American Journal of Epidemiology
Copyright © 1998 by The Johns Hopkins University School of Hygiene and Public Health
All rights reserved
Vol. 147, No. 11
Printed in U.S.A.
Pulmonary Function Levels as Predictors of Mortality in a National Sample
of US Adults
Lucas M. Neas and Joel Schwartz
Single breath pulmonary diffusing capacity for carbon monoxide (DLCQ) was examined as a predictor of
all-cause mortality among 4,333 subjects who were aged 25-74 years at baseline in the First National Health
and Nutrition Examination Survey (NHANES I) conducted from 1971 to 1975. The relation of the percentage
of predicted DLC O to all-cause mortality was examined in a Cox proportional hazard model that included age,
sex, race, current smoking status, systolic blood pressure, serum cholesterol, alcohol consumption, body
mass index, percentage of predicted forced vital capacity (FVC), and the ratio of forced expiratory volume at
1 second (FEV.,) to FVC. Mortality had a linear association with the percentage of predicted FVC (rate ratio (RR)
= 1.12, 95% confidence interval (Cl) 1.08-1.17, for a 10% decrement) and a significantly nonlinear association
with the percentage of predicted DLC O with an adverse effect that was clearly evident for levels below 85% of
those predicted (RR = 1.24, 95% Cl 1.12-1.37 for a 10% decrement). The relative hazard for the percentage
of predicted DLCQ below 85% was not modified by sex, smoking status, or exclusion of subjects with clinical
respiratory disease on the initial examination. This association with the percentage of predicted DLC O was
present among 3,005 subjects with FEV, levels above 90% of those predicted. Thus, pulmonary diffusing
capacity below 85% of predicted levels is a significant predictor of the all-cause mortality rate within the
general US population independent of standard spirometry measures and even in the absence of apparent
clinical respiratory disease. Am J Epidemiol 1998;147:1011-18.
adult; mortality; pulmonary diffusing capacity; respiratory function tests
Pulmonary function testing has been recognized as
an important indicator of life expectancy since 1846
(1). In recent prospective studies in Boston, Massachusetts, and Honolulu, Hawaii, lower levels and rapid
declines of forced expiratory flows were associated
with increased rates of all-cause mortality even after
adjustment for cigarette smoking (2, 3). In a US national cohort, a ratio of forced expiratory volume at 1
second (FEV,) to forced vital capacity (FVC) of less
than 70 percent was associated with increased mortality (4). Single breath pulmonary diffusing capacity for
carbon monoxide (DLCO) is a measure of the lung's
capacity for gas exchange that has not previously been
evaluated as a determinant of all-cause mortality. Each
measurement involves the inhalation of air containing a known concentration of carbon monoxide, a
10-second breath-hold, and the measurement of the
remaining concentration of carbon monoxide in the
exhaled breath. Compared with simple spirometry, the
measurement of D L C O is complex and time consuming
and has a coefficient of variation that is considerably
larger than the more standard forced expiratory maneuver. The current study examines pulmonary diffusing capacity as a predictor of all-cause mortality in a
representative sample of US adults. The analysis controls for forced expiratory volumes and flows and
considers the potential for nonlinear relations and the
effects of collinear variables in the model.
DLCO
MATERIALS AND METHODS
Received for publication April 14, 1997, and accepted for publication December 4, 1997.
Abbreviations: Cl, confidence interval; 0LC O , pulmonary diffusing
capacity for carbon monoxide; FEV,, forced expiratory volume at 1
second; FVC, forced vital capacity; NHANES I, First National Health
and Nutrition Examination Survey; OR, odds ratio; RR, rate ratio.
From the Departments of Epidemiology and Environmental
Health, Harvard School of Public Health, and Channing Laboratory,
Brigham and Women's Hospital, Harvard Medical School, Boston,
MA.
Reprint requests to Dr. Lucas M. Neas, Department of Environmental Health, Harvard School of Public Health, 665 Huntington
Ave., Boston, MA 02115.
Sampling frame
The First National Health and Nutrition Examination Survey (NHANES I) collected medical and nutritional information on US residents between 1971 and
1975 (5), providing a representative sample of the
entire adult, noninstitutional, civilian population of the
United States with special emphasis on minority populations. The sampling procedures have been published elsewhere (6, 7).
