CLIN. CHEM. 39/7, 1495-1503 (1993)
Variability of Lipid Measurements: Relevance for the Clinician
Gordon Schectman”3
and Edward
Sasse2
Decreasing the large test variability associated with measurements of blood cholesterol, triglyceride, and highdensity lipoprotein (HDL)- and low-density lipoprotein
(LDL)-cholesterol is likely to improve the classification of
coronary heart disease (CHD) risk and allow improved
monitoring of lipid-lowering treatments. However, improv-
ing test precision will benefit the clinician only if (a) the
analytical test variability is high relative to the biological
test variability and (b) detecting subtle responses to diet
or drug therapy is clinically important. Improving HDL- and
LDL-cholesterol test precision can be expected to increase the clinical usefulness of these measurements
because values for HDL- and LDL-cholesterol correlate
closely with CHD risk; are associated with small, yet
clinically important, changes in response to diet and (or)
drug therapy; and have substantial analytical test variability relative to biological variability. On the other hand,
measurements of both blood cholesterol and triglyceride
have high biological relative to analytical variability, and
do not correlate as closely with CHD risk. Therefore,
further improvements in precision for these measurements are less likely to be useful to the clinician.
IndexIng
cison
Terms: cholesterol
heart disease
triglycerides . accuracy
. pee-
ilpoproteins
Each year, >500 000 people in the US die from
coronary heart disease (CHD).4 Cigarette
smoking, hypertension, and hypercholesterolemia
are the three
principal
preventive risk factors for this disease. In
1984, the National
Heart, Lung and Blood Institute
convened a consensus
panel that recommended
lower
cholesterol threshold values to trigger
treatment
of
hypercholesterolemia. The National Cholesterol Education Program
(NCEP), created by the Institute
in 1985
to “reduce the prevalence of elevated blood cholesterol in
the United States” (1), carried the recommendations of
the 1984 consensus panel one step further.
The report of the Adult Treatment Panel of the NCEP
(2) represented
a major change over earlier guidelines
for several reasons. First, low-density lipoprotein
(LDL)
cholesterol was recognized as a more sensitive predictor
of CHD risk than total cholesterol, and assay of total
Department of Medicine, ‘Division of Endocrinology, Metabolism and Nutrition, and2 Department
of Pathology, Medical College of Wisconsin, Milwaukee, WI 53211.
8Mdress
correspondence
to this author at Division of General
Internal
Medicine, 8700 W. Wisconsin Ave., Milwaukee,
WI
53211.
4Nonstandard
abbreviations:
HDL, high-density lipoprotein;
LDL, low-density lipoprotein; VLDL, very-low-density lipoprotein;
CHD, coronary heart disease; NCEP, National Cholesterol Education Program; CVI, total test variability; CVa, analytical variability; CVb, biological variability; and apo, apolipoprotein.
Received December 4, 1992; accepted February 3, 1993.
was largely relegated
to the status of a
test. Use of LDL-cholesterol
assay was recommended to evaluate
CHD risk and to monitor response
to diet and (or) drug therapy. Second, treatment goals
were well defined, encouraging
the use of frequent
laboratory
tests to determine
whether such goals had
been achieved. Finally, the clinical repercussions of test
inaccuracy
and imprecision were recognized,
and a laboratory standardization panel was created with the aim
of improving accuracy and precision of cholesterol measurements (3). These guidelines,
widely publicized,
stimulated
mass cholesterol screening
programs around
the country. According
to NCEP guidelines,
>41% of
American
adults were estimated
to have abnormally
high screening
values for cholesterol
and to require
further evaluation (4).
These recommendations
placed increased demands on
total cholesterol, serum triglyceride, and high-density
lipoprotein (HDL) cholesterol measurements
to be as accurate and reproducible as possible. Imprecise
measurements will impair the correct classification
of CHD risk
increasing
the frequency of either over- or under-treatment. Further, high test variability
may limit the clinician’s ability to detect an important response to therapy or
to determine a meaningful
reduction in CHD risk.
Our purpose here is to assess whether the analytical
and biological variability
usually associated
with these
lipid measurements
interferes with their clinical usefulness, and to evaluate the extent that reductions in test
variability
will improve the clinician’s ability to evaluate and treat lipid disorders.
cholesterol
screening
Upld and Upoproteln Test VarIability: DefInition and
ClInical SIgnIficance
The test variability
associated
with lipid and lipoprotein measurements has two components. The first component is the natural biological variability
(Vb), which
may be influenced
by many factors, including
recent
diet, day of the week or season, illness, and postural
changes (Table 1). The second component is the analytical variabifity
(Va) due to the imprecision of the assay.
These values are usually expressed
as a coefficient of
variation (CV, CVb) around the test mean. The total
test variability
is also expressed
as a coefficient
of
variation
(CVi). The total test variability
(CVi) is defined (6) as (CV)2 = (Ta)2
+ (7)2
To reduce the CV of any test, three approaches can be
considered.
First, the CV can be decreased by improving the reliability
of the assay in the clinical laboratory.
The Laboratory
Standardization
Panel of the NCEP has
set standards
for total-cholesterol
measurements
(3),
and others have suggested that similar requirements
be
established for other lipid measurements
(7). Figure 1
CLINICAL
CHEMISTRY,
Vol. 39, No. 7,
1993
1495
Table 1. Factors Affecting the Biological VariabilIty of
UpId Measurementsa
Smoking
Exercise
Day of week
Seasonalvariations
Medicationchanges
Fastingstatus
Illness
Pregnancy
Alteration of diet habits
Weight change
Postural changes
Alcohol intake
Conditions of specimen storage
For an excellent detailedreview, see Cooper et al. (5).
Tourniquet time
shows the effects of reducing
CVa on the CV for total
cholesterol,
triglycerides,
HDL-cholesterol,
and LDLcholesterol, in which it is assumed that the CVb for each
test remains
constant.
