Lung function and chronic exposure to air pollution: A cross-sectional analysis of NHANES II

ENVIRONMENTAL RESEARCH 50, 309--321

(1989)

Lung Function and Chronic Exposure to Air Pollution" A Cross-Sectional Analysis of NHANES II1 JOEL SCHWARTZ U.S. Environmental Protection Agency, Washington, D.C. 20460 Received February 21, 1989 The effect of air pollution on lung function in children and youths ages 6-24 years was examined, after controlling for age, height, race, sex, body mass, cigarette smoking, and respiratory symptoms. Forced vital capacity (FVC), forced expiratory volume at 1 see (FEV1), and peak expiratory flow all showed statistically significant (P < 0.05) negative correlations with annual concentrations of total suspended particulates (TSP), nitrogen dioxide, and ozone. The ozone and NO 2 relationships were highly significant. The TSP relationship was less significant. No relationship was found with sulfur dioxide. The relationships held whether or not children with respiratory conditions, or smokers were included. Demographic and geographic variables had little or no impact on the pollution relationships, which also held when only persons still residing in their state of birth were considered. Essentially identical relationships were found using pollution averaged over 2 years. The relationships held across most N O 2 concentrations, but were only apparent at high particulate and ozone concentrations. © 1989AcademicPress, Inc.

INTRODUCTION

Whether long-term exposure to air pollution causes decrements in pulmonary function remains an unresolved question. In population-based studies at relatively high levels of pollution, Lunn et al. (1967) found lower lung function in children in areas with high particulate concentrations. Similar associations were seen in adults studied in Cracow, Poland (Sawicki and Lawrence, 1977) and in Berlin, New Hampshire (Ferris et al., 1971). Bouhuys and associates (1978), in contrast, in a two community study, found no such association in children. For air pollutants other than particulates, the results are even less clear. The Chattanooga study (Shy et al., 1973) found a weak association with decrements in forced expiratory volume in 0.75 s e c (FEVo.75) in children living in higher N O 2 areas. Detels and co-workers (1981) reported lower lung function in areas that had higher exposure to oxidants, N O 2 , and sulfates, but with no clear association with individual pollutants. More recently, the Six City Study (Ware et al., 1986) has reported no association between lung function and air pollution across the six cities, but a strong association between chronic respiratory symptoms and chronic exposure to particulates. Ostro (1983) reported an association between particulates and acute respiratory symptoms. In another acute study, Lioy and colleagues (1985) have

This work was done while on leave as a visiting scientist at the Department of Biostatistics and the Respiratory Epidemiology Program, Harvard School of Public Health. This research was supported in part by the Electric Power Research Institute under Contract RP1001. 309 0013-9351/89 $3.00 Copyright © 1989 by Academic Press, Inc. All fights of reproduction in any form reserved.

310

JOEL SCHWARTZ

reported short-term decrements in lung function associated with ozone exposure. While it is possible that such short-term insults could have cumulative impacts, no clear evidence for a chronic ozone effect exists. One method for assessing the chronic effect of air pollutants is to correlate levels of pulmonary function with levels of exposure, after adjusting for cigarette smoking. Assuming that the measured levels of exposure represent chronic exposure, they may be reflected in the level of pulmonary function. Practical considerations may limit the number of sites, however. This makes the results highly sensitive to idiosyncratic variations in other risk factors that may occur across the geographic locations to be compared. A cross-sectional sample with many locations is clearly helpful in seeking to untangle this problem. The Second National Health and Nutrition Examination Survey (NHANES II) is a cross-sectional sample of persons drawn from multiple locations across the country. It provides a good database for examining the relationship between chronic air pollution exposure and respiratory function. The basic hypothesis tested was whether there was an association between lung function and chronic exposure to air pollutants. We defined chronic exposure to be the average of the exposure measurements taken in the previous 365 days. For ozone, we used the average of the ozone levels between 11 AM and 5 PM each day as well as the daily maximum. This is more consistent with the hypothesis of Rombout et al. (1986) that daylight-hour exposure is of greater concern than maximum exposure. For other pollutants we used the annual average for all hours, consistent with the current definitions of Annual Ambient Air Quality Standards. The lung functions examined were forced vital capacity (FVC) and forced expiratory volume in 1 sec (FEV1), which have been used in previous analyses, as well as peak flow rate, which some studies have found associated with ozone (Lioy et al., 1985). Several subhypotheses were also explored. The pollutants examined were NO2, SO2, total suspended particulates (TSP), and ozone. Because of the lack of coincident monitoring and high degree of collinearity between the pollutants, each was examined separately. Smokers, or persons with chronic respiratory conditions may have different sensitivities to air pollution. This was tested in analyses that excluded those subpopulations. We also tested whether a nonlinear or threshold relationship existed between the various pollutants and lung function level. MATERIALS AND METHODS

