Various studies show a relation between the levels of particles in the surrounding air and various health measures. The levels of ambient particles have been associated with days of restricted activity and work loss, daily emergency room visits, emergency hospital admissions, and death rates. Studies have shown a relation between different levels of air pollution, lung function and prevalence of chronic respiratory disease. Lung function is a specific measure of respiratory health and may be more appropriate than restricted activity days or work loss days in assessing the effects of air pollution. The relation between lung function and the levels of total suspended particulates (TSP), which refer to particles within the surrounding air, was assessed. The study involved adults who had never smoked and who lived in any of the 49 sites where the First National Health and Nutrition Survey (NHANES I) was conducted. TSP levels were shown to be associated with two measures of lung function, the forced vital capacity (FVC) and forced expiratory volume at one second (FEV1). Anthropometric measurements, or measurements of the dimensions of the body, and socioeconomic characteristics of the subjects were also evaluated. The results showed that an increase in TSP levels was associated with a decrease in FVC. However, the findings also suggest that there is a threshold level of TSP below which TSP levels do not affect lung function. (Consumer Summary produced by Reliance Medical Information, Inc.)
A RELATIONSHIP between ambient particulate matter and a variety of health measures has been evidenced in recent epidemiological studies. Many of these studies have included adult subjects nationwide, and a relationship between ambient particulate matter and restricted activity days and/or work loss days has been established. [1-4] Other studies have reported an association between daily emergency room visits or emergency hospital admissions and daily particulate matter levels of coefficient of haze levels. [5-8] (Coefficient of haze is a measure of visual air pollution that can be correlated with ambient particulate matter). Mortality rates in cross-sectional and time-series studies have been associated with ambient particulate matter levels. [9-11]
Pulmonary function and clinical respiratory health measures have been used in some epidemiological studies in which air pollution levels across two to six cities were compared. [12-18] After age, smoking, and occupation are considered, several--but not all--of these studies reported significant differences in pulmonary function or in prevalence of chronic respiratory symptoms or diseases between communities with different levels of air pollution. If comparison is limited to a few communities, it is difficult to confidently attribute the observed differences in respiratory health to variations in air pollution rather than to some other unidentified factors that might differ between the communities. Pulmonary function measures are also sensitive to short-term pollution exposures. [19-21]
Pulmonary function is a measure specific to respiratory health and may reflect more subtle effects of air pollution than are captured by symptom measures, e.g., restricted activity days or work loss days. If pulmonary function, rather than restricted activity days or work loss days, is used as the health outcome, the previous nationwide studies could be extended.
In this study, we examined the relationship between ambient particulate matter levels and pulmonary function in adults across 49 cities in the United States. Pulmonary function data were gleaned from the First National Health and Nutrition Examination Survey (NHANES I) and were combined with ambient particulate matter data from the national database maintained by the United States Environmental Protection Agency (EPA), Office of Air Quality Planning and Standards.