As illustrated
in Figure 1, decreasing CVa improves CVI. However, a point is reached
for each lipid measurement where further
improvements in test precision
no longer significantly
reduce
CV. For total-cholesterol
and triglyceride
measurements, where the CVa achieved in many clinical laboratories is reported at 3% or less, further improvements
in test precision do not yield meaningful
benefits in CVI.
On the other hand, the CVa reported
for HDL- and
LDL-cholesterol
measurements
currently is high relative to CVb, and further reductions
in test imprecision
decrease CV and improve test performance.
To improve
test precision, an alternative
to reducing CV,, directly is
to perform the assay in duplicate or triplicate and report
the mean value-an
approach commonly used in the
research
setting,
but rarely cost-effective when routinely used in the clinical laboratory.
25
2S
I
Is
II
S
2
4
4
1
ii
12
Analytic VariabilIty (%)
FIg. 1. Effect of test precision on total variability for total cholesterol,
triglycerides, I-IDL-cholesterol, and LDL-choiesterol
Assumes biological variability for blood triglycerides (top line) = 21%, LDLcholesterol (second linefromthe top) = 9.6%, total cholesterol (thi,rI linefrom
the top) = 6.5%, and HDL-cholesterol (bottom llne) = 4.8%. The expected
anat4lcalvariability frequentlyachieved In clinical laboratories is shown by the
x for each laboratory test
1496 CUNICALCHEMISTRY,
Vol. 39, No. 7, 1993
A second approach to reduce total test variability is to
decrease
CVb, such as by assessing
differences
in patients’ characteristics
or behavior that may affect repeated lipid measurements (Table 1). Third, serial specimens can be drawn
1 or 2 weeks apart
and the
individual
values averaged.
The extent to which this
approach will decrease test variability
can be estimated
by dividing CV by the square root of the number of
repeated samples (8). Acknowledging
the large variability of total- and LDL-cholesterol
measurements,
the
NCEP encouraged
that mean values derived from at
least two measurements
be obtained before classifying a
patient at high, borderline, or low risk for CHD (2).
Because the total test variabifity
of most lipid measurements is large, the physician should be aware of the
CV and the potential impact that this variance
may
have on clinical
decision-making.
Reliable
measurements may be particularly
important when:
(a) there is a strong correlation
with CHD risk. Because a primary goal of lipid-disorder
management
is to
reduce CHD risk, an accurate assessment
of risk reduction in response to treatment is necessary. Therefore,
the clinician may accept a larger CV for blood triglyceride values than for HDL-cholesterol values, which
correlate much more strongly with CHD risk. On the
other hand, even small changes in HDL-cholesterol
may
convey important information
regarding
Clii) risk reduction and therefore may be clinically important.
(b) the effect of therapy is expected to be relatively
small. For example, to assess a patient’s response
to
diet, the clinician
may need to detect a decrease
in
LDL-cholesterol
of only 10%. For one to reliably discern
this difference from baseline measurements
(assumed,
for the sake of simplicity, to reflect the patient’s true
LDL-cholesterol value), the CV would need to be less
than one-half of the expected response
to therapy, or
<5%. A 10% reduction in LDL-cholesterol in response to
diet would then reflect a true difference from baseline
within the 95% confidence interval
of the test. This
simple approach to estimating
the CV necessary
to
assess response to treatment will be helpful for individuals who respond at least as well as expected to the
intervention.
A lower CV may still be required to detect
an effect of the intervention
for those who show less
than average responses.
(c) the test is used to monitor response to therapy,
rather than to classify CHD risk status. In the latter
setting,
repeat
measurements
are more practical
because they are usually required
only during the initial
evaluation
of the patient. However, when a test is used
to monitor response to therapy, multiple measurements
may be required
after each therapeutic
change. In our
own clinic, less than half of patients achieve goal LDLcholesterol
concentrations
after treatment
with their
initial therapeutic
regimen, and several drugs and (or)
dose changes are necessary until an optimal response is
achieved.
Obtaining
repeat
measurements
tic intervention
is inconvenient,
expensive and may be difficult
after each therapeutime-consuming,
and
for the physician in
private practice to do. Therefore, reducing test variability to obviate the need for multiple measurements
is a
desirable goal.
Interpreting Test VarIabIlIty: ImplIcations for
Measurement of Total Cholesterol, TriglycerIde, and
HDL- and LDL-Cholesterol
Total Cholesterol
Importance.
Although values for total cholesterol now
are rarely used to monitor response to therapy, careful
attention to accuracy and precision of the total cholesterol measurement
remains important for several reasons. First, it is the primary screening tool for detecting
hypercholesterolemia,
therefore
suboptimal
accuracy
and precision will result in significant misclassification
and result in normocholesterolemic
individuals
receiving diet and (or) drug therapy, and in hypercholesterolemic individuals not receiving correct treatment (9, 10).
Second, most clinical laboratories
use the Friedewald
equation to calculate LDL-cholesterol
(11). Because this
formula includes dividing the triglyceride value by five,
and because values for HDL-cholesterol are usually less
than a third of total-cholesterol
values, the variability
of
the total-cholesterol
measurement
usually contributes a
large part of the variability
observed for LDL-cholesterol. Therefore, accurate decisions regarding classification of CIII) risk and treatment of above-normal
LDLcholesterol
concentrations
depend upon reliable totalcholesterol
measurements.
For these reasons,
the
laboratory
standardization
panel was established to set
appropriate accuracy and precision standards
for total
cholesterol (3). Acceptable standards
for inaccuracy and
imprecision
were considered to be <5%, with further
reductions
to 3% by 1993.
Variability. Figure 1 shows the effect of increasing
analytical
precision of total-cholesterol measurements
on the total variability. Assuming
a biological CV of
6.5% (12), decreasing the CVa from 5% to 3% will reduce
the CV by 1.2%. However, when the analytical
vanability is lowered to <3%, negligible reductions in total
test variability
are achieved.
Figure 2 shows the effect of replicate measurements
on the CVI, assuming a test precision of 3%. At this level
of precision, the analytical
variability
contributes
only
18% of the total test variability,
and therefore replicate
measurements
do not significantly reduce total test
variability.
