is somewhat simplistic for a
number of reasons. The measurand may be unstable, and
examinations must be performed soon after the collection of
specimens (eg, for some hematological measurands, such as
mean cell volume, and numbers of erythrocytes and leuko-
cytes per volume). In this case, to obtain the necessary statisti-
cally unconfounded estimate of CVI, the CVA can be estimated
by analyzing all of the samples in duplicate, but this is a
within-run CV. Thus, quality control materials have to be
analyzed between each run during the examinations to ascer-
tain that variance due to systematic deviations in the analyti-
cal procedure between each examination is not introduced.
If the CVA estimated from examination of the duplicates is
less than the CVA estimated from the between-run control
material, the CVI might be overestimated. Using this strategy,
it must be assured that the concentrations or activities of the
quality control materials are similar to those of the specimens
from the subjects studied because CVA often varies with con-
centration or activity. Moreover, it must be assured that the
analytical variation of the examinations of specimens from
the individuals and the quality control materials are not sig-
nificantly different. This can be assessed by examining a
number of the specimens from the individuals in duplicate,
calculating the analytical SD, using the formula: SD = [(sum
of differences between duplicates − Σd)2/2 × number of pairs
− n]1/2, SD = (Σd2/2n)1/2, and then comparing this SD with
that obtained with the quality control materials, using the
F-test for comparison of variances.
There is a school of thought that most biological data are
naturally logarithmically distributed, and if this is the case,
the calculations must be performed on the natural logarithms
of the observations to make the data distribution closer to
normal.6 Furthermore, it may be that the measurand is not
present in matrices from apparently healthy subjects (such as
unusual proteins found in myeloma) or that it would be
unacceptable or unethical to collect specimens from the indi-
viduals in whom the measurand is most interesting, such as
children. In such cases, specimens could be collected from
 CHAPTER 7 Biological Variation 159

patients with stable disease, as discussed later and described
in recent reviews and editorials.7-9
Estimates in laboratory medicine are usually accompanied
by confidence intervals (CIs); this is rarely done in reports
that provide estimates of the components of biological varia-
tion. The determination of CIs for different balanced designs
for a two-level nested variance analysis model with varying
analytical imprecision has been examined in detail.5 Data sets
based on the model were created to calculate the power of
different study designs for estimation of CVI. It was found
that the reliability of an estimate for CVI and the power are
greatly influenced by the study design and by the ratio
between CVA and CVI. For a fixed number of measurements,
it is preferable to have a high number of specimens from each
individual. If the CVA is high compared with the CVI, the
number of replicate examinations should be increased. The
study provided tables that indicated the effects of increasing
the number of individuals studied and the number of repli-
cates at different levels of imprecision. This work is manda-
tory reading before studies generating data on the components
of biological variation are performed. In addition, estimates
of the components of biological variation should always be
reported with CIs.
Methods for Data Analysis
The currently accepted best method for data analysis is that
described by Fraser and Harris.3 Duplicate examinations are
performed on specimens from a cohort of individuals.
However, before any estimation of components of variation
are made, it is important to examine the data set for outliers,
that is, values that do not belong to the set. This assessment
of outliers is important because this process might detect
either contamination of specimens or an error in the analysis
(eg, insufficient sampling); such aberrant values will lead to
erroneous inflated values. This assessment is done at three
levels. First, the duplicate variances from the individuals
are examined; each variance is calculated by squaring the
difference between the duplicate results and dividing by two
(as per the preceding formula for calculation of the SD of the
set of duplicate results). To decide whether an extreme value
is different from the remainder of the distribution, the ratio
of the maximum variance to the sum of the variances is cal-
culated, and the Cochrane test is applied. This assumes that
each variance is based on an equal number of observations
(two in this case), and that only one variance appears to be
an outlier. Failure to examine for outliers of the duplicates
can result in a falsely high CVA and a falsely low CVI, with
wide CIs. Second, again using the Cochrane test, outliers in
the variances of the specimen mean values are examined to
assess whether any duplicate values are different from the
remainder; this is vital—because if the difference between
any duplicate results is not significantly different from the
rest, it does not mean that both duplicates are not different
from those from the other individuals. Although the Cochrane
test of the variances of the mean values of the specimens is
primarily an outlier test, it can also be used as a simple alter-
native for a homogeneity test, such as Bartlett’s test (see later).
Failure to examine for outliers in the mean results from each
individual can result in a falsely increased CVI. Finally, outli-
ers among the mean values of the individuals are assessed; a
simple strategy to perform this process is to use Reed’s
criterion—that the difference between any mean value and
the next value in the series should be less than one-third of
the overall range of all values. In the assessment of outliers,
one of the most useful tools to use is a simple graphical
approach in which the mean values and the range of all these
values are plotted for each individual (on the y-axis) against
concentration or activity (on the x-axis); an example is pro-
vided in Fig. 7.110 and discussed later in this chapter. Failure
to exclude outliers of the mean values will result in a falsely
large CVG, and because the mean value will be different, it will
also affect the CVI. After exclusion of the outliers, a nested
analysis of variance is used to derive the components of ana-
lytical and biological variations.

27
26
25
24
23
22
21
20
19
18
17
16
Subject

15
14
13
12
11
10
9
8
7
6
5
4
3
2
1
60 70 80 90 100 110 120 130 140 150
Creatinine (μmol/L)
FIGURE 7.1 Means and extreme values for serum creatinine in 27 older adults.33 Note: 100 μmol/L
= 1.13 mg/dL.
160 SECTION I Basics of Laboratory Medicine
Many have not used formal analysis of variance but instead
have used simply subtracted variances. The thesis is that
because pre-examination sources of variation have been min-
imized and can be considered negligible, the total CV (CVT)
of a set of results from each cohort of individuals includes
CVA, CVI, and CVG. Then, because
CVT = [(CVA )2 + (CVI )2 + (CVG )2 ]1 2
the components can be calculated. Outliers are often not
looked for, and a normal distribution is assumed, but the
detection of the former and the checking of the latter assump-
tion (using Kolmogorov-Smirnov or Anderson-Darling or
other techniques for assessment of normality) should now be
considered mandatory. Usually what is done in practice is
that, first, the results from each individual are taken and the
mean, SD, and CV are calculated. This CV (CVB) is comprised
of CVA and CVI:
CVB = [(CVA )2 + (CVI )2 ]1 2
Overall, CVA is then calculated from the duplicate or
replicate assays, or from assessment using quality control
materials (with the previously noted caveats). Then, by
simple subtraction, an estimate of CVI = [(CVB)2 − (CVA)2]1/2
is generated for each individual. If replicates are used and
the mean of the values calculated for subsequent examina-
tion, then the analytical imprecision is reduced by n1/2, where
n is the number of replicates and the correct formula is
CVI = [(CVB)2 − (CVA)2/n]1/2. This is not required if only the
first result is used, but this also appears to be a waste of half
of the results generated. Then an overall estimate of the CVI
is calculated by taking the individual CVI, squaring all of
these, adding the squares (the variances), dividing by the
number of subjects and taking the square root; thus, average
CVI = (ΣCVI2/n)1/2. Following this, then CVG can be generated
using the formula:
CVG = [(CVT )2 − {(CVA )2 + (CVI )2 }]1 2.
Many publications dealing with the generation of esti-
mates of the components of biological variation do not
dissect out CVA and simply report the previously noted CVB,
which is [(CVA)2 + (CVI)2]1/2, as the “within-subject biological
variation”; this is clearly incorrect. Pure unconfounded esti-
mates of the components of analytical and biological varia-
tion are required so that they can be applied correctly, as
described in the following.
Homogeneity and Heterogeneity
The calculation of biological variation data assumes that the
individuals examined are in a “steady state,” that is, the mea-
surand does not change during the time span of the study.
Moreover, data on within-subject biological variation can be
applied, particularly for estimation of RCVs, only if the esti-
mates are homogeneous and do not show heteroscedasticity.
