Passive Smoking, Air Pollution, and Acute Respiratory Symptoms in a Diary Study of Student Nurses1- 3

JOEL SCHWARTZ and SCOTT ZEGER Introduction

Since the realization that environmental pollutants may adversely affect the respiratory system, numerous epidemiologic studies have attempted to discover whether chronic exposure to various air pollutants is associated with increased risk of mortality (1, 2), chronic respiratory disease (3), and decreased lung function (4). Acute effects of air pollution on lung function have also been noted (4-6). Daily diaries represent a conceptually attractive approach to examining the potential relationship between air pollutants and acute symptoms, although the serial correlation in such data makes analysis more difficult (7). One valuable diary study was conducted in a nursing school in Los Angeles. The first report from this study by Hammer and coworkers (8), found associations between photochemical oxidants and respiratory symptoms. However, that analysis ignored smoking and serial correlation in the data, used linear regression to model the probability of a respiratory incident, ignored other potentially colinear air pollutants, and assumed a pollutant would have the same impact on starting an episode of symptoms as on prolonging the episode. Because of limited computational facilities at the time, the data on individual subjects were collapsed to rates per day. A more recent analysis of these data by Schwartzand colleagues (7)used logistic regressions, examined other pollutants, and used models that incorporated the serial correlation in the data. However,this study was still limited to the data on daily prevalence rate. Therefore, new incidence of symptoms and persistence of symptoms were still lumped together, and individual risk factors such as smoking and allergies had to be ignored. We have reexamined the original diaries from this study. They contain smoking and allergy histories as wellas symptom reports. In addition, several symptoms reported in the diaries have never been analyzed, and the individual datum 62

SUMMARY A cohort of approXimately 100 student nurses In Los Angeles was recruited for a diary study of the acute effects of air pollution. Smoking histories and presence of asthma and other allergies were determined by questionnaire. Diaries were completed dally and collected weekly for as long as 3 yr. Air pollution wes measured at a monitoring location within 2.5 miles of the school. Incidence and duration of a symptom were modeled separately. Pack-yearsof cigarettes were preclletlve of the number of episodes of coughing (p < 0.0001) and of bringing up phlegm (p < 0.0001). Current smoking, rather than cumulative smoking, was a better predictor ofthe duration of a phlegm episode (p < 0.0001). Controlling for personal smoking, a smoking roommate Increased the risk of an episode of phlegm (odds ratio [OR) = 1A1, P < 0.001), but not of cough. Excluding esthmatlcs (who msy be medicated), Increased the odds ratio for passive smoking to 1.76 (p < 0.0001). In logistic regression models controlling for temperature and serial correlation between days, an Increase of 1 SOIn carbon monoxide axposure (6.5 ppm) wes associated with Increased risk of headache (OR 1.09, P < 0.001), photochemical oxidants (7.4 pphm) were assoclsted with Increased risk of chest discomfort (OR = 1.17, P < 0.001) and eye Irritation (OR = 1.20 P < 0.001), and nitrogen dioxide (9.1 pphm) wesassoclated with Increased risk of phlegm (OR = 1.08 P < 0.01), sore throats (OR = 1.26, P < 0.001), and eye Irritation (OR = 1.16, P < 0.001). S02 wes not significantly assoclsted with any symptom. Weconclude thst passive smoking and short-term exposure to NO, and photochemical oxidants Increase the Incidence rates of respiratory Illness.

=

AM REV RESPIR DIS 1990; 141:82-67

allowed us to examine whether air pollution and smoking are associated primarily with increased risks of new episodes of respiratory (and other) symptoms, with increased duration of episodes, or both (or neither). The individual datum also allowed the examination of symptom severity. Methods Health Data A population beginning nursing school was recruited for a study of viral diseases and other risk factors for acute illness (8). Smoking histories and presence of asthma, hay fever, and other allergic conditions were obtained on all subjects. To be eligible for the study, students had to be residents of the nursing school. Daily diaries of acute symptoms were handed out and collected each Monday for 3 yr. Symptoms were graded as mild, moderate, or severe. The diaries for each nurse were double keypunched, and all discrepancies were validated. Symptoms keypunched were presence of headache, cough, sore throat, phlegm or sputum, chest discomfort, and eye irritation. Subjects who moved off campus or dropped out of school were dropped from the study at that point. Further detail on the study design has been published previously (7, 8). Tho types of potential risk factors for acute

symptoms were available: time-varying factors such as air pollution, and fixed subject covariates such as allergy and smoking history. Our study focused on the time-varying covariates (air pollution). Subject covariates (smoking and allergies) were examined primarily as potential covariates whose omission might increase the noise in the relationships with air pollution. The exception was passive smoking, whose relationship to acute symptoms was examined after controlling for personal smoking. The factors that increase the risk of acquiring an illness are not necessarily the same as those that increase its duration. Therefore sep-