Methods
NHANES I Data. The NHANES I was conducted at 90 locations across the United States and included 23 808 children and adults who were 1-75 y of age. Respondents participated in several interviews and were examined by a physician. A subsample of 6 913 adults ,(24-75 y of age) was given a more detailed examination, including spirometry (pulmonary function testing), and was asked more extensive medical history questions. Interviewers and medical examiners who conducted the NHANES I were specially trained to ensure that conduct of the survey at each site, questionnaires, and techniques were standardized. [22,23]
The following instruments were used in the NHANES I to collect and store spirometric data: an Ohio Medical Instruments Corporation Model 800 electronic spirometer; a storagye x-y oscilloscope that, for monitoring purposes, displayed the flow-volume curve; a single-channel linear strip chart recorder; and a data acquisition unit to encode, convert, and record on digital tape the spirometry volume signals. [22] A pneumatic 5-1 calibration procedure was used in NHANES I. These calibrations were performed at the beginning of each day and each session on days when split spirometry test sessions occurred. [23] The electronic portion of the instrumentation system was calibrated before each subject was tested to ensure reliability of these components. Spirometric data were recorded on a Beckman Digicorder Model DRS-1000 digital tape recorder system. The computer used temperature and pressure data to develop a BTPS correction factor. The unit also encoded each signal that the computer identified as the recording location, subject identification number, age, sex, race, height, and technician code. Further details about instrumentation, calibration, and analysis have been published elsewhere. [22]
Examinees stood during the test. Nose clips, if needed, and disposable mouthpieces were provided. All persons were examined by a physician before they participated in the spirometry test, and anyone who evidenced serious respiratory disease or who manifested other forms of poor health was excluded from the study population. In the spirometry sample, a test set was complete if at least 5 spirograms were recorded for each individual. A 20-min rest period was required between each set of 5 trials. Reproducibility was achieved by repeating, up to 3 times, each set of 5 trials. [23] The best trial was defined as that which demonstrated a maximum forced vital capacity (FVC) and forced expiratory volume at one second ([FEV.sup.1.0]) from at least 2 reproducible and error-free trials. A trial was "reproducible" if the FVC in a second trial was within (a) 5% of a FVC that was at least 3 I obtained during the best trial or (b) within 10% of the FVC if the best FVC was less than 3 I. [23] If reproducibility was not established among error-free trials of if fewer than two error-free trials were obtained, the subject was excluded.
The NHANES I excluded all persons whose best trial had a duration of less than 3 s because this possibly represents a less-than-miximal effort. Error can be introduced into spirometry data sets if subjects do not make maximal effort in trials. We compared the NHANES I data to spirometry data for adults in the Harvard Six Cities study [24] to determine if there was any evidence of error. Regression equations for men and women were used to adjust both groups to an age of 50 y and to heights of 1.7 m for men and 1.6 m for women. The mean FVC values of the two adjusted samples were identical for men (4.22 I), and the NHANES I sample provides a slightly higher mean for women than what was provided in the Six Cities sample.
Individuals who have respiratory diseases are commonly excluded from predictive equations for pulmonary function. Our goal was to determine if air pollution was associated with reductions in pulmonary function in the population (which includes these persons); therefore, we adopted a different approach. All persons who were not excluded as a result of the physicians' examinations and who presented reproducible error-free trials were considered. Of course, as in any analysis in which spirometric endpoints are used, we risked missing a true association between ambient particulate matter and pulmonary function because persons likely to be most affected by air pollution are those with serious respiratory diseases. The correlation between ambient particulate matter levels and the percentage of subjects who were excluded from the spirometry tests by examining physicians was determined so that the potential impact of these exclusions could be assessed. We found a statistically significant positive correlation (p < .01), which suggests a greater prevalence of pulmonary disease in areas with higher levels of particulate matter. No statistically significant association was found between the percentage of subjects capable of providing reproducible trials and the levels of ambient particulate matter at each location. This suggests that if the exclusions have any effect on the results, it will be to reduce the observed relationship between pulmonary function and levels of ambient particulate matter.
Only white and black subjects were included in the analyses, which controlled for physiological variation . In the sensitivity analysis, a subsample of "whites only" was considered. Subjects were divided into never-smokers and smokers, depending on whether they responded negatively or affirmatively to, "Have you smoked at least 100 cigarettes in your lifetime?" We focused on never-smokers so that respiratory effects of smoking would not confound our analysis. When we considered smokers, cigarette-years of exposure was defined as the duration of exposure multiplied by the response to the question, "When you were smoking the most, how many cigarettes per day were you smoking?"
Five pulmonary function parameters were selected for use in this analysis. Diminished FVC is commonly found in individuals who have restrictive lung diseases (e.g., pulmonary fibrosis) and in individuals with obstructive lung diseases (e.g., emphysema). [FEV.sub.1.0] isa also expected to be decreased in patients who have restrictive or obstructive lung diseases. FVC and [FEV.sub.1.0] are two of the most frequently used measure of lung function in epidomiologic studies. Patiets with obstructive diseases tend to show a reduced [FEV.sub.1.0]/FVC ratio, whereas those with restrictive diseases usually have a normal or sightly increased ratio ([FEV.sub.1.0] and FVC are reduced in similar proportions). Maximum and mid-expiratory flow (MMEF) may be decreased in early stages of obstructive lung disease. The ratio of measured FVC to its predicted value was also used as a measure of the degree of impairment. Ambient particulate matter may be associated with more severe impairment of lung function; therefore, we searched for an association between ambient particulate matter and FVC less than 70% of predicted.