If the data are not homogenous, the results are not represen-
tative of the population, and ubiquitous application of the
estimates is fraught with difficulties. Although estimates of
CVI and RCVs can clearly be calculated, these cannot be
generalized for the entire population. It is therefore impor-
tant to know that the variances of samples drawn from one
and the same population are ‘‘homogeneous’’ by definition,
and consequently, the ranked cumulative distributions of
these variances are distributed around the true variance of
the population according to χ2/df (χ2 distribution for degrees
of freedom [df] according to the individual sample sizes). In
contrast, when a series of different variances have a dispersion
around the pooled variance according to a χ2/df distribution,
they are considered to be heterogeneous. This can be illus-
trated by plotting the cumulated ranked fractions of within-
subject variation values as a function of the within-subject
variation estimates on a rankit scale. If homogeneous, this
curve will fit to the theoretical of the square root of the
pooled variance times χ2/df. Variance homogeneity can be
tested further by Bartlett’s test.6,10
Alternatively, an index of heterogeneity (IH) has been pro-
posed, and the simple mathematical estimation of this and
its interpretation are given in the review of Fraser and Harris.3
The IH provides a means of determining whether individuals
within a population have similar within-subject variation for
a given analyte. It is defined as the ratio of the overall CV of
the (SDA2 + SDI2)1/2 of the subjects to [2/(n − 1)]1/2, where
SDA and SDI are the examination and within-subject biologi-
cal variations as standard deviations, and n is the number of
specimens per subject. The higher the index of heterogeneity
is, the greater the heterogeneity of within-subject biological
variation. It is important to test for homogeneity (eg, with
the Bartlett or Cochrane test) and indicate how many indi-
viduals have had to be removed to obtain homogeneity of the
estimate of CVI. This again would provide an indication of
the representative nature of the data and underscore its suit-
ability for wide application.
Many data have been generated over the last 45 years on
the components of biological variation in a broad range of
measurands. Before considering the uses of the existing data,
it is necessary to consider their reliability. A small number of
the measurands that have been assessed have had a number
of studies done, which has allowed a few reviews to be per-
formed on the robustness of estimates; examples are blood
glycated hemoglobin (HbA1c),11,12 serum C-reactive protein
(CRP),13 and three serum enzymes.14 Published studies on the
biological variation of HbA1c were examined to check the
consistency of the available data to accurately define analyti-
cal performance specifications; the authors found nine studies
and considered that these were limited in a number of ways,
including choice of analytical methodology, population selec-
tion, protocol application, and statistical analyses.12 A similar
evaluation of the 11 available studies on CRP again found
deficiencies in all aspects of the generation of the estimates,
and only 1 study fulfilled all major preexamination, examina-
tion, and postexamination requirements.13 A search of the
literature found 10 publications with data on the components
of biological variation of alanine aminotransferase (ALT), 14
on aspartate aminotransferase (AST), and 9 on γ-glytamyl
transferase (GGT).14 The protocols used for the derivation of
the components were varied. The ranges of CVI reported were
11.1 to 58.1% for ALT, 3.0 to 32.3% for AST, and 3.9 to 14.5%
for GGT. The range of values is shown in Fig. 7.2. Available
CIs are shown: the dark diamond is the estimate in the current
database. The median values (ALT: 18.0%, AST: 11.9%, and
GGT: 13.8%) were, possibly as expected, similar to those
listed in a database commonly used as a reference source.
These three studies, and other similar studies, suggest that

60
Within-subject variation CV, %
50
40
30
20
10
0
ALT AST GGT
FIGURE 7.2 Estimates of within-subject biological variation
for three enzyme activities as components of variations [CV]
(%). ALT, Alanine aminotransferase; AST, aspartate amino-
transferase; GGT, γ-glutamyl transferase. (From Carobene A,
Braga F, Roraas T, et al. A systematic review of data on biological
variation for alanine aminotransferase, aspartate aminotransfer-
ase and γ-glutamyl transferase. Clin Chem Lab Med 2013;51:
1997–2007.)
there are some concerns regarding the usefulness of the cur-
rently available data.
DATABASES
Current Databases and Their Merits
and Disadvantages
As stated previously, derivation of the components of biologi-
cal variation is not without difficulties and requires expendi-
ture of considerable resources of various types, so the need
for a database of published information found in the litera-
ture was perceived some years ago. In view of the concern
over some of the data available on the components of biologi-
cal variation, examination of these available databases is a
necessary prerequisite to their use. Compilations of data were
generated,15-17 but these simply provided lists of published
data. It was believed that publication of a database giving one
value for each measurand for which data were available, based
on the objective assessment of the reliability of the data in
the literature and updated regularly, would be of major
benefit. Generation of such a comprehensive database was
initiated in 1997 by the Analytical Quality Commission of the
Spanish Society of Clinical Chemistry (SEQC), and it was
first published in 1999.18 An update of this table is made
available biannually on the Internet,19 and the information
documented contains a brief introduction concerning the
aims of provision of the database, the changes that have been
made compared with the previous edition, and three impor-
tant appendices, namely, a list of the measurands studied with
the within-subject and between-subject components of bio-
logical variation. These are expressed as CV (CVI and CVG,
respectively), the derived examination quality specifications
for imprecision, bias, and total allowable error at three differ-
ent levels (desirable, minimum, and optimum), the number
of publications examined for each measurand, a list of the
publications examined by measurand, and documentation of
the full citation for every publication. Full details of the struc-
ture and derivation of the database have been recently pub-
lished,20 including the criteria for inclusion of published
CHAPTER 7 Biological Variation 161
estimates into the database. The criteria for acceptance state
that the ratio CVA/0.5 CVI, originally called the index of fidu-
ciality,21 must be less than 2.0, which ensures that the esti-
mates of CVI are not confounded by CVA.
The database, which has been much cited and widely
used, has a number of advantages. CVI has been determined
for 358 measurands in 247 articles, and this large resource has
been developed and refined over nearly 20 years. Data are
available on measurands in a number of matrices, namely,
serum (n =185), plasma (n = 74), whole blood (n = 55), and
urine (n = 47). In addition, as stated previously, data are
systematically updated every 2 years and made available on
the Internet.19 Moreover, an introduction informs of changes
made to existing estimates and information on included new
measurands.
The disadvantages include lack of data for many measur-
ands of interest in laboratory medicine. There is a paucity of
new publications available each year for inclusion in the data-
base, with only 25% published since 2000; few data are docu-
mented for some measurands because 202 were found in a
single publication, 129 had data from 2 to 9 publications, and
27 had data from 10 or more publications. Further, as dis-
cussed previously, there appears to be heterogeneity for some
measurands, including blood HbA1c,12 serum CRP,13 AST, ALT,
and GGT,14 prostate-specific antigen,22 and urinary albumin.23
However, it must be realized that these data are estimates, and
as such, it should not be expected that they will be identical
across studies in numerical terms, but they will have a distri-
bution. Therefore, in future, CIs for the estimates should be
provided to allow for comparison of these. In the current
database, in which most of the published papers have no CIs
given, the robustness of the data is difficult to assess objec-
tively. However, this was examined in the current database by
calculation of the ratio of the maximum CVI to the minimum
CVI (CVI max/CVI min), and it was considered that a ratio
below seven indicates robustness. This criterion was achieved
or surpassed by 86% of the measurands included in the
database.20 It has been suggested, in an editorial about the
database, that it is difficult to use the data for setting perfor-
mance specifications or RCVs if the estimates can vary with
a factor of seven, and that a ratio of no more than two is more
than likely to indicate a significant difference between the
estimates.6
It is generally assumed that the data on the components
of biological variation are robust, and that the estimates are
representative for the specific population and setting in which
they will be applied. In view of the growing concerns about
the apparent heterogeneity of the estimates, which is proba-
bly due in part to the less than ideal methods used in many
of the studies, and therefore, their general applicability,
an Expert Working Group on Biological Variation of the
European Federation of Clinical Chemistry and Laboratory
Medicine (EFLM) has produced a checklist to enable stan-
dardized assessment of existing and future publications of
biological variation data.24 The checklist identifies key ele-
ments to be reported in studies to enable safe, accurate, and
effective transfer of biological variation data sets across
healthcare systems. The checklist is mapped to the domains
of a minimum data set required to enable this process.
Following this work, a new EFLM Task and Finish Group has
evolved the checklist into a practical tool that will enable
existing studies to be classified according to how well the
162 SECTION I Basics of Laboratory Medicine
work fulfils all the required attributes. In addition, work is
already in progress to examine the data in the existing data-
base; a new database with high-quality estimates generated in
studies that fulfill the criteria laid down will be generated and
made available on the EFLM website.25 Moreover, compliance
with the checklist for new studies will enable authors, review-
ers, and journal editors to ensure that studies are fit for
purpose, appropriately powered, share common terminology,
and deliver robust estimates of CVI and CVG, accompanied
by the key metadata required to enable valid application of
the data described.26 It is anticipated that this ongoing work
will define a standard for the reporting of studies on biologi-
cal variation akin to the well-known standard for reporting
of studies on diagnostic accuracy (STARD),27 and will mean
that only studies accompanied by a complete checklist will be
considered acceptable by reviewers and editors. It is hoped
that this standard will be included in the requirements docu-
mented in instructions for authors.
Biological Variation in Health and Disease
Some argue that many of the data on the components of
biological variation are inappropriate for wide use in labora-
tory medicine because they have been derived, in general,
from studies on healthy individuals, and not on patients with
disease who are the source of most requests for examina-
tions.28 A database on the components of biological variation
in disease has been published29; the group from the Analytical
Quality Commission of the SEQC have continued to collect
such data and have prepared an update that has recently been
made available on the Internet.30 The 2014 database has infor-
mation on 97 measurands in subjects with 41 disease states.