1011
1012
Neas and Schwartz
A subsample of 6,913 adults aged 25 through 74
years who were cleared for the tests by the examining
physician was given a more complete examination that
included spirometry, pulmonary diffusing capacity,
and an extensive medical history. All subjects completed a medical history questionnaire and were examined by a physician, who administered supplemental respiratory or cardiovascular questionnaires to
subjects with positive histories and noted any positive
clinical findings. The examining physicians excluded
from spirometry a small proportion of subjects with
very severe respiratory disease or other health conditions that would preclude performing spirometry or the
inhalation of carbon monoxide. Subjects with clinical
respiratory disease who could safely perform spirometry were retained in the study.
The NHANES I Epidemiologic Followup Study followed all subjects who were 25—"7r4 years of age at the
time of the baseline examination (8). Vital status was
determined by tracing each subject using telephone
contacts, direct mail, US Postal Service address information requests, National Death Index checks, state
department of motor vehicle listings, state vital statistics files, and field visits to neighbors at last known
address. By the end of the 1986 follow-up, 95 percent
of the original cohort were successfully traced for a
determination of their vital status, with an average
duration of follow-up of 12.8 years (9). This analysis
is based on subjects aged 25-74 years at the baseline
examination who were successfully followed for vital
event status.
Spirometric and DLCO
methods
Pulmonary function measurements were made with
a model 800 electronic spirometer (Ohio Medical Instruments, Inc., Cincinnati, Ohio). Each subject performed five maneuvers for each test set. The subject's
forced vital capacity was defined as the maximum
forced vital capacity from at least two reproducible
and error-free trials. Details of the measurement procedure and a comparison with other studies of pulmonary function level have been published elsewhere
(10, 11). FEV, and FVC are measured in liters.
Measurements of DLCO were performed with a Collins modular lung analyzer (Warren E. Collins, Inc.,
Braintree, Massachusetts) using a gas mixture that
contained air, 10 percent helium, and 0.3 percent carbon monoxide (12). In each of three maneuvers, each
subject performed a maximal inspiration of the gas
mixture, held his or her breath for 10 seconds, and
expired rapidly and completely to residual volume. All
D L C O measurements within 10 percent of the largest
acceptable measurement were averaged to determine
the subject's DLCO. The DLCO measurement procedure
has been published elsewhere (13). The measurement
technique has since been standardized by the American Thoracic Society (14) and the European Respiratory Society (15), but the methods used in the
NHANES I follow a similar framework. The inspired
vital capacities used to calculate DLCO were not
available in the data set. DLCO is measured in
ml/min/mmHg.
Other covariates
Cigarette smoking was assessed by a questionnaire
administered by a trained interviewer, leading to a
classification as never smoker, former smoker, or current smoker. Body mass index was calculated as the
subject's weight (kg)/height (m)2. Samples of venous
blood were collected and analyzed by the Centers for
Disease Control for standard hematologic parameters,
including serum cholesterol. Moderate alcohol consumption was defined as more than one drink each
week but less than one drink each day.
Statistical methods
The percentage of predicted pulmonary function
levels was obtained for FVC, FEV,, and DLCO by
dividing the observed values by the predicted values
based on internal prediction equations and then multiplying the results by 100. Internal prediction equations were derived for FVC and FEV, based on sexspecific models for the natural logarithm of pulmonary
function level as a function of race, the natural logarithms of height and age, the square of the natural
logarithm of age, and an interaction between sex and
the natural logarithm of height. Sex- and race-specific
prediction equations for DLCO were obtained from a
previously published analysis of this data set (16).
The all-cause mortality hazard ratios were estimated
with a Cox proportional hazard model using an S-plus
function that permitted restricted cubic spline functions as well as linear functions of the predictors (17).
The restricted cubic splines are a flexible approach
that can detect nonlinearities in relations. Relations
that appeared linear using spline models were refit as
a linear function. The basic Cox model included sex,
race, age, serum cholesterol, systolic blood pressure,
and current smoking status. Former smoking status
and the natural logarithms of pack-years smoked and
current number of cigarettes smoked were not strong
predictors of all-cause mortality in this data set after
adjustment for age and current smoking status and did
not confound the association of mortality with pulmonary function level.
Previous studies have shown a relation between
all-cause mortality and standard spirometric measures
Am J Epidemiol
Vol. 147, No. 11, 1998
Mortality and Pulmonary Function in US Adults
such as the FEV, and peak expiratory flow rate. Since
diffusing capacity clearly varies with lung surface area
and alveolar volume, any factor associated with decreased alveolar volume, such as the declines in forced
vital capacity associated with age and smoking status,
will be associated with diminished diffusing capacity.