Description o f Data: Spirometry and Health Data The N H A N E S II was conducted from February 1976 through February 1980. The National Center for Health Statistics (NCHS) and the Bureau of the Census selected representative samples of the civilian, noninstitutionalized U.S. population, ages 6 months to 74 years. A total of 20,322 people were examined, and spirometric measures were obtained for a subsample of 4300 people, ranging from 6 to 24 years of age. Budgetary constraints limited spirometry collection to this age group. Details of the complex survey design, the examination procedures, and the

AIR P O L L U T I O N AND L U N G F U N C T I O N

311

laboratory measurements have been published (U.S. DHHS, 1981; NCHS 1981a). Interviewers visited the subjects at home, before the medical examinations, and gathered demographic information and administered a medical history questionnaire. Subjects between the ages of 12 and 17 were asked in a private interview room about their smoking habits. Medical examiners took anthropometric measurements, radiographs, spirograms, and electrocardiograms. Height was always measured with examinees standing erect, facing directly forward (head in the Frankford Horizontal Plane) and wearing special slippers with very thin foam bottoms (NCHS, 1980). Interviewers and medical examiners who worked on the N H A N E S II were specially trained to ensure standardization of the conduct of the survey at each site, and used standardized questionnaires. The N H A N E S II survey selected 64 primary sampling units (counties or county clusters). Approximately 20 neighborhood clusters within each sampling unit were then chosen from which to draw examinees. These neighborhoods were often distant enough to give variation in pollution exposure within a single sampling unit. Sixteen of the sampling units were rural counties, and no pollution data are available for these sites. Spirometric data in the NHANES II were collected using an Ohio Medical Instruments Corp. Model 800 dry rolling seal electronic spirometer, a storage x - y oscilloscope to display the flow-volume curve for monitoring purposes, a singlechannel linear strip chart recorder, and a data acquisition unit to encode, convert, and record on digital tape the spirometry volume signals (NCHS, 1981b). The examinees stood during the test. The N H A N E S II used the pneumatic 3-liter calibration procedure recommended by the American Thoracic Society (1979). These calibrations were performed at the beginning of each day and of each session on days when split spirometry test sessions occurred (16). The electronic portion of the instrumentation system was calibrated before each subject was tested (NCHS, 1981b). Spirometry technicians who worked on the NHANES II were specially trained to administer pulmonary function tests to adults and to children. For all individuals in this study, a complete test set required the recording of at least five tracings. Further details about the instrumentation and measurements have been published (NCHS, 1981b; Schwartz et al., 1988a). Trials were deleted if they were of such poor quality that no spirometry data could be obtained. All measurements were corrected to body temperature saturated with water vapor (BPTS) and taken from the best trial. The best trial was defined as that which demonstrated the maximum sum of FVC and FEV~ from at least two reproducible and error-free trials. A trial was defined to be reproducible if a second trial existed for which the FVC was either within 5% of the FVC of the best trial, if the best FVC was over 3 liters, or within 10% of the best trial, if the best FVC was under 3 liters, ff reproducibility was not established among error-free trials, or if fewer than two error-free trials existed, the subject was excluded from this study. Similarly, we deleted all trials that resulted in an inadequate FVC (less than 1.8 liters for individuals 12 years and older or less than 1 liter for those under 12) or that produced no stable baseline. Total forced expiration had to last at least 1 sec and measurements for the

312

JOEL SCHWARTZ

FEV1/FVC ratio had to be greater than 0.5. Because of these deletions, the study population was reduced to 3922 (92%) of the 4300 for which the N H A N E S II spirometric data exist. Air Pollution Measurements