(Receivedin originalform February 17, 1989 and in revised form June 26, 1989) 1 From the U.S. Environmental Protection Agency, Washington, DC; the Respiratory Epidemiology Program, Harvard School of Public Health, Boston, Massachusetts; and the Department of Biostatistics, Johns Hopkins School of Hygiene and Public Health, Baltimore, Maryland. 2 Supported in part by Contract No. RPlOOIfrom the Electric Power Research Institute and by Grant No.5 R29-AI-25529-02 from the National Institutes of Health. 3 Correspondence and requests for reprints should be addressed to Joel Schwartz, Ph.D., U.S. Environmental Protection Agency, PM221, 401 M Street SW, Washington, DC 20460.

63

PASSIVE SMOKING, AIR POLWTION, AND RESPIRATORY SYMPTOMS

arate analyses were carried out on incidence and duration. Incidence was defined as the presence of a symptom when the previous day was symptom-free. If the symptom was present on the previous day, the subject was not eligible to become incident, and incidence was coded as missing. An episode was defined as a set of days in which a symptom was continuously present, and duration was defined as the number of such days. To allow time for recruitment and acclimation to the diary questionnaire, the first 2 months of diaries were ignored in these analyses.

Aerometric Data Air pollution data wereobtained from a monitor located within 2.5 miles of the nurses' residence. Because the nurses lived, studied, and worked in the same location, the effects of mobility upon exposure were smaller than would be found in a general population sample. The highest l-h concentrations of sulfur dioxide (S02), nitrogen dioxide (N02), photochemical oxidants, and carbon monoxide (CO) were obtained daily. Temperature was obtained from a NOAA site located about a mile from the air pollution monitor. Daily particulate levels were not avaliable. The distribution of these daily pollution values, along with their units of measurement, are shown in table 1. Statistical Methodology Duration and incidence were modeled separately. Incidence was modeled in two stages. First, the number of incidents of each symptom for each subject was regressed on subject covariates (smoking and allergies). Poisson regression is commonly used to model counts of relatively rare events (9), and we adapted that strategy. More specifically, we assumed: In[E(Yi)] = Z,' a

+ InNi

where Zj is the vector of risk factors for the ith subject, Yi is the number of incidents for the ith subject, N; is the number of days the ith subject was eligible to be incident, and E denotes the expected value; maximal likelihood estimation was used. These models identified subject covariates to include in our analysis of air pollution. Subjects were then stratified into subgroups using these subject covariates. In the second step, the data were reduced to the proportion of subjects with the symptom incidence per strata per day. Once the dataset was stratified by the ap-

propriate subject covariates, weexamined the time-dependent variables using logistic regressions to model the probability of a subject experiencing a symptom incidence on a given day. Although incidence by definition implies the subject was symptom-free on the previous day, there is nevertheless a strong correlation between the percent of subjects with new incidents on successive days. This serial correlation is likely due to epidemic effects and the failure to control for all timedependent covariates (e.g., weather), which are themselves serially correlated (autocorrelated). Such autocorrelation between repeated measurements is a common problem in the Social Sciences where models for continuous response variables have been developed (10). The autocorrelation means that the observations are not truly independent, and analyses ignoring this autocorrelation will at a minimum misestimate the standard errors of the regression coefficients. Methods of estimating logistic regressions for autocorrelated data have been discussed elsewhere (11, 12). In this analysis, the methods proposed by Zeger and Liang (12) for logistic regression of autocorrelated data were used to estimate effects. As in logistic regression, this method models the logarithm of the odds of a symptom as a linear function of fixed covariates. Hence, we assume: Logit[E(Ytj)] = Xt'/}