The component of DLCO that is independent of alveolar volume, that is, the proportion of an inhaled bolus
of carbon monoxide that is absorbed during a breathhold of measured duration, is the specific diffusing
capacity. To control for differences in alveolar volume
and to differentiate between diffusing capacity and
spirometry as measures of lung health, we repeated
our regression analysis using forced vital capacity as a
surrogate for alveolar volume, since more precise estimates of alveolar volume were not available in the
data set.
RESULTS
The initial sample of 6,913 adults selected for the
pulmonary function examination included 6,830 subjects aged 25-74 years whose race was reported as
white or black (excludes 72 subjects of other races and
11 subjects of other ages). Subjects were excluded
from spirometry and DLCO measurements by the examining physician on the basis of their health history
and a physical examination. The final data set included
4,333 subjects aged 25-74 years whose race was reported as white or black, consisting of 1,726 never
smokers, 975 former smokers, and 1,632 current
smokers (table 1). Their pulmonary function characteristics are described in table 2.
Nonpulmonary predictors of all-cause mortality
Sex, race, age, serum cholesterol, systolic blood
pressure, and current smoking status were considered
as predictors of the all-cause mortality hazard ratio
prior to the introduction of pulmonary function variTABLE 1. Number of subjects and deaths by sex, clinical
respiratory disease at enrollment, and smoking status at
enrollment among 4,333 adults aged 25-74 years, First
National Health and Nutrition Examination Survey, 1971-1975
Women
Total
With clinical respiratory
disease
Never smokers
Former smokers
Current smokers
Am J Epidemiol
Men
Subjects
(no.)
Deaths
(no.)
Subjects
(no.)
Deaths
(no.)
2,333
241
2,000
362
150
15
212
60
1,216
350
767
136
26
79
510
625
865
64
131
167
Vol. 147, No. 11, 1998
1013
TABLE 2. Distribution of pulmonary function levels at
enrollment by sex, among 4,333 adults aged 25-74 years,
First National Health and Nutrition Examination Survey,
1971-1975
Pulmonary
drftusing
capacity
capacity
(FVC)
(liters)
Forced
expiratory
(low at
1 second
(FEV)
(liters)
(ml/minute/
mmHg)
Women
Third quartile
Mean
First quartile
3.71
3.23
2.76
2.95
2.54
2.14
25.80
22.95
19.90
0.83
Men
Third quartile
Mean
First quartile
5.08
4.41
3.76
3.95
3.34
2.77
34.70
30.26
0.81
0.75
0.71
Forcsd
vital
VlleU
25.75
FEV,/FVC
ratio
0.79
0.74
ables into the model. Age was the strongest predictor
with a 10-year increment of age associated with a rate
ratio of 2.55 (95 percent confidence interval (CI)
2.34-2.78). Current smoking (rate ratio (RR) = 1.88,
95 percent CI 1.58-2.23), male sex (RR = 1.95, 95
percent CI 1.64-2.32), and black race (RR =1.16, 95
percent CI 0.91-1.49) were associated with increased
mortality rates. Moderate alcohol consumption (less
than one drink each day but more than one drink each
week) was associated with decreased mortality rates
(RR = 0.85, 95 percent CI 0.69-1.05). A 15-mmHg
increment in systolic blood pressure was associated
with a rate ratio of 1.13 (95 percent CI 1.07-1.19), and
a 30-mg/dl increment in serum cholesterol was associated with a rate ratio of 1.06 (95 percent CI 1.001.12). Restricted cubic spline models indicated that
these associations were linear for the continuous predictors except for the association with serum cholesterol, which showed a distinct quadratic relation. Inclusion of a quadratic term for serum cholesterol did
not significantly improve the overall model fit, and the
quadratic term did not confound the subsequent pulmonary function associations. Body mass index
showed a nonlinear association with mortality and was
modeled with both linear and quadratic terms. An
increase in the body mass index from 25 to 30 kg/m2
was associated with a rate ratio of 1.45. A 10-cm
increment in height showed an adverse effect in men
(RR = 1.15, 95 percent CI 0.99-1.33) and a protective
effect in women (RR = 0.86, 95 percent CI 0.69-1.06).
