SOURCES OF BIOLOGIC VARIABILITY
The two components of biologic variability are
interindividual variability, which is the variability
due to the heterogeneity of physiologic influences
among individuals, and intraindividual variability,
which is due to the variability in individuals over
time. As discussed in Chapter 1, both components
of biologic variability contribute to the distribution
of study values that constitute reference frequency
distributions.
Sex, race, age, and biorhythms are the sources
of biologic variability that will be considered here.
Sometimes one of these sources of variability can be
identified as an important determinant of the refer-
ence frequency distribution. In such circumstances
it is usually desirable to account for that source of
variability when constructing the distribution. This
can be done in two ways. If the source of variability
has a discrete pattern of heterogeneity, the clinical
population can be partitioned into subgroups based
upon that pattern. For instance, sex shows discrete
heterogeneity; if an individual’s sex strongly influ-
ences the value of an analyte, two reference
frequency distributions should be established, one
for males and one for females. If the source of
variability has a continuous pattern of heterogeneity,
such as age, the reference frequency distribution can
be constructed as a continuous function of that
pattern.
Discrete heterogeneity
A number of different approaches can be taken
to determine if a source of discrete heterogeneity
should be taken into account when constructing
reference frequency distributions. One approach is
to decide that partitioning of the population into
subgroups is justified if the difference in the location
or width of the subgroup reference frequency distri-
butions is statistically significant. Because a differ-
ence of any size, even a minute difference, can be
made statistically significant as long as adequate
numbers of individuals are studied, this approach is
not satisfactory. A second approach is to utilize
subgroup frequency distributions if the width of any
of the separate frequency distributions is appreciably
smaller than the width of the frequency distribution
derived from the whole population. This criterion is
appealing because it is based directly on the degree
of variability attributable to an identifiable source of
that variability. There is a practical problem with
the criterion, however, in that the width of the
population frequency distribution may not be much
larger than the widths of the subgroup frequency
distributions, even when the subgroup distributions
are widely separated. This is illustrated for a two
subgroup population in Figure 6.1.
Another approach to subgroup partitioning of
reference frequency distributions is to base the
decision to partition upon the magnitude of the effect
that such partitioning would have upon the diagnos-
tic performance of the study. If partitioning signifi-
cantly improves the performance of a study,
subgroup frequency distributions should be used.
Consider, for example, the two subgroup population
of Figure 6.1. At a study value that yields a speci-
ficity of 0.95 in the combined population, the likeli-
hood ratios differ by 6.3-fold between the two
subgroups. This is a large disparity, one that would
clearly be clinically significant arising as it does in
the range of study values that is frequently pertinent
to diagnostic decision making. Thus, it would be
preferable to utilize subgroup frequency distributions
in this population.
Harris and Boyd (1990, 1991) consider this last
approach to be the most appropriate clinically and
suggest two decision rules based upon it: for a
population with two subgroups, the frequency distri-
bution for a study should be partitioned if either,
mean
1
−
mean
2
var
1
n
1
+
var
2
n
2
>
3
n
avg
120
or
0.67
>
SD
2
SD
1
>
1.5
where
n
1
and
n
2
, are the number of values in the
frequency distribution arising from individuals in the
first and second subgroups, respectively,
n
avg
is the
Biologic Variability
6-1
Chapter 6
BIOLOGIC VARIABILITY
© 2001 Dennis A. Noe