The potentially confounding physiological factors that were expected to be strong determinants of pulmonary function included age, height, sex, race, and obesity (as judged by the examining physician). The respiratory hazards to which individuals were exposed at the workplace and average ambient levels of total suspended particulates (TSP) in the areas where the individuals lived were assessed. FVC was regressed on all of these factors, except TSP, so that estimates of the predicted FVC value could be development for each subject. Additional variables that reflected differences in weather, geographic regions, and income--all of which might also be related to respiratory health--were included in the sensitivity analyses. Definitions and means for the variables used to obtain the final results reported herein are provided in the Appendix.
Particulate data. Confidentiality restrictions limited location information for individuals who lived in areas populated by less than 100 000. However, information was available for 57 of the 90 NHANES I locations that covered one or more counties. A consistent pollution data set was developed by identifying the central urban area in each NHANES I location, and TSP data were obtained from the population-oriented monitors in that area. Data were averaged if TSP data for a 3-mo period were available for more than one population-oriented monitor in a city. If there were insufficient data available form EPA to calculate quarterly summary statistics, the monitoring data were excluded. The TSP data were matched to the NHANES I subjects in accord with the location and quarter during which the subjects were examined. Acceptable TSP data were available for 49 of the 57 NHANES I locations in which populations exceeded 100 000.
Average annual TSP exposure levels for each NHANES I respondent were calculated by averaging the mean level for the quarter in which the individual was examined with the mean TSP levels measured during the three preceding quarters. In this cross-sectional sample, the annual means correlated strongly (0.93) with the quarterly means. Therefore, the quarterly values were used in most analyses because these more closely matched the time of the examination. The annual average could not be calculated for one location because data were missing.
The TSP data are the most widely available and reliable ambient air pollutant data for the time period and locations of the NHANES I survey. Measurement techniques for sulfur dioxide have changed since commencement of the NHANES I survey, and validity of some earlier measurements has been questioned. Ozona was not measured in most areas during that period. It is acknowledged that TSP may be correlated with the presence of other air pollutants, and any association found between TSP and pulmonary function may reflect the effects of tSP and/or any other pollutants with levels correlated with TSP levels.
Another limitation of the TSP data is that they include some particles that are too large to be deeply inhaled into the lugs. For this reason, the federal primary ambient air quality standard for particulate matter has been recently changed from a TSP measure to a measure of particles with diameters of 10 [mu] or less ([PM.sub.10]). We expect TSP and [PM.sub.10] levels to be strongly correlated across most of the NHANES I locations, but differences in this ratio will introduce some error in the analysis. Whether any of the locations is likely to have a relatively high level of fugitive dust from natural conditions is of particular concern. Fugitive dust usually consists of fairly large particles, and it can cause high reads of TSP even when levels of respirable particles remain low. Tucson, Arizona, was the only city in the sample of 49 NHANES I locations that EPA cited as being likely to have a high level of fugitive dust compared with the other cities. In fact, the measured level of TSP for Tucson was the second highest in the sample, whereas the average pulmonary function levels for Tucson residents were close to the sample means. This concern about the Tucson TSP data was considered in the sensitivity analysis.
Quarterly and annual average temperature and humidity data for the NHANES I locations were compiled from the monthly Climatological Data published by the National Oceanic and Atmospheric Administration. The weather data were included to control for climate variations that may affect TSP levels and the individuals' respiratory conditions.
Data analysis. The analysis of the potential relationship between TSP levels and pulmonary function in adults involved two components: (1) use of a graphical technique and regression analysis that explored the existence and shape of any potential relationship between the pulmonary function measures and TSP levels, and estimation of a multivariate regression with a function form consistent with the graphical results; and (2) extension of the regression analyses to test the stability of the results by considering alternative independent variables, adjustments in the sample of cities, threshold models, and potentially different sensitivities to TSP in subsamples of individuals.