The N H A N E S II sampling units were often quite large, with topographic and meteorologic features that may have led to substantial differences in exposure between locations within the sampling unit. To account for this possibility, all the individuals in each census tract in the study were assigned the averages of pollution values from all the monitors located within 10 miles of the population centroid of the census tract. All population-oriented monitors in the Environmental Protection Agency's SAROADS database were used. Significant variations in exposures were found among different locations within sampling units. For example, ozone levels varied substantially in Los Angeles between locations along the coast and inland. The location in Los Angeles County with the highest ozone levels had average exposure 60% higher than for the lowest pollution neighborhood sampled. Similar variability in particulate levels occurred in Allegheny County, Pennsylvania, where local sources, hills, and valleys combine to create considerable disparity in exposures. Each person in the sample was assigned the average value of each air pollutant for the 365 days preceding the spirogram. For a day's values to contribute to a monitor's mean SO2 or NO 2 value, at least 18 hourly readings were required to be present. For ozone, 6 of the 7 hourly readings between 11 AM and 5 PM were required. For ozone, the annual average was used rather than the average over the ozone season because the hypothesis being tested was that chronic exposure was related to lung function. Some parts of the country have ozone seasons that last twice as long as others. The annual average incorporates that longer exposure period into the exposure index. The monitoring techniques that were judged acceptable and comparable, and the units used, are shown in Table 1. The density of monitoring sites varied by pollutant, and in some locations all of the pollutants were not monitored during this period. This yields a varying sample size by pollutant. Table 2 summarizes, for each pollutant, the number of samples obtained. TSP was measured at the most sampling sites (68% of the sampling TABLE 1 AIR POLLUTION METHODS AND UNITS

Pollutant SO2

NO2 TSP Ozone

Method

Units

Colorimetric Flame photometric West Gaeke colorimetric Conductimetric Colorimetric Chemiluminescence High volumegravimetric Chemiluminescence

ppm

uv

ppm p~g/m3 ppm

313

AIR POLLUTION AND LUNG FUNCTION TABLE 2 SAMPLE SIZE Pollutant

Neighborhoods

Sampling units

Subjects

TSP NO2 Ozone SO2

378 60 60 166

44 11 24 29

1849 535 1005 832

units). However, only about 70% of the persons living in those sampling units lived close enough to a monitor for us to assign its value to them, giving a total sample size of 1849. Again, not all subjects in each county were close enough to the gaseous monitors to meet our criteria for inclusion in the study. NO 2 measurements were only available in half as many counties as ozone, for 535 subjects in total. Table 3 shows, for each pollutant, the 10th, 25th, 50th, 75th, and 90th percentiles for the annual pollution values used in the study. For most of the pollutants, there was approximately a twofold range between the 10th and 90th percentile values. Statistical Methodology Data were analyzed in two stages. First regression equations including factors known to affect lung function were fitted. These regression models have already been described (Schwartz et al., 1988a, b). Second, the residuals from these regressions were regressed on air pollution. The predictors used were age, race, sex, standing height, sitting height, and Q u e t e l e t ' s body mass index (weight/height2). Because of the heteroscedastic relationship between lung function and some of these predictors, we log-transformed both the lung function measures and height and age (Schwartz et al., 1989a). Since this age range covers the period of growth spurt when lung function changes nonlinearly with age, we fitted separate models for children, teenagers, and young adults. In addition, we found that a sex-by-height interaction term for teenagers was required to control for the different periods of peak growth in the two sexes. This interaction term indicated that the sex difference in lung function increased with increasing height in teenagers. These analyses are discussed in more detail elsewhere (Schwartz et al., 1988a, b). Some of the persons examined were suffering from asthma or other respiratory conditions. Excluding them from the study would have eliminated the potentially TABLE 3 DESCRIPTIVE STATISTICS FOR POLLUTANTS Pollutant

10%

25%

50%

75%

90%

TSP (~g/m 3) NO 2 (ppm) Ozone (ppm) SO z (ppm)

