## Abstract

Spirometric lung function is partly determined by sex, age and height (Ht). Commonly, lung function is expressed as a percentage of the predicted value (PP) in order to account for these effects.

Since the PP method retains sex, age and Ht bias, forced expiratory volume
in 1 s (FEV_{1}) standardised by powers of Ht and
by a new sex-specific lower limit (FEV_{1} quotient (FEV_{1}Q)) were investigated to determine which method best predicted
all-cause mortality in >26,967 patients and normal subjects.

On multivariate analysis, FEV_{1}Q was the best predictor, with
a hazard ratio for the worst decile of 6.9 compared to 4.1 for FEV_{1}PP.
On univariate analysis, the hazard ratios were 18.8 compared to 6.1, respectively;
FEV_{1}·Ht^{−3} was the next-best predictor of
survival. Median survival was calculated for simple cut-off values of FEV_{1}Q and FEV_{1}·Ht^{−3}. These survival
curves were accurately fitted (r^{2} = 1.0)
by both FEV_{1}Q and FEV_{1}·Ht^{−3}
values expressed polynomially, and so an individual's test result could
be used to estimate survival (with sd for median survival
of 0.22 and 0.61 yrs, respectively).

It is concluded that lung function impairment should be expressed in a
new way, here termed the FEV_{1}Q, or, alternatively, as FEV_{1}·Ht^{−3}, since these indices best relate spirometric
lung function to all-cause mortality and survival.

From the first scientific recording of lung function data 1, 2,
it was appreciated that the values obtained were dependent upon the subject’s
sex, age and height (Ht). This led to the practice of trying to
take these influences into account by using prediction equations and then
relating the subject's result to the expected value (percentage
of the predicted value (PP)). In the second edition of *Respiratory Function in Disease* by Bates *et al.* 3, it was suggested that, if a lung function index
value were <80% pred, then it was likely to be abnormal. This method
was widely embraced 4 and has
endured, but there are several reasons why this is not a helpful rule of thumb 5, 6.
The European Respiratory Society (ERS) was the first to recommend
the use of standardised residuals (SRs), which are in essence a *z*-score, for determining whether or not an index is outside the normal
range 7. The lower limit of normal (LLN)
is -1.645 SRs, which is an estimate of the lower 5th percentile, and
this method has been recommended by the most recent American Thoracic Society (ATS)
and ERS statement 8 for determining
whether or not a result is abnormal, with PP suggested as the method for expressing
severity.

Whenever lung function is related to a predicted value, it requires accurate
prediction equations against which to compare the subject's data. The
equation must be obtained from a relevant population of subjects, using comparable
equipment and with rigorous technical standards applied. Exactly how normal
these subjects are can be hard to define, and a population that is too pure
may be unrepresentative. Even the best prediction equation has quite wide
95% confidence limits for its predicted value, and so this, in itself,
is an inexact science. An alternative approach has been to standardise spirometric
lung function using a power relationship with Ht that helps to account for
some size and sex difference. This method was shown, in the Framingham study 9, to be helpful in relating lung function
data to subsequent survival, and was found by Fletcher *et al.* 10 to be the best method for evaluating
longitudinal decline in function in chronic obstructive pulmonary disease (COPD).

The purpose of the present article is to explore the limitations and advantages
of these various methods, and to explore other and perhaps better methodologies
so that these can be tested by other researchers in order to help determine
the best way forward. The relationship between lung function and all-cause
mortality was used to explore this since this was information that was readily
available. It has previously been shown that forced expiratory volume in 1 s (FEV_{1}) is a predictor of all-cause mortality in the general population 9, 11–13, and that
this relationship is even stronger for mortality caused by respiratory diseases
related to airflow obstruction 11.