A basic set of independent variables expected to be related to pulmonary function was selected for initial analysis. These included physiological variables, potential occupational dust exposure, and TSP (see Appendix for definitions and specific variables). There is no a priori information about the expected shape of a possible relationship between pulmonary function and particulate exposure; therefore, a graphical technique was used to investigate the shape of this relationship. We initiated this analysis by taking the residuals from a partial regression analysis of pulmonary function on our basic set of predictor variables, excluding TSP. The residuals from regressing TSP on the other predictor variables were also obtained to control for potential covariates. Graphical techniques of Exploratory Data Analysis were used to compare unexplained variations in pulmonary function with TSP residuals. Cleveland's locally weighted smoothing algorithm (LOWESS) is a robust nonparametric approach that is capable of demonstrating the trend in data and detecting nonlinearities, piecewise linear relationships, threshold, etc. [25] LOWESS fits a separate, locally wieghted regression for each data point to produce a local smoothing that can fit any smooth nonlinear curve. A check for influential observations provides robustness weights for a second iteration. Further details and provided by Cleveland [25] and Chambers. [26]
After we controlled for the basic covariates, LOWESS and some initial linear regression analyses were used to detect a possible relationship between TSP and each of the pulmonary function measures in our data set. If a relationship was found, the shape of the smoothed curve was considered when selecting the type of regression model (i.e., linear or nonlinear) to use for subsequent examination of the significance and stability of the estimated relationship. The Marquardt algorithm in SAS [27] was used to estimate the nonlinear models. Additional nonlinearities were explored in the extended analyses by (1) estimating the regressions over different ranges of TSP levels (selected on the basis of the plots) and (2) estimating a threshold model that allows for a TSP level below which no relationship with pulmonary function is observed. Logistic regressions were used to determine if TSP is associated with prevalence of FVC less than 70% of predicted.
Linear and log-transformed models were used in the analyses. The log transformation involved use of the natural logarithm of the pulmonary function measure as the dependent variable in the regression. Heteroscedasticity is common in pulmonary function data because the variance in the population increases with age, height, and potentially with other factors such as exposure to respiratory hazards. [28] The log transformation achieved homoscedastic residues with respect to height and age, which was not the case with the simple linear estimates, and was, therefore, used in all subsequent regression analyses. the log transformation indicates] that each equation predicts proportionate differences in pulmonary function rather than absolute differences. The physiological variables, age and height, were also converted to natural logs in the log-transformed models. The estimated regression coefficients thus give the percentage difference in pulmonary function that is associated with a 1% difference in these explanatory variables. In addition to the log of age, the square of the log of age was included, which allowed for potential nonlinearity in the relationship between pulmonary function and age.
Pulmonary function measures that showed a significant relationship with TSP in the initial regression analyses were subjected to further analyses to determine the stability of the estimated relationships. The stability was tested by exploring the effects of additional potentially confounding factors. (e.g., weather, income) on the estimated regression coefficients. The estimated regression coefficients for a few cities that had very high or very low TSP levels were also tested. The city of Scranton, Pennsylvania, had a very high TSP level compared with the remainder of the sample. Tucson, Arizona, was identified by EPA as a city in the NHANES I sample that had a potential for an unusually high amount of fugitive dust, which would probably be reflected in the measured TSP level.
Results
Initial analyses. The smoothed curve that illustrates the relationship between FVC and TSP in never-smokers, after controlling for covariates, is shown in Figure 1. FVC, expressed as the percentage difference from predicted, is plotted against TSP. A nonlinear relationship is clearly indicated. A similar curve was obtained for [FEV.sub.1.0]. It would appear that no relationship existed between FVC and TSP at low TSP levels, and a negative relationship existed at higher TSP levels, starting between 60 and 80 [mu]g/[m.sup.3]. Approximately one-half of the locations in the sample had TSP levels that exceeded 80 [mu]g/[m.sup.3]. The linear regression results (not shown) indicate that a statistically significant negative relationship existed between FVC and TSP and between [FEV.sub.1.0] and TSP.