+ Sj'a

Where X, is the vector of air pollution and temperature values at time t, Ytj is the proportion of subjects in strata j who have an incidence at time t, 10git(p)=log(p/(I-p), the log odds ratio, and Sj indicates the jth strata. However, the correlation among outcomes is accounted for when estimating the regression coefficients and, more importantly, when estimating their uncertainty. Similar analyses looked at the proportion of subjects with moderate or severe symptoms, or with symptoms with fever. In our approach, previous outcomes are not explicitly used to predict the current value as they are in Markov models, as described, for example, by Korn and Whittemore (11). Air pollution (and temperature) may induce symptoms on the day of exposure or on following days. To investigate this, all analyses were repeated looking at current pollution, pollution on the previous day, and pollution 2 days before the reporting day. When an association with more than 1 day was found, the means of 2 or 3 days of pollution were

tried. We also considered the possibilities of a nonlinear relationship between pollution and symptoms by examining logarithmic and square transformations of each time-dependent covariate. In addition, indicator variables were used to examine day of the week effects. The relationship between subject covariates and duration was investigated by linear regressions of the logarithm of duration against cigarette years, cigarettes/day, asthma, pollen allergy, etc. Because of the low rate of symptom incidence in the population, the sample size of episodes whose length could be modeled was much smaller than the number of subject days where an incident could occur. Therefore, we modeled only potentially strong factors such as smoking and allergies and did not examine the relationship between duration and air pollution. To help elucidate the shape of the mean dependence of outcome on pollutant in some of the models, we used LOWESS (13), a robust locally weighted nonparametric smoothing algorithm. LOWESS can be heuristically thought of as a generalization of a moving average.

Results

For each symptom shown in table 2, the median, 25th and 75th percentiles of incidence rate for the subjects, and the median duration are given. Cough, phlegm, and sore throat had daily incidence rates of 2 to 30/0, with median durations of 2 days. Chest discomfort was quite rare, with no incidents on more than 50% of the days. Prevalence rates for these symptoms (table 3) have been published previously (8).

Subject Covariates Smoking. In Poisson regressions of subject-specific covariates, pack-years of smoking was a better predictor of the number of incidents of cough (p < 0.0001)and phlegm (p < 0.001)than was current cigarettes smoked per day (table 4). Neither was a significant predictor of chest discomfort. Current cigarettes per day was a superior predictor of the duration of an episode of phlegm (p < 0.0001) and chest discomfort (p < 0.001), al-

TABLE 2 INCIDENCE RATES FOR SYMPTOMS (PERCENT)

TABLE 1

Symptom

25%

Median

75%

Median Duration

Cough Phlegm Sore throat Headache Chest Eye irritation

0 0 0 5.26 0 1.79

2.56 1.89 3.03 8.70 0 3.74

4.62 3.45 5.00 12.50 2.27 5.88

2 2 2 1 1 1

DISTRIBUTION OF POLLUTION AND TEMPERATURE Variable CO, ppm Oxidants, pphm N02, ppm S02' ppm Maximal temperature, OF

25%

Median

75%

Mean

12 4 0.06 0.010

15 9 0.11 0.020 71

20 15 0.17 0.040 79

16.71 10.24 0.13 0.034 71.8

64

64

SCHWARTZ AND ZEGER

TABLE 3 PREVALENCE RATES FOR CHRONIC CONDITIONS (PERCENn

22 23 18 5 17

Hay fever Sinusitis Smoking Asthma Pollen allergy

TABLE 4 DISTRIBUTION OF CIGARETTES SMOKED PER DAY AMONG SMOKERS

25% 6.5

Median

75%

Mean

12.5

20

13

though pack-years was a superior predictor of duration of cough episodes (p < 0.00(1). Controlling for personal smoking, a smoking roommate increased the risk of an episode of phlegm (RR = 1.41; 950/0 CI, 1.08 to 1.85) but not of cough. Excluding asthmatics (four subjects) increased the relative risk of phlegm incidence associated with passive smoking to 1.76 (95% CI, 1.33 to 2.33). Allergies. Pollen allergies were significantly associated with eye irritation (RR = 1.92;95% CI, 1.77to 2.08) and headache (RR = 1.55;95% CI, 1.46to 1.65). Chronic sinusitis was associated with increased risk of sore throat (RR = 1.95; 95% CI, 1.76102.15). Asthma wasnotsignificantly associated with any outcome.

