Biologicaland AnalyticComponentsof Variationin Long-Term

Studiesof Serum Constituentsin Normal Subjects

Ill. Physiological and Medical Implications

Ernest Cotlove,’ Eugene K. Harris, and George Z. Williams2

Biologicalcomponents of variationforeach of 15 serum constituents have

been estimatedin68 normal subjectsby weekly measurements oversev-
eralmonths, and concurrentmeasurements on invariant human serum
pools.Intra-individualcoefficient was smallest, less than 2%,
of variation
forNa, Cl,Ca, Mg; 3 to4% forC02,albumin,and totalprotein; 5 to7% forK,
glucose,cholesterol, and phosphate;and largest, 9 to 12%, foruricacid,
urea nitrogen, SGOT, and LDH. lnter.individualvariation was largerthan
intra-individual forall15testsexceptCl, C02,and K. Analytic devi-
variation
ation was as large as or larger than biological (intra- or inter-individual) vari-
ability for constituents, such as calcium, which are under precise homeo-
static control, thus producing artifactual widening in the observed normal
rangeand obscuringthe borderlinebetween normal and abnormal results.
Limitsof tolerable
variability
inlaboratoryproceduresareproposed,based
on composite intraand inter-individual
variation,
thatwould permitmed-
ically
significant to be detected.Such limitswere ex-
deviationsin results
ceeded by current methodology in at least half the 15 tests.

Additional Keyphrases homeosfasis #{149} normal values for common clinical lab-
oratory tests #{149} single vs. repetitive measurements group vs. personal norms
#{149}

T HIS REPORT 1S the third in the present series in important implications for the interpretation of
which we explore the biological control of laboratory data. They indicate that “normal
blood constituents over prolonged periods of time ranges” of test values must he redefined and that
in normal persons (1, 2). Weekly measurements of the borderline between normal and abnormal must
the concentration of 15 serum constituents were be examined closely (3).
used to study the individual degree of homeostatic
control over serum composition. Statistical anal- Materials and Methods
yses of variance were used to isolate analytic con-
tributions to variation in the observed data. The procedures we used have been described in
Factors contributing to variability of data arise detail (1, 2). The group of 68 normal subjects com-
from the process of measurement, biological vari- prised 35 females and 33 males, 47 Caucasians and
ation within each individual, and differences among 21 Negroes, 36 younger than 30 years, and 32 who
individuals owing to genetic or other factors. were 30 or older. Each subject was studied for
Results of this study of 68 normal subjects have several months, with no restrictions on usual
activity or diet except for overnight fasting before
morning blood collections at weekly intervals for
From the Clinical Pathology 1)epartment (E.C., G.Z.W.) and 10 to 12 weeks. On the day of collection, the sepa-
Laboratory of Applied Studies, l)ivision of Computer Research rated serum was analyzed in duplicate for 15 chem-
and Technology (E.K.H.), National Institutes of Health, Be-
thesda, Md. 20014.
ical constituents (Table 1), along with a sample of
‘Deceased Sept. 15, 1970. specially prepared frozen pooled serum (“serum
2 Present address, Research Institute of Laboratory Medicine, pool”). Ten subgroups of six to eight subjects each
The Institute of Medical Sciences, Pacific Medical Center, San
Francisco, Calif. 94115.
were studied during successive three-month periods
Received Sept. 11, 1970; accepted Oct. 5, 1970. for 39 months (1).

1028 CLINICAL CHEMISTRY, Vol. 16, No. 12,1970

 Table 1. Values for Serum Components in a Normal Population Groupa
Test Mean 80’ TestMean 80’ Test Mean 80’

Sodium 139.4 1.9 Magnesium 081 0.07 Uric acid 4.62 1.16
mmol/liter mmol/liter mg/100 ml
Potassium 4.12 0.29 Phosphate 3.49 0.51 Urea nitrogen 13.5 3.2
mmol/Iiter mg/100 mlb mg/lOU ml
Chloride 104.6 2.2 Total protein 6.97 0.50 Cholesterol 205.0 36.0
mmol/liter g/100 ml mg/lOU ml
CO2 27.3 1.9 Albumin 4.20 0.35 SGOT 14.5 4.7
mmol/Iiter g/100 ml Karmen units
Calcium 2.55 0.12 Glucose 94.5 9.7 LDH 328.0 72.0
mmol/Iiter mg/lOU ml Henry units
a The means and standard deviations (se’) are based on all results for each test on 68normalsubjects ofabout 1500individual
(total
analyses for each test).
Phosphate values are in terms of phosphorus content of inorganic phosphate.