TSP exposure was also significantly associated with a FVC of less than 70% of predicted (p < .05 in the logistic regression [not shown]). The smoothed curve for percentage of subjects (in groups of 20) with a FVC less than 70% of predicted versus TSP is shown in Figure 2. The percentage was calculated for each group of 20 consecutive individuals, who were sorted by increasing order of TSP exposure.
The nonparametric curve illustreated in Figure 1 suffices for a dose-response relationship, but it does not allow direct testing of the statistical significance of the relationship. Given that the curve indicates an S-shaped relationship, we estimated a nonlinear regression where the relationship between the log of FVC and TSP follows a logistic curve. Such a model can be expressed as LN (FVC) = A + B/[1 + exp(C + D x TSP)] + [B.sub.2] x LN(AGE) + [B.sub.3] x [LN(AGE)][sup.2] + [B.sup.4] x LN (HEIGHT) + [B.sub.5] x (OBESE) + ... [1] and the other covariates can be entered linearly. The association between TSP and pulmonary function can be reflected in B, C, and D, and it is, therefore, difficult to use the equation above to test the statistical significance and robustness of the estimated relationship. We estimated all three parameters for both FVC and [FEV.sub.1.0] and the fixed B and C at the initial estimates for each pulmonary function measure for all subsequent analyses. If B and C are fixed, D can be interpreted as the TSP coefficient, and changes in D associated with changes in the model specification can be interpreted as changes in the estimated magnitude of the association between TSP and pulmonary function. The regression results for LN (FVC) and LN ([FEV.sub.1.0]) are shown in Tables 1 and 2. The TSP coefficients were statiscally significant (p < .001) for both pulmonary function measures.
The graphical analysis failed to reveal a relationship between TSP and the [FEV.sub.1.0]/FVC ratio or MMEF. This was confirmed in initial linear regression results, which also showed no significant relationship between TSP and [FEV.sub.1.0]/FVC or MMEF. This suggests that TSP is associated with a volume effect. Equation 1 was also estimated for MMEF, and TSP remained nonsignificant. Our subsequent analyses, therefore, focus primarily on FVC and [FEV.sub.1.0].
In the regression results reported in Tables 1 and 2, and in subsequent analyses described below, the estimated coefficients for physiological variables were quite stable when changes in the specifications were made. The positive and negative coefficients shown were as expected. The two age coefficients indicate declining pulmonary function with increasing age over the range of ages included in this sample. The coefficients indicate that for persons aged 50 y, the average rate of decrease in FEV per year of age is 32 ml. The estimated coefficients for height show that pulmonary function increases with height and confirmed the expected quadratic dependence of FVC on height. Males and whites had approximately 15% higher pulmonary function than females and blacks. Obese individuals have lower pulmonary function.
Extended regression analyses. The first goal of the extended regression analyses was to explore whether the observed inverse relationship between pulmonary function and TSP, as seen in the initial regression results and in the plots, might be the result of correlations between TSP and factors that were excluded in the estimations, e.g., weather or socioeconomic variables. The second regressions in Tables 1 and 2 show the results obtained when several such variables (i.e., average temperature for the quarter, household income, and region of residence) were added to the equation. Region of residence might reflect general climatic and socioeconomic differences that may influence respiratory health. The only significant factor for both FVC and [FEV.sub.1.0] was residence in the South, which was associated with lower pulmonary function. The TSP coefficients in both the FVC and [FEV.sub.1.0] regressions were slightly larger than in the first regressions, which did not include these additional variables. A few other variables (i.e., average relative humidity, whether
[TABULAR DATA OMITTED]
population of area exceed 1 million, whether spirometry test was done in winter) were also considered, but none was found to be statistically significant or to have any appreciable impact on the estimated TSP coefficient.
The next question explored was whether the TSP coefficients would change if locations with the lowest levels of TSP were eliminated from the sample. The plots discussed above suggest that the inverse relationship between pulmonary function and TSP may not begin until TSP exceeds 50 or 60 [mu]g/[m.sup.3]. This curvature is, to some extent, captured by tghe logistic specification, but such a specification may not be exact. Results of regression 3 (Tables 1 and 2) show the effects of dropping the two locations in the sample that had TSP levels less than or equal to 50 [mu]g/[m.sup.3]. The estimated coefficients for TSP were unchanged for both pulmonary function measures.
