of the series is removed further still from that of the
single study.
All of the foregoing calculations have been
based upon the condition that there is no classifica-
tion correlation between repeat studies. In reality,
classification correlation usually exists (Politser
1982). Because intraindividual variability in study
results is usually fairly small, a repeat study in an
individual is likely to yield a result close to a previ-
ous result and, therefore, to give a similar diagnostic
classification, even if it is a misclassification. When
classification correlation is present, the actual
diagnostic performance of a test series will differ
from that computed under the assumption that the
classification correlation is zero. For the "believe-
the-positive" rule, the sensitivity will be greater than
predicated and the specificity will be less; for the
"believe-the-negative" rule, the sensitivity will be
greater than predicted and the specificity will be
less.
Combination testing.
Two popular positivity
rules for combination testing are analogous to those
used for repeat testing. The "any-test-positive" rule,
for which the test combination is considered positive
if any of the constituent study results are positive, is
the same as the "believe-the-positive" rule. The
"all-tests-positive" positivity rule is equivalent to the
"believe-the-negative" rule for repeat testing. And,
just as for repeat testing, the first rule leads to an
increased sensitivity and decreased specificity
compared to the individual studies and the second
rule results in decreased sensitivity but increased
specificity (Cebul
et al.
1982). This is shown in
Figure 3.6 for the combination of ferritin and trans-
ferrin saturation as comarkers of iron deficiency.
Using the data of Dallman
et al.
(1981), at a critical
value of 10%, transferrin saturation has a sensitivity
of 0.53 and a specificity of 0.75 and, at a critical
value of 10 µg/L, plasma ferritin concentration has a
sensitivity of 0.28 and a specificity of 0.87. If there
is no classification correlation between the studies,
using the any-test-positive rule gives the test combi-
nation a sensitivity of 0.65 and a specificity of 0.65;
using the all-tests-positive rule yields a sensitivity of
0.15 and a specificity of 0.97. The any-test-positive
rule is frequently used in multiphasic health screen-
ing. A multiphasic health screen is a combination of
12, 18, 24 and sometimes more laboratory studies
performed upon a single blood specimen for the
purpose of detecting clinically silent disease in
asymptomatic individuals. The presence of any
study result outside of its reference interval is
supposed to identify persons who should be evalu-
ated further for subclinical disease. The diagnostic
specificity of such testing is low, however. Very
low! For one laboratory study with a reference
interval based upon a specificity of 0.95, the
probability that a healthy person will have a test
result outside of the reference interval is 0.05. For a
combination of
j
uncorrelated laboratory studies,
each of which has a reference interval chosen to give
a specificity of 0.95, the chance of one or more
positive results in a healthy individual is
1 - (0.95)
j
This means that for multiphasic screens of 12, 18,
and 24 tests the probability of at least one positive
result is 0.46, 0.60, and 0.71, respectively, for an
individual who is, in fact, absolutely healthy. In
clinical practice most physicians deal with this
problem by ignoring positive results that represent
only small deviations outside of the reference range.
They do respond to larger deviations. This is appro-
priate because larger deviations are associated with
greater specificities among the individual tests and,
therefore, with a reasonable level of overall specific-
ity for the multiphasic screen.
Diagnostic and Prognostic Classification
3-6
iron-replete
patients
iron-deficient
patients
19
21
>10% <10%
transferrin saturation
72 25
>10% <10%
transferrin saturation
10
3
7
8
>10
<10
>10
<10
ferritin
µg/L
any-test-positive
19 21
>10% <10%
transferrin saturation
72 25
>10% <10%
transferrin saturation
10 3
7
8
>10
<10
>10
<10
ferritin
µg/L
all-tests-positive
Figure 3.6
Application of different positivity rules to combi-
nation testing using transferrin saturation (critical value,
10%) and ferritin concentration (critical value, 10 µg/L). A
classification correlation of zero is assumed.