On the other hand, drawing additional specimens to reduce the CVb has much greater impact.
Factors affecting the biological variability
of lipid and
lipoprotein measurements
are shown in Table 1. Careful
attention to patients’ characteristics
and blood-drawing
and specimen-handling
procedures can be helpful to
reduce
preanalytical
and (or) biological
variability.
Large interpersonal
fluctuations
in these values can
sometimes
be explained
by between-sampling
changes
in the patients’ behavior or life style, or by differences in
blood-drawing
procedures, information that can be elicited either from the patient or laboratory
personnel
(preanalytical
variability).
However,
the extent
to
which these factors affect lipid and lipoprotein measure-
Shugle
Duplicate
Triplicate
Number of ReplIcateMeasurements
Fig. 2. Effectof repeatedmeasurements to reduce total cholesterol
variability
The number of replicate measurements (number of times the assay is
repeated) performedto reduce anal,4lcalvariability is shown on the x-ads. 0,
one bloodspecimendrawn;, two blood specimens drawn; 0, three blood
specimensdrawn; , four blood specimens drawn; +, five bloodspecimens
drawn
ments is usually difficult to quantify in the individual
patient.
In summary,
many laboratories
have been able to
decrease analytical
variability
to <3% for total-cholesterol measurements.
Further limiting
analytical
variability will confer little additional decrease in total test
variability.
In contrast, reducing biological variability
may stifi significantly
improve test performance
(Table
1). Attention to uniform blood-drawing
procedures
and
specimen handling may also be helpful. Alternatively,
multiple
samplings
of serum cholesterol
may partly
overcome this biological variability.
Reducing total test
variability
to <6% by improving
test precision (Figure
2) should lead to fewer misclassification
errors when
individuals are evaluated according to NCEP guidelines
(10). However, the clinical significance of further improvements in test performance
is unclear.
Triglycerides
Importance.
Evaluation
of serum triglycerides
is important
for three reasons. First, values exceeding
11.3
mmol/L (1000 mg/dL) are associated
with pancreatitis,
and individuals
with values >5.65 mmol/L (500 mg/dL)
should be identified and targeted for evaluation
and
therapy (2). Second, in patients with milder increases of
blood triglycerides
(2.80-5.65
mmoIIL; 250-500 mg/dL),
who have fRmilisl combined hyperlipidemia or concomitant abnormalities
in LDL and (or) HDL concentrations, diet and (or) drug therapy can be considered to
reduce CHI) risk (2). In such patients, serial triglyceride
determinations
are necessary
to monitor response to
therapy. Third, values for serum triglyceride provide an
estimate of very-low-density
lipoprotein (VLDL)-cholesCUNICAL CHEMISTRY,
Vol. 39, No. 7, 1993
1497
terol concentrations,
which are necessary
for the calculation of LDL-cholesterol.
VLDL from frankly hypertriglyceridemic
subjects has
been shown to convert monocytes to foam cells and is
toxic to endothelial cells (13); therefore, abnormal VLDL
has atherogenic
properties, at least in vitro. Blood triglyceride values are also associated with CHD in most
prospective epidemiological
studies. However, the extent
to which this relationship
exists independently
of HDLcholesterol is unclear (14). In most studies, the association between
high triglyceride
values and CHD loses
significance
when HDL-cholesterol
and either total- or
LDL-cholesterol
are included in a multivariate
analysis.
Intervention
trials do not show a relationship
between
triglyceride
reductions
and decreased
CIII) outcomes
(15-i?).
For these reasons, LDL- and HDL-cholesterol,
rather than blood triglyceride concentrations,
are used to
classify and monitor CHD risk status.
Variability.
Blood triglyceride
measurements
are associated with much biological variabifity,
and have a
CVb usually
>20% (7, 18-20). Because the CVa of the
test in most clinical laboratories
is usually <5% (7),
>95% of the total test variability
is ascribable to the
biological variability.
Therefore, further improving the
test precision has little effect on reducing CV1, (Figure
1). Drawing
multiple samples, rather than performing
replicate measurements
on the same specimen,
is an
effective approach to reduce CV (Figure 3).
Drug therapy effectively diminishes blood triglyceride
concentrations.
Gemfibrozil
may reduce triglyceride
concentrations
by >40% (15). In hypertriglyceridemic
subjects, niacin and fish oil decrease values for blood
triglyceride
by >25% (21, 22). To reliably detect response to an agent that decreases values by 25%, a test
with a CV of <12.5% would be desirable, according to
the criteria discussed earlier. To achieve this degree of
precision,
three or more blood triglyceride measurements will be necessary,
assuming
typical analytical
and biological variabilities
of 3% and 21%, respectively
(Figure 3). Ensuring
that the patient has fasted before
the test and that uniform
patient preparation,
blooddrawing procedures, and specimen handling
are maintained may also be helpful in minimizing
the large
biological
variability
present in these measurements
(Table 1) and thereby in limiting the number of consecutive blood specimens necessary
to evaluate a response
to these interventions.
In summary, values for blood triglyceride are not used
to classify CHD risk. Because the biological variability
of blood triglyceride
relative to the analytical
variability is high, increasing the analytical
precision of the test
will not improve test performance
notably. Drugs used
to treat high triglyceride
concentrations
frequently decrease values for blood triglyceride by >25%; to reliably
detect differences of this magnitude,
a CV of <12.5%
would be preferable.
Because of the high biological
variability,
the CV usually exceeds 20% (Figure
1).
Multiple samplings,
or minimizing
biological variability, or both, are therefore necessary
to reduce total test
variability.
1498
CUNICAL CHEMISTRY, Vol. 39, No. 7, 1993
22
0-
-_
-o
2$
1$
a..
=
z
a
1’
‘4
ci-
12
-c
p
‘I
SIngle
Duplicate
Triplicate
Number ef Replicate Measurements
Fig. 3. Effect of repeatedmeasurementsto reduceblood triglyceride
variability
Symbols as in Fig. 2
HDL-Cholesterol
Importance.