## DATA AND METHODS

### Data sets

In order to help explore various methods for using spirometric lung function
data, three sets of data were used. One was obtained from routine lung function
tests performed at the University Hospitals Birmingham National Health Service
Trust (Birmingham, UK). These data comprise the results of the most
recent attendance tests obtained from 11,972 patients (53% male)
referred for whatever reason for lung function tests, and all had their survival
registered up to October 2008 in UK National Health Service data records.
Tests were performed on equipment validated to conform to ATS/ERS specifications 14, and the data were obtained following
these test criteria. The second set of data were from the Copenhagen City
Heart Study (CCHS) 15,
kindly released to us by P. Lange (Hvidovre Hospital, Hvidovre, Denmark)
in order to facilitate exploration of a novel approach to using lung function
data 11. The lung function data
from subjects entered into the CCHS during the period 1976–1978 and
their survival up to December 2002 were released for analysis. The methods
and background to this large project have been described previously 12, 15. FEV_{1} and forced vital capacity (FVC)
were recorded without prior brochodilatation whilst sitting using a Vitalograph
bellows spirometer (Maids Moreton, UK). Only subjects with at least
two measurements within 5% of each other were included, and the highest
value obtained was recorded. Any subject whose recorded FEV_{1} was <0.3 L (n = 4)
or whose FEV_{1} exceeded FVC (n = 19)
were excluded, leaving data from 13,900 subjects (46% male)
for analysis. The third set of data comprised 1,095 patients (41%
male) with COPD who had had their post-bronchodilator FEV_{1}
recorded and were then followed for 15 yrs in order to explore predictors
of survival 16. When all three
data sets were combined and only those aged ≥20 yrs retained, this
left 26,967 subjects for analysis with regard to FEV_{1} and survival
prediction.

### Methods for expressing FEV_{1} impairment

FEV_{1} has been expressed in a number of ways, as PP (FEV_{1}PP), using European Coal and Steel Community (ECSC)
reference equations, and as FEV_{1} divided by Ht squared (FEV_{1}·Ht^{−2}) 13, 17 and Ht cubed (FEV_{1}·Ht^{−3}) 10. FEV_{1} has also been presented as a SR (FEV_{1}SR), which is derived from:

FEV_{1}SR = (observed FEV_{1}−predicted
FEV_{1})/RSD

where RSD is the residual standard deviation of the prediction equation
used 7. Although the ECSC reference
equations are only relevant for people aged ≤70 yrs, they are frequently
used beyond this age, and we have found the equations to be as good at prediction
as other more age-specific equations up to an age of 95 yrs 18. Plotting the patient data showed that,
regardless of age, there was a flat lower limit to FEV_{1}, as shown
in figure 1⇓. It was found
that the lower 1st percentile of FEV_{1} in the patient group of nearly
12,000 subjects differed between the sexes (0.5 L for males and
0.4 L for females), but did not vary significantly with age at
ages >50 yrs, where more reliable estimates of the 1st percentile
were possible, as shown in figure 2⇓.
It was then decided to standardise FEV_{1} using these sex-specific
lowest 1st percentiles, and this index was termed the FEV_{1} quotient (FEV_{1}Q). It is an index of the number of turnovers of a nominal lower
limit of lung function remaining, and takes into account some sex and size
differences in lung function. When using data from a single sex, FEV_{1}Q has no advantage over raw FEV_{1}.

### Statistics

All analysis was undertaken using Stata/SE version 9.1 (StataCorp,
College Station, TX, USA). Cox's regression models for predicting
survival from FEV_{1} were derived together with age and sex as predictors,
and then without these predictors, since the object was to determine the best
method for using lung function data in a clinical setting in which other factors
would not be explicitly accounted for in decision making. All models were
confirmed to abide by the assumptions implicit in proportional hazard analysis.

## RESULTS

Each method of expressing lung function impairment is taken in turn, with results provided to support or reject its use in this context. Finally, results on survival in the present large data set are explored.