Forced expiratory volumes and flows
A 1-liter decrement in FEV, was associated with
increased mortality (RR = 1.56, 95 percent CI 1.35—
1.79) after adjustment for sex, race, age, serum cholesterol, systolic blood pressure, current smoking
1014
Neas and Schwartz
status, body mass index, and moderate alcohol consumption. The introduction of FEV, into the model
reduced the adverse effect of black race (RR = 1.07)
but did not greatly alter the effects of the other covariates. After FEV, was introduced to the model, a 10-cm
increment in height was associated with increased
mortality in men (RR = 1.32, 95 percent CI 1.13—
1.54), but height was not associated with mortality in
women (RR = 0.93). The slope for a 1-liter decrement
in FEV, was steeper in women (RR = 1.81) than in
men (RR = 1.50). This sex difference in the effect of
a unit change in FEV, disappeared when each sex was
evaluated in terms of the sex-specific interquartile
ranges. An interquartile change in FEV, had the same
effect on mortality in men (RR =1.61 for a 1.18-liter
decrement in FEV,) as in women (RR = 1.63 for a
0.82-liter decrement in FEV,). The results for FVC
were similar to those for FEV,. These sex differences
in the effects of FEV, and height suggested that the
model might be improved by substitution of the sexspecific percentage of predicted pulmonary function
based on internal prediction equations that included
height.
Sex-specific internal prediction equations were derived for pulmonary function levels of never-smoking
subjects without clinical respiratory disease using logarithmic models that adjusted for race, the natural
logarithms of height and age, and the square of the
natural logarithm of age (table 3). The percentage of
predicted FVC and FEV, computed from these predicted values had similar distributions for men and
women (table 4). The mean percentage of predicted
pulmonary function levels was less than 100 percent
because of the inclusion of current and former smokers
and subjects with clinical respiratory disease. Height
was uncorrelated with the percentage of predicted
FVC (Pearson's correlation coefficient (r) = —0.04)
or the percentage of predicted FEV, (r = —0.07).
A 10 percent decrement in the percentage of pre-
TABLE 4. Distribution of the percentage of predicted
pulmonary function levels at enrollment by sex among 4,333
adults aged 25-74 years, First National Health and Nutrition
Examination Survey, 1971-1975
%of
predicted
forced vital
capacity
(% FVC)
%of
predicted
forced
expiratory
flow at 1
second
(% FEV,)
Women
Third quartile
Mean
First quartile
109.1
99.8
89.6
109.3
99.1
89.2
106.0
95.8
85.8
Men
Third quartile
Mean
First quartile
108.1
98.6
89.2
106.5
95.5
85.8
105.2
93.7
82.1
%of
predicted
pulmonary
diffusing
capacity
(% DL CO )
dieted FEV, was associated with increased mortality
(RR = 1.15, 95 percent CI 1.10-1.19) after adjustment for sex, race, age, serum cholesterol, systolic
blood pressure, current smoking status, moderate alcohol consumption, and body mass index. Association
of the percentage of predicted FEV, with mortality did
not differ by sex {p = 0.60) or race (p = 0.70). A
restricted cubic spline model revealed that association
of the percentage of predicted FEV, with the natural
logarithm of the rate ratio (ln(RR)) was essentially
linear (figure 1). The exclusion of those subjects with
clinical respiratory disease at enrollment did not alter
the effect of a 10 percent decrement in the percentage
of predicted FEV, (RR = 1.16). The effect was lower
for never smokers (RR = 1.13, 95 percent CI 1.06—
1.21) and current smokers (RR = 1.10, 95 percent CI
1.03-1.17) and higher for former smokers (RR = 1.23,
95 percent CI 1.15-1.32).
A 10 percent decrement in the percentage of predicted FVC also was associated with increased mor-
TABLE 3. Sex-specific internal prediction equations for the natural logarithm of pulmonary function
among 1,642 never smoking adults aged 25-74 years without clinical respiratory disease, First National
Health and Nutrition Examination Survey, 1971-1975
of
subjects
Multiple
R*
Intprppnt
IIIICI l/C^I
Black
race
Natural
logarithm
of height
(cm)
Natural
logarithm
of age
(years)
Squared
natural
logarithm
of age
1,158
484
0.46
0.52
-13.29
-12.35
-0.16
-0.12
1.88
2.61
3.00
0.53
-0.45
-0.11
1,158
0.52
0.50
-11.94
-11.71
-0.19
-0.12
1.61
2.20
2.98
1.28
-0.46
-0.22
bin
INU.