The types of variation examined in this study Discussion

lie in two categories-biological and analytic. Com-
ponents of variation were derived by statistical Sources of Variation
analysis of variance, assuming that the variance
components are both independent and additive.3 Analytic variation. The total analytic variance of
tests performed on human serum pools on different
days over several months was substantially larger
Results than that measured within a day as duplicate
Serum values in normal population group. Table 1 variance (1). The long-term variance component
presents the mean and standard deviation of results arising over many days was significant and, in most
for each of the 15 chemical constituents measured. tests, was a major contributor to the total analytic
The population mean is the average of 68 “subject variance, although it has usually been ignored as
means” (each of which is the mean of 10 to 12 it is difficult to detect and to measure reliably. The
pairs of weekly duplicate analyses for each indi- use of an invariant source of pooled human serum
vidual). permitted us to estimate the total analytic variance
Types of variation in serum constituents. The in a subject’s results, and also his underlying
magnitudes of five types of variation are charted personal variation (1, 2).
as coefficients of variation (cv) for all 15 tests in Biological variation. An important finding of the
Figures 1, 2, and 3. The tests fall into three sets. present study is that the “noise” of analytic devi-
(cv)0 is 1 to 4% in the tests of Figure 1, 5 to 10% in ation may often mask the true biological vari-
Figure 2, and 15 to 25% in Figure 3. (CV)p is less ability, both within an individual and among
than 2% in Na, Cl, Ca, and Mg; 3 to 4% in CO2, individuals. In each individual, the “subject mean”
total protein, and albumin; 5 to 7% in K, glucose, of a particular constituent represents the central
cholesterol, and phosphate; 9 to 10% in uric acid tendency, control level, or “set-point” concentra-
(in men) and LDH; and about 12% in urea ni- tion of homeostatic regulation. The isolated per-
trogen, and in uric acid among women. The magni- sonal variation, separated from analytic deviations,
tude of (cv) for SGOT is uncertain because of the measures the extent of fluctuation above and below
large relative analytic variability. the set-point that is allowed by the subject’s
Analytic variation exceeds isolated group vari- homeostatic mechanism. The isolated inter-
ation in Na and Ca, hut (cv)A is about half or less individual (group) variance, separated from both
of (cv)0 for K, phosphate, glucose, urea nitrogen, personal and analytic components, reflects differ-
uric acid, and cholesterol. Analytic variation is ences in homeostatic set-points among different
greater than twice the isolated personal variation individuals, arising from such factors as genetic
in Na, Ca, Mg, and SGOT, hut (cv)A is about half characteristics, diet, physical activity, and age.
of (CV)p in K and uric acid. As inferred from estimates of isolated personal
variation, normal homeostatic regulation of the
serum electrolytes Na, Cl, Ca, and Mg is remark-
Abbreviations used: Symbols and definitions of variance com- ably precise even over periods of several months
ponents common to this series of papers are listed together iii the
previous paper (5, Table 1). Additional symbols used here refer
(Figures 1 and 2); average coefficients of variation,
to coefficient of apparent group variation (cv)G’ and coefficient of (cv), are less than 2%. The highly precise homeo-
apparent personal variation (CV)p’. SGOT, serum glutamic-oxalo- static regulation of serum Na in each normal indi-
acetic transaminase (L-aspartate : 2-oxoglutarate aminotransfer-
ase, EC 2.6.1.1); LDH, lactate dehydrogenase (i.-lactate: NAD ox-
vidual [cv < 1.1% in 95% of the normal subjects
idoreductase, EC 1.1.1.27); cv, coefficient of variation. (2)] is paralleled by the precise control of the set-

CLINICAL CHEMISTRY, Vol. 16, No. 12, 1970 1029

 I I I I -.