This work has led to further evidence that estimates of CVI
are generally independent of the state of health, except when
the measurand is one that is pathologically changed, such as
tumor markers in patients with cancers.
It is generally assumed that CVI is independent of age,
sex, time span of study (unless very frequent specimens are
collected when serial correlation of data exist), geograph-
ical area, health, and disease, as well as the measurement
procedure used. This is hardly surprising because CVI are
numerical estimates of the homeostatic mechanisms of the
measurands.
As a consequence, it can be considered that estimates of
the components of biological variation should be relatively
easily available and can be used in a large number of applica-
tions fundamental to laboratory medicine. However, because
the quality of published papers varies, it is strongly recom-
mended that all users investigate the sources and quality of
the data selected before application in their individual labo-
ratories (eg, by using the proposed check list).24
Generation of Data From Patient Populations
Because estimates of CVI, a measure of the homeostatic
mechanisms in individuals, seem constant in general, this
characteristic can be used to determine this component of
biological variation in difficult settings. For example, estima-
tion of CVI of HbA1c was done in specimens taken routinely
from children with cystic fibrosis who had no evidence of
diabetes mellitus or impaired glucose tolerance.31 Similarly, it
would be difficult to justify taking multiple samplings from
apparently healthy individuals for examinations such as arte-
rial pH, gas partial pressures, and electrolytes; derivation of
CVI from results of a number of examinations in diseased
individuals performed regularly in the intensive care unit
provided reliable estimates.32 These interesting examples
showed that, although it might be ideal to undertake the well-
documented, previously noted experimental protocols and
statistical analyses, further studies using novel approaches
such as these might be required for examinations for which
the generation and application of traditional estimates of CVI
are difficult.
INTERPRETATION AND USE OF DATA
Individuality
The results of most examinations in laboratory medicine are
compared with conventional population-based reference
intervals or sometimes fixed clinical decision-making limits.
This is mandatory when previous results are unavailable, as
is often the case in clinical settings of diagnosis, case finding,
and screening. However, reference intervals represent the
values found in a fractile (usually 0.95) of the reference popu-
lation rather than the values found in a single individual. The
ramification of biological variation on the use of reference
intervals is determined by the individuality of the measurand;
this has been explained in detail.33 The example therein is
reproduced in part here.
Fig. 7.1 shows a graph, which as stated earlier, should be
prepared by all who are generating data on the components
of biological variation to assess the presence of outlying
observations visually. This shows the means and extreme
values on a cohort of 27 older adults for serum creatinine
concentration. Subjects 1 to 13 were women and subjects 14
to 27 were men. The conventional reference intervals for cre-
atinine in individuals older than 55 years of age generated in
the laboratory were 60 to 98 μmol/L (0.68–1.10 mg/dL) for
women and 66 to 128 μmol/L (0.75–1.45 mg/dL) for men. As
documented previously,10 analysis of the data in Fig. 7.1
allows the following conclusions to be drawn:
• No individual has observed values that span the entire
reference interval, and the range of values from each
individual occupies only a small part of the dispersion
of the reference interval.
• Most individuals have all observed values within the
reference interval.
• The means of the observed values of most individuals
lie within the reference interval, but they are different
from each other.
• A few individuals have observed values that span the
lower reference limit, and these individuals have values
that change from usual to unusual over time.
• A few individuals have observed values that span the
upper reference limit, and these individuals also have
values that change from usual to unusual.
It is clear that the CVI of creatinine (the variation around
the homeostatic setting points) is smaller than the CVG (the
difference among homeostatic setting points). In numerical
terms, CVI was 4.3% and CVG was 18.3%. Thus, CVI is less
than CVG, and in such situations, the measurand is said to have
marked individuality. This characteristic can be expressed
mathematically as an index of individuality (II) and is best
calculated as the ratio of examination plus within-subject
biological variations to between-subject biological variation,

mathematically: II = (CVA2 + CVI2)1/2/CVG. However, it is
common now for II to be simply calculated as CVI/CVG.
This is satisfactory if CVA is less than 0.5CVI, as is often the
case with modern analytical technology and methodology,
because CVA will then contribute little analytical noise to
the numerator (CVA2 + CVI2)1/2. Calculation of II from the
most recent database shows that most commonly examined
measurands have a low II, meaning that they have marked
individuality.
This example provides a biological explanation for the fact
that serum creatinine concentration, compared with conven-
tional reference intervals, even if partitioned by age or sex,
does not have high sensitivity for the detection of mild renal
impairment, and provides the reason for the undoubted ben-
efits of estimated glomerular filtration rate. The estimated
glomerular filtration rate uses formulas that take age, sex, and
ethnicity into account. This example also provides a sound
rationale for the well-established fact that most measurands
examined in laboratory medicine are not very useful in popu-
lation screening or in case finding.
Consequences for Population-Based
Reference Intervals
The consequences of individuality were first postulated by
Harris34,35 who showed that, when CVI/CVG is high (the cri-
terion usually applied is CVI/CVG >1.4), the distribution of
values from any single individual will cover much of the
entire dispersion of the reference interval derived from values
found in reference individuals. In contrast, if CVI/CVG is low
(especially when CVI/CVG is <0.6), the dispersion of values
for any individual will span only a small part of the conven-
tional population-based reference interval.
The ramifications of this individuality on the interpreta-
tion of the results of examinations are profound. When II is
low, most individuals can have values that are unusual for
them, but these will often lie within the reference interval. As
a consequence, these results would not be flagged by labora-
tories as deserving of further attention because, although they
are unusual for that individual, they are within the reference
interval. Moreover, the users of the laboratory results would
be highly unlikely to pay attention to such unusual values.
Thus, taking one specimen from an individual and compar-
ing the result of an examination with the population-based
reference interval will not, as shown previously for creatinine,
be an effective way of picking up the small changes often seen
in early pathological processes. However, when only one
sample is examined in an individual, the II will have no influ-
ence on the percentage of false positives and true positives
detected, irrespective of whether the upper reference limit or
a selected clinical decision-making value is used. However, if
a “confirmatory” measurement is performed, the II is of
importance. For quantities with a very low II, which is the
usual situation in laboratory medicine, a new result of mea-
surement will be close to the first and only provide limited
new information. For quantities with a high II, a repeat mea-
surement will decrease the number of true positives and false
positives. In a low prevalence situation (eg, in screening and
case finding), in which it is important to prevent healthy
individuals being incorrectly labelled, a positive result will
“confirm” the first. In a relatively high prevalence situation
(eg, in diagnosis) in which the number of false positives is
CHAPTER 7 Biological Variation 163
TABLE 7.2 Within-Subject (CVI)
and Between-Subject (CVG) Biological
Variation of Urine Creatinine and Indices
of Individuality (II)
Group CVI (%) CVG (%) II
Men (n = 7) 11.0 6.0 1.83
Women (n = 8) 15.7 11.0 1.42
Total 13.0 28.2 0.46
low, and it is important to discover most of the diseased
patients, the measurement should not be repeated. Thus, the
only clear reason for a “confirmation” measurement is in a
low prevalence situation when the II is high; this is rare in
laboratory medicine.36,37
If II is defined as CVI/CVG, this ratio must be made larger
to make conventional population-based reference intervals of
higher clinical usefulness, especially in diagnosis, case finding,
and screening, when no previous results on an individual are
available. This is actually sometimes easily achieved because
CVG can be made smaller by stratifying (or partitioning) the
data.36 An example is shown in Table 7.2.
Table 7.2 shows that, for the total cohort, II is 0.46, and
therefore, the reference intervals will be of low usefulness,
especially for monitoring individuals. However, for men
and women taken separately, the II are 1.83 and 1.42,†
respectively, and reference intervals will be very useful.
Stratification according to gender has vastly increased the
usefulness of conventional population-based reference inter-
vals. Because most measurands have low II, stratification
must be considered when reference intervals are being devel-
oped (see Chapter 8). Knowledge of individuality gives a
sound scientific basis for stratification; II must be high for
conventional, population-based reference intervals to be of
high usefulness. Because CVI is generally considered constant
and rarely stratified, it must be CVG that is made smaller, if
possible.
REFERENCE CHANGE VALUES AND
DIFFERENCES IN SERIAL RESULTS
The results of examinations in laboratory medicine are used
for many purposes; most are used for monitoring, either of
acute disease in the short term in secondary and tertiary care
institutions, or for chronic disease, which is often done in
primary care. Monitoring, by definition, means assessment of
results over time. Because most measurands have marked
individuality and low II, conventional population-based ref-
erence intervals have disadvantages as aids to interpretation
of serial results in an individual.
