example, Figure 1.3 (data taken from Ali
et al.
1978) shows values of the concentration of ferritin in
plasma in patients with iron deficiency and in
patients who are iron replete. There is obviously
considerable measurement variability in both groups.
This variability results in a substantial overlap of
values: the ferritin concentrations in patients with
iron deficiency overlap those in patients with normal
iron stores. Because ferritin concentrations in the
region of overlap can be found in individuals in
either clinical class, for such study results there is
uncertainty as to the correct classification. As will
be discussed in later chapters, to deal with this kind
of classification uncertainty, clinicians must have
access to quantitative descriptions of the distribution
of study results in persons with and without the
disorder. These descriptions are called frequency
distributions. A frequency distribution that arises
from the performance of a defined laboratory study
upon a sample of subjects from a defined, or refer-
ence, clinical population is called a reference
frequency distribution.
Establishing reference frequency distributions
The determination of a reference frequency
distribution for a laboratory study proceeds in the
order listed in Table 1.6 (Solberg 1987a). A
satisfactory statement of the inclusion criteria for a
population of individuals afflicted by a disorder is of
paramount importance. The inclusion criteria,
which amount to the basis for the diagnosis of the
disorder, must rely upon a universally accepted
method for identifying the disease. Such methods
are called reference methods or "gold standards." A
related concern is that the stage or severity of the
disorder be considered when constructing the refer-
ence criteria (Ransohoff and Feinstein 1978). The
inclusion and exclusion criteria applied to popula-
tions free of a disorder will determine the clinical
settings in which these reference distributions will be
useful. The criteria should define persons similar to
those upon whom the laboratory study will be
performed in practice. For instance, if the study is
to be used to screen for a disorder in the general
population, a sample of the general population
should be used. On the other hand, if the study is to
be used exclusively to identify the disorder in a
select subset of patients, the sample should consist of
members of that subset. At the same time, the
sample subjects should represent as broad a
spectrum of biologic variability as is possible within
the confines of the stipulated criteria (Ransohoff and
Feinstein 1978). In particular, sex and age should
be considered when defining the desired subject
composition of the sample because these characteris-
tics contribute so much to the biologic variability of
most laboratory studies. Alternatively, separate
reference distributions can be constructed for males
and females or for certain intervals of age. Parti-
tioning of the clinical population into subsets should
be considered when the subsets show significant
differences in the location of their frequency distri-
butions (Harris 1975, Harris and Boyd 1990).
The calculation of the frequency distribution of
study results involves, at a minimum, the
Laboratory-based Medical Practice
1-7
Bone marrow iron stores
1
10
100
1000
Ferritin (µg/L)
Absent
Present
Figure 1.3
Relationship between bone marrow iron stores
and ferritin concentration in the plasma.
Table 1.6
Steps in the Determination of a Reference Frequency
Distribution
1. Definition of the analytic procedure, equipment, and
reagents that generate the measurement
2. Definition of the conditions under which specimens are
obtained and the procedure for specimen collection,
handling, and storage
3. Definition of the reference population by specification of
the criteria for subject inclusion and exclusion and the
criteria for partitioning subsets of the population
4. Solicitation of subjects and performance of the study
5. Calculation of the study result frequency distribution