### Percentage of the predicted value

PP methodology has sustained itself over the years, but it has no statistical
basis and can be misleading when comparing different lung function indices.
Figure 3⇓ shows idealised
data for males and females indicating that the true LLN is at different PPs
for different ages, and that this differs between the sexes. It also varies
with Ht. Thus, if the procedure of relating to a predicted value is an attempt
to account for age, Ht and sex, this method indeed conceals influences for
each of these three domains that potentially corrupt the result. The problems
with PP get worse if it is desired to compare results from different indices
because the PP that relates to the estimated 5th percentile is very different
according to the index considered. Table 1⇓
shows the LLN (estimated 5th percentile) expressed as PP for both
sexes for a variety of indices using the ECSC equations 7. The values for LLN range from 58% pred for
residual volume to 87% pred for FEV_{1} as a percentage of
FVC. If the PP 80% rule 3
were used, it might be falsely assumed that the result for residual volume
was extremely low but that the FEV_{1} as a percentage of FVC was
acceptable, whereas they are indeed both equivalent and at the LLN. This table
indicates how difficult and potentially misleading it is to use PP to look
for patterns of abnormality amongst lung function indices.

If PP were a valid method of expressing severity of impairment, then it
might be expected that the lowest FEV_{1}PP seen in patients would
be roughly the same irrespective of sex, age and Ht. Looking at this another
way, if FEV_{1}PP were a valid measure of severity, then young patients
with cystic fibrosis would die with a larger raw FEV_{1} than older
people because their predicted value is larger than that found in older subjects.
Figure 1⇑ shows the FEV_{1} data from all of the present patients plotted against age expressed
as raw FEV_{1}, FEV_{1}PP, FEV_{1}·Ht^{−2}, FEV_{1}SR and FEV_{1}Q. For raw FEV_{1}, it is striking that the lower boundary was roughly the same irrespective
of age. For FEV_{1}PP, the lower limit in the young subjects was lower
than that seen in older subjects, *i.e.* young subjects can survive
with an FEV_{1} that is a much lower PP than can older subjects. For
FEV_{1}·Ht^{−2} and FEV_{1}Q, the lower
boundary is flat, much as for raw FEV_{1}. Figure 1⇑ confirms what is known from clinical practice, *i.e.* that young cystic fibrosis patients can survive with an absolute
FEV_{1} just as low as can 70 yr olds, and so can survive with a much
lower PP 19. This suggests that
FEV_{1}PP is also not the best method for estimating severity.

### Standardised residuals

Use of SRs is the method endorsed by the ATS and ERS in their recommendations
for determining whether or not an individual's lung function is outside
the normal range 7, 8. The SR is commonly used in statistical
analysis, with the term being synonymous with a *z*-score, and was
first used in the context of lung function data in a study looking at patterns
of abnormality in smokers 20.The
advantage of this technique is that the units are the same for all types of
index, and the SR indicates where a subject's result lies with regard
to the Gaussian distribution of the normal population. Since 1.645 SRs
below the predicted value is an estimate of the lower 5th percentile (1.96 SRs
below estimates the 2.5th percentile), a level of deviation from predicted
where clinical interest is to be directed can be decided upon. The ATS and
ERS 8 have suggested that 1.645 SRs
below the predicted value is the level to use in patients to define the LLN.
Implicit in this is that 5% of people who have been judged to be completely
normal would now be considered as abnormal (*i.e.* they are false
positives). For patients or symptomatic subjects, this may be acceptable,
but, if an asymptomatic population of nondiseased subjects were being tested,
the estimated 2.5th percentile might be chosen instead in order to minimise
the number of false positive results.