Time-Dependent Covariates On the basis ofthe above results, the data for coughing and phlegm were stratified into never smokers, subjects with 2 pack-years or less, and subjects with greater than 2 pack-years of cigarette use. (The median number of pack-years among smokers was 2.5.) The data for eye irritation and headache were stratified by pollen allergy, and the data for sore throat was stratified by presence of chronic sinusitis. Daily incidence rates were computed for each strata. Logistic regressionanalysis (weighted by the number of subjects in each strata) was used to examine the association of air pollution, temperature, and day of the week. Indicator variables were used for each strata. Carbon monoxide seemed only plausibly related to headaches, and only that relationship was examined. SO], NO], and oxidants wereexamined for association with the other symptoms individually, and then in multiple pollutant combination, using stepwise selection.

Initial analyses considered a positive re- predictive ofheadache. Eye irritation and sponse regardless of severity; we then ex- chest discomfort were also correlated amined only the more severe symptoms with the same-day exposure. In contrast, (moderate or severe) and the incidence . sore throat and chest discomfort were of symptoms with fever. Logistic regres- correlated with pollution on the prevision coefficients and standard errors for ous day as well. The average of today's the significant pollutant symptom rela- and yesterday's pollution was more pretionships, after controlling for tempera- dictive than either day individually, and ture, subject covariates, day of the week was used in table 5. The linear form of pollution was aleffects, and serial correlation between days, are shown in table 5. On the basis ways more significant than the logarithof a 1 SD change in pollution, these co- mic transform. However, for eye irritaefficients corresponded to odds ratios of tion and chest discomfort, the square of 1.09 (95% CI, 1.05 to 1.12) for head- oxidant concentration wasmore predictive aches with CO of 1.17 (950/0 CI, 1.07 to than a linear term. This suggestsa nonlin1.29) for chest discomfort and oxidants ear relationship, with small effects of low of 1.08(95% CI, 1.015 to 1.15) for phlegm level exposure but greater proportional and NO], and of 1.26 (95% CI, 1.18 to effects of high level. These relationships 1.35) for sore throat and NO]. Both ox- are illustrated in figures 1 and 2. In conidants (OR = 1.20;95% CI, 1.15 to 1.25) trast, figures 3 and 4 show the more nearand NO] (OR = 1.16; 95% CI, 1.10to ly linear relationships between NO] and 1.21) were significant predictors of eyeir- eye irritation and sore throat. These figritation when included in the model to- ures are nonparametric smooths of the gether. All the symptoms showed in- raw data, using LOWESS, a robust, locreased rates on Mondays. Chest discom- cally weighted smoothing algorithm (13). fort also had a significantly lower risk LOWESS can be heuristically thought of on Saturday. as a generalization of a moving average. In simple logistic regressions that igThe current day of exposure to CO, rather than any lagged exposure, was nored serial correlation, smoking, and al-

TABLE 5 LOGISTIC REGRESSION RESULTS FOR SYMPTOM INCIDENCE Symptom Chest discomfort Intercept Oxidants· Temperature Monday Saturday Phlegm Intercept NO. Temperature Monday Smokingt Sore throat Intercept NO. Temperaturet Monday Sinusitis Eye irritation Intercept NO.§ (Oxidant)' Pollen allergy Monday Headache Intercept CO Monday Pollen allergy

Coefficient

Standard Error

p Value

-2.905 0.000684 -0.0208 0.5146 -0.246

0.3309 0.000205 0.00497 0.0902 0.1202

< 0.01 < 0.001 < 0.0001 < 0.0001

-2.379 0.843 -0.0169 0.626 0.207

0.244 0.343 0.00369 0.0740 0.059

< 0.0001 < 0.0001 < 0.0001 < 0.001

-2.311 2.571 -0.0238 0.4750 0.7514

0.2163 0.3635 0.0033 0.0713 0.0619

< 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001

-3.705 1.604 0.000767 0.5066 0.527

0.0474 0.257 0.000090 0.0602 0.0603

< 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001

-2.218 0.0125 0.386 0.328

0.138 0.0025 0.047 0.0471

< 0.0001 < 0.0001 < 0.0001 < 0.0001

• Mean of today's plus yesterday's oxidants, squared. 0 = nonsmoker, 1 = as much as 2 pack-years, 2 = > 2 paCk-years. Yesterday's temperature. § Mean of today's plus yesterday's NO,.

t

*

0.040

0.0140

65

PASSIVE SMOKING, AIR POLWTION. AND RESPIRATORY SYMPTOMS

0.010

i

I I 25 ;

ti•

0.01

0.10

-;.