A strong inverse relationship
exists between HDL-cholesterol concentrations
and CIII) risk
(23-25).
Increases in HDL-cholesterol
were correlated
with a reduction in CHD endpoints (mortality)
in the
Helsinki
Heart
Study (15,26),
and have been associated
of atherosclerosis
as
determined
by sequential
angiography (17). Because
HDL-cholesterol
is a strong independent
predictor
of
CHD risk, its accurate measurement
is essential before
intervention
with diet or drug therapy (2). Although the
NCEP has yet to recommend
specific intervention
to
increase HDL-cholesterol
concentrations
if they are initially low, consideration
to increase
low HDL-cholesterol concentrations
has been suggested if the situation
is associated with undesirable
values for blood triglyceride or LDL-cholesterol
(27). A major multicenter trial
is currently under way to determine whether increasing
HDL-cholesterol
in subjects with normal values for
triglyceride
and LDL-cholesterol
will reduce CIII) endpoints (28).
Unlike interventions
affecting LDL-cholesterol,
diet
and drug therapy are usually associated with only small
changes
in HDL-cholesterol
concentrations.
Therapy
with gemfibrozil
produced a 10% increase
in HDLcholesterol
in the Helsinki
Heart Study (15). Gemfibrozil and lovastatin were recently compared in normolipidemic
men with
subnormal
values
for HDLcholesterol, and neither drug increased HDL-cholesterol
by >12% (29). Niacin has been associated
with somewhat larger increases in HDL-cholesterol,
ranging from
13% to 31%, but frequently is poorly tolerated, owing to
with stabilization and regression
severe flushing
(30). Therapy with estrogen will also
increase HDL, but the increases usually are also in the
range of 10% to 15% (31).
Small changes in HDL-cholesterol
may be clinically
important, The Framingham
risk equation (32), which
accurately predicts the reduction in CHD risk obtained
in intervention
studies (33), can be used to compare the
impact of changes in HDL on expected CHD incidence.
For a nonhypertensive,
nondiabetic
60-year-old
man
with an LDL-cholesterol
of 4.65 mmol/L (180 mg/dL), an
HDL-cholesterol
of 0.80 mmoliL (30 mg/dL), and a
triglyceride
value of 1.70 mmol/L (150 mg/dL), the
likelihood
of developing CHD in the next 10 years is
21%. Figure 4 illustrates
the relative changes in CHD
risk that could be expected by modifying HDL-cholesterol concentrations.
A relatively large (40%) change in
CHD risk occurs when HDL-cholesterol
is modified by
only 0.25 mmol/L (10 mg/dL) in either direction. For this
reason, small changes in HDL-cholesterol
produced by
diet or drug therapy have clinical relevance
to the
clinician. The reliable detection of changes in HDLcholesterol of 0.15 mmol/L (5 mg/dL) or less may be an
important characteristic
of a clinically useful assay.
Variability.
The imprecision of HDL-cholesterol
measurements
and their potential impact on clinical decision-making
have been reviewed (34). HDL-cholesterol
measurements
require two separate manipulations:
the
isolation of the HDL-containing
fraction from plasma or
serum, and then the measurement
of cholesterol in this
fraction. The former step appears to be associated
with
much of the error in precision (35). Interlaboratoiy
analytical
variability
ranging from 9% to 38% has been
reported (36). The CAP Comprehensive
Chemistry Survey showed large interlaboratory
coefficients of variation for HDL cholesterol,
ranging from 11% to 16%,
I
21
I
0.5
regardless
of the precipitation technique used (7). This
may be improving,
although
in the Canadian
interlaboratory proficiency study, almost a third of the laboratories tested were more than ± 10% from the target
HDL-cholesterol
values (37). Some clinical laboratories
have successfully reduced the within-laboratory
analytical variability
for HDL-cholesterol to 5% or less (20),
although test precision at many laboratories may be
considerably higher (34). Although office-type analyzers
for total cholesterol
have achieved accuracy and precision standards
rivaling those of the clinical laboratory,
these techniques
for HDL-cholesterol have been associated with unacceptable
degrees of imprecision (38). As
has been pointed out, the extent of deviation
from
acceptable performance
in most routine clinical laboratories is unknown,
because
participation
in qualitycontrol surveys is usually voluntary (34).
Figure 5 shows the clinical impact of the large analytical variability
of HDL-cholesterol
measurements.
For an analytical
variability
of 8%, the total test variability will be 10% [assuming
biological
variability
of
5.8%, as reported by Mogadam
et al. (20)]; two-thirds
of
the total variabifity
is ascribable
to the poor test precision. A CV this large will make difficult the reliable
detection of HDL-cholesterol
responses to many diet and
drug interventions.
In addition, the accurate
assessment of CHD risk may be impaired.
Decreasing
the CV
to <5% would be necessary to detect a true difference
between baseline and repeat HDL-cholesterol
values of
10% with >95% confidence. To decrease the CV to this
extent, either four separate blood specimens
drawn on
different dates, or three specimens,
each performed
in
triplicate,
would be required (Figure 5).
The right-hand
side of Figure 5 shows results from an
HDL assay with improved analytical precision (analytical CV = 3%). With this improved test, only 21% of total
test variability
is ascribable
to analytical
variability.
Because the analytical
imprecision
is already low, performing the assay in duplicate or triplicate confers little
benefit. However, the total test variability
has been
decreased
to <7% with one specimen,
and only two
specimens are necessary to decrease it to <5%. Improving analytical precision in this case does reduce the total
test variability,
limiting the need for multiple specimens.
In summary,
a significant
component of variability
of
HDL-cholesterol
measurements
is due to the analytical
imprecision
of the test. Because HDL-cholesterol
is a
strong inverse predictor of CIII) risk, and because most
drug and diet interventions
produce
only small increases ( 15%) in HDL-cholesterol, the ability to reliably detect small changes in HDL-cholesterol
is desirable. Improving the precision of the test will improve
the clinician’s ability to accurately
assess CHD risk, and
will also allow a more reliable assessment
of response to
diet and drug therapy.