Natural logarithm of forced
vital capacity (FVC)
Women
Men
Natural logarithm of forced
expiratory flow at 1
second (FEV,)
Women
Men
484
Am J Epidemiol
Vol. 147, No. 11, 1998
Mortality and Pulmonary Function in US Adults
a
a:
a
0.775%
85%
95%
100%
105%
Percentage of Predicted FEV1
FIGURE 1. The rate (or hazard) ratio for all-cause mortality as a
function of the percentage of predicted forced expiratory flow at 1
second (FE\A,) among 4,333 adults aged 25-74 years, First National
Health and Nutrition Examination Survey, 1971-1975. The function
is a restricted cubic spline adjusted for sex, race, age, serum
cholesterol, systolic blood pressure, current smoking status, moderate alcohol consumption, and body mass index.
tality (RR = 1.15, 95 percent CI 1.10-1.20) after
adjustment for sex, race, age, serum cholesterol, systolic blood pressure, current smoking status, moderate
alcohol consumption, and body mass index. Addition
of the percentage of predicted FVC to a model with the
percentage of predicted FEV, indicated that the former
had no independent explanatory power (p = 0.49).
However, the effect of the percentage of predicted
FEV, on mortality could be partitioned into an effect
of volume (RR = 1.14, 95 percent CI 1.09-1.19 for a
10 percent decrement in the percentage of predicted
FVC) and an effect of relative flow (RR = 1.12, 95
percent CI 1.02-1.23 for a 10 percent decrement in the
FEV,/FVC ratio) in the same model.
1015
1.97). The slope for women (RR = 1.98) was higher
than for men (RR = 1.49, p = 0.09). Subjects with
D L C O at enrollment in the lowest decile for their sex
(below 21.8 for men and 17.5 for women) had greatly
increased mortality among both women (RR = 1.93,
95 percent CI 1.44-2.57) and men (RR = 1.59, 95
percent CI 1.23-2.04).
A 10 percent decrement in the sex- and race-specific
percentage of predicted D L C O also was associated with
increased mortality (RR = 1.11, 95 percent CI 1.051.16) after adjustment for sex, race, age, serum cholesterol, systolic blood pressure, current smoking status, moderate alcohol consumption, and body mass
index. Association of the percentage of predicted
D L C O with the natural logarithm of the rate ratio for
all-cause mortality was significantly nonlinear (p <
0.001 for a restricted cubic spline model) with a higher
rate ratio that was clearly evident for DLCO below 85
percent of the predicted value (figure 2). The shape of
this empiric curve suggested that the cubic spline
function could be approximated by a piecewise linear
model with an arbitrary breakpoint chosen at 85 percent of that predicted.
Below 85 percent of the predicted value, each 10
percent decrement in the percentage of predicted DLCO
was linearly associated with increased mortality
(RR = 1.33, 95 percent CI 1.21-1.45), while levels
above 85 percent of the predicted value were not
associated with all-cause mortality (p = 0.86). The
slopes were the same for men and women (p = 0.89).
Adjustment for either a linear or a cubic spline function of hemoglobin concentration did not alter the
association. The exclusion of those subjects with clin1.5-
Pulmonary diffusing capacity
A 5-ml/min/mmHg decrement in D L C O was associated with increased mortality (RR = 1.17, 95 percent
CI 1.07-1.27) after adjustment for sex, race, age, serum cholesterol, systolic blood pressure, current smoking status, moderate alcohol consumption, and body
mass index. The association of DLCO with the natural
logarithm of the rate ratio for all-cause mortality was
significantly nonlinear (/? = 0.002 for a quadratic
term) with a higher rate ratio only for DLCO below 25
ml/min/mmHg. A piecewise regression with a separate
term for DLCO in the lowest quartile (sex specific: 25.8
for men and 19.9 for women) indicated that D L C O
above the lowest quartile had no association (p =
0.59). In the lowest quartile, each 5-ml/min/mmHg
decrement in D L C O was associated with greatly increased mortality (RR = 1.57, 95 percent CI 1.25Am J Epidemiol
Vol. 147, No. 11, 1998
m
<r
S
ira
1.0-
75%
85%
95%
100%
105%
Percentage of Predicted DLco
FIGURE 2. The rate (or hazard) ratio for all-cause mortality as a
function of the percentage of predicted pulmonary diffusing capacity (DLCQ) among 4,333 adults aged 25-74 years, First National
Health and Nutrition Examination Survey, 1971-1975. The function
is a restricted cubic spline adjusted for sex, race, age, serum
cholesterol, systolic blood pressure, current smoking status, moderate alcohol consumption, and body mass index.
1016
Neas and Schwartz
ical respiratory disease at enrollment did not alter the
effect of a 10 percent decrement in the percentage of
predicted DLCO below 85 percent of the predicted
value (RR = 1.32). Low DLCO was not simply a
surrogate for current or cumulative cigarette smoking.