0.025

I

0.020

i

0.04

I..

j

~

J

0.011

0.010

0.01

• '"

10

20

10

40

10

0.08

0.04

0.02

0.00

0.02 0

O.oe

0.0

0.1

0.2

0.3

0.4

0.0

0.5

0.1

N02 (ppm)

Fig. 1. The line shows the LOWESS smooth ofthe probability (in percent) of having an incident of chest discomfort versus oxidant levels (pphm).

0.10

i !

f

:•;

0.08

0.01l

0.04

0.02

0.00 L..._--'-_--'-_---'_ _...L-_--l

a

10

20

30

40

50

Oxidant (pphm)

Fig. 2. The line shows the LOWESS smooth of the probability (in percent) of having an incident of eye irritation versus oxidant levels (pphm).

lergies, the same pollutants were predictive of the same outcomes. This was also true after controlling for smoking and allergies. The regression coefficients for the significant pollutants in simple logistic regressions, after stratifying by smoking or allergies, and for the final models are shown in table 6. The coefficients were generally stable across model specification; however, as expected, the variances of the coefficients were larger in the model that accounted for serial correlation. When we redefined symptom incidence to require the presence of fever, none ofthe pollutants were predictive of respiratory illness, although oxidants, but not NO], remained significant for eye irritation (results not shown). In contrast, when incidence was redefined to require moderate or extreme severity, NO] remained predictive of sore throat (P = 2.19, t = 2.43) and carbon monoxide re-

0.2

0.3

0.4

0.1

N02(ppm)

Fig. 3. The line shows the LOWESS smooth of the probability (in percent) of having an incident of sore throat versus NO. levels (ppm).

Fig. 4. The line shows the LOWESS smooth of the probability (in percent) of having an incident of eye irritation versus NO. levels (ppm).

mained predictive of headaches (P = 0.0267, t = 4.89); both NO] (P = 3.35, t = 5.44) and oxidants (squared) (P = 0.0008, t = 4.72) were predictive of eye irritation.

problems and may be more sensitive at detecting increased frequency of relatively rare acute episodes in subjects without chronic respiratory symptoms. Kaufman and coworkers (24) have noted that the density of persons per room is higher in Europe than in the United States, which may partially account for the disparate findings. They note that particle concentration in the Six Cities Study (25) for homes with one and two smokers were less than half those found in homes with one or two smokers in the Netherlands (26). College dorm rooms, with two roommates per room, are surely closer to European housing density levels than to American levels. This is somewhat mitigated, however, because much, although not all, social activity on campuses occurs outside of dorm rooms. We conclude that the passive exposure levels seen in this study likely represent higher concentrations than do those commonly seen in U.S. homes. Exposure in many occupational settings, where the ratio of persons/square foot is much higher than in homes, may be much more representative of the levels seen in this study.

Discussion The effects of passive smoking on chronic respiratory symptoms in children has been noted in several studies (14-17). Studies of respiratory symptoms in adults are much rarer. Studies in Europe and Japan (18-20) have tended to fmd associations between passive smoking and chronic respiratory symptoms, whereas most, but not all, U.S. studies have not found such associations (14, 21-23). Most of these studies used questionnaires derived from the British Medical Research Council questionnaire (e.g., ATS, NHLBI). Such annual questionnaires may be subject to diminished recall for events more than a few months prior to administration, and are probably better at determining the presence of chronic cough or phlegm than at determining the frequency of acute events. A daily diary, in contrast, avoids recall

TABLE 6 POLLUTION COEFFICIENTS FOR DIFFERENT MODEL SPECIFICATIONS Outcome

Pollutant

Basic'

Phlegm Sore throat Chest Headache Eye

NO. NO. Oxidant:!: CO NO. OXidant:!:

0.756 1.76 7.04 x 10'" 0.0117 1.52 0.0025

• Normal logistic regression. Normal logistic regression. stratified by smoking/allergy status. Square of oxidants.