IIDL Ckoiesterol (mniol/L)
Fig. 4. Effect of a change in HDL-cholesterol concentration on the
percentage change in risk of coronary heart disease
Framingham risk equation used to estimate CHD risk In a 60-year-old
nonsmoking, nonhypertensive, nondlabetic man with a baseline HDL- and
WL-choIesterol of 0.80 and 4.65 mmolIL (30 and 180 moJdL), respectively
LDL-Cholesterol
Importance.
LDL-cholesterol
concentrations
are important determinants
of CHD risk. Multiple intervention trials
demonstrate
that lowering
cholesterol,
CUNICAL CHEMISTRY, Vol. 39, No. 7,
1993
1499
I.
A. Carrent
Performance
II
B. Ideal
Perferinance
$
I
*
4
-
Slalle
Duplks*s
Numberof R,$kate
Tripikat,
emeuts
-
Slugk
Duplicate
TrlpIIc*le
Number .f ReplicateMeasure.eute
Fig. 5. Effect of repeated measurements to reduce HDL-cholesteroi vanability
A: Current test performance, assuming analytical variability of 8%, biologicalvariability of 5.8%. B: Ideal test pertormance if theanalytical vanablllty was reduced
to 3%. Symbols as in Fig. 2
whether by diet or drug therapy, decreases the incidence
of CIII). These findings have been consistent with the
rule of thumb that diminishing
LDL-cholesterol by 1%
decreases
the incidence of CHD by 2% (39). NCEP
guidelines
stress the importance
of setting individual
goal values for LDL-cholesterol,
and of utilizing both
diet and drug therapy,
if necessary,
to achieve these
goals. Therefore, determination
of LDL-cholesterol is an
essential
part of dyslipidemia
evaluation,
and LDLcholesterol determinations
are necessary
to assess response to therapy (2).
Therapy by diet modification in the compliant patient
can produce decreases in LDL-cholesterol in the range of
10% to 15% (40). Psyllium reportedly lowers LDLcholesterol
by 5% to 15% (41, 42). Inhibitors
of hydroxymethylglutaryl-CoA
reductase,
the drug therapy
that most effectively lowers LDL-cholesterol,
can decrease values by 30% or more (43). Niacin and bile acid
sequestrants,
the two first-line
agents in therapy of
hypercholest.erolemia,
usually decrease LDL-cholesterol
by 15% to 25% (16, 44, 45). Because
of the strong
association
between LDL-cholesterol
and CHD risk, and
because important
interventions,
such as diet therapy,
may decrease LDL-cholesterol
by 10% or less, reliable
detection
of small changes in LDL-cholesterol concentrations
is necessary. To reliably detect a difference of
this magnitude,
a reduction
of the total test variability
to <5% would be desirable.
Variability. LDL-cholesterol
is usually estimated
by
use of the Friedewald
equation, which assumes that the
amount
of cholesterol in VLDL can be estimated
by
dividing
the blood triglyceride
concentration
by five
(10). If blood triglyceride
values are <4.52 mmol/L (400
mg/dL), results by use of the Friedewald equation cor1500
CUNICAL
CHEMISTRY,
Vol.39, No. 7,
1993
relate well with Ll)L-cholesterol concentrations
as determined
by ultracentrifugation
(46-48). However, several difficulties
are associated
with this calculation.
First, blood triglyceride
values >4.52 mmoIJL (400 mg/
dL) do not accurately predict VLDL-cholesterol
concentrations; therefore, LDL-cholesterol cannot be estimated
in the setting of hypertriglyceridemin.
Second, the reliability of the LDL calculation depends on the accurate
measurement
of total-cholesterol,
HDL-cholesterol, and
triglyceride concentrations. Poor analytical
precision in
any or all of these measurements will contribute to the
variability
observed in values for LDL-cholesterol. Finally, the biological variability
inherent in each of these
three measurements
will also contribute to the total
variability
of the LDL cholesterol value. The high biological variability
of blood triglyceride,
in particular,
may interfere
with the reliability
of the LDL-cholesterol
calculation.
In laboratories
with good precision for total-cholesterol (CV 3%), HDL-cholesterol (<5%), and triglyceride (<3%) assays, the analytical
variabffity for LDLcholesterol as calculated
by the Friedewald formula has
been reported as <5%, with a biological variability
of
-6% to 10% (19, 20). These components of variability
contribute to a total test variability
of -11% (Figure 6).
To decrease the CV to <5%, evaluation of five blood
specimens
is necessary
(Figure 6A). Because the high
biological variability contributes 80% of the total test
variability,
replicate measurements
of total cholesterol,
HDL-cholesterol, and triglyceride to minimize analytical variability
offer little additional benefit.
To overcome these limitations of the Friedewald equation, direct measurements of LDL-cholesterol are being
developed: selective precipitation of LDL apolipoprotein
12
A. Current Perforaance
8. Ideal
12
Performance
1C
*
$
4
4
4
Sl.Ie
Duplicate
Slule
Triplicate
Number if ReplicateMesauremesta
Duplicate
Thplkate
Numberif Replicate M,asvr,meuls
Fig.6. Effect of repeated measurements In reducing variability in LDL-choiesterol
A: Current test performance, assuming analytical variability of 4.8%, biologIcal variability of 9.6%.
decreasedto 3% and the biologicalvariability to 5.0%. Symbols as in Fig. 2
(apo) B and then quantification
of the cholesterol coprecipitated with the apo B. Several different precipitation
techniques
have been evaluated, including precipitation
with heparin (49), polyvinyl sulfate (50), and dextran
sulfate (51). The accuracy of these methods has been
compared with those of standard ultracentrifugal
techniques and of the Friedewald equation, and the results
have recently been reviewed (52, 53). Results obtained
by use of the Friedewald formula agreed better with
those by ultracentrifugation than did those by any of the
precipitation
methods.