Using SRs to express the degree of abnormality below the LLN is more problematic
since the predicted values for younger subjects are higher and thus, in terms
of the number of RSDs available to fall, the younger are able to go lower.
This can be seen in figure 3a⇑
for males, where point A is the predicted value for a male aged 25 yrs
of Ht 1.77 m and point B is for a male of the same age and Ht with
an FEV_{1} of 0.6 L. Points C and D are the equivalent for
a male aged 70 yrs. The baseline of zero FEV_{1} is 8.6 and
6.1 SRs below the predicted values for these males aged 25 and 70 yrs,
respectively. The FEV_{1} at B represents 7.44 SRs below the
predicted value and point D represents an FEV_{1} of 4.9 SRs
below the predicted value. It is not possible for the male aged 70 yrs
to have an FEV_{1} of 7.44 SRs below the predicted value since
this would require a negative FEV_{1}, which is nonsensical. If the
two subjects at B and D were indeed equivalently disabled, showed equivalent
symptoms and had similar survival projections, then the SR method would not
appear to reflect properly the degree of impairment. This is borne out in
figure 1⇑, where the SRs
go much lower in the younger subjects than in the older subjects.

### Standardising by powers of height

The Framingham study 9 showed
that FEV_{1} divided by Ht gave a reasonable prediction of long-term
survival. Fletcher *et al.* 10
showed that FEV_{1}·Ht^{−3} as a means of standardising
FEV_{1} was the best method for evaluating lung function decline.
This form of standardisation by Ht takes some sex and size differences into
account, and it is these differences that make use of raw FEV_{1}
problematic, especially when considering data from both sexes together.

Regression of log FEV_{1} against log Ht in the CCHS
data gives a slope of 3.7, but only 0.33 of the variance in ln FEV_{1} was explained by ln Ht. The fit was not very good and this
slope for both sexes suggested Ht to the power of three or four might provide
the best fit. Figure 4⇓
shows histograms of CCHS FEV_{1} data expressed as FEV_{1}PP,
FEV_{1}SR, FEV_{1}Q and FEV_{1}·Ht^{−3}. Since these data were randomly acquired from a normal population with
respect to their lung function, the distribution for a satisfactory method
of expressing lung function for both sexes together should be normal. For
raw data, the histogram would indeed show two separate distributions for males
and females, with their known size differences (skewness of 0.57).
When expressed as PP the distribution was negatively skewed (-0.27),
and the same was true for SRs (-0.26). FEV_{1}Q had a skewness
of 0.23, but FEV_{1}·Ht^{−2} gave a better fit
for a normal distribution, with skewness of 0.15, and the fit was best for
FEV_{1}·Ht^{−3}, with skewness of 0.00.

### Testing lung function impairment and survival

The hardest end-point for lung function impairment to be tested against
is survival, and this is a clearly defined end-point. Table 2⇓ shows the mean ages and mean survival for
each component of this large data set, and table 3⇓ shows the number of subjects, split into 10-yr age
bands, with their mean±sd FEV_{1}SR and survival,
and the percentage of subjects who had died. Receiver operating characteristic (ROC)
curves were calculated to investigate which method of expressing FEV_{1} was best, on its own, at predicting survival, and the area under the
curve was best for FEV_{1}Q (0.631 (95% confidence
limit 0.624–0.637)), with FEV_{1}·Ht^{−3} almost the same (0.626 (0.619–0.633)); next
best was FEV_{1}·Ht^{−2} (0.621 (0.614–0.628)),
followed by raw FEV_{1} (0.606 (0.599–0.612))
and FEV_{1}PP (0.586 (0.579–0.592)), with
FEV_{1}SR being worst at 0.571. Figure 5⇓ shows the ROC curves for FEV_{1}Q and FEV_{1}PP, with FEV_{1}Q being more specific and no less sensitive
than FEV_{1}PP.