Adjustment for linear and quadratic functions of packyears and current cigarettes smoked per day only
slightly reduced the association of all-cause mortality
with the percentage of predicted DLCO below 85 percent of the predicted value (RR = 1.29). However, the
effect was greater for former smokers (RR = 1.56, 95
percent CI 1.32-1.84) than for either current smokers
(RR = 1.25, 95 percent CI 1.11-1.41) or never smokers (RR = 1.33, 95 percent CI 1.02-1.74).
Almost one quarter of the NHANES I cohort had
DLCO measurements below 85 percent of predicted
values. In this subcohort with a low percentage of
predicted DLCO, there were 223 never smokers (13
percent of all never smokers), 194 former smokers (20
percent of all former smokers), and 772 current smokers (47 percent of all current smokers). The predictors
of a binary indicator of £>L CO measurements below 85
percent of predicted values were examined among the
2,328 current and former smokers without clinical
respiratory disease at enrollment. The predictors of
low D L C O measurements were the natural logarithm of
current cigarettes smoked per day (odds ratio (OR) =
2.98 for the first 20 cigarettes per day), the natural
logarithm of pack-years smoked (OR = 1.78 for the
first 20 pack-years), black race (OR = 1.54), and lean
body mass (OR = 1.11 for a unit decrement in body
mass index). Of the 223 never smokers with DLCO
measurements below 85 percent of predicted values,
83 (37 percent) reported a spouse who smoked cigarettes. When the sample was restricted to 3,005 subjects whose FEV, was greater than 90 percent of
predicted values, the percentage of predicted D L C O
below 85 percent of the predicted value showed the
same association with increased mortality (RR = 1.30
for a 10 percent decrement, 95 percent CI 1.08-1.56).
In a model with both the percentage of predicted
FEVj and the percentage of predicted DLCO below 85
percent of predicted values, the effect of a 10 percent
decrement in the percentage of predicted FEV, was
unchanged (RR = 1.12, 95 percent CI 1.08-1.17),
while the effect of a 10 percent decrement in the
percentage of predicted D L C O below 85 percent of the
predicted value was reduced (RR = 1.22, 95 percent
CI 1.11-1.35). When the effect of the percentage of
predicted FEV, was partitioned into the effects of the
percentage of predicted FVC and the FEV,/FVC ratio
in a model with the percentage of predicted DLCO
below 85 percent of that predicted, a 10 percent decrement in the FEV,/FVC ratio was not associated with
all-cause mortality among women (p = 0.99) and only
weakly associated with mortality among men (RR =
1.09, 95 percent CI 0.98-1.22). Even for men, the
model with the percentage of predicted FVC and the
FEV,/FVC ratio was not as predictive of all-cause
mortality as the model with the percentage of predicted FEV,, after adjustment for sex, race, age, serum
cholesterol, systolic blood pressure, current smoking
status, moderate alcohol consumption, body mass index, and the percentage of predicted DLCO below 85
percent of that predicted. Association of the percentage of predicted £>L CO below 85 percent of the predicted value with mortality was unaltered by substitution of the percentage of predicted FVC and the FEV,/
FVC ratio for the percentage of predicted FEV, in the
model.
DISCUSSION
In the NHANES I cohort of 4,333 men and women
aged 25-74 years who were followed for an average of
12.8 years, a 1-liter decrement in FEV, was associated
with increased mortality (RR = 1.59, 95 percent CI
1.38-1.83) after adjustment for sex, race, age, serum
cholesterol, systolic blood pressure, current smoking
status, moderate alcohol consumption, and body mass
index. In the Normative Aging Study of 1,956 men
aged 21-80 years who were followed for 30 years,
Weiss et al. (2) found that a 1-liter decrement in FEV,
at enrollment resulted in a similar association with
all-cause mortality (RR = 1.67, 95 percent CI 1.252.22) after adjusting for age, white cell count, serum
cholesterol, systolic blood pressure, and current smoking status. Among the 2,000 men in the NHANES I
cohort, the association was similar (RR = 1.61, 95
percent CI 1.36-1.90).
Based on an internal prediction equation, a 10 percent decrement in the percentage of predicted FEV,
was associated with a rate ratio of 1.13 among never
smokers (95 percent CI 1.06-1.21) after adjustment
for sex, race, age, serum cholesterol, systolic blood
pressure, current smoking status, moderate alcohol
consumption, and. body mass index. In a previous
analysis of the same NHANES I cohort using an
external prediction equation, Bang et al. (4) reported
that a 10 percent decrement in the percentage of predicted FEV, was associated with a rate ratio of 1.14
among never smokers after adjustment for sex, age,
body mass index, clinical respiratory disease at enrollment, and high school graduation. The trivial difference in these two values suggests that neither the
prediction equation nor the choice of model covariates
is important.