0
C
Isolated Group Vanotion
5 ‘ “ ‘-“- Apparent Personal Variation “f(((((((r11(i1U((((((” ‘a ‘b
C
P
- Apparent Group Variation
Isolated Personal Variat,on U P
Analytic Variation A

w
z
L__
____________________
z

Apparent Group VarIation

9 6 3G tsoiat.d Group Variation
p’ ///‘/(jz’’., C d Apparent Personal Variation
_J D Isolated Personal VariatIon
I ________ Analytic Variation
c,A

r a
A A

________________
2 Gti’ ‘i’i
#{176}P IW/lEilllhia1/1/,1Zv’
p V/W/hM/LMM?kS8AA/’A
#{176}Al
I I I I I I
I I I 0 5 tO IS 20 25 30 35 40
0 I 2 3 4 5 6 7 8
COEFFICIENT OF VARIATION - PERCENT
COEFFICIENT OF VARIATION - PERCENT
Fig.3. Types of variationin serum constituents.(See
Fig.1.Types of variationin serum electrolytes
legend to Fig.1)
The extentofvariation for each type is expressedas coefficient
of variation in percent, based on the mean value of the normal group variation
The extentof isolated is shown for the total
group for each test (Table 1). The magnitudes of the apparent normal group and also for demographic classes: uric acid (a,
personal (P’), isolated personal (P), and total analytic (A) Co- women, and b,men), cholesterol (d,women, atorover30years,
efficients of variation are the averages of these values in the and c, allothers),
and SOOT (e, women andf, men).The derived
68 normal subjects. Note that these Coefficients of variation valueforaverageisolated personal variation in SOOT isuncertain
are related by the following two equations: (CV)’ =
(dashed line) because of the large analytic component
V(cv)e + (Cv)p’2, and (CV)p’ V(cv)p’ + (CV)A’
gesting the action of homeostatic control. A con-
trol mechanism for Mg, however, has not yet been
identified (6).
U, Personal variation in the data does not include
any contribution from circadian (diurnal) rhythms
(2), although biological rhythms of longer period-
icity are not excluded.
The best homeostatic regulation within an indi-
vidual, and also the smallest inter-individual
I Apparent
Isolated
GroupVariation
GroupVariation
differences, were found for substances that are
6,
,-t- P Apparent Personal riotion important for the stability of composition and
00 p Ioaiated Personal ‘rIatlon
A Analytic Variation
volume of the extracellular and intravascular
z Cr fluids: Na, Cl, C02, Ca, Mg, albumin, and total
protein. Intermediate degrees of regulation and
-J P
4 A biological variability were found for K and sub-
stances involved in anabolic processes: glucose,
cholesterol, and phosphate. rfhe largest biological
-JP _______
“A variability, both intra-individual (9 to 12%) and
I I I I inter-individual (15 to 25%), were found for serum
0 2 4 6 8 10 12 14 16
constituents that are end-products of catabolism
COEFFICIENT OF VARIATION - PERCENT
(uric acid and urea) or are released from tissues
Fig. 2. Types of variation in serum constituents
(sGoT and LDH enzymes) (Figures 1 to 3).
See legendto Fig.1
Implications for Medical Interpretation

points of serum Na concentration among different The “normal range” of results for a particular
individuals [estimated (cv)G, 0.8%]. test (traditionally the population mean ± 2 SD) is
The close homeostatic control of serum Ca in an customarily obtained from analysis of a single
individual [av (cv), 1.7%1 is understandable, blood collection from each individual in the normal
since this cation is regulated through several group. The standard deviation of these “single-
mechanisms involving parathormone, thyrocal- time” results is the apparent group standard devi-
citonin, and vitamin D (4, 5). A striking finding, ation (sG’) which is affected by all three sources of
however, is the small personal variation of serum variation: group, personal, and analytic (as shown
Mg in individual subjects [av (cv)p, 1.5%] sug- in Figure 4, curve 1, based on s’ for serum Ca).