t

*

Stratat

0.948 2.08 NA

0.0129 1.63 0.0032

Final

0.848 2.57 6.84 x 10-' 0.0125 1.61

o.ooon

66

In summary, we interpret the increased risk of episodes of phlegm caused by passive exposure to tobacco smoke as measuring a somewhat different and more sensitive response than previous studies with annual questionnaires. Nevertheless, these findings appear broadly consistent with the positive associations found in those studies conducted in population where smoke concentrations appear to be higher. The lack of association between symptoms and asthma, and the higher relative risk for passive smoking when asthmatics are excluded, may be due to the use of medication by the asthmatics. The finding of a stronger association of current smoking with duration of episode (for chest and phlegm) and of cumulative smoking with number of incidents (for cough and phlegm) is also of interest. It suggests that even for these low pack-years smokers, cumulative impacts of smoking on the lung are increasing the overall risk of respiratory morbidity. Current irritation, by contrast, seems to be playing a role in extending the length of the episodes. Studies of the effects of N01 on respiratory symptoms have yielded mixed results. Comstock and coworkers (21) found an increased incidence of chronic respiratory symptoms among men, but not among women, in homes using gas stoves. A later analysis restricted to nonsmokers (27) found an association in both sexes. Others have reported similar associations in children (28, 29). However, other studies did not find associations (3D, 31). No association was found between gas stove use and the number of acute respiratory illnesses as determined by biweekly telephone calls (32, 33). However, a significant association between number of days of illness in a longitudinal study and hours of use of kerosene heaters has been reported by Berwick and colleagues (34). The hours of use measure was shown to correlate with indoor N01 levels. The above results, although mixed, indicate that our findings are plausible. Our analysis, by the use of daily diaries, should be more sensitive to detect effects of N01 than annual questionnaires. Conversely,the use of outdoor N01 measurements decreases the sensitivity. Even when smoking, allergies, day of the week, temperature, and serial correlation have been controlled for, the possibility of unmeasured confounders exists in any epidemiologic study. The finding of no significant association with symptoms of fever, which likely represent infectious

SCHWARTZ AND ZEGER

diseases, is reassuring since it suggests the association is not due to confounding with epidemics. N01 remains significant as a predictor of more severe sore throats, indicating that the lack of association with fever plus symptoms is not due merely to the lower frequency of the complex. Taken together, the results for all symptoms, symptoms with fever, and severe symptoms show a pattern consistent with a causal relationship caused by irritation. The association of photochemical oxidants with chest discomfort is consistent with the results of Korn and Whittemore (11)for asthmatics. It suggests that ambient oxidants are likely to increase the risk of respiratory symptoms even in nonasthmatics. However, the greater significance of the square of oxidants in the regression suggests a nonlinear relationship on the logistic scale. This finding, suggesting no noticeable elevation in risk below 20 pphm, is confirmed in figure 1. This is well above the current National Ambient Air Quality Standard of 12 pphm. Schwartz and coworkers (7), in analyzing prevalence rates in these data, found a greater association between chest symptoms and SOl than for oxidants, and no association with oxidants when SOl was controlled for. In contrast, we found no association between incidence rates and SOl' with or without oxidants in the model. The differences may be due to an association between SOl and persistence of chest discomfort. However, given the much lower autocorrelation in incidence data than in prevalence data, the differences observed may also be due to model misspecification in the earlier analysis. Previous analysis of these data showed a relationship between carbon monoxide and the prevalence of headaches (7). We have confirmed this relationship for incidence in models that control for allergies. Carbon monoxide exposure has been linked to headaches in garage workers (35) and in experimental settings (36). The hypoxia of carbon monoxide shows similarities to the hypoxia of altitude (37), which has been linked to headaches, adding credence to these findings. Because CO levels are associated with automobiles, it is possible that there is some confounding with noise exposure, however. Eye irritation has long been associated with photochemical oxidant (38). We have found a dose-dependent relationship that appears nonlinear. The square of oxidants was a better predictor of eye irritation than a linear term, suggesting