43 0.025 0.023 0.0060

53 0.030 0.027 0.0106

62 0.033 0.032 0.0131

73 0.041 0.038 0.0159

95 0.063 0.040 0.0193

JOEL SCHWARTZ

314

most sensitive population. Therefore, they were included in the sample, and dummy variables were inserted in the initial regressions for asthma, wheezing, cough with phlegm, and current cold. Asthma was the response to the question "Did a doctor ever tell you that you had asthma?". Wheezing was the response to "Not counting cold or flu, have you frequently had trouble with wheezing?", and cough with phlegm the response to "During the past three years have you ever had cough with phlegm for a period of three weeks or more?". Excluding them, however, assures that any correlation between air pollution and lung function is not driven by a correlation between respiratory conditions and air pollution (whether causal or not). Therefore the analysis was repeated with this exclusion. This may be a conservative approach, since air pollution may be associated with these conditions. Smokers were likewise included in the analysis, after inclusion of number of cigarettes smoked per day as a variable in the first stage. Again, imperfect control for smoking may introduce distortion into the pollution-lung function relationships, and the analyses were therefore repeated excluding smokers. Data on parental smoking were not available. Because there was insufficient overlap between the locations where particulates, ozone, and N Q were measured, it was not possible to look at models with more than one pollutant. Therefore the second stage analysis looked at each pollutant separately to see if an association existed between pollution and lung function. This examination consisted of several parts. First, linear regressions were used to test for a relationship. Then the robustness of the relationship was tested by looking at different specifications and subgroup analyses. Persons living in the same neighborhood or metropolitan area are often more alike than persons randomly selected from the population. This may induce correlations (called design effects) in the residuals of subjects living in the same city or neighborhood. To account for this, a two-stage random effects model was used. In this, the covariance of the regression residuals is assumed to be in the form

COV(rok,rlmn) =

Ve

8il gjm ~kn

q- Wnhbhd~jm ~kn "~- Wcity ~kn,

where rij~ is the residual for the ith subject in thejth neighborhood in the kth city (sampling unit). This is similar to the approach of Ware et al. (1986), but with a second level of nesting to account for neighborhood correlations.

Robustness Analyses Linearity. Nonparametric smoothing was used to test for nonlinearities in the relationships. Logarithmic and quadratic transforms of each pollutant were also tested. A quadratic transform tests for a relationship with little change at low exposure levels and increasingly larger effects for higher exposure. A logarithmic transform tests for a saturation type model, with a large initial response followed by declining incremental effect. Finally, where the smoothed plots suggested the possibility of a threshold, the regressions were repeated only using the data above the inflection point. Interactions and subgroup analyses. Interaction terms were used to test individually for different pollution-lung function relationships in persons with asthma,

315

AIR POLLUTION AND LUNG FUNCTION

wheezing, cough with phlegm, and cigarette smokers. Interaction terms were also tested for different relationships by sex, race, and age. In addition, all pollution models were reestimated on a subgroup excluding subjects with chronic conditions and on a subgroup excluding smokers. Socioeconomic and demographic factors. All pollution models were reestimated with family income, head of household educational level, residence inside central city, region of the country, and degree of urbanization added individually. Alternative exposure measures. All pollution models were reestimated using the mean of pollution for the previous 2 years. Models were also reestimated using only subjects still residing in the state of their birth, and using pollution averaged over a 20-mile, instead of a 10-rnile, radius. Finally, to test whether pollution was associated with increases in the number of subjects with significant impairment of lung function, logistic regressions were estimated on whether a subject had lung function less than 70% of predicted. Formally, we assumed logit(Pr Y/ = 1) = X i' [3, where i denotes the ith subject, Yi = 1 if FVC < 70% of predicted, and zero otherwise, logit means the log odds ratio, Xi is the vector of risk factors, and [3 the vector of regression coefficients. To account for the design effects in logistic models, the generalized estimating equations of Liang and Zeger (1986) were used.

RESULTS TSP, ozone, and N O 2 w e r e all significantly associated with FVC, FEV1, and peak flow in the basic regressions. SO2 was not associated with any outcome (Table 4). Figures 1, 2, and 3 show the smoothed plots of FVC (percentage deviation from predicted) versus TSP, ozone, and N O 2. The plots were produced TABLE 4 REGRESSION RESULTS: AIR POLLUTION AND LUNG FUNCTION

TSP* FVC FEV1 Peak flow NO2~" FVC FEV1 P e a k flow Ozonet FVC FEV 1 Peak flow * ~g/m 3. ppm.