The effects of several different changes in the sample versus TSP coefficients are shown in Table 3. The TSP coefficients for regressions 1 and 3 are repeated in Table 3 for comparison. All the results in Table 3 are based on estimations in which the seven covariates included in regression 1 were used. Regression 4 was estimated exclusive of Tucson. The impact of eliminating Tucson can be examinated if regession 4 coefficients are compared with regression 1 coefficients (the entire range of TSP is included in both coefficients). No noticeable change in the TSP coefficients occurred. Scranton, Pennsylvania, presented another concern because its TSP level was approximately 100 [mu]g/[m.sup.3] higher than that found in the city with the next highest TSP levels. It is possible that data points for Sranton could have had an inordinate influence on the regression coefficients. Therefore, regression 5 was estimated without data from Scranton. The coefficients of TSP in the FVC and [FEV.sub.1.0] equations were slightly changed in regression 5 and remain statistically significant (p < .001). Regression 6 was estimated without data from Scranton or Tucson. The results, when compared with regression 3, indicate little sensitivity to the deletion of the two higyhest TSP exposure cities in the sample.
Threshold analyses. An alternative specification to fit the changing slope of the FVC-TSP relationship (Fig. ) is to fit a piecewise linear model. Whereas this seems less likely than the curvilinear model the reflect the true picture,
[TABULAR DATA OMMITTED]
[TABULAR DATA OMITTED]
it serves two purposes: (1) it tests the robustness of the relationship to an alternative specification; and (2) because the optimal joint point for the two regions can be estimated in such models, [29] it provides a quantitative estimate of a level below which there is effectively no relationship between FVC and TSP. When such a model was fit to the data and a slope was fixed at 0 for low TSP, the estimated optimal joint point was 60 [mu]g/[m.sup.3].
Other subject samples. An additional senstivity test was conducted. Regressions for FVC and MMEF in the population of current and ex-smokers were repeated, and cigarette-years of exposure were controlled for as defined previously. For ease of comparison, B and C were fixed at the same values as for never-smokers. The results indicated that TSP is a significant predictor of FVC (p < .05) for smokers. The estimated coefficient was .0243, which was approximately one-half the magnitude of the estimated coefficient for never-smokers. The results also indicated a significant (p < .01) relationship between TSP and MMEF for smokers.
We also estimated the FVC equation for whites only to determine if any misspecification of the relationship between race and pulmonary function may have distorted the estimated relationship between TSP and FVC. The estimated TSP coefficient for whites only was .059 (p < .001), which is very similar to the result obtained for whites and blacks combined. We also considered estimating the FVC equation for white males only, but found that the sample size was too small to obtain meaningful results with the nonlinear estimation procedure. No significant correlation was found between sex and TSP (p > .75), which suggested that confounding problems between percentage of the sample that was male in each city and TSP levels in each city were not likely to be significant.
Discussion
This analysis of the NHANES I data on pulmonary function suggests the presence of a statistically significant relationship between ambient TSP levels and the FVC and [FEV.sub.1.0] levels of adults who reside in 49 of the NHANES I locations. The regression results showed a statistically significant relationship between TSP and pulmonary function that was robust across several specification and sample changes, e.g., deletion of the cities with the two highest and two lowest TSP levels, restriction of sample to whites only. The results of the analysis also suggest a threshold level of TSP (quarterly average of 60 [mu]g/[m.sup.3]) below which a relationship with pulmonary function ceases to exist.
The study had some aspects of an ecological study because the average air pollution level in each city was assigned to each subject. However, all the major predictors of pulmonary function (height age, race, sex, smoking) were available for each individual, which would not be the case in a pure ecological design. The use of city-specific rather than individual-specific air pollution measurements can cause errors in variables. In general, if any bias in the estimated coefficients results from such errors, it can be expected to be toward zero.