The best FEV_{1} predictor of survival on multivariate analysis
was determined from Cox's regression models, which were derived using
each index, sex and age as predictors, with lung function in deciles. The
best model for predicting survival was with FEV_{1}Q, followed by
FEV_{1}·Ht^{−2}, FEV_{1}·Ht^{−3} and then FEV_{1}PP, with each model significantly better
than the next (p<0.05 (likelihood-ratio test)), with
the hazard ratios for the results shown in table 4⇓. The FEV_{1}Q column in table 4⇓ shows that the hazard ratios for older
age groups were smaller for FEV_{1}Q than for FEV_{1}PP, but
that the opposite was true for the hazard ratios associated with worsening
lung function. This indicates that age *per se* plays a smaller part
in prediction of survival for the model using FEV_{1}Q than for the
model with FEV_{1}PP. The data were then split into data set A, comprising
12,181 subjects with an FEV_{1}SR ranging from 0.0 to -1.645 (mean±sd -0.82±0.46), and data set B with 9,630 subjects with
an FEV_{1}SR of <-1.645 (mean±sd -2.81±0.95), *i.e.* all were below the LLN. Cox's models were generated for each
of these two data sets using sex and quintiles of both function and age to
determine whether or not the ability of the various FEV_{1} indices
to predict survival was different in those with better (data set A)
or worse (data set B) lung function. The best model for predicting
survival in both data sets was with FEV_{1}Q, followed, in order,
by FEV_{1}·Ht^{−3}, FEV_{1}·Ht^{−2} and then FEV_{1}PP. For each data set, the FEV_{1}Q model was significantly better than the other three models, and the
FEV_{1}PP model was significantly worse than the others (p<0.001 (likelihood-ratio
test)). Models with FEV_{1}SR were very much worse with
little utility. Thus the superiority of FEV_{1}Q in predicting survival
was not affected by the range of lung function being considered.

Each method of expressing lung function was then split into the top quartile,
as the reference group for normal survival, and then the remaining values
for the index were divided into a further nine bins of subjects using cut-off
levels derived as follows. If Xq were the value defining the upper quartile,
the other bins were defined at Xq, Xq×9/10, Xq×8/10 .
. . Xq×1/10. The worst two groups were combined since the number
of subjects was <10 in the lowest groups. Regression models using only
these bins as predictors were derived without sex or age as predictors, with
the results shown in table 5⇓.
Again, FEV_{1}Q was the best predictor, with FEV_{1}·Ht^{−3} being the next best. Lastly, Cox's regression models were
derived for simple numerical cut-off levels of FEV_{1}Q and FEV_{1}·Ht^{−3} that might easily be applied in lung
function laboratories (<1.0, 1.0–1.9, 2.0–2.9 . . . 6.0–6.9
and ≥7.0 for FEV_{1}Q, with the cut-off levels for FEV_{1}·Ht^{−3} being numerically a tenth of these). Median survival was
calculated for each group, with the results shown in figure 6⇓. The survival curves in figure 6⇓ predicted from FEV_{1}Q and FEV_{1}·Ht^{−3} could each be accurately fitted by polynomial
functions, as presented in table 6⇓.

## DISCUSSION

The present study has shown that the currently widely used PP method is
significantly inferior to other methods for expressing FEV_{1} when
considering the relation between lung function and subsequent survival. In
this context, it has been shown that standardisation using a power of Ht is
much better than using FEV_{1}PP or FEV_{1}SR. Overall FEV_{1}·Ht^{−3} has a slight edge as the best power of
Ht for removing sex and size bias, and, in a random normal population, this
measure was normally distributed. However, the best way of expressing impairment
was with use of FEV_{1}Q, a novel method we propose for expressing
lung function data. Like FEV_{1}PP and FEV_{1}SR, however,
it depends on sex, since the denominator is sex-dependent. Choosing the best
method in a given circumstance is dependent upon the aspect of clinical care
or management that is relevant. We chose the method that was best for predicting
all-cause mortality as this information was readily available to us, and it
is a well-defined end-point that is ultimately the most important outcome
in any medical condition. The results presented here may not be correct for
an alternative end-point, such as symptoms like breathlessness, and this aspect
needs testing in other appropriate data sets. Focussing on respiratory mortality
alone might further enhance the prediction; it has previously been shown,
in a general population sample, that the hazard ratio of cut-off levels of
FEV_{1}·Ht^{−2} for predicting death caused by
respiratory disease were 10 times higher than for all-cause mortality in the
more severely affected subjects 11.