a smaller relative effect at low levels. This appears to be borne out in figure 2. In contrast, the relationship with N01 appeared more linear on the logistic scale. This study could not determine whether N01 had a direct effect, or showed a correlation because of its precursor role in the formation of peroxyacetyl nitrate. The latter seems more likely (38). Whether the impact is direct or indirect, these results again suggest that controlling daily N01 concentrations would reduce minor morbidity. The lack of daily particulate measurements creates a potential for confounding since particulate levels have been associated with short-term respiratory effects (4). The significant gaseous pollutants in this study are not usually highly collinear with particulates, but the omission must caveat the results of this study. In conclusion, given the greater sensitivity of diaries of acute symptoms compared with annual questionnaires to detect respiratory effects of passive smoking, and the strength of the association in these data, we believe these results strongly suggest an impact of passive smoking. The air pollution results are also relatively strong. The oxidant effects are expected, and primarily occur at high concentrations. The N01 associations, in contrast, occur at levels that were seen frequently in Los Angeles and are common elsewhere in urban areas, even in ones attaining the current annual standard. Although no epidemiologic study can prove causation, these results strongly suggest that the current National Ambient Air Quality Standards, which only set a limit for annual average N01 , need to be changed to incorporate a short-term standard. Acknowledgment Ed Fu supervised the keypunching of the individual diaries and prepared the database used in this analysis.Douglas Dockery provided valuable comments on the manuscript. References 1. Martin AE, Bradley WHo Mortality and morbidity statistics and air pollution. Proc R Soc Med 1964; 57:979. 2. Mazumdar S, Schimmel H, Higgins lIT. Relation of daily mortality to air pollution: an analyses of 14 London winters 1958/59-1971/72. Arch Environ Health 1982; 37:213-20. 3. Ware JH, Ferris BO Jr, Dockery OW, Spengler JO, Stram 00, Speizer FE. Effects of ambient sulfur oxides and suspended particles on respiratory health of preadolescent children. Am Rev Respir Dis 1986; 133:834-42. 4. Dockery OW, Ware JH, Ferris BO Jr, Speizer FE, Cook NR, Herman SM. Changes in pulmo-

PASSIVE SMOKING, AIR POLWTION, AND RESPIRATORY SYMPTOMS

nary function in children associated with air pollution episodes. JAPCA 1982; 32:937-42. 5. Dasen W, BrunekreefB, Hoek G, etaL Decline in children's pulmonary function during an air pollution episode. JAPCA 1986; 36:1233. 6. Spektor DM, Lippman M, Lioy PJ, et al. Effects of ambient ozone on respiratory function in active normal children. Am Rev Respir Dis; 1988; 137:313. 7. Schwartz J, Haselblad V, Pitcher H. Air pollution and morbidity: a further analysis of the Los Angeles student nurse data. JAPCA 1988; 38:158-62. 8. Hammer Dl, Hasselblad V, Portnoy B, Wehrle PF. Los Angeles student nurse study. Daily symptom reporting and photochemical oxidants. Arch Environ Health 1974; 28:255-60. 9. McCullagh D, NeIder JA. Generalized linear models. London: Chapman and Hall, 1983. 10. Box GEP, Jenkins GM. Time series analysis: forecasting and control. San Fransico: Holden-Day, 1970. ll. Korn EL, Whittemore AS. Methods of analyzing panel studies of acute health effects of air pollution. Biometrics 1979; 35:795-802. 12. Zeger SL, Liang K-Y. Longitudinal data analysis for discrete and continuous outcomes. Biometrics 1986;42:121-30. 13. Cleveland WS. Robust locally weighted regression and smoothing scatterplots. J Am Stat Assoc 1979; 74:829-36. 14. Lebowitz MD, Burrows B. Respiratory symptoms related to smoking habits of family adults. Chest 1976; 69:48-50. 15. WeissST, lager IR, Speizer FE, Rosner B. Persistent wheeze. Its relationship to respiratory illness, cigarette smoking, and level of pulmonary function in a population sample of children. Am Rev Respir Dis 1980; 122:697-707. 16. Leeder SR, Cookhill R, Irwig LM, Holland WW, Colley JIIT. Influence of family factors on the incidence of lower respiratory illness during the first year oflife. Br J Prev Soc Med 1976;30:203-12. 17. Ware JH, Dockery DW, Spiro A III, Speizer FE, Ferris BG Jr. Passive smoking, gas cooking,