For values below 85 percent of those predicted, each
10 percent decrement in the percentage of predicted
Am J Epidemiol
Vol. 147, No. 11, 1998
Mortality and Pulmonary Function in US Adults
was associated with increased mortality (RR =
1.33, 95 percent CI 1.21-1.45) after adjustment for
sex, race, age, serum cholesterol, systolic blood pressure, current smoking status, moderate alcohol consumption, and body mass index. This association was
present even for subjects without clinical respiratory
disease at enrollment and for never-smoking subjects.
While somewhat reduced, this association remained
after controlling for the percentage of predicted FEV,
or the percentage of predicted FVC and the FEV ,/FVC
ratio. This association still remained when the sample
was restricted to subjects whose FEV, levels were
clinically unremarkable (an FEV, greater than 90 percent of that predicted). Hence, pulmonary diffusing
capacity contains additional information about health
status that is not contained in standard spirometric
measurements.
The choice of a prediction model in the calculation
of the percentage of predicted FEV, or the percentage
of predicted DLCO is not a decisive factor in the
analysis. We used an internal prediction equation, but
other prediction equations would have performed
equally well. In fact, a simple prediction equation
based solely on height (£)Lco/height2) had the same
^-statistic and overall model fit for survival as the more
detailed prediction equation. Hence, the predictive
power of the percentage of predicted DLCO for survival is not dependent on the specific nature of our
prediction equation for DLCO.
These findings are biologically plausible. The standard spirometric measurements (FVC, FEV,, and the
FEV,/FVC ratio) primarily describe the ability to mechanically exchange air. Controlling for those factors,
D L C O provides information about the ability of the
alveolar region to diffuse gases and the effective perfusion of the alveolar surface. We have previously
shown that cigarette smoking provides larger and earlier effects on DLCO than on FVC (12). Thus, in the
current study of pulmonary function as a predictor of
life expectancy, DLCO measured at enrollment may
serve as a better indicator of the initial damage due to
cigarette smoke and other environmental factors than
FVC or FEV,. While active cigarette smoking is a risk
factor among current and former smokers, the association between D L C O and mortality among never smokers suggests that other environmental factors such as
environmental tobacco smoke play a substantial role.
Damage to the alveolar region can occur with little
effect on FEV, such as occurs in the early stages of
pulmonary interstitial diseases. Isolated DLCO deficits
can be seen in the early stages of interstitial pulmonary
fibrosis, even with normal vital capacity (18), and can
occur before the disease is clinically manifest. In patients with isolated DLCO reduction and normal FEV,
DLCO
Am J Epidemiol
Vol. 147, No. 11, 1998
1017
and FVC, the flows and the FEV ,/FVC ratio are
unusually high and reflect increased elastic recoil (19).
Normal subjects have considerable reserve diffusing
capacity above the minimum required to fully saturate
the pulmonary blood circulation, and all-cause mortality was not associated with increments of the percentage of predicted DLCO for the 73 percent of our subjects with the percentage of predicted D L C O levels
above 85 percent of predicted values. For the 27
percent of our subjects with the percentage of predicted DLCO levels below 85 percent of predicted
values, decrements in the percentage of predicted
DLCO levels were linearly associated with increased
all-cause mortality even after adjustment for spirometric measures.
These results indicate that decrements in pulmonary
function (RR = 1.15 for a 10 percent decrement in the
percentage of predicted FEV,) or diffusing capacity
(RR = 1.33 for a 10 percent decrement in the percentage of predicted D L C O below 85 percent of that predicted) are as important a determinant of all-cause
mortality as increments in serum cholesterol (RR =
1.05 for a 30-mg/dl increment) and systolic blood
pressure (RR = 1.14 for a 15-mmHg increment).
These results indicate that pulmonary diffusing capacity plays a role in all-cause mortality within the general US population independent of standard spirometry
measures and even in the absence of apparent clinical
respiratory disease.
ACKNOWLEDGMENTS
Dr. Neas was supported in part by National Research
Service Award HL-07427 from the National Institutes of
Health and by National Institute of Environmental Health
Sciences Occupational and Environmental Health Center
grant ES-00002. Dr. Schwartz is supported in part by a John
D. and Catherine T. MacArthur Fellowship.
REFERENCES
1. Hutchinson J. On the capacity of the lungs, and on the respiratory functions, with a view of establishing a precise and easy
method of detecting disease by the spirometer. Med Chir
Trans 1846,29:132-252.