If important covarieties have not been identified or properly specified in analyses of this type, a spurious correlation could be interpreted as meaningful. We attempted to minimize this risk by restricting the sample to never-smokers, and we controlled for anthropometric measurements and socioeconomic characteristics of the subjects.
The results of this analysis imply that a 1 standard deviation increase (about 34 [mu]g/[m.sup.3]) in TSP from its sample mean value of 87 [mu]g/[m.sup.3] is associated with about a 2.2% decrease in FVC. This effect is no bigger than the expected variation in FVC measurement for an individual at different times, but as difference in the average across two population groups, it may reflect a significant health impact. For example, IQ measurements may vary by approximately five points for an individual at different times, but a 5-point difference in the mean IQ for two ethnic groups would be of considerable social importance.
No significant association between TSP and MMEF was found in never-smokers, even though a significant relationship was found in smokers. The decline in [FEV.sub.1.0] paralleled the decline in FVC; no statistically significant relationship was found between the ratio of [FEV.sub.1.0] to FVC and TSP levels. This suggests that TSP may not be specifically associated with obstructive lung disease. The weaker association between TSP and flow measures (MMEF, [FEV.sup.1.0]/FVC) may have resulted from the greater unexplained variability in these measures. Whether the observed decrements should be classified as restrictive, or whether they represent a more complex phenomenon, is not clear from the data used in this analysis.
The results of this analysis are consistent with the findings of previous epidemiological studies that have investigated the association between particulate matter and chronic respiratory symptoms [18] and acute symptoms. [3] The demonstration of a similar association between particulate matter and pulmonary function lends strength to the hypothesis that exposures to ambient particulate matter contribute to respiratory health problems. However, the exact biological mechanism that underlies these observed relationships requires further exploration before causation can be established. TSP was significantly related to the probability of FVC being less than 70% of predicted, which suggests that a link exists between particulate matter exposures and chronic respiratory disease. The much smaller estimated effect of TSP at levels less than 80 [mu]g/[m.sup.3] suggests that the absence of a relationship between particulate matter and lung function reported by Ware et al. [18] may have resulted from the lower exposure levels in their sample.
The premise of this study was that many communities with different air pollution levels (i.e., cities in the NHANES I sample) provide a valuable data set with which the cross-sectional relationship between long-term air pollution exposures and pulmonary function can be examined. Our results substantiated this premise. This premise was also supported by the confirmation of several expected associations (height, age, etc.) with pulmonary function at magnitudes similar to those found in other studies. This finding of expected associations, within the limits of the partial ecologic design, lends, credibility to the observed relationship between air pollution and pulmonary function. Our study suggests that further analyses of the NHANES data sets are warranted.
[TABULAR DATA OMITTED]
References
[1] Ostro B. The effects of air pollution on work loss and morbility. J Environ Econ Manage 1983; 10:371-82.
[2] Ostro B. Urban air pollution and morbility; a retrospective approach. Urban Studies 1983; 20:343-51.
[3] Ostro B. Air pollution and morbility revisited: a specification test. J Environ Econ Manage 1987; 14:87-98.
[4] Crocker TD, Schulze W, Ben-David S. Kneese AV. Methods development for assessing air pollution control benefits. I. Experiments in the economics of air pollution epidemiology. EPA-600/5-79-0010. Washington, DC: United States Environmental Protection Agency.
[5] Graves P, Krumin RJ. Health and air quality: evaluating the effects of policy. Washington, DC: American Enterprise Institute for Public Policy Research, 1981.
[6] Samet JM, Bishop Y, Speizer FE, Spengler JD, Ferris BG, Jr. The relationship between air pollution and emergency room visits in an industrial community. J Air Pollut Control Assoc 1981: 31:236-40.
[7] Mazumdar S. Sussman N. Relationship of air pollution to health: results from the Pittsburgh study. Arch Environ Health 1983; 38:17.
[8] Bates DV, Sizto R. Relationships between air pollutant levels and hospital admissions in Southern Ontario. Can J Public Health 1983; 74:117-22.
[9] Lave LB, Seskin EP. Air pollution and human health. Baltimore, MD: Johns Hopkins University Press for Resources for the Future, 1977.