Harris and Yasaka38 introduced the concept of the RCV (as
discussed earlier), which is also sometimes called the critical
difference; the former term is much preferred to indicate an
analogy with population-based reference intervals. The gen-
eration and application of RCVs have been recently reviewed
in depth.6-8
The result of one examination will have: dispersion =
Z × (CVA2 + CVI2)1/2, where Z is the Z-score equal to the
164 SECTION I Basics of Laboratory Medicine
number of standard deviations appropriate for the probabil-
ity desired. The result of a second examination will have the
same dispersion, and so the total dispersion of two results
will be 21/2 × Z × (CVA2 + CVI2)1/2. Thus, for two results on
the same individual to be different, this inherent difference
due to CVA and CVI must be exceeded, and this is the RCV.
It is assumed that preexamination sources of variation are
considered negligible, and in clinical and laboratory prac-
tice, this means having well-documented standard operating
procedures for patient preparation and specimen collection,
transport, and handling before examination, and also good
training of healthcare staff performing these tasks. Moreover,
it is important to realize that changes in the bias of the exami-
nation between the collections of the serial specimens can
also add to the RCV; if these can be quantitated, as a dif-
ference due to bias in percentage terms, ΔB, the formula
becomes RCV = ΔB + 21/2 × Z × (CVA2 + CVI2)1/2. However,
in a single laboratory in practice, the main source of ΔB over
time is due to recalibration, and this random bias is usually an
integral component of the longer term CVA as estimated from
replicate examinations of internal quality control materials.
Thus, it can be assumed that the bias of the examination does
not change during the period between the two examinations,
and the simpler formula applies.
It is often assumed that a Z-score of 1.96 for P <0.05 (and
sometimes also 2.58 for P <0.01) are appropriate. It is almost
ubiquitously stated in studies on biological variation that the
RCV is calculated as 2.77 × (CVA2 + CVI2)1/2. This is incorrect
for a number of reasons.
First, these Z-scores are termed bidirectional (or two-tailed
or two-sided), and this infers that the difference between the
two serial results can be either an increase or a decrease.
However, in most clinical situations, the decision-making is
the assessment of a significant fall (decline, decrease or reduc-
tion, for example, HbA1c after treatment for diabetes mellitus
or in blood glucose after adjustment of insulin dosage), or a
significant rise (for example, an increase in serum creatinine
to assess acute kidney injury or serum troponin after acute
chest pain). Thus, unidirectional (one-tailed or one-sided)
Z-scores must be used in most clinical situations to facilitate
correct interpretation; these are 1.65 for P <0.05 and 2.33 for
P <0.01. Correct definitions of the clinical decision-making
context and the major differences between the terms “change”
and “rise or fall,” and their synonyms, are required for correct
calculation of appropriate RCVs.
Second, clinical decision-making is not always done at
P <0.05, which is the most commonly used probability in
analysis of research data. The semantics used are crucial to
understanding the probability that is appropriate. An example
was given in a recent study on RCVs for dehydration markers,
which, in addition to graphs that showed probability against
change for plasma osmolality, urine specific gravity, and body
mass, integrated a series of semantic interpretative anchors,
namely, change was likely at P >0.80, more likely at P >0.90,
very likely at P >0.95, and virtually certain at P >0.99.39 RCVs
should be used in a spectrum of post-examination processes,
including provision of graphs and tables of change versus
probability, Δ-checking, and flagging of significant changes at
different levels of probability on electronic and paper reports
of results of examinations.1
In addition, because RCV is calculated as RCV = 21/2 ×
Z × (CVA2 + CVI2)1/2, the magnitude of what the RCV for each
laboratory is will depend on the CVA achieved. Furthermore,
generation of data on CVI is not easy, and it is important
that the CVI used is obtained from a population with a
homogenous CVI and that is similar to the population for
whom the RCV is being created. In addition, the time interval
used for obtaining the CVI must be comparable to the one
used in practice. An overview of the source publications on
CVI can be found in the available database,19 and these should
be examined for their usefulness23 until the new EFLM data-
base (which is being developed) is available. Thus, laborato-
ries can calculate relevant RCVs by using CVA derived from
their own internal quality control programs, using data close
to clinical decision-making concentrations or activities, along
with the CVI estimates from publications cited in the most
up-to-date database to create a RCV to use for a variety of
purposes. The RCV is commonly calculated assuming that
CVI for the both examinations is identical (ie, CVI is con-
stant). A formula has been developed, which specifies that,
even if CVA is constant, because the concentrations or activi-
ties will be different, the SD of the two examinations can also
differ.40 Using the proposed more complex formula, the RCV
becomes larger for increases than for decreases. This interest-
ing proposal does not seem to have been translated into
routine practice.
Moreover, it is obvious that the RCV generated using the
traditional formula will depend on the number of significant
figures to which the results of examinations are reported;
these should be determined by consideration of the CVA
achieved in practice.41 However, it has been suggested that,
for examinations in which small changes in results may be of
clinical significance for monitoring purposes, the effect of the
number of significant figures reported on the RCV should
also be taken into consideration.42
Traditional RCVs are believed by some to be rather sim-
plistic, especially because they only address how likely it is
that a certain change can be explained by CVA and CVI, but
not the probability that a change in the disease state has
occurred. It has been suggested that a tool for better under-
standing and interpretation of measured differences in moni-
toring is needed; the concepts of sensitivity, specificity,
likelihood ratios, and odds used for diagnostic test evalua-
tions were applied to monitoring by substituting measured
concentrations with measured differences.43 It was suggested
that this idea expanded the earlier concept of RCV by making
it possible to have an estimate of the post-test odds for a
certain difference to occur. Consequently, the likelihood ratio
for change increases with a larger measured difference, and
when used together with the pretest odds or pretest probabil-
ity, the post-test odds and post-test probability, which are
related to the clinical situation, can be calculated.
It has been proposed6-8 that the probability of significance
of any difference seen in clinical practice between two results
can be readily calculated using a simple rearrangement of
the RCV equation making the Z-score (and thus, the prob-
ability) the unknown, namely: Z = difference/[21/2 × (CVA2 +
CVI2)1/2]. This does not seem to have been applied in practice,
but there is clearly scope for adoption of this technique to
enhance interpretation of serial results. Exactly as for RCVs,
this application would have important consequences for CVA:
the smaller the CVA, the smaller the RCV will be for any prob-
ability, and the significance of any difference seen will be of
higher probability.

Reference Change Values When the Distribution
Is Not Normal
Some would argue that biological variables are not normally
distributed but have a natural logarithm distribution. The
traditional approach to generation of RCV does assume that
CVA is random and also assumes that CVI is a random fluc-
tuation around a homeostatic setting point, both distributed
normally. However, recently, a number of examinations have
been evaluated, making the assumption that the distributions
of results in an individual were ln-normal rather than normal.
The strategy proposed44 has since been adopted by others
who are investigating examinations such as serum tropo-
nin.45,46 CVI was calculated from the total imprecision (CVT)
for duplicate examinations by the often used formula: CVI =
(CVT2 − CVA2)1/2. With the ln-normal approach, the median
normal deviation of the n-normal distribution (σ) was cal-
culated as σ = [ln(CVT2 + 1)]1/2. The asymmetrical limits for
the upward (positive) value for the ln-normal RCV (RCVpos)
and for the downward (negative) value for the ln-normal
RCV (RCVneg), were calculated as: RCVpos = [exp(1.96 × 21/2
× σ) − 1] × 100 and RCVneg = [exp(1.96 × 21/2 σ) − 1] × 100,
respectively. This approach gives different RCVs for increases
in concentration or activity to the RCV for decreases. This
approach is likely to be used more often in the future when
the measurand has a ln-normal rather than a normal distri-
bution.47 The CVT term used in this calculation is equal to
(CVA2 + CVI2)1/2, and where it is difficult to dissect out numer-
ical estimates of CV from the total variation of individuals,
then calculation of RCV as RCV = 21/2 × Z × CVT seems
appropriate and does provide an estimate of RCV that can be
used in clinical and laboratory practices, assuming the CVA is
similar between that derived in the assessment of (CVA2 +
CVI2)1/2 and in the practice in which it is going to be used.
Reference Change Values for More Than
Two Serial Examinations
Frequently, in practice, more than two results of examina-
tions are available for the individuals over time. Using the
traditional RCVs described previously, it is only possible to
calculate the significance of changes between each of the two
consecutive examinations. Thus, a RCV method including all
available serial results might be useful for interpretation of
significant differences over time.
Mathematical methods have been developed to assess
serial results from an individual, which is sometimes called
methods for time series analysis.34 One model is called the
“homeostatic model.” The model assumes that the measur-
and varies randomly around a homeostatic setting point.