B

t

P va l ue

- 0.00059 - 0.00049 - 0.00113

2.68 2.13 3.83

0.0074 0.0332 0.0001

- 2.94 - 3.09 - 3.23

3.57 3.64 3.11

0.0004 0.0003 0.0019

- 3.20 - 3.19 - 4.15

5.67 5.29 4.44

0.0001 0.0001 0.0001

316

JOEL SCHWARTZ 0.10

c)

:5

0.05

E o ~o 0.00

@-0.05 J

-0 10 i 0

i

i

i

50

100

150

,

200

TOTAL SUSPENDED PARTICULATES (#g/m 3)

FIG. 1. The line shown represents the nonparametric LOWESS smooth of the percentage difference between the observed FVC and the FVC predicted by a regression controlling for the effects of standing height, sitting height, age, race, sex, body mass index, smoking, and respiratory conditions smoothed against TSP (ixg/m3).

using LOWESS, a robust locally weighted smoothing algorithm. LOWESS has already been used to show the relationship between lung function and age (Schwartz et al., 1988a) in these data. Figures 1 and 3 show strong signs of nonlinearity, with an indication of a nonlinear relationship for NO 2 as well (Fig. 2). When square and log transforms of the pollutants were tested instead of their linear form the log transform was always a poorer predictor. The t statistics for the square transform for NO2 was slightly higher (less than 10%) than for the linear case. For TSP and ozone, however, there was a more marked improvement in the 0.10 o

0.05

u~ 0.00

"o

@-0.05 c) > u_

-0.10 0.00

i__

0.02

i

i

0.04

0.06

0.0

NO2(ppm)

FIG. 2. The line shown represents the nonparametric LOWESS smooth of the percentage difference between the observed FVC and the FVC predicted by a regression controlling for the effects of standing height, sitting height, age, race, sex, body mass index, smoking, and respiratory conditions smoothed against NOz (ppm).

317

AIR POLLUTION AND LUNG FUNCTION 0.10

0.05 E o •

0.00

"~ - 0 . 5 > IL -0.10 0.00

i

i

0.02

I

0.04

0.06

0.08

OZONE (ppm)

FIG. 3. The line shown represents the nonparametric LOWESS smooth of the percentage difference between the observed FVC and the FVC predicted by a regression controlling for the effects of standing height, sitting height, age, race, sex, body mass index, smoking, and respiratory conditions smoothed against ozone (ppm).

fit, s u g g e s t i n g t h a t a n o n l i n e a r r e l a t i o n s h i p , w i t h s m a l l r e s p o n s e a t l o w p o l l u t i o n l e v e l s b u t a l a r g e r s l o p e a t t h e h i g h l e v e l s , is a m o r e a p p r o p r i a t e d e s c r i p t i o n o f t h e d a t a . T h i s a c c o r d s w e l l w i t h F i g s . 1 a n d 3. When interaction terms for respiratory conditions and smoking were tested (data not shown), none of the terms were significant for any endpoint in regressions with any pollutant. When subjects with chronic respiratory conditions were e x c l u d e d f r o m t h e s a m p l e ( T a b l e 5) t h e c o e f f i c i e n t s o f all o f t h e p o l l u t a n t s w e r e r e d u c e d b y a b o u t 1 0 - 2 0 % . T S P w a s n o l o n g e r a s i g n i f i c a n t p r e d i c t o r o f F E V 1. A l l