Further,
all three
methods
proved less reliable in the setting of hypertriglyceridemia, and thus failed to overcome a central limitation
of the Friedewald equation.
Figure 6B demonstrates the potential advantages of
using an improved method of LDL separation to measure LDL-cholesterol directly. Direct measurement of
LDL-cholesterol offers the potential
to improve both
analytical
and biological variability, because the precision of the LDL measurement would not depend upon
the analytical and biological variability
present in measurements of triglyceride, total-, and HDL-cholesterol.
In this hypothetical example, the analytical variability
is decreased to 3%, the biological variabifity to 5%. The
total test variability
of one measurement is decreased to
6%, and only two measurements
are required to reduce
it to <5% (Figure 6).
In s1Immsry,
LDL-cholesterol
assay plays a central
role in the evaluation and management of hypercholesterolemia. Detecting
small differences
in LDL-cholesterol is clinically important, both to quanti1y changes in
cardiovascular
risk and to assess response to diet and
drug therapy. The current approach to estimating
LDLcholesterol, the Friedewald calculation, is limited by the
imprecision and biological variability
of the three other
B:
Ideal test performance it the analytical variability was
lipid measurements.
Improved direct approaches
to
LDL-cholesterol
measurement
are needed to overcome
these limitations
and allow a simple, reliable,
and
precise assessment of CHD risk.
Conclusion
Lipid measurements are associated with substantial
test variability.
To treat lipid disorders, the clinician
must fully understand
the implications of this variability. For example, if the expected effect of the intervention is small relative to the CVI, then multiple samplings may be necessary
if a true response to the
treatment is to be reliably detected.
To determine whether improvements in test performance will improve the clinical usefulness of the test,
the clinician must also grasp the relationship between
analytical
and biological test variability. Improving the
precision of the test will be clinically useful only if
analytical
variability
is high relative to biological variability. Even then, this approach will be relevant to the
clinician only if (a) further reductions in test variability
allow a better assessment of clinically important
differences in cardiovascular
risk, or (b) it allows more precise
determination of clinically important
responses
to an
intervention.
Further improvements in HDL- and LDLcholesterol measurements are necessary
for both reasons. On the other hand, improving the precision of the
blood triglyceride assay will not be as useful clinically,
and, in any case, cannot be achieved by reducing analytical variability
alone. Because understanding
the
laboratory characteristics (including
both analytical
and biological variability) and the clinical relevance of a
test are both crucial for an accurate assessment of test
performance,
close communication between clinician
and clinical chemist is mandatory.
Through this collabCLINICAL CHEMISTRY, Vol. 39, No. 7, 1993
1501
oration,
a critical
of the strengths and limiin test
health care delivery.
evaluation
tations of a test will allow specific improvements
performance to optimize
References
1. Cleeman JI, Lenfant C. New guidelines for the treatment of
high blood cholesterol in adults from the National Cholesterol
Education Program. From controversy to consensus. Circulation
1987;76:960-2.
2. The Expert Panel. Report of the National Cholesterol Education Program Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol
in Adults. Arch Intern Med
1988;148:36-69.
3. National Cholesterol Education Program. Current status of
blood cholesterol measurement
in clinical laboratories in the
United States: Report from the Laboratory Standardization Panel.
Clin Chem 1988;34:193-201.
4. Sempos C, Fulwood R, Haines C, Carroll M, Anda R, Wffliamson DF, et al. The prevalence of high blood cholesterol levels
among adults in the United States. JAmMed Assoc 1989;262:4552.
5. Cooper GR, Myers GL, Smith J, Schiant RC. Blood lipid
measurements.
Variations and practical utility. J Am Med Assoc
1992;267:1652-60.
6. Harris BK, Kanofsky P. Shakaiji G, Cotlove E. Biological and
analytical components of variation in long-term studies of serum
constituents in normal subjects, II: estimating biological components of variation. Clin Chem 1970;16:1O-22.
7. Naito HK. Reliability of lipid, lipoprotein, and apolipoprotein
measurements.
Clin Chem 1988;34(Suppl):B84-B94.
8. Jacobs DR., Barrett-Connor
B. Retest reliability of plasma
cholesterol and triglyceride. Am J Epidemiol 1982;116:878-85.
9. Naughton
MJ, Luepker RV, Strickland D. The accuracy of
portable cholesterol analyzers in public screening programs [see
comments]. JAm Med Assoc 1990;263:1213-7.
10. Weisafeld JL, Holloway JJ. Precision of blood cholesterol
measurement
and high blood cholesterol case-finding and treatment. J Clin Epidemiol 1992;45:971-84.
11. Friedewald WT, Levy RI, Fredrickson DS. Estimation
of the
concentration
of low-density
lipoprotein
cholesterol in plasma,
without the use of the preparative centrifuge. Clin Chem 1972;18:
499-502.
12. Cooper GR, Myers GL, Smith SJ, Sampson LI. Standardization of lipid, lipoprotein, and apolipoprotein
measurements.
Clin
Chem 1988;34(Suppl):B95-B105.
13. Gianturco
SH, Bradley WA. Lipoprotein-mediated
cellular
mechanisms
for atherogenesis
in hypertriglyceridemia.
Semin
Thromb Hemost 1988;14:165-9.
14. Austin MA. Plasma triglyceride
and coronary heart disease.
Arterioscier Thromb 1991;1l:2-14.
15. Prick MH, Elo 0, Haapa K, Heinonen OP, Heinsalini P, Helo
P, et al. Helsinki Heart Study: primary-prevention trial with
gemflbrozil in middle-aged men with dyslipidemia.
N Engl J Med
1987;317:1237-45.
16 The Lipid Research Clinics Program.
The Lipid Research
Clinics Primary
Prevention
Trial results. I. Reduction in the
incidence of coronary heart disease. J Am Med Assoc 1984;251:
351-64.