1030 CLINICAL CHEMISTRY, Vol. 16, No. 12, 1970

Subtraction of the analytic variance component (personal, Sp), inter-individual (group, s0), or com-
from the apparent group variance yields the derived posite personal and group (SB). Since s, is less than
value for (personal plus group) biological variance for most of the tests (Figures 1 to 3), the relative
(SB2) and the corresponding standard deviation SB magnitudes of the three variations usually are:
(which yields curve 2, Figure 4). This narrower Sp < S0 < 8B#{149}
curve represents the distribution of “error-free, Which of these biological elements is particularly
single-time” values that would be obtained if mea- relevant is determined by the interpretive require-
surements could be made without any analytic de- ments. For example, studies of homeostasis within
viation. The narrowest distribution (curve 3, Figure an individual-e.g., circadian patterns-or his
4) represents the homeostatic set-points, or true response to experimental or therapeutic influences
means, based on “error-free, multiple-time” mea- involve the isolated personal variation, s. Eval-
surements on each subject in the group. Thus, the uation of an individual in relation to others or to a
observed normal range can be substantially larger population norm, based on single-time measure-
than the true normal range of either single-time or ments, primarily involves composite biological
multiple-time measurements. Analytic variability variation, Sfi. Similar evaluation based on multiple-
can artif actually widen the distribution of ab- time measurements in each individual involves the
normal as well as of normal values, and thus blur isolated group variation,
the borderline between normal and abnormal re- Medical interpretations usually are based on
sults (3, 7). single-time measurements in patients and also in
To define the degree of analytic resolution in an normal subjects (to establish normal ranges). Thus,
individual’s result that is required for medical use- the most suitable reference for medical use is the
fulness, two types of criteria have been proposed composite biological variation, sA. To minimize
which have stimulated interest in this important artifactual increase in biological variation, the
aspect of laboratory test interpretation: the “allow- analytic standard deviation should be less than
able error” (8) and the “medically significant” half that of the relevant biological variation.4
value (9). The allowable error, defined as a fixed Thus, for medical interpretation a tolerable ana-
fraction (one-fourth) of the normal range, has the lytic variability equal to 0.5 (SB) should be used to
limitation that it becomes less meaningful if revise the previously suggested values for medi-
greater degrees of analytic deviation produce ex- cally significant variation (9). rFhese proposed
pansion of the normal range ol)served. The values values of analytic variability are substantially
for medically significant variation initially pro- lower than the previous values in the case of most
posed were based on subjective judgments derived tests, and are not higher in any of the 12 compari-
from experience with test results that incorporated sons (Table 2).
undetermined degrees of analytic deviation (9). In the present study, the analytic deviations
A more objective criterion can be established by observed in eight of the 15 tests exceeded the
relating analytic deviation to one or more of the tolerable variabilities (up to more than twofold in
underlying biological variations: intra-individual Na and Ca). The analytic deviations (sA in Table
2) are overall averages obtained during several
SERUM CALCIUM, mmol/L years under the usual working conditions of a busy
2.2 2.3 2.4 25 2.6 2,7 2.8 29 hospital laboratory. The real analytic variability
over long periods, involving accuracy and compa-
>.
rability of reference standards, as well as precision,
C-)
z
Li
is exceedingly difficult to determine reliably, and
0
few realistic measures have been reported (7, 8).
Li
a
C.-
The present study emphasizes the need for further
Li improvement in assay instrumentation and pro-
>
I- cedures for quality assurance in the laboratory, as
-J
Li well as for evaluation of the actual analytic vari-
a
ability in relation to the degree of resolution re-
quired for biological or medical interpretation.
50 52
The concepts developed in this report provide a
mEq/i rationale for the clinical practice of obtaining mea-
surements in an individual on several occasions, as
Fig.4.Frequency distribution
ofserum calcium innormal
subjects
The three idealized normal distribution curves are based on
values determined in 68 normal subjects for serum calcium: 1, The effect of analytic variation on artifact.ual increase of

observed, apparent group standard deviation (se’) of 0.235 biological variation (SB) is given by \/2 + The art.ifactual
mEq/liter (0.118 mmol/liter); 5, derived, composite personal. increase in biological variation is only 12% if SA = O.)8B, but
group standard deviation (SB) of 0.166mEq/liter (0.083 mmol/ is 40% or more if SA S. In the present study, at. least half the
liter); 3, derived, isolated group standard deviation (so) of 0.142 tests showed analytic deviations nearly equal to or larger than
mEq/liter(0.071 mmol/liter) the isolated biological variations.