TABLE

5

POLLUTION COEFFICIENTS UNDER ALTERNATIVE SPECIFICATIONS

TSP* FVC FEV 1 PEF NOzt FVC FEV 1 PEF Ozone? FVC FEV 1 PEF

Basic

No conditions

No smokers

2-year

State born

20 mile

-

- 0.59 0.49 - 1.13

- 0.47 - 0.39 -0.95

- 0.76 - 0.78 - 1.38

- 0.54 - 0.39 -0.95

- 0.46 - 0.27 - 1.11

- 0.38 - 0.29 - 1.00

- 2.94 -3.09 - 3.23

- 2.56 -2.43 - 2.53

- 3.76 -3.75 - 3.83

- 2.74 -3.04 - 3.11

- 2.96 -2.84 - 2.39

- 2.96 -3.11 - 3.36

- 3.20 -3.19 -4.15

- 2.91 -2.51 -2.51

- 3.34 -3.38 -4.40

- 3.38 -3.40 -4.01

- 3.11 -2.70 -3.46

- 3.25 -3.80 -5.04

Note. No conditions: excluding subjects with chronic respiratory conditions; No smokers: excluding smokers; 2-year: average of previous two years pollution; State born: only subjects still residing in state of birth; 20 mile: average of all pollution monitors within 20 miles. * p~g/m3, 1000x coefficient. t ppm.

318

JOEL SCHWARTZ

the other pollution-lung function relationships remained significant. Excluding smokers, in contrast, increased the coefficients by about 10-20%, and all of the pollutants remained significant for all three outcomes. Sex, race, and age interaction terms were likewise insignificant. Both family income and head of household educational level were insignificant for all outcomes. Residence inside of a central city was negatively associated with all three outcomes. Depending on the pollutant in the model, residence inside a central city was associated with a 2-3% decrease in FVC and FEV1 and with a 3-4% decrement in peak flow. However, inclusion of this term had little effect on the air pollution correlations. For the ozone and NO2 regressions, the coefficients of the pollution measure increased by 10% or less after including the center city term; for TSP regressions, the coefficients decreased by about 10%. None of these changes were statistically significant. Dummy variables for the Northeast, South, Midwest, and West were used to control for potential regional differences. They were insignificant for FVC and FEVI in all regressions. However, for peak flow, the Northeast had a statistically significant decrement of about 5%. The coefficients of TSP and NO2 increased slightly, and those of ozone decreased slightly, after controlling for this factor. Using a 2-year averaging time for pollution instead of 1 year produced little change in the pollution coefficients (Table 5). Restricting the analysis to subjects still residing in the state they were born in reduced the coefficients somewhat (Table 5). TSP was no longer a significant predictor of FEV~ in these regressions. All other pollution relationships remained significant. Use of a 20-mile radius reduced the coefficients for TSP, but increased them for NO 2 and ozone. These results are shown in Table 5. In the logistic regressions, both NO2 and ozone were significantly associated with an increased risk of having a FVC less than 70% of predicted (P < 0.001 for both).

DISCUSSION Highly statistically significant relationships exist between chronic pollution measures and lung function in these data. For ozone, Fig. 3 suggests there is little or no correlation except for the highest 20% of ozone exposures. This is in contrast to findings for acute exposure, where levels as low as half the current ambient standard produce decreases in lung function. However, the annual exposure levels associated with these effects (0.05 ppm) are similar to the levels producing acute responses. This may indicate a true threshold effect for chronic ozone exposure, or it may indicate there is an omitted confounder that resulted in lower lung function in a few areas with very high ozone levels. Support for the hypothesis of a chronic effect of ozone can also be found in recent studies (e.g., Grose et al., 1989) reporting persistent decreased total lung capacity and morphologic changes in animals chronically exposed to ozone at similar levels. This result deserves further study. Confounding appears less likely for NOz since its relationships continue across most of the sites. TSP appears intermediate in this regard. None of the interaction terms with chronic respiratory conditions or smoking were significant, indicating no evidence for a differential impact of air pollution on