After collection of results from a small number of examina-
tions, the mean and standard deviation are calculated. The
next result should fall within the range calculated from the
mean and the standard deviation. Then, new data from
the same individual is used in an ongoing manner to refine
the estimates of the mean and standard deviation for that
individual to interpret whether the next new result has
changed significantly from previous data. In contrast, the
“random walk” model assumes that the measurand behaves
randomly, and there is no homeostatic setting point. The
dispersion of each result is calculated. However, instead of
calculating a mean for the individual and recalculating this
mean as further data are gathered, it is assumed that the most
CHAPTER 7 Biological Variation 165
recent result is the starting point from which to assess whether
change has occurred. These models have not been widely
applied; they generate many false positive signals that change
has occurred, when there is no obvious usual or pathological
change. They also seem too complex and time consuming to
be used in everyday practice.
Recently, studies on both unidirectional and bidirectional
changes in serial results that used computer simulations have
been performed.48,49 Factors used to multiply the first result
from an individual were calculated to create the limits for
constant cumulated significant differences. The factors were
shown to become a simple function of the number of results
and the total CV. The first result is multiplied by the appropri-
ate factor for an increase or a decrease, which gives the limits
for a significant difference. It remains to be seen if such
apparently simple techniques become used in practice.
ANALYTICAL PERFORMANCE SPECIFICATIONS
BASED ON BIOLOGICAL VARIATION
Analytical performance specifications are the numerical
standards of examination performance required to facilitate
optimum patient care. There are many strategies available to
set these specifications, and these have been described over
time as this facet of laboratory medicine has evolved.50,51 A
concise historical perspective has been published recently.52
Following pioneering studies done in the United States53
on the definition of the components of biological variation,
a College of America Pathologists conference held in 1976
supported the concept that specifications should be best
based on biology.54 The consensus statement, restated in
current terms and symbols, was: “For group screening, in
which an individual is to be selected from a population, a
specification for imprecision (CVA) is defined as: CVA = 0.5 ×
(CVI2 + CVG2)1/2” and “For individual single and multipoint
testing, in which an individual is evaluated on the basis of
discrimination values: CVA = 0.5 × CVI.”
Specifications for Examination Imprecision
This approach became of even greater interest as the quantity
of data on the components of biological variation increased.
In most cases, the examination performance specification for
CVA was simply taken as CVA = 0.5 × CVI, with the rationale
that the same technology and methodology were used to
examine specimens in both of the previously described clini-
cal settings (and other situations). In consequence, the most
stringent situation would allow both specifications to be met.
Specifications for Examination Bias
Examination bias was not mentioned at that time, possibly
because the thesis was that laboratories all had their own
reference intervals to which the results of their examinations
were compared. However, as interest in harmonization of
data across time and geography developed, it was realized that
harmonization of reference intervals was important, and it
was proposed that the examination performance specifica-
tion for bias (B), to allow use of harmonized reference inter-
vals, was B <0.25 × (CVI2 + CVG2)1/2.55
Specifications for Total Error Allowable
As the concepts of total laboratory quality management
evolved, and the idea that random and systematic sources of
166 SECTION I Basics of Laboratory Medicine
variation (imprecision and bias) were both important, it
became clear that total analytical error was the most impor-
tant clinically.56 It was proposed that the linear model of
combining imprecision and bias could be used to set analyti-
cal performance specifications for total error allowable (TEa)
as a simple linear addition; for P <0.05: TEa <0.1.65 × 0.5 CVA
+ 0.25 (CVI2 + CVG2)1/2. It should be noted that this model for
combining bias and imprecision is only one of the models
available; the advantages and disadvantages of these have
been discussed in detail,57 and the disadvantages of this par-
ticular model have been recently restated.58 EFLM has estab-
lished a Task and Finish Group to address the difficulties with
the total error concept and to try to evolve new models that
are sounder. However, the present model is widely used in
current practice in laboratory medicine, albeit sometimes
with a different multiplier for the imprecision term, to deliver
different levels of probability.
Application of this has become very widespread because
the formula is simple, much data on biological variations
exist, and they are directly related to the general use of
results of examinations in laboratory medicine. However,
concerns have been expressed. First, because the biological
variation data currently used for some measurands vary,6
and second, because the analytical performance standards
derived using biological variation are unattainable for
some measurands with current technology and methodol-
ogy, including serum sodium, chloride, and calcium. How-
ever, some analytical performance standards were easy to
obtain for others, including serum urea, triglycerides, and
many enzyme activities. A three-tier model to cater for this
was proposed, giving minimum, desirable, and maximum
examination performance specifications based on biological
variation using 0.25 CVA, 0.5 CVA, and 0.75 CVA for impreci-
sion and 0.125 (CVI2 + CVG2)1/2, 0.25 (CVI2 + CVG2)1/2, and
0.325 (CVI2 + CVG2)1/2 for bias, respectively.59 Numerical data
on these three levels of analytical performance specifica-
tions are included for the 358 measurands in the eighth
edition of the biological variation database on the Internet.19
Furthermore, because analytical technology has evolved
rapidly of late, and analytical performance has improved,
it has been suggested that a fourth “ideal” level should be
introduced with multipliers of 0.1 for both imprecision
and bias.60 Combinations of the three (or four) levels can be
done to obtain relevant analytical performance specifications
for TEa.
Specifications Derived From Opinions of Clinicians
Because it is the users of the results of examinations in labo-
ratory medicine that make use of the data provided, it might
be believed that they should be able to inform about the
analytical quality required to facilitate decision-making. One
way of attempting to do this is with clinical vignettes, whereby
the clinician provides information about the change in serial
results that is required to stimulate a clinical decision. Some
early studies were carried out, but these were completely
unsatisfactory, in that it was assumed that changes were all
due to analytical imprecision, and within-subject biological
variation was not considered. The probability used was
bidirectional, and always at P <0.05 despite the very large
differences in probability attributable to words used in the
statements.61,62 The principle behind the studies should be
that a difference in serial results is due to preexamination
sources of variation (CVP), CVA, CVI, and changes in
bias (ΔB):
difference = 21 2 × Z × [CVP2 + CVI2 + CVA 2 ]1 2 + ΔB
now, let CVP and ΔB be zero and rearrange the equation:
CVA = [difference (21 2 × Z + CVI2 )]1 2.
Studies have been done using this model on a variety of
measurands, and have included some studies that involved
patients undertaking self-testing. An early example is pro-
vided by the studies on blood hemoglobin63; other studies
performed since then have investigated HbA1c, glucose, pro-
thrombin time, erythrocyte sedimentation rate, and urinary
albumin.64 It was found that the derived analytical per-
formance specification depends on the clinical setting, the
semantics of the questions posed, how close the result was to
a decision limit or reference limit, and probably on current
examination performance. Moreover, there was large intercli-
nician variation in responses. This approach is probably pos-
sible only for assessment of a single measurand in a specific
clinical situation. In addition, it has been suggested that this
strategy is possible only for analytes that have a major role in
monitoring and/or diagnosis.
Specifications for Measurement Uncertainty
Because there is now a requirement for medical laboratories
to document the measurement uncertainty to comply with
ISO 15189, there should be consideration of the setting of
analytical performance specifications for this characteristic.
Because the fundamental principles are that bias should be
eliminated (if possible) and all sources of variation should be
added linearly as variances, then the only possible analytical
performance specifications for measurement uncertainty are
that CVA < f × CVI, where f = 0.10, 0.25, 0.50, 0.75 for ideal,
optimum, desirable, and minimum levels of performance,
respectively,60 but again this is dependent on reliable data
being available for CVI.