CLINICAL CHEMISTRY, Vol. 16, No. 12, 1970 1031

 Table 2. Comparison of Analytic (84) with “Medically Significant” (SM) Values of Variationa
Test 0.5 (se) Test si 0.5 (au)

Sodium 1.7 2.0 0.5 Total 0.24 0.30 0.22

mmol/liter Protein g/lOO ml
Potassium 0.09 0.25 0.14 Albumin 0.18 0.25 0.15
mmol/Iiter g/100 ml
Chloride 1.2 2.0 0.9 Glucose 3.5 5.0 4.5
mmol/liter mg/lOU ml
Co2 0.9 1.0 0.8 Uric acid 0.22 0.5 0.57
mmol/liter mg/lOO ml
Calcium” 0.085 0.065 0.04 Urea 1.2 2.0 1.5
mmol/Iiter Nitrogen mg/lOU ml
Magnesium” 0.04 0.03 Cholesterol 10.0 20.0 17.0
mmol/liter mg/lOU ml
Phos phate 0.22 0.25 0.23 SGOT 2.7 1.9
mg/lOU ml Karmen units
LDH 38.0 30.0
Henry units
54 = total analyticalstandarddeviation; 8M = “medically significant” standard deviation (9); 0.5 (SB) = “tolerable analytic vari-
ability” = half the composite biological variation (personal and group) of the normal population.
Calcium and magnesium concentration in mmol/liter is the concentration in mEq/liter.

in evaluating suspected hyperparathyroidism by References

determinations of serum calcium at periodic inter-
1. Williams, G. Z., Young, D. S., Stein, M. R., and Cotlove, E.,
vals. The resulting “subject mean” can be a more Biological and analytic components of variation in long-term
discriminating measure to detect milder states of studies of serum constituents in normal subjects. I. Objectives,
abnormality than any single value. Repetitive selection of subjects, laboratory procedures, and estimation of
analytic variance. Curs. CHF:M. 16, 1016 (1970).
measurement alone is inadequate to compensate
2. Harris, E. K., Kanofsky, P., Shakarji, G., and Cotlove, E.,
for poor analytic methodology, but even error-free Biological and analytic components of variation in long-term
analyses performed only once cannot assess per- studies of serum constituents in normal subjects. II. Estimating
biological components of variation. CLIN. CHEM. 16, 1022
sonal variability. Reliable estimates of isolated
(1970).
intra-individual variation may also prove useful,
3. Cotlove, E., Observations on laboratory’ screening of normal
since an impaired regulatory mechanism may be individuals. In Multiple Laboratory Screening; Benson, E. S.,
manifested by an abnormal degree of variation as and Strandjord, P. E., Eds. Academic Press, Inc., New York,
well as by an abnormal mean concentration value. N.Y. 1969, p207.

There is a common tendency in the clinical 4. Maclntyre, I., Calcitonin: A review of its discovery and an
account of purification and action. Proc. Roy. Soc., B. 170, 49
interpretation of laboratory test results either to (1968).
have complete faith in a result as normal or ab- 5. Foster, G. V., and Maclntyre, I., The endocrine control of
normal, or to discount the result as an error (10). calcium metabolism. In Recent Advances in Endocrinology;
Which of these alternative views is adopted in any James, V. H. J., Ed. Churchill, London, 1968, p 95.

instance is inevitably influenced by the most cur- 6. Funk, E. B., and Jones, J. E., Eds., The pathogenesis and
clinical significance of magnesium deficiency. Ann. N.Y. A cad.
rent clinical impression or diagnosis and by prior Sci. 162, 705 (1969).
information on test correlations. The present study 7. Gowenlock, A. H., The influence of accuracy and precision on
examines laboratory data objectively, considering the normal range. Ann. Gun. Biochem. 6, 3 (1969).
the test result not as a value which is either com- 8. Tonks, D. B., A study of the accuracy and precision of clinical
pletely correct or completely erroneous but as one chemistry determinations in 170 Canadian laboratories. CLIN.
CHEM. 9,217 (1963).
which is influenced by sources of variation that can
9. Barnett, R.. N., Medical significance of laboratory results.
be defined and quantified. With this approach, Amer. J. Gun. Pathol. 50, 671 (1968).
correlations of laboratory data with clinical con- 10. Zieve, L., Misinterpretation arid abuse of laboratory tests by
ditions can be re-examined more rationally. clinicians. Ann. N.Y. Acad. Sci. 134, 563 (1966).

1032 CLINICAL CHEMISTRY, Vol. 16, No. 12, 1970