mick
of the fifth percentile as the LLNwhich corresponded to a z score of
definition of abnormal depends entirely on correlation with
Paediatric Respiratory Reviews 12 (2011) 206–207
Contents lists available at ScienceDirect
Paediatric Respirrange of normal is much greater and an forced expiratory volume in
one second (FEV1) that was 15% lower than the mean would be twicenormal is roughly 5 meq/L around 140 meq/L which is less than
4% of the absolute mean value. Hence for someone whose serum
sodium is 130, the z score would be4 or 4 standard deviations away
from the mean although the absolute value would be about 7% less
than of the mean. The implication is that less than 1 in 1000 people
would have a serum sodium this low and it would almost certainly be
associated with a disease state. For lung function measurements, the
1.64 was also identical. There is one major advantage to the new
reference equations in that they cover a wide age range from 4 to
80 years but one major disadvantage in that they are applicable
only to Caucasians. However, it must be recognized that this type
of population study with such a large exclusion criteria and a LLN
as a z score of 1.64 is very different from other biological
measurements used to define normal and abnormal where the
definition of the LLN is a z score of 2. Any validity of such a ‘‘liberal’’Review
Using Reference Values to Interpret Pul
Allan L. Coates
Division of Respiratory Medicine, Physiology and Experimental Medicine, Hospital for S
Toronto Ontario M5G 1X8, Canada
INTRODUCTION
Clearly one of the first decisions that anyone needs to make in a
medical situation is the question of normal or abnormal. This may
be blatantly obviouswith a displaced fracture of a limb or itmay be
muchmore subtle and perhaps not all that well understood such as
a borderline sweat test in a patient with a relative with cystic
fibrosis where the definition of normal becomes a problem. For
many if not most but by no means all biological observations, the
values are normally distributed around the mean. Examples of this
would be height in relationship to age, blood electrolyte values and
spirometric values in relationship to height at a given age. Formost
biological measurement such as blood electrolytes, the standard
assumption is that values within 2 standard deviations or 2 ‘‘z-
scores’’ represent 95% of the population and are normal. It is
implied that the 2.5% above and 2.5% below are ‘‘abnormal.’’
Obviously, such an assumption has serious limitations and is
totally dependent on the population sampled. In other words, if a
measurement of some biological variable was made in 200
absolutely healthy normal individuals, 5 would have to be below
the 95% range of ‘‘normal’’ and another 5 would have to be above.
Themean andmedian of a normally distributed set of valueswould
be the same and could be called the ‘‘predicted value.’’ For the vast
majority of biologicalmeasurements, there is a predicted value and
a range of ‘‘normal’’ around this predicted value. Nowhere in any of
the statistical definitions is there a definition of disease state. That
assumption can only have any validity if being outside the range of
normal is strongly correlated with a defined disease state. For
biological values such as serum sodium, the standard deviation of
the measurement is small, approximately 2.5 and the range ofas great a proportional difference than the personwith hyponatremia
but still well within the normal range. Hence themajor advantage of a
z score is that it defines a degree of ‘‘abnormality’’ in relationship to
E-mail address: allan.coates@sickkids.ca.
1526-0542/$ – see front matter  2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.prrv.2011.01.012onary Function Tests
Children Research Institute, University of Toronto, 555 University Avenue,
the variability of value being measured. This variability is both due to
biological differences and the coefficient of variation around the
measurement. For example, for children, an FEV1 of 80%would be just
at the limit of the normal range (to be defined later) but an FEF25-75 of
70% would be well within the normal range due to the larger
coefficient of variation within this measurement.
When it comes to developing reference values for something
like height, the standard practice would be to measure all children
over a large range of ages and develop a distribution curve for each
age. Perhaps those with some severe height altering disease such
as scoliosis or acromegalymight be omitted but only a tiny fraction
would likely fit into these categories. On the other hand, when the
reference values for spirometry that came out of the NHANES III
American populations study, those subjects who answered yes to
smoking, wheezing, shortness of breath or physician diagnosed
lung disease were excluded1. This meant that from the over 15,000
adults who performed lung function tests, less than 5,000were left
after the exclusion. While the loss of subjects in the paediatric
group was much lower, still more than 25% of the population was
excluded from the final analysis. From this data, the lower limit of
normal (LLN) was defined as the lower 5% of the population in
relation to age, sex and ethnicity (Figure 1). Using regression
analysis, it was possible to develop a series of polynomial curves
specific for race and sex that allowed the prediction of the
spirometric values of interest and the LLN from age and height.
While the new reference equations presented by Sanojevic and
colleagues2 uses a somewhat more sophisticated statistical
analysis3, the majority of the subjects included in that study are
the same subjects as in theNHANES III reference equations. The use
atory Reviewsdisease states.
One significant difference between the interpretation of
spirometric values and most other biological tests is that
descriptors of severity are based on where the patient is in
relation to the percent predicted of a particular value4. In other
words, a patient whose forced expiratory volume in one second