and respiratory health of children living in six cities. Am Rev Respir Dis 1984; 129:366-74. 18. Kojima N, Kasuga H, Osaka F, Matsuk H. Passive smoking effects to non-smoking adults. Jpn J Hyg 1981; 36:200. 19. Gillis CR, Hole DJ, Hawthore VM, Boyle P. The effect of environmental tobacco smoke in two urban communities in the west of Scotland. Eur J Respir Dis 1984; 65(Suppl 133):121-6. 20. Simecek C. Reflection of passive exposure to smoking in the home on the prevalence of chronic bronchitis in non-smokers. Czech Med 1980; 3: 308-10. 21. Comstock GW, Meyer MB, Helsing KJ, Tockman MS. Respiratory effects of household exposure to tobacco smoke and gas cooking. Am Rev Respir Dis 1901; 124:143-8. 22. Shilling RSF, Letai AD, Hui SL, Beck GT, Schoenberg JB, Bouhys A. Lung function, respiratory disease and smoking in families Am J Epidemiol 1977; 106:274-83. 23. Schenker MB, Samet JM, Speizer FE. Effect of cigarette tar content and smoking habits on respiratory symptoms in women. Am Rev Respir Dis 1982; 125:684-90. 24. Kaufman F, Dockery DW, Speizer FE, Ferris BG Jr. Respiratory symptoms and lung function in relation to passive smoking. A comparative study of American and French women. Int J Epidemiol 1989; 18:334-44. 25. Spengler JD, Dockery DW, Thrner WA, Wolfson JM, Ferris BG Jr. Long-term measurements of respirable sulfates and particles inside and outside homes. Atmos Environ 1981; 15:23-30. 26. Brunekreef B, Boleig JSM. Long-term average suspended particulate concentrations in smoker's homes. Int Arch Occup Environ Health 1982; 50: 299-302. 27. HelsingKT,ComstockGW,MeyerMB, Tockman ML. Respiratory effects of household exposure to tobacco smoke and gas cooking on non-smokers. Environ Int 1982; 8:365-70. 28. Melia RJW, Florey CduV, Chinn S. The relation between respiratory illness in primary school

67 children and the use of gas for cooking. 1. Results from a national survey. Int J Epidemiol 1978; 8: 333-8. 29. Speizer FE, Ferris BGF Jr, Bishop YMN, Spengler J. Respiratory disease rates and pulmonary function in children associated with NO. exposure. Am Rev Respir Dis 1980; 121:3-10. 30. Ogston SA, Florey CV, Walker CH. The Thyside Infant Morbidity and Mortality Study: effect on health of using gas for cooking. Br Med J 1985; 290:957-60. 31. Schenker MB, Samet JM, Speizer FE. Risk factors for childhood respiratory disease: the effect of host factors and home environmental exposures. Am Rev Respir Dis 1983; 28:1038-43. 32. KellerMD, Lanese RR, Mitchell RI, Coke RW. Respiratory illnessin households using gas and electricity for cooking. I. Survey of incidence. Environ Res 1979; 19:495-503. 33. KellerMD, Lanese RR, Mitchell RI, Coke RW. Respiratory illness in households using gas and electricity for cooking. II. Symptoms and objective fmdings. Environ Res 1979; 19:504-15. 34. BerwickMA, Zagraniski RJ, Leadere BP, Stolwick JA. Respiratory illness in children exposed to unvented combustion sources. In: Berglund R, Lindvall T, Sundell J, eds. Indoor air. Vol. 2: Radon, passive smoking, particulates, and housing epidemiology. Stockholm: Swedish Council for Building Research, 1984; 255-60. 35. Fristedt B, Akesson B. Health hazards from automobile exhaust exposure at service facilities in multistory garages. Hyg Rev 1971; 6O:112. 36. Stewart RD, Peterson JE, Baretta Ed, Bachard lIT, Hosko MJ, Hermann AA. Experimental human exposure to carbon monoxide. Arch Environ Health 1970; 21:154. 37. Astrup P, Pauli HG. Comparison of prolonged exposure to carbon monoxide and hypoxia in man: report of a joint Danish-Swiss study. Scand J Clin Lab Invest [Suppl) 1968; 103:1. 38. National Research Council. Ozone and other photochemical oxidants. Washington, DC: National Academy of Sciences, 1977.

Passive smoking, air pollution, and acute respiratory symptoms in a diary study of student nurses.

A cohort of approximately 100 student nurses in Los Angeles was recruited for a diary study of the acute effects of air pollution. Smoking histories a...
701KB Sizes 0 Downloads 0 Views