FEV_{1} has been found by many authors to relate to survival in
the general population 9, 11–13, but the exact reason for this is not clear. It is possible
that this link occurs because genes that are associated with worse lung function
are in some way co-located with genes that determine susceptibility to common
diseases, such as cancer or cardiovascular disease. In support of this, the
Framingham study found that FEV_{1} did not relate to survival in
the elderly 9. However, it has
recently been shown that lung function still predicts survival in a cohort
of 95 yr olds 18, when a putative
link between lung function and other disease risks would have been mainly
spent. This suggests that having lower lung function may mean that other diseases
are more likely to be fatal, for example by predisposing to pneumonia following
a stroke, but a firm causal link has not yet been proven.

The problems with the PP method relate to the proportional assumption implicit in this expression and the numerical aspects that lead to PP retaining unwanted age, sex and Ht bias. A further issue is that all assessments of impairment are currently based on looking at how far a subject has fallen from an estimated predicted point that is deemed appropriate for the subject's sex, age and Ht. This predicted value includes a lot of uncertainty, and so the resulting index includes this uncertainty, and perhaps contributes to why the PP method is not the best index. We chose to turn the issue on its head and concentrate on looking at how far a subject is above the bottom line. Using a zero value as the bottom line is no good since, with data for both sexes, size differences obscure the signal. Using Ht standardisation is effective in this respect and it seems, from the combined data here, that standardisation by Ht cubed is better than using lower powers of Ht.

We here propose a new concept for expressing spirometric data, which was
suggested to us from the observation in figure 1⇑ that there is an absolute lower limit of FEV_{1} seen in laboratory testing. It was then found that this limit is slightly
lower in females and, if a subject's FEV_{1} is standardised
by the relevant sex-specific lower limit, this gives the number of turnovers
of FEV_{1} left for the subject. This index, FEV_{1}Q, is
the best overall predictor for use with regard to predicting survival. This
is true with multivariate (table 4⇑),
as well as univariate, analysis (table 5⇑). This represents a change in thinking to concentrate
on what function it is known a subject has left to survive on, rather than
on what it is thought that they might have lost. In a clinical setting, there
is always awareness of the age and sex of the patient and appreciation that
survival is related to both of these attributes. However, an index may be
of greater utility to a clinician if it does not require any additional manipulation
in order to take these into account. With FEV_{1}Q and FEV_{1}·Ht^{−3}, this is possible, but, if FEV_{1}PP were to be used,
then this index must be manipulated with a complex function in order to take
into account the age and sex of the subject before it can accurately be used
to predict survival potential. Since an age effect is retained in FEV_{1}Q and FEV_{1}·Ht^{−3}, it is possible
that these indices might not be so suitable if it were necessary to focus
solely on the extent of lung function abnormality *per se* and totally
avoid any age effect, or, alternatively, if there were a research need to
tease out the exact effect of age, as distinct from lung function, on an aspect
of medical interest. Although FEV_{1}PP attempts to account for age
effects on lung function by use of prediction equations, this method introduces
other age, sex and Ht biases from the equation used, and the assumption of
proportionality and these effects introduce noise in the signal and reduce
the overall ability of a researcher to predict mortality. In analysis of the
effect of time-related exposures on lung function, such as in occupational
medicine, there may be an advantage in using FEV_{1}Q or FEV_{1}·Ht^{−3} as it does not hide such biases. It has
been shown here that FEV_{1}Q and FEV_{1}·Ht^{−3} are the best indices for investigating all-cause mortality, but it
is another question that remains to be tested as to how well they relate to
symptoms and other markers of lung disease.