2. Weiss ST, Segal MR, Sparrow D, et al. Relation of FEV, and
peripheral blood leukocyte count to total mortality. The Normative Aging Study. Am J Epidemiol 1995;142:493-8.
3. Rodriguez BL, Masaki K, Burchfiel C, et al. Pulmonary function decline and 17-year total mortality: the Honolulu Heart
Program. Am J Epidemiol 1994;140:398-408.
4. Bang KM, Gergen PJ, Kramer R, et al. The effect of pulmonary impairment on all-cause mortality in a national cohort.
Chest 1993;103:536-40.
1018
Neas and Schwartz
5. National Center for Health Statistics, US Public Health Service. Plan and operation of the Health and Nutrition Examination Survey: United States, 1971-73. Washington, DC: US
Government Printing Office, 1973. (Vital and health statistics,
Series 1: Programs and collection procedures, no. 10a)
(DHEW publication no. (PHS) 73-1310).
6. National Center for Health Statistics, US Public Health Service. Plan and initial program of the Health Examination
Survey. Washington, DC: US Government Printing Office,
1965. (Vital and health statistics, Series 1: Programs and
collection procedures, no. 4) (DHEW publication no. (PHS)
1000).
7. National Center for Health Statistics, US Public Health Service. Plan and operation of the HANES I augmentation survey
of adults 25-74 years: United States, 1974-75. Hyattsville,
MD: US Department of Health, Education, and Welfare, 1978.
(Vital and health statistics, Series 1: Programs and collection
procedures, no. 14) (DHEW publication no. (PHS) 78-1314).
8. National Center for Health Statistics, US Public Health Service. Plan and operation of the NHANES I Epidemiologic
Followup Study, 1982-84. Hyattsville, MD: US Department
of Health and Human Services, 1987. (Vital and health statistics, Series 1: Programs and collection procedures, no. 22)
(DHHS publication no. (PHS) 87-1324).
9. National Center for Health Statistics, US Public Health Service. Plan and operation of the NHANES I Epidemiologic
Followup Study, 1987. Hyattsville, MD: US Department of
Health and Human Services, 1992. (Vital and health statistics,
Series 1: Programs and collection procedures, no. 27) (DHHS
publication no. (PHS) 92-1303).
10. National Center for Health Statistics, US Public Health Service. Computer-assisted spirometry data analysis for the National Health and Nutrition Examination Survey, 1971-80.
Washington, DC: US Government Printing Office, 1981. (Vi-
11.
12.
13.
14.
15.
16.
17.
18.
19.
tal and health statistics, Series 2: Data evaluation and methods
research, no. 86) (DHHS publication no. (PHS) 81-1360).
Chestnut LG, Schwartz J, Savitz DA, et al. Pulmonary function and ambient paniculate matter: epidemiologic evidence
from NHANES I. Arch Environ Health 1991;46:135-44.
National Center for Health Statistics, US Public Health Service. Instruction manual: data collection, part 15a. HANES
examination staff procedures manual for the Health and Nutrition Examination Survey, 1971-1973. Washington, DC: US
Government Printing Office, 1972. (GPO publication no.
722-554/89).
National Center for Health Statistics, US Public Health Service. Chest X-ray, pulmonary diffusion, and tuberculin test
results, ages 25-74. National Health and Nutrition Examination Survey, 1971-75. (Public use data tape documentation).
Washington, DC: National Technical Information Service.
(Tape no. 4251).
American Thoracic Society. Single breath carbon monoxide
diffusing capacity (transfer factor): recommendations for a
standard technique. Am Rev Respir Dis 1987;136:1299-307.
Cotes JE, Chinn DJ, Quanjer PH, et al. Standardization of the
measurement of transfer factor (diffusing capacity). Eur Respir J 1993;6(suppl 16):41-52.
Neas LM, Schwartz J. The determinants of pulmonary diffusing capacity in a national sample of U.S. adults. Am J Respir
Crit Care Med 1996;153:656-64.
Harrell FE. Predicting outcomes: applied survival analysis and
logistic regression. Durham, NC: Duke University Medical
Center, 1994.
Epler GR, McLoud TC, Gaensler EA, et al. Normal chest
roentgenograms in chronic diffuse infiltrative lung disease.
N Engl J Med 1978;298:934-9.
Miller A. Pulmonary function tests in clinical and occupational lung disease. Orlando: Crane & Stratton, 1986:276-7.
Am J Epidemiol
Vol. 147, No. 11, 1998