Specifications for Other Applications
Data on biological variation have been used to generate ana-
lytical performance specifications in other than general clini-
cal settings, including an evaluation of systems,65 reference
methods,66 and for the allowable difference in bias (ΔB)
between two systems used to generate results on the same
individuals, which is ΔB < 0.33 CVI.67
Role of Biological Variation in Setting Analytical
Performance Specifications
The setting of analytical performance specifications in labo-
ratory medicine has been a topic of discussion and debate
for more than 50 years; 15 years ago, as this topic matured
and a profusion of recommendations appeared, a number of
leading professionals in laboratory medicine realized that
there was a need for a global consensus on the setting of such
specifications. The Stockholm Conference held in 1999 on
“Strategies to set global analytical quality specifications in
laboratory medicine” achieved this and advocated the ubiq-
uitous application of a hierarchical structure of approaches,
based on a model proposed by Fraser and Petersen.68 The
hierarchy has five levels, namely: (1) evaluation of the effect

of analytical performance on clinical outcomes in specific
clinical settings; (2) evaluation of the effect of analytical
performance on clinical decisions in general, using (a) data
based on components of biological variation or (b) analysis
of clinicians’ opinions; (3) published professional recom-
mendations from (a) national and international expert bodies
or (b) expert local groups or individuals; (4) performance
goals set by (a) regulatory bodies or (b) organizers of external
quality assessment (EQA) schemes; and (5) goals based
on the current state of the art as (a) demonstrated by data
from EQA or proficiency testing scheme, or (b) found in
current publications on methodology.69 This approach has
mostly been used, and there is considerable evidence that the
specifications advocated in level 2, which are based on bio-
logical variation, are widely implemented. A number of EQA
schemes, including those in Australia,70 base their acceptable
standards of examination performance on the Stockholm
consensus conference recommendations, especially the use of
biological variation data. Three convocations of experts in
quality control in laboratory medicine discussed the genera-
tion and application of analytical performance specifications,
and these strategies, based on biological variation, are used
by many.60,71,72 Furthermore, a recent survey with more than
450 responses from professionals in more than 80 countries
demonstrated the wide use of specifications based on biologi-
cal variation.73 Since the Stockholm conference, there have
been further proposals for setting performance specifications
for examination performance characters,74,75 but it is interest-
ing that all of these include biological variation data in their
considerations.76
Because laboratory medicine has evolved considerably
over the last few years, the EFLM organized the 1st Strategic
Conference on defining analytical goals 15 years after the
Stockholm conference; the many insightful presentations are
available on the Internet.77 The consensus statement and
formal papers emanating from the speakers are included in a
Special Issue of Clinical Chemistry and Laboratory Medicine,78
and represent an invaluable up-to-date resource on all aspects
of setting analytical performance standards, and on genera-
tion and application of data on biological variation. The con-
sensus was that the Stockholm hierarchical approach was
supported, but could be simplified. In this revision, the hier-
archy was simplified and represented by three different
models to set analytical performance specifications. Model 1
is based on the effect of analytical performance on clinical
outcomes using direct (1a) or indirect (1b) outcome studies,
which basically investigate the impact of analytical perfor-
mance of the test on clinical outcomes. Model 2 is based on
components of biological variation of the measurand. Model
3 is based on state of the art. The three models use different
principles, and some models will be better suited for certain
measurands than for others. Application of model 1 is diffi-
cult and will probably be limited to a few measurands. It has
been considered that generally applicable analytical per-
formance specifications will be best based on biological vari-
ation data for some time to come. In view of the caveats
regarding the reliability of certain data on the components of
biological variation, it will be significant when the EFLM
group on biological variation fulfills its remit and evaluates
the current database, creates a database of high quality, and
its recommendations on standards of generation and report-
ing of data are carried through into practice.
CHAPTER 7 Biological Variation 167
OTHER APPLICATIONS
Calculation of Reliability Coefficient
The individuality of tests and the consequences of the marked
individuality (low II) of most measurands in laboratory
medicine in diagnosis, case finding, screening, and monitor-
ing has been discussed. Epidemiologists and others use
similar information to the II, but in a slightly different way.
The reliability coefficient is used, which is calculated as
CVG2/(CVA2 + CVI2 + CVG2)1/2, which is the between-subject
variance divided by the total variance.1 The reliability coef-
ficient, usually called R, is numerically equal to the correla-
tion coefficient of repeated measurements. It can be between
0 and 1. If R approached 0, II would be high, and if R
approached 1, then II would be low. Most measurands have
high values for R.
Number of Samples Needed
In usual clinical practice, only one sample is taken.
Examination result variation can be reduced by multiple
sampling (or multiple examinations), and the variation is
made smaller by the square root of the number of replicates.
To estimate the number of specimens needed to determine
the homeostatic setting point within a certain percentage
error with a stated probability, a simple rearrangement of the
usual standard error of the mean formula is used, namely:
n = (Z × [CVA 2 + CVI2 ] 12 D)2 , where Z is the Z-score appro-
priate for the probability, and D is the desired percentage
closeness to the homeostatic setting point. It is important
to note that taking multiple specimens and undertaking
replicate analyses does affect the overall variability of the
individual examination result; the dispersion (expressed as
1 CV) can be calculated as: dispersion = Z × [(CVA2/nA) +
(CVI2/nS)]1/2, where Z is the number of standard deviations
appropriate to the probability selected, nA is the number of
replicate examinations, and nS is the number of specimens.
The relative magnitudes of CVA and CVI are important in
deciding if a lower dispersion is required, whether it is better
to undertake replicate examinations on one specimen or
singleton examinations on multiple specimens. Calculators
for this can be found, along with a more detailed discussion
of this topic, with real examples.79 A review has documented
further examples and detailed the reasons why knowledge of
numerical data on the components of biological variation is
of crucial importance.80
Reporting Results, Selecting the Best Specimen
to Collect, and Choosing the Best Examination
It is sometimes possible to report the results of examinations
in different ways. For example, measurands in urine such
as creatinine can be reported as concentration or output per
day, and many are reported as a ratio with creatinine concen-
tration. Moreover, for some measurands, it is possible to
collect different samples for the same clinical purpose (eg,
early morning or random or timed urine specimens for low
concentration albumin and protein examinations). In certain
clinical situations, examinations that might be considered
to have a somewhat similar purpose are available, such as
serum creatinine and cystatin-C, or blood HbA1c and serum
fructosamine. Knowledge of the components of biological
variation can assist in making decisions about reporting
168 SECTION I Basics of Laboratory Medicine
results, selecting the best specimen to collect, and choosing
the best test.1
To undertake such comparisons, the influences of biologi-
cal variation should be considered. The ideal measurand
would have low CVI so that a single examination will give a
good measure of the true value for that individual. Moreover,
this would allow easy monitoring over time and detection
of significant differences, because the RCV would be low,
provided that the CVA was also low. In addition, the ideal
measurand would have no heterogeneity of CVI among indi-
viduals and across studies, and would not be dependent on
age and gender and other possible confounding factors so
that the simple general formulas given in this chapter would
hold for all.
Method Development and Evaluation
Introduction of new examinations is an ongoing task for
most medical laboratories. Some years ago, Zweig and
Robertson suggested that the introduction of a new proce-
dure should be similar to the structured evolution of a new
drug through phase trials.81 They suggested that the phases
should be the following: analytical investigation and assess-
ment of reliability and practicability characteristics; overlap
investigation involving generation of reference intervals and
assessment of values in disease; clinical investigation and
evaluation of sensitivity, specificity, and predictive value;
outcome investigation of whether individuals gain an advan-
tage; and, finally, investigation of usefulness, which is a cost–
benefit analysis with respect to individuals and the population.
In contrast to the linear models, another approach has been
proposed in which the essential components of analytical and
clinical performance, clinical and cost-effectiveness, and the
broader impact of testing are assembled in a dynamic cycle.
This approach emphasizes the interaction of the different
components, and that clinical effectiveness data should be fed
back to refine analytical and clinical performances to achieve
improved outcomes.82
One aspect that is not covered in any of these phases,
however, is the need to generate and apply data on biological
variation early in the evolution of any examination. Data
can be generated on the components of biological variation
through duplicate analysis of a small number of samples from
a small cohort of healthy individuals. This allows the setting
of analytical performance specifications, calculation of the
significance of changes in an individual (otherwise unobtain-
able and not mentioned in any of the phases), and assessment
of the use of the generated population-based reference
intervals. Generation and application of data on biological
variation are essential prerequisites for introducing new
examinations.83 Moreover, the data are also necessary for
objective analysis of the often somewhat subjective guide-
lines from professional bodies that give recommendations on
interpretation of the numerical results of examinations and
on examination performance specifications; unfortunately,
these recommendations are often flawed.80
OVERALL CONCLUSIONS
Numerical estimates of within-subject and between-subject
biological variation are best generated by examination of a
series of specimens taken from a cohort of individuals, fol-
lowed by statistical analysis of the sources of variation. The
proper design and performance of such studies is complex.
However, databases of estimates are available that facilitate
application in determining the individuality of a measurand
and the usefulness of conventional population-based refer-
ence intervals, the statistical significance of changes in serial
results from an individual, analytical performance specifica-
tions for imprecision, bias, total error allowable, measure-
ment uncertainty, and other characteristics. The estimates
have many other uses. There are current concerns regarding
the robustness of certain of these data and estimates because
some measurands show considerable heterogeneity. Recent
recommendations should be followed in the generation and
application of biological variation data.
POINTS TO REMEMBER
Biological variation may be daily, monthly or seasonal, but
most measurands have random variation around homeostatic
setting points.
It is challenging to generate data on within-subject and
between-subject components of biological variation, but
available databases facilitate application; users should ensure
applicability of published data to their applications.
Data on biological variation are used, inter alia, to determine:
• the individuality of a measurand and the usefulness of
conventional population-based reference intervals
• the statistical significance of changes in serial results
from an individual and the probability that any change
documented is significant
• examination performance specifications for impreci-
sion, bias, TEa, measurement uncertainty, and other
characteristics.