[10] Evans JS, Tosteson T, Kenney PL. Cross-sectional mortality studies and air pollution risk assessment. Environ International 1984; 10:55-83.
[11] Mazundar S, Schimmel H, Higgins ITT. Relationship of daily mortality to air pollution: a reanalysis of data on London winters, 1958/59 - 1971-72. Arch Environ Health 1982; 37:213-20.
[12] Ferris BG Jr, Anderson DO. Epidemiological studies related to air pollution: a comparison of Berlin, New Hampshire and Chiliwack, British Columbia. Proceedings of the Royal Society of Medicine 1964; 57:979.
[13] Detels R, Rokaw SN, Coulson AH, Tashkin DP, Sayre JW, Massey FJ Jr. The UCLA population studies of chronic respiratory disease. I Methodology and comparison of lung function in areas of high and low pollution. Am J Epidemiol 1979; 109(1):33.
[14] Rokaw SM, Detels R, Coulson AH, Sayre JW, Tashkin DP, Allwright SS, Massey FJ Jr. The UCLA population studies of chronic obstructive respiratory disease. III. Comparison of pulmonary function in three communities exposed to photochemical oxidants, multiple primary pollutants, or minimal pollutants. Chest 1980; 78:252-62.
[15] Sawicki F. Chronic non-specific respiratory diseases in the city of Cracow. IX. The cross-sectional study. Epidemiol Rev 1969;23:242-52.
[16] Mostardi RA, Martell R. The effects of air pollution on pulmonary function in adolescents. Ohio J Sci 1975; 75:65.
[17] Bell KA, Linn WS, Hazucha M, Hackney JD, Bates DV. Respiratory effects of exposure to ozone plus sulfur oxide in southern Californians and eastern Canadians. Am Ind Hyg Assoc J 1977; 38:696.
[18] Ware JH, Ferris BG Jr, Docker DW, Spengler JD, Strom DD, Speizer FE. Effects of ambient sulfur oxides and suspended particles on respiratory health of preadolescent children. Am Rev Respir Dis 1986; 133:834-42.
[19] Motley HL, Smart RW, Leftwich CI. Effect of polluted Los Angeles Air (smog) on lung volume measurements. J Am Med Assoc 1959; 171:1469.
[20] Bates DV, Hazucha M. The short-term effects of ozone on the lung. In: Proceedings of the Conference on Health Effects of Air Pollutants, National Academy of Sciences, Washington, D.C. Washington, DC:Senate Committee on Public Works, 1973; 507-40 (serial no. 93-15).
[21] Linn WS, Jones MP, Bachmayer EA, Spier CE, Mazur SF, Arol EL, Hackney JD. Short-term respiratory effects of polluted ambient air: a laboratory study of volunteers in a high-oxidant community. Am Rev Respir Dis 1980; 121:243-52.
[22] National Center for Health Statistics. Computer-assisted spirometry data analysis for the National Health and Nutrition Examination Survey 1971-1980. Vital and health statistics, series 2, no. 86, DHHS publication no. (PHS) 81-1360.
[23] National Center for Health Statistics. Public use data tape documentation: spirometry--best trials only. Tape no. 4250; 1981.
[24] Dockery DW, Speizer FE, Ferris BG Jr, Ware JH, Louis TA, Spiro A. Cumulative and reversible effects of lifetime smoking on simple tests of lung function in adults. Am Rev Respir Dis 1988; 137:286-92.
[25] Cleveland WS. Robust locally weighted regression and smoothed scatterplots. J Am Stat Assoc 1979; 74:829-36.
[26] Chambers JM, Cleveland WS, Kleiner B, Tukey PA. Graphical methods for data analysis. Boston, MA: Duxbury Press, 1983.
[27] SAS users guide: statistics. Cary, NC: Statistical Analysis System, 1985.
[28] Fletcher C. The natural history of chronic bronchitis and emphysema. New York: Oxford University Press, 1976.
[29] Hudson DJ. Fitting segmented curves whose joint points have to be estimated. J Am Stat Assoc 1966; 61:873-80.