The results in figure 6⇑,
based on simple cut-off points of the lung function index alone, can be used
clinically to judge the severity of a situation with regard to survival by
using the appropriate polynomial prediction from table 6⇑. This represents a potential benefit to
patients for initiating treatment strategies and for disease severity stratification
in future research into outcomes of lung diseases. Thus, for example, a subject’s
FEV_{1} would first be tested to see whether or not it was outside
the expected range (*e.g.* below the lower 90% confidence
limit, *i.e.* an FEV_{1}SR of <-1.645), and then the
severity of any abnormality could be judged using the estimated median survival
for the FEV_{1}Q. This survival could, if needed, be related to the
estimated survival for the predicted FEV_{1}Q. In relating any other
measurements that might arise in research to severity, using the FEV_{1}Q itself would suffice. Table 7⇓
shows some examples of how lung function results can be expressed using the
present results.

The method for standardising FEV_{1} using the lowest sex-specific
FEV_{1} (FEV_{1}Q) is a way of avoiding the difficulties
of using raw FEV_{1} because of sex and size differences between individuals.
This method expresses an individual's FEV_{1} as the number of
turnovers of the bottom line level of lung function that remain. It has been
found that decline in FEV_{1} in normal subjects is greater in males
than females, and is defined in terms of absolute loss of volume 21. For never-smokers aged >50 yrs
the annual FEV_{1} decline has been estimated to be 28 mL in
males and 22 mL in females 22,
giving a female to male ratio of 0.78, which is approximately the same as
the ratio of a female to male 1st percentile FEV_{1} of 0.8. In patients
with asthma, FEV_{1} decline has been found to be ∼50 mL·yr^{−1} in intrinsic asthmatics and ∼23 mL·yr^{−1} in extrinsic asthmatics 23, and ∼32 mL·yr^{−1} in asthmatic
nonsmokers aged 40–60 yrs and ∼26 mL·yr^{−1} in those aged >60 yrs 21. None of these studies suggested that this loss was a proportional
effect. These longitudinal findings can be easily applied to FEV_{1}Q
data in that, for nonsmokers aged 40–60 yrs, a decrease in FEV_{1}Q of 1.0 would take ∼18 yrs 22. In smokers aged >60 yrs, their accelerated loss
would equate to a decrease in FEV_{1}Q of 1.0 every 10 yrs 21. These estimates of FEV_{1}Q
decline are independent of sex because the proportional difference in lung
function decline between males and females appears to be an approximate fit
for the observed sex difference in 1st percentile FEV_{1}.

As FEV_{1} declines, the likelihood of terminal hypoventilatory
failure increases, and the fact that females appear to have a smaller absolute
lower limit of FEV_{1} can be explained in two ways. Ventilatory dead
space 24, 25 and airway volume 26, which are closely correlated to anatomical dead space, have
been shown to be smaller in females and are related to their overall smaller
stature. Secondly, females also exhibit a lower basal metabolic rate than
males 27 and so have a lower
basal ventilatory demand. Thus females may be able to survive to a lower absolute
FEV_{1} than males.

It is concluded that the FEV_{1}PP method is not ideal for expressing
lung function impairment and should be dropped in favour of a new method of
expressing FEV_{1} impairment called FEV_{1}Q, with FEV_{1}·Ht^{−3} being the next best alternative. Future
work should determine how these expressions of spirometric lung function impairment
relate to symptoms and whether other lung function indices can be managed
in a similar way.

## Statement of interest

None declared.

## Acknowledgments

We thank P. Lange (Hvidovre Hospital, Hvidovre, Denmark) and the research group of the Copenhagen City Heart Study (Epidemiological Research Unit, Bispebjerg University Hospital, Copenhagen, Denmark) for releasing the spirometric data for this analysis. We thank A. Dirksen (Gentofte University Hospital, Hellerup, Denmark) for the inclusion of the 1,095 chronic obstructive pulmonary disease patients followed for 15 yrs.

- Received February 14, 2009.
- Accepted August 27, 2009.

- © ERS Journals Ltd