SELECTED REFERENCES
For a full list of references for this chapter, please refer to
ExpertConsult.com.
1. Fraser CG. Biological variation: from principles to practice.
Washington, DC: AACC Press; 2001.
3. Fraser CG1, Harris EK. Generation and application of data on
biological variation in clinical chemistry. Crit Rev Clin Lab Sci
1989;27:409–37.
5. Røraas T, Petersen PH, Sandberg S. Confidence intervals and
power calculations for within-person biological variation:
effect of analytical imprecision, number of replicates, number
of samples, and number of individuals. Clin Chem 2012;58:
1306–13.
8. Fraser CG. Reference change values. Clin Chem Lab Med
2011;50:807–12.
19. Minchinela J, Ricós C, Perich C, et al. Biological variation
database and quality specifications for imprecision, bias and
total error (desirable and minimum). The 2014 update.
http://www.westgard.com/biodatabase-2014-update.htm>.
20. Perich C, Minchinela J, Ricós C, et al. Biological variation
database: structure and criteria used for generation and
update. Clin Chem Lab Med 2015;53:299–305.
24. Bartlett WA, Braga F, Carobene A, et al. A checklist for critical
appraisal of studies of biological variation. Clin Chem Lab
Med 2015;53:879–85.

29. Ricos C, Iglesias N, Garcia-Lario JV, et al. Within-subject
biological variation in disease: collated data and clinical
consequences. Ann Clin Biochem 2007;44:343–52.
33. Fraser CG. Inherent biological variation and reference values.
Clin Chem Lab Med 2004;42:758–64.
37. Petersen PH, Sandberg S, Fraser CG, et al. Influence of index
of individuality on false positives in repeated sampling from
healthy individuals. Clin Chem Lab Med 2001;39:160–5.
43. Hyltoft Petersen P, Sandberg S, Iglesias N, et al. Likelihood-
ratio and odds applied to monitoring of patients as a
supplement to reference change value (RCV). Clin Chem
Lab Med 2007;46:157–64.
59. Fraser CG, Hyltoft Petersen P, Libeer JC, et al. Proposals for
setting generally applicable quality goals solely based on
biology. Ann Clin Biochem 1997;34:8–12.
68. Fraser CG, Petersen PH. Analytical performance characteristics
should be judged against objective quality specifications. Clin
Chem 1999;45:321–3.
CHAPTER 7 Biological Variation 169
69. Hyltoft Petersen P, Fraser CG, Kallner A, et al., editors.
Strategies to set global analytical quality specifications in
laboratory medicine. Scand J Clin Lab Invest 1999;59:
475–585.
75. Klee GG. Establishment of outcome-related analytic
performance goals. Clin Chem 2010;56:714–22.
78. Sandberg S, Fraser CG, Horvath AR, et al. Defining analytical
performance specifications: Consensus Statement from the 1st
Strategic Conference of the European Federation of Clinical
Chemistry and Laboratory Medicine. Clin Chem Lab Med
2015;833–5.
80. Fraser CG. Test result variation and the quality of evidence-
based clinical guidelines. Clin Chim Acta 2004;346:
19–24.
83. Fraser CG. Data on biological variation; essential prerequisites
for introducing new procedures. Clin Chem 1994;40:
1671–3.

REFERENCES
1. Fraser CG. Biological variation: from principles to practice.
Washington, DC: AACC Press; 2001.
2. Simundic AM, Kackov S, Miler M, et al. Terms and symbols
used in studies on biological variation: the need for
harmonization. Clin Chem 2015;61:438–9.
3. Fraser CG1, Harris EK. Generation and application of data on
biological variation in clinical chemistry. Crit Rev Clin Lab Sci
1989;27:409–37.
4. Plebani M, Padoan A, Lippi G. Biological variation: back to
basics. Clin Chem Lab Med 2015;53:155–6.
5. Røraas T, Petersen PH, Sandberg S. Confidence intervals and
power calculations for within-person biological variation:
effect of analytical imprecision, number of replicates, number
of samples, and number of individuals. Clin Chem 2012;58:
1306–13.
6. Aarsand AK, Røraas T, Sandberg S. Biological variation—
reliable data is essential. Clin Chem Lab Med 2015;53:
153–4.
7. Fraser CG. Improved monitoring of differences in serial
laboratory results. Clin Chem 2011;57:1635–7.
8. Fraser CG. Reference change values. Clin Chem Lab Med
2011;50:807–12.
9. Fraser CG. Making better use of differences in serial
laboratory results. Ann Clin Biochem 2012;49:1–3.
10. Fraser CG. Biological variation in the elderly; implications for
reference values. In: Faulkner WR, Meites S, editors. Geriatric
clinical chemistry reference values. Washington, DC: AACC
Press; 1994. p. 44.
11. Carlsen S, Petersen PH, Skeie S, et al. Within-subject biological
variation of glucose and HbA1c in healthy persons and in
type 1 diabetes patients. Clin Chem Lab Med 2011;49:1501–7.
12. Braga F, Dolci A, Mosca A, et al. Biological variability of
glycated haemoglobin. Clin Chim Acta 2010;411:1606–10.
13. Braga F, Panteghini M. Biological variability of C-reactive
protein: is the available information reliable? Clin Chim Acta
2012;413:1179–83.
14. Carobene A, Braga F, Roraas T, et al. A systematic review of
data on biological variation for alanine aminotransferase,
aspartate aminotransferase and γ-glutamyl transferase. Clin
Chem Lab Med 2013;51:1997–2007.
15. Fraser CG. The application of theoretical goals based on
biological variation data in proficiency testing. Arch Pathol Lab
Med 1988;112:404–15.
16. Fraser CG. Biological variation in clinical chemistry. An
update: collated data, 1988 – 1991. Arch Pathol Lab Med
1992;116:916–23.
17. Sebastián-Gámbaro MA, Lirón-Hernández FJ, Fuentes-Arderiu
X. Intra- and inter-individual biological variability data bank.
Eur J Clin Chem Clin Biochem 1997;35:845–52.
18. Ricós C, Álvarez V, Cava F, et al. Current databases on
biological variation: pros, cons and progress. Scand J Clin Lab
Invest 1999;59:491–500.
19. Minchinela J, Ricós C, Perich C, et al. Biological variation
database and quality specifications for imprecision, bias and
total error (desirable and minimum). The 2014 update.
<http://www.westgard.com/biodatabase-2014-update.htm>.
[accessed 15.02.14].
20. Perich C, Minchinela J, Ricós C, et al. Biological variation
database: structure and criteria used for generation and
update. Clin Chem Lab Med 2015;53:299–305.
CHAPTER 7 Biological Variation 169.e1
21. Fraser CG, Browning MC. The “index of fiduciality” proposed
for use in evaluation and comparison of methods. Clin Chem
1988;34:1356–7.
22. Sölétormos G, Semjonow A, Sibley PE, et al. Biological
variation of total prostate-specific antigen: a survey of
published estimates and consequences for clinical practice.
Clin Chem 2005;51:1342–51.
23. Miller GW, Bruns DE, Hortin GL, et al. Current issues in
measurement and reporting of urinary albumin excretion.
Clin Chem 2009;55:24–38.
24. Bartlett WA, Braga F, Carobene A, et al. A checklist for critical
appraisal of studies of biological variation. Clin Chem Lab
Med 2015;53:879–85.
25. European Federation of Clinical Chemistry and Laboratory
Medicine. <http://www.efcclm.org>. [accessed 15.02.14].
26. Simundic AM, Bartlett WA, Fraser CG. Biological variation: a
still evolving facet of laboratory medicine. Ann Clin Biochem
2015;52:189–90.
27. STARD Guidelines. <http://www.stard-statement.org>.
[accessed 15.02.14].
28. Lawson N. Is variation in biological variation a problem? Ann
Clin Biochem 2007;44:319–20.
29. Ricos C, Iglesias N, Garcia-Lario JV, et al. Within-subject
biological variation in disease: collated data and clinical
consequences. Ann Clin Biochem 2007;44:343–52.
30. Westgard QC. Quality requirements. <http://
www.westgard.com/biodatabasedisease.htm>. [accessed
15.02.15].
31. Desmeules P, Cousineau J, Allard P. Biological variation of
glycated haemoglobin in a paediatric population and its
application to calculation of significant change between
results. Ann Clin Biochem 2010;47:35–8.
32. Cembrowski GS, Trani DV, Higgins TN. The use of serial
patient blood gas, electrolyte and glucose results to derive
biologic variation: a new tool to assess the acceptability of
intensive care unit testing. Clin Chem Lab Med 2010;48:
1447–54.
33. Fraser CG. Inherent biological variation and reference values.
Clin Chem Lab Med 2004;42:758–64.