AIR P O L L U T I O N A N D L U N G F U N C T I O N

319

these populations. However, the small number of subjects with asthma gave this study low power to detect small effects. All the associations with air pollution were stronger in the nonsmokers, suggesting that inadequate control for smoking may mask the relationship. Focussing on nonsmokers eliminates any such confounding. A similar effect has been noted previouslywComstock et al. (1981) reported an increased incidence of chronic respiratory symptoms in men, but not in women living in homes with gas stoves. A latter analysis (Helsing et al., 1982) restricted to nonsmokers found a stronger effect in both sexes. Socioeconomic factors also do not appear to confound the relationship. Income was not significantly related to any lung function, nor was parental educational level. Residence inside a central city was significantly (negatively) related to lung function, but its inclusion only changed the coefficients of the air pollutants by a few percentage points. The terms for region of the country were generally insignificant and had little impact on the pollution-lung function relationships. The pollution relationships were also reasonably stable when 2-year averages of pollution were used, and when only long-term residents were included in the analysis. Overall, Table 5 shows relationships between pollution and lung function that are quite stable to alternative specifications. This does not prove that the relationships are causal, but it does strongly suggest such a possibility. Of the three significant pollutants, TSP had the most fragile relationships, becoming insignificant for FEV 1 in some models. However, in the models excluding smokers, TSP always had a P < 0.005. TSP may be subject to more errors in variables than the other pollutants since it includes coarse particles as well as respirable ones. In addition, the annual averages are noisier, since fewer measurements of TSP are taken in a year than for the gaseous pollutants. The significance of NO2 and ozone as risk factors for an FVC less than 70% of predicted indicates that pollution is associated with a shift of the entire distribution of lung functions to lower levels. This means that changes of a few percentage points in lung function are also associated with increases in the number of subjects whose lung function is noticeably impaired. In such persons, moreover, a small additional loss of lung function due to air pollution can cause decreases in the ability to engage in normal activities. Since this was a general population study, these associations were occurring at levels commonly reached in U.S. urban areas. For nitrogen dioxide, these decreases are associated with exposure levels at or below the current National Ambient Air Quality Standard of 0.05 ppm. The ozone relationship appears primarily at levels above the current ambient standard, but as the data indicate, those levels are commonly reached. The TSP relationship seems strongest at levels above the previous ambient standard (the current standard regulates particulates smaller than 10 p~m). But again, millions of people live in cities where these concentrations are reached. If these associations are causal, they are of public health concern. Given the highly significant, and reasonably stable, results obtained in this analysis, an obvious question is why it differs from the Six City Study, which also looked across communities for correlations in the residuals of lung function measurements (Ware et al., 1986). Particulate levels in the six cities ranged from 30 to 150 p,g/m3. Given the weak or absent relationship between lung function and TSP

320

JOEL SCHWARTZ

at levels below 85 tzg/m 3, a study of only six cities, with only one consistently above that level, has much less power to detect an effect than a study of 44 cities, with 8 above 85 t~g/m3. The much larger number of sites and greater pollution range in the N H A N E S II survey makes it less susceptible to chance variation in omitted covariates. In addition, in this study we had more detailed socioeconomic data to control for possible omitted covariates. The insensitivity of the pollution effects to those covariates, as well as the larger number of pollution points, makes the observed relationships more plausible. Interestingly, the regressions indicate that a given change in pollution is associated with a larger percentage reduction in peak flow than in either FVC or FEV 1. This suggests that peak flow may be a more sensitive indicator of the effects of pollution. In summary, we have found in a large cross-sectional study a highly significant relationship between air pollution exposure and decrements in lung function. These results are not obviously confounded by either differences in respiratory health or socioeconomic factors across the cross section. The large number of cross-sectional units in this study, and the greater range of pollution exposure, has probably given it more power to detect an effect than previous studies. Without other explanations for the results, the possibility that chronic exposure to air pollution causes decrements in lung function deserves serious consideration.

ACKNOWLEDGMENTS The author thanks Robert Murphy, Terry Drizd, and Willard Hadden of the National Center for Health Statistics for their assistance; Michael Grossman of the National Bureau of Economic Research for the distances between census tracks and SAROADS monitors; and Diane Younkman and Ed Fu for producing the annual average pollution values and merging them to individuals. I have profited greatly from the comments of Douglas Dockery, Frank Speizer, and James Ware on this manuscript. Regrettably, the usual disclaimers apply.