17. Brown BG, Albers JJ, Fisher LI), Schaefer SM, Lin JT,
Kaplan C, et al. Regression of coronary artery disease as a result
of intensive lipid-lowering
therapy in men with high levels of
apolipoprotein
B. N Engl J Med 1990;323:1289-98.
18. Friedlander Y, Kark JD, Stein Y. Variability of plasma lipids
and lipoproteins: The Jerusalem Lipid Research Clinic Study. Clin
Chem 1985;31:1121-6.
19. Bookstein L, Gidding SS, Donovan M, Smith PA. Day-to-day
variability
of serum cholesterol,
triglyceride,
and high-density
lipoprotein cholesterol levels. Impact on the assessment
of risk
according to the National Cholesterol Education Program guidelines [see comments]. Arch Intern Med 1990;150:1653-7.
20. Mogadam M, Ahmed SW, Mensch AH, Godwin ID. Withinperson fluctuations of serum cholesterol and lipoproteins. Arch
Intern Med 1990;150:1645-8.
21. Clofibrate and niacin in coronary heart disease. The Coronary
Drug Project Research Group. J Am Med Assoc 1975;231:361-80.
1502 CUNICAL CHEMISTRY, Vol. 39, No. 7, 1993
22 Schectman G, Kaul S, Kissebah AH. Heterogeneity
of low
density lipoprotein responses to fish oil concentrate in hypertrig-
lyceridemic subjects. Arteriosclerosis 1989;9:345-54.
2& Miller GJ, Miller NE. Plasma high density lipoprotein concentration and development of ischaemic heart disease. Lancet 1975;
i:16-9.
24 Goldbourt V, Holtzman E, Neufeld HN. Total and high density
lipoprotein cholesterol in the serum and risk of mortality: evidence
of a threshold effect. Br Med J 1985;290:1239-43.
25. Castelli WP, Garrison RJ, Wilson PWF, Abbott RD, Kalousdian S, Kannel WE. Incidence
of coronary
heart disease and
lipoprotein cholesterol levels. J Am Med Assoc 1986;256:2835-6.
26. Manninen V, Elo MO, Prick MH, et al. Lipid alterations
and
decline in the incidence of coronary heart disease in the Helsinki
Heart Study. JAm Med Assoc 1988;260:641-51.
27 Grundy SM, Goodman DS, Rifkind BM, Cleeman JI. The place
of HDL in cholesterol management: a perspective from the National Cholesterol Education Program. Arch Intern Med 1989; 149:
505-10.
28. Rubin HE, Robins SJ, Iwane MK, Boden WE, Elam MB, Fye
CL, et al. The VA HDL Intervention
Trial (HIT): secondary
prevention
of coronary
heart
disease
in men with
low HDL-
cholesterol and desirable LDL-cholesterol:
rationale and design.
Am J Cardiol 1993;71:759-65.
29. Vega GL, Grundy SM. Comparison of lovastatin and gemfibrozil in normolipidemic patients with hypoalphallpoproteinemia.
J Am Med Assoc 1989;262:3148-53.
30. Schectman G, Hiatt J, Hartz A. Evaluation of the effectiveness
of lipid-lowering
therapy for treating hypercholesterolemia
in
veterans (bile acid sequestrants,
niacin, psyllium, and lovastatin).
Am J Cardiol 1993;71:759-65.
31. Kissebah All, Schectman G. Hormones and lipoprotein metabolism. In: Shepherd J, ed. Lipoprotein
metabolism, VoL 1(3).
London: WE Saunders Co., 1987:699-725.
32 Anderson
KM, Wilson PWF, Odell PM, Kannel WE. An
updated coronary risk profile. A statement for health professionals. Circulation 1991;83:356-62.
33. Grover BA, Abrahamowicz
M, Joseph L, Brewer C, Coupal L,
Suissa S. The benefits of treating
hyperlipidemia
to prevent
coronary heart disease. J Am Med Assoc 1992;267:816-22.
34 Superko HR.,Bachorik PS, Wood PD. High-density lipoprotein
cholesterol measurements.
J Am Med Assoc 1986;256:2714-7.
35. Bachorik PS, Most B, Lippel K, Albers JJ, Wood PDS. Plasma
lipoprotein
analysis: relative precision of total cholesterol
and
lipoprotein-choleaterol measurements in 12 Lipid Research Clinics
laboratories.
Clin Chem 1981;27:1217-22.
36. Warnick GR, Albers JJ, Teng-Leary
E. HDL cholesterol:
results of interlaboratory
proficiency test. Clin Chem 1980;26: 169-
70.
37. McQueen MJ, Henderson AR, Patten RL, Krishnan S, Wood
DE, Webb S. Results of a province-wide quality assurance program
assessing the accuracy of cholesterol,
triglycerides,
and highdensity lipoprotein cholesterol measurements
and calculated lowdensity lipoprotein
cholesterol in Ontario, using fresh human
serum. Arch Pathol Lab Med 1991;115:1217-22.
38. Bachorik PS, Cloey TA, Finney CA, Lowry DR., Becker DM.
Lipoprotein-cholesterol
analysis during screening: accuracy and
reliability. Ann Intern Med 1991;114:741-7.
39. Lipid Research Clinics Program: the Lipid Research Clinics
Coronary Primary Prevention Trial results: H. The relationship
of
reduction in incidence of coronary heart disease to cholesterol
lowering. JAm Med Assoc 1984;251:365-74.
40. Tyroler HA. Review of lipid-lowering clinical trials in relation
to observational
epidemiologic studies. Circulation
1987;76:515-
22.
41. Bell LP, Hectorne
DB. Cholesterol-lowering
K, Reynolds
H, Balm
TK, Hunninghake
effects of psyffium hydrophilic mucilloid.
Adjunct therapy to a prudent diet for patients with mild to
moderate hypercholesterolemia. JAm Med Assoc 1989;261:341923.
42 Levin EG, Miller VT, Muesing RA, Stoy DB, Balm TK,
LaRosa JC. Comparison
of psyffium hydrophilic mucilloid and
cellulose as adjuncts to a prudent diet in the treatment of mild to
moderate
hypercholesterolemia.
