cite a previously published description of the method
or to refer to the manufacturer’s instructions when a
commercial method is being studied. When any
options exist in the performance of a method, the
options chosen should be indicated.
Study population
Method comparison studies are performed using
clinical specimens that have been submitted to the
laboratory in the course of the medical care of
patients. Most often, the specimens that are used are
those that have been submitted for determination of
the analyte measured by the methods under study.
This is an obvious necessity in the case of xenobiot-
ics but it also makes sense for other kinds of analytes
because the range of values for the analyte is usually
largest among those patients in whom the analyte is
being measured. For instance, HbA
1c
concentrations
are only measured in patients with diabetes. These
patients have values that range from normal to
greatly increased with the majority being slightly to
moderately elevated. A random sampling of labora-
tory specimens would be expected to uncover only a
few specimens with elevated concentrations. In
keeping with this logic, in their comparison study,
Turpeinen
et al.
utilized specimens obtained from
patients with diabetes:
For method comparison we used 123 blood
samples obtained mainly from diabetics sent to
our routine laboratory for HbA
1c
analysis.
A wide clinical spectrum should be represented
by the specimens used in a method comparison
study. This is important for two reasons. First, the
range of values of the analyte often relates to the
spectrum of disease in the patients from whom the
specimens are obtained. For instance, if the speci-
mens that are submitted to the routine laboratory for
HbA
1c
analysis come almost exclusively from outpa-
tients whose disease is well controlled, the values
will be for the most part normal or slightly
increased. The concordance of the methods in this
range may not reflect the concordance at the high
values seen in patients whose disease is out-of-
control. The HbA
1c
values reported by Turpeinen
et
al.
in their article include many above 10% of total
hemoglobin, indicating moderate to severe chronic
hyperglycemia, so their clinical population clearly
encompasses a broad spectrum of glycemic control.
The second reason to seek a broad clinical spectrum
is to guarantee a wide spectrum of biochemical
variability. The wider the biochemical spectrum,
the more likely it is that measurement differences
due to differential method specificity will be
detected.
Evaluation of concordance
There are two general approaches for the evalua-
tion of concordance between the results of field
methods, regression analysis and difference analysis.
Correlation analysis is not a useful approach for
evaluating concordance for a number of reasons
(Bland and Altman 1986, Hollis 1996). First and
foremost of these is that the correlation coefficient is
not a measure of agreement between data pairs but,
rather, is a measure of the goodness-of-fit of a linear
model of the data pairs. Thus, for example, data
pairs that are aligned along the line,
result
method 2
= 10 + 2 result
method 1
will have a perfect correlation coefficient despite the
fact that the data pairs do not agree at all. Another
problem with the correlation coefficient is that its
value depends upon the range of the data analyzed.
The wider the range, the larger the correlation
coefficient. In this way, the inclusion of extreme
data pairs, even pairs with less than average agree-
ment, will inflate the value. Yet another problem
with the correlation coefficient is that it reflects data
variability as well as data linearity. As a result, the
correlation coefficient of two highly precise methods
that, on average, do not agree particularly well may
be larger than the correlation coefficient of two less
precise methods that, on average, agree very well.
Regression analysis.
The goal of regression
analysis, which has been the standard approach for
concordance evaluation for decades, is to define the
functional relationship between the results of the
methods so that it can be compared to the relation-
ship that characterizes ideal concordance. In prac-
tice, this means using a linear regression technique
to find the equation of the line that best fits the
paired result data,
result
method 2
= b0 + b1 result
method 1
The estimated values of the intercept and the slope
are compared to the values expected for perfect
concordance, i.e. an intercept of zero and a slope of
one.
Turpeinen
et al.
used the Deming technique of
linear regression in their comparison study. The
Laboratory Methods
2-24