REFERENCES ATS Statement (1979). Snowbird workshop on standardization of spirometry. Amer. Rev. Respir. Dis. 119, 831-838. Bouhuys, A., Beck, G. J., and Schoenberg, J. B. (1978). Do present levels of air pollution affect respiratory health? Nature (London) 276, 466. Comstock, G. W., Meyer, M. B., Helsing, K. J., and Tockman, M. S. (1981). Respiratory effects of household exposure to tobacco smoke and gas cooking. Amer. Rev. Respir. Dis. 124, 143-148. Detels, R., Sayre, J. W., Coulson, A. H., Rokaw, S. N., Massey, F. J., Jr., Tashking, D. P., and Wu, M. (1981). Respiratory effect of long term exposure to two mixes of air pollutants in Los Angeles County. Chest 80, 27S. Ferris, B. G., Jr., Higgins, I. T. T., Higgins, M. W., Peters, J. M., Van Ganse, W. F., and Goldman, M. D. (1971). Chronic non-specific respiratory disease, Berlin, New Hampshire, 1961-1967: A cross-sectional study. Amer. Rev. Respir. Dis. 104, 232. Grose, E. C., Costa, D. L., Hatch, G. E., Crapo, J. D., Chang, L. Y., Stevens, M. J., Jaskot, R. J., and Tepper, J. S. (1989). Chronic exposure to ozone causes restrictive lung disease. Air Pollut. Control Assoc. Paper No. 89-12.3. Helsing, K. T., Comstock, G. W., Meyer, M. B., and Tockman, M. S. (1982). Respiratory effects of household exposure to tobacco smoke and gas smoking on non-smokers. Environ. Int. 8, 365-370. Liang, K.-Y., and Zeger, S. L. (1986). Longitudinal data analysis using generalized linear models. Biometrika 73, 13-22.

AIR POLLUTION AND LUNG FUNCTION

321

Lioy, P. J., Vollmuth, T., and Lippmann, M. (1985). Persistence of peak flow decrements following ozone exposure exceeding the national ambient air quality standard. JAPCA 35, 1068. Lunn, J. E., Knowelden, J., and Roe, J. W. (1967). Patterns of respiratory illness in Sheffield infant school children. Brit. J. Prev. Soc. Med. 21, 7. National Center for Health Statistics (NCHS) (1980). "Instruction Manual Part 15A HANES II: Examination Staff Procedures for the Health and Nutrition Examination Survey, 1976--1979." Hyattsville, MD. National Center for Health Statistics (NCHS) (1981a). "Plan and Operation of the Second National Health and Nutrition Examination Survey, 1976-1980." Vital and health statistics, Series 1, No. 15. DHHS Publ. No. 81-1317. National Center for Health Statistics, Washington, DC. National Center for Health Statistics (NCHS) (1981b). "Computer-assisted spirometry data analysis for the National Health and Nutrition Examination Survey, 1971-80." Vital and health statistics, Series 2, No. 86. DHHS Publ. No. (PHS) 81-1360. National Center for Health Statistics, Washington, DC. Ostro, B. D. (1983). The effects of air pollution on work loss and morbidity. J. Environ. Econ. Man 10, 371. Rombout, P. J. A., Lioy, P. J., and Goldstein, B. D. (1986). Rationale for an eight-hour ozone standard. JAPCA 36, 913. Sawicki, F., and Lawrence, P. S. (1977). "Chronic Nonspecific Lung Disease in the City of Crackow." National Institute of Hygiene, Warsaw, Poland. Schwartz, J., Katz, S. A., Fegley, R. W., and Tockman, M. S. (1988a). Analysis of spirometric data from a national sample of healthy 6-24 year olds (NHANES II). Amer. Rev. Respir. Dis. 138, 1405-1414. Schwartz, J., Katz, S. A., Fegley, R. W., and Tockman, M. S. (1988b). Sex and race differences in the development of lung function. Amer. Rev. Respir. Dis. 138, 1415-1421. Shy, C. M., Hasselblad, V., Burton, R. M., Nelson, C. J., and Cohen, A. (1973). Air pollution effects on ventilatory function of U.S. school children. Results of studies in Cincinnati, Chattanooga, and New York. Arch. Environ. Health 27, 124. U.S. Department of Health and Human Services, Public Health Service, Centers for Disease Control (U.S. DHHS) (1981). "Laboratory Procedures Used by the Clinical Chemistry Division, Centers for Disease Control, for the Second Health and Nutrition Examination Survey (HANES II) 1976-1980." Ware, J. H., Ferris, B. G., Dockery, D. W., Spengler, J. D., Stram, D. O., and Speizer, F. E. (1986). Effects of ambient sulfur oxides and suspended particles on respiratory health of pre-adolescent children. Amer. Rev. Respir. Dis. 133, 834.