Arch Intern
Med 1990;150:
1822-7.
43. Bradford RH, Shear CL, Chremos AN, Dujovne C, Downton M,
Franklin F, et al. Expanded clinical evaluation of lovastatin (E
XEL) study results. I. Efficacy in modifying plasma lipoproteins
and adverse event profile in 8245 patients with moderate hypercholesterolemia. Arch Intern Med 1991;151:43-9.
44. Henkin Y, Oberman A, Hurst DC, Segrest JP. Niacin revisited: clinical observations on an important but underutilized drug.
Am J Med 1991;91:239-46.
45. Knopp RH, Ginsberg J, Albers JJ, et al. Contrasting effects of
unmodified and time-release forms of niacin on lipoproteins in
hyperlipidemic subjects: clues of mechanism of action of niacin.
Metabolism 1985;34:642-50.
46. DeLong DM, DeLong ER, Wood PD, Lippel K, Rifkind BM. A
comparison of methods for the estimation of plasma low- and very
low-density lipoprotein cholesterol. J Am Med Assoc 1986;256:
2372-7.
47. Warnick GR, Knopp RH, Fitzpatrick V, Branson L. Estimating low-density lipoprotein cholesterol by the Friedewald equation
is adequate for classifying patients on the basis of nationally
recommended cutpoints. Clin Chem 1990;36:15-9.
48 McNamara JR, Cohn JS, Wilson PWF, Schaefer EJ. Calculated values for low-density lipoprotein cholesterol in the assessment of lipid abnormalities and coronary disease risk. Clin Chem
1990;16:36-42.
49. Wieland
H, Seidel D. A simple specific method for precipita-
tion of low density lipoproteins.
J Lipid Res 1983;24:904-9.
50. Assmann G, Jabs HU, Kohnert U, Nolte W, Schriewer H. LDL
cholesterol determination
in blood serum following precipitation
of
LDL with polyvinylsulfate. Clin Chim Acta 1985;140:77-83.
51. Armstrong VW, Seidel D. Evaluation of a commercial kit for
the determination
of LDL cholesterol in serum based on precipitation of LDL with dextran sulfate. Arztl Lab 1985;31:325-30.
52. Seikmeier R, Marz W, Grob W. Insufficient accuracy and
specificity of polyanion precipitation
methods for quantifying lowdensity lipoproteins. Clin Chem 1990;36:2109-13.
53. Rifai N, Warnick GR, McNamara JR, Belcher JD, Grinstead
GF, Frantz ID. Measurement of low-density-lipoprotein cholesterol
in serum: a status report. Clin Chem 1992;38:150-60.
Discussion
Bradley Copeland: Dr. Schectman, I appreciate
the
directness
of your clinical judgment
relating to laboratories. My concept of clinicians includes what I call the
innate
wisdom of clinicians.
You implied
that your
formula came from, I think, this source-innate
wisdom. The only thing in this I would change is I would
divide by 2.8 or 3 instead of by 2. Now, in your slides, the
y-axis demonstrates
total test variability.
How do you
determine
total test variability?
Gordon Schectman:
Total test variability was determined from the equation,
(CV)2 = (CVa)2 + (CVb)2,
assuming that the biological and analytical variabilities
were known. To determine
the effect of replicate measurements,
I used the simple approach of dividing the
analytical
test variability
by the square
root of the
number of replications.
For repeated sampling, the total
test variability
was divided by the square root of the
number of blood samples collected.
Bradley Copeland: My rule of thumb is three times
the standard
deviation.
Neil Greenberg:
I would like to address the case of
high-density
lipoprotein
(HDL). You contrasted
the
ideal performance
of 3% for the analytical
variability
with the current typical performance
of 8%. It still
seemed to me that, because of the large biological
variability,
your graphs indicated substantial
incremental improvement
in the value of the test by doing
replicate samples from the same patient, even when the
analytical
variation
was 3%. Therefore, my question is,
why would we even bother investing
in improvements
in the analytical
variability
if you stifi need to do
multiple sampling on the aptients to get the sensitivity
down to where you could detect real changes of 3-5%,
which seem to be quite significant
in terms of the overall
risk assessment?
Gordon Schectman:
The answer, in my opinion, relies
on the importance of the HDL cholesterol measurement,
and the fact that even improvements
that appear to be
small may be clinically important
if you need to repeat
the test less often. Small improvements
in the total test
variability
for HDL due to improvements
in analytical
precision will make clinical decision-making
easier as
we try to assess small differences from baseline. As you
say, you cannot overcome biological variabifity by improving analytical
variability;
for HDL, however, reducing the analytical
test variabifity
will provide an improvement
in total test variability
that will, in my
opinion, translate
into clinical benefit.
Callum Fraser:
Cholesterol
and lipids provide
an
interesting
example of the generation
of performance
standards.
The biological variability
of cholesterol,
triglycerides,
etc., was all published many years ago. The
messages about the significance
of changes, numbers of
specimens required, and so forth were all quite widely
disseminated
in the literature
of clinical chemistry.
However, after the NCEP guidelines were issued, it was
very interesting
to see a flurry of clinical interest and
papers on biological variability
that did not cite the
literature
of clinical chemistry
but appeared in the
medical journals along with editorials that merely restated what clinical chemists had been saying for years.
This provides the interesting
message that, if clinicians
can provide guidelines
and algorithms,
the data on
biological and analytical variabffity are available, and
the impact of performance
on clinical decision-making
and desirable standards
of performance
can easily be
calculated.
For calculated low-density
lipoprotein
cholesterol, it
is important
that we consider the biological variability
of the calculated
variable
and then work back and
assess the imprecision and inaccuracy
required
for the
components
that make up the variable. Simply to assume that triglycerides
are very poorly controlled homeostatically
and therefore imprecision
of 10% is satisfactory is nonsense.
The result is used to calculate
another result.
CUNICAL
CHEMISTRY,
Vol. 39, No. 7, 1993
1503