34. Harris EK. Some theory of reference values. II. Comparison of
some statistical models of intraindividual variation in blood
constituents. Clin Chem 1976;22:1343–50.
35. Harris EK. Statistical aspects of reference values in clinical
pathology. Prog Clin Pathol 1981;8:45–66.
36. Hyltoft Petersen P, Fraser CG, Sandberg S, et al. The index of
individuality is often a misinterpreted quantity characteristic.
Clin Chem Lab Med 1999;37:655–61.
37. Petersen PH, Sandberg S, Fraser CG, et al. Influence of index
of individuality on false positives in repeated sampling
from healthy individuals. Clin Chem Lab Med 2001;39:
160–5.
38. Harris EK, Yasaka T. On the calculation of a “reference
change” for comparing two consecutive measurements. Clin
Chem 1983;29:25–30.
39. Cheuvront SN, Fraser CG, Kenefick RW, et al. Reference
change values for monitoring dehydration. Clin Chem Lab
Med 2011;49:1033–7.
40. Jones GRD. Critical difference calculations revised: inclusion
of variation in standard deviation with analyte concentration.
Ann Clin Biochem 2009;46:517–19.
41. Hawkins RCW. The significance of significant figures. Clin
Chem 1990;36:824.
169.e2 SECTION I Basics of Laboratory Medicine
42. Jones GRD. The effect of the reporting interval size on critical
difference estimation – beyond ‘2.77’. Clin Chem 2006;52:
880–5.
43. Hyltoft Petersen P, Sandberg S, Iglesias N, et al. Likelihood-
ratio and odds applied to monitoring of patients as a
supplement to reference change value (RCV). Clin Chem Lab
Med 2007;46:157–64.
44. Fokkema MR, Herrmann Z, Muskiet FAJ, et al. Reference
change values for brain natriuretic peptides revisited. Clin
Chem 2006;52:1602–3.
45. Frankenstein L, Wu AH, Hallermayer K, et al. Biological
variation and reference change value of high-sensitivity
troponin T in healthy individuals during short and
intermediate follow-up periods. Clin Chem 2011;57:
1068–71.
46. Vasile VC, Saenger AK, Kroning JM, et al. Biologic variation of
a novel cardiac troponin I assay. Clin Chem 2011;57:1080–1.
47. Braga F, Ferraro S, Ieva F, et al. A new robust statistical model
for interpretation of differences in serial test results from an
individual. Clin Chem Lab Med 2015;53:815–22.
48. Lund F, Petersen PH, Fraser CG. Sölétormos G. Calculation of
limits for significant unidirectional changes in two or more
serial results of a biomarker based on a computer simulation
model. Ann Clin Biochem 2015;52:237–44.
49. Lund F, Petersen PH, Fraser CG. Sölétormos G. Calculation of
limits for significant bidirectional changes in two or more
serial results of a biomarker based on a computer simulation
model. Ann Clin Biochem 2015;52:434–40.
50. Fraser CG. Desirable performance standards for clinical
chemistry tests. Adv Clin Chem 1983;23:299–339.
51. Fraser CG, Hyltoft Petersen P. Desirable standards for
laboratory tests if they are to fulfil medical needs. Clin Chem
1993;39:1447–55.
52. Fraser CG. The 1999 Stockholm Consensus Conference on
quality specifications in laboratory medicine. Clin Chem Lab
Med 2015;53:837–40.
53. Cotlove E, Harris EK, Williams GZ. Biological and analytic
components of variation in long-term studies of serum
constituents in normal subjects. III. Physiological and medical
implications. Clin Chem 1970;16:1028–32.
54. Elevitch FR, editor. College of American Pathologists Conference
II (1976): analytical goals in clinical chemistry. Skokie, IL:
College of American Pathologists; 1977.
55. Gowans EM, Hyltoft Petersen P, Blaabjerg O, et al. Analytical
goals for the acceptance of common reference intervals for
laboratories throughout a geographical area. Scand J Clin Lab
Invest 1988;48:757–64.
56. Westgard JO, Hunt MR. Use and interpretation of common
statistical tests in method-comparison studies. Clin Chem
1973;19:49–57.
57. Petersen PH, Fraser CG, Jørgensen L, et al. Combination of
analytical quality specifications based on biological within-
and between-subject variation. Ann Clin Biochem 2002;39:
543–50.
58. Oosterhuis WP. Gross overestimation of total allowable error
based on biological variation. Clin Chem 2011;57:1334–6.
59. Fraser CG, Hyltoft Petersen P, Libeer JC, et al. Proposals for
setting generally applicable quality goals solely based on
biology. Ann Clin Biochem 1997;34:8–12.
60. Adams O, Cooper G, Fraser C, et al. Collective opinion paper
on findings of the 2011 convocation of experts on laboratory
quality. Clin Chem Lab Med 2012;50:1547–58.
61. Elion-Gerritzen WE. Analytic precision in clinical chemistry
and medical decisions. Am J Clin Pathol 1980;73:
183–95.
62. Skendzel LP, Barnet RN, Platt R. Medically useful criteria for
analytic performance of laboratory tests. Am J Clin Pathol
1985;83:200–5.
63. Thue G, Sandberg S, Fugelli P. Clinical assessment of
haemoglobin values by general practitioners related to
analytical and biological variation. Scand J Clin Lab Invest
1991;51:453–9.
64. Thue B, Sandberg S. Analytical performance specifications
based on how clinicians use laboratory tests. Experiences from
a post-analytical external quality assessment scheme. Clin
Chem Lab Med 2015;53:857–62.
65. Fraser CG, Hyltoft Petersen P, Ricos C, et al. Proposed quality
specifications for the imprecision and inaccuracy of analytical
systems in clinical chemistry. Eur J Clin Chem Clin Biochem
1992;30:311–17.
66. Thienpont L, Franzini C, Kratochvila J, et al. Analytical
quality specifications for reference methods and operating
specifications for networks of reference laboratories. Eur J Clin
Chem Clin Biochem 1995;33:949–57.
67. Petersen PH, Fraser CG, Westgard JO, et al. Analytical
goal-setting for monitoring patients when two analytical
methods are used. Clin Chem 1992;38:2256–60.
68. Fraser CG, Petersen PH. Analytical performance characteristics
should be judged against objective quality specifications. Clin
Chem 1999;45:321–3.
69. Hyltoft Petersen P, Fraser CG, Kallner A, et al., editors.
Strategies to set global analytical quality specifications in
laboratory medicine. Scand J Clin Lab Invest 1999;59:
475–585.
70. RCPAQAP Chemical Pathology. Assessment of performance.
<http://dataentry.rcpaqap.com.au/chempath/index.cfm>.
71. Burnett D, Ceriotti F, Cooper G, et al. Collective opinion
paper on findings of the 2009 convocation of experts on
quality control. Clin Chem Lab Med 2010;48:41–52.
72. Cooper G, DeJonge N, Ehrmeyer S, et al. Collective opinion
paper on findings of the 2010 convocation of experts
on laboratory quality. Clin Chem Lab Med 2011;49:
793–802.
73. Westgard S. Global analytical goal survey results. <http://
www.westgard.com/global-goal-results.htm>.
74. Haeckel R, Wosniok W. A new concept to derive permissible
limits for analytical imprecision and bias considering
diagnostic requirements and technical state-of-the-art. Clin
Chem Lab Med 2011;49:623–35.
75. Klee GG. Establishment of outcome-related analytic
performance goals. Clin Chem 2010;56:714–22.
76. Hyltoft Petersen P, Sandberg S, Fraser CG. Do new concepts
for deriving permissible limits for analytical imprecision and
bias have any advantages over existing consensus? Clin Chem
Lab Med 2011;49:637–40.
77. <http://www.eflm.eu/index.php/educational-material.html>.
78. Sandberg S, Fraser CG, Horvath AR, et al. Defining analytical
performance specifications: Consensus Statement from the 1st
Strategic Conference of the European Federation of Clinical
Chemistry and Laboratory Medicine. Clin Chem Lab Med
2015;833–5.
79. Westgard QC. Dispersion calculator and critical number of
test samples. <http://www.westgard.com/dispersion-calculator
-and-critical-number-of-test-samples.htm>.

80. Fraser CG. Test result variation and the quality of evidence-
based clinical guidelines. Clin Chim Acta 2004;346:
19–24.
81. Zweig MH, Robertson EA. Why we need better test
evaluations. Clin Chem 1982;28:1272–6.
CHAPTER 7 Biological Variation 169.e3
82. Horvath AR, Lord SJ, StJohn S, et al. From biomarkers to
medical tests: the changing landscape of test evaluation. Clin
Chim Acta 2014;427:49–57.
83. Fraser CG. Data on biological variation; essential prerequisites
for introducing new procedures. Clin Chem 1994;40:1671–3.