Air pollution and daily hospital admissions in metropolitan Los Angeles.

We used daily time-series analysis to evaluate associations between ambient carbon monoxide, nitrogen dioxide, particulate matter [less than and equal to] 10 microm in aerodynamic diameter (PM(10)), or ozone concentrations, and hospital admissions for cardiopulmonary illnesses in metropolitan Los Angeles during 1992-1995. We performed Poisson regressions for the entire patient population and for subgroups defined by season, region, or personal characteristics, allowing for effects of temporal variation, weather, and autocorrelation. CO showed the most consistently significant (p<0.05) relationships to cardiovascular admissions. A wintertime 25th-75th percentile increase in CO (1.1-2.2 ppm) predicted an increase of 4% in cardiovascular admissions. NO(2), and, to a lesser extent, PM(10) tracked CO and showed similar associations with cardiovascular disease, but O(3) was negatively or nonsignificantly associated. No significant demographic differences were found, although increased cardiovascular effects were suggested in diabetics, in whites and blacks (relative to Hispanics and Asians), and in persons older than 65 years of age. Pulmonary disease admissions associated more with NO(2) and PM(10) than with CO. Pulmonary effects were generally smaller than cardiovascular effects and were more sensitive to the choice of model. We conclude that in Los Angeles, atmospheric stagnation with high primary (CO/NO(2)/PM(10)) pollution, most common in autumn/winter, increases the risk of hospitalization for cardiopulmonary illness. Summer photochemical pollution (high O(3)) apparently presents less risk.

Time-series analyses of daily mortality or morbidity have shown statistical associations with air pollution in cities throughout the world. Physiologic/toxicologic mechanisms of these phenomena remain unknown, and time-series analyses have not clearly linked specific pollutants with specific health outcomes (1)(2)(3); thus, their application to pollution-control policy decisions remains controversial (4). Combustion-related particulate matter, the only pollutant common to virtually all locations of time-series studies, has been the focus ofscientific and regulatory attention (1,2,(4)(5)(6). However, recent studies in a number of North American cities also associate cardiovascular and/or pulmonary disease incidence with pollutant gases such as carbon monoxide, nitrogen dioxide, and/or ozone (7-14. Where they are not highly correlated, gas and particulate pollutants appear to have separate statistically and medically significant influences on cardiopulmonary morbidity (9,11,14M. The Los Angeles metropolitan area has been studied relatively little by time-series analysis, but is a good candidate for study because of its large diverse population (a 14 million); detailed monitoring of air quality and hospital admissions; mild climate, which should limit confounding of pollution effects by weather stresses; and severe but widely variable air pollution (with maximum levels of primary pollutant gases, secondary photochemical oxidant gases, and particulate pollution occurring at somewhat different times and places). Powerful tests of pollution effects should be possible in the entire metropolitan population and in subpopulations defined geographically, demographically, or clinically. We hypothesized that regional and/or seasonal differences in time-series analysis results in the general population and/or in particular subgroups, would allow us to distinguish effects associated with primary pollutants (CO or NO2), photochemical oxidants (03), or particulate matter more clearly than has been possible elsewhere. If so, we could rank these categories of pollution in terms of their public health impact, and thus provide useful guidance for regulatory policymaking and for future research on mechanisms. To test this hypothesis, we analyzed daily admission data for 1992-1995 from the South Coast Air Basin (Los Angeles, Riverside, San Bernardino, and Orange Counties in California, excluding mountain and desert regions of the first three counties) in relation to daily levels of CO, NO2, 03, and particulate matter < 10 pm in aerodynamic diameter (PM0O).
Methods Data acquisition and management. After its institutional review board verified confidentiality protection, the California Office of Statewide Health Planning and Development (OSHPD) (Sacramento, CA) provided records of hospital admissions in the metropolitan counties for 1992-1995 (the only years with adequately comparable PM10 data). The records included hospital identifier, date, principal and additional diagnoses as International Classification ofDiseases (ICD; World Health Organization, Geneva) codes, All-Patient-Refined Diagnosis-Related Group (APR-DRG; 3M Inc., Murray, UT)a broader classification based on Medicare diagnosis-related groups, sex, age, ethnic group, and residence zip code. Daily counts after 21 December 1995 were excluded from analysis because the records for numerous patients not discharged until 1996 were missing, and all 1995 data were excluded from ethnic-group analyses because of changes in OSHPD ethnic classifications. Broad principal-diagnosis categories used in analyses were cardiovascular (APR-DRG 103-144); cerebrovascular (APR-DRG 14-17 and 22); pulmonary (APR-DRG 75-101); and abdominal-a negative control category thought to be unrelated to pollution (APR-DRG 146-207). More-specific principal diagnoses, thought likely to associate with air pollution on the basis of previous epidemiologic or toxicologic evidence, were also analyzed: congestive heart failure (CHF) (APR-DRG 127); myocardial infarction (APR-DRG 1 1 1, 1 1 5, and 121); cardiac arrhythmia (APR-DRG 138); occlusive stroke (APR-DRG 14); asthma (ICD 493); and chronic obstructive pulmonary disease (COPD) (APR-DRG 88). Analyses excluded patients younger than 30 years of age (with exceptions noted in "Results") and prescheduled admissions.
We obtained air pollution and meteorologic data from the South Coast Air Quality Management District (SCAQMD; Diamond Bar, CA) and from the National Weather Service. These data included hourly PMIO from six SCAQMD stations with continuous monitors; hourly CO, 03, NO2, temperature, and relative humidity from those stations plus others; 24-hr-average PMIO measured every sixth day by high-volume samplers at or near each continuous PMIO station; and barometric pressure and rainfall at the Los Angeles International Airport. Figure 1 shows monitor locations. Analyses related daily admission counts with 24-hr averages of environmental variables. For 03, maximum hourly concentrations were also analyzed; they correlated highly with 24-hr averages in all seasons (r 2 0.79) and showed similar relationships to daily morbidity ("Results"). We did not analyze relative humidity because many data were missing or out of range. Stations differed in their continuous PM1O monitoring techniques and their relationships of continuous to high-volume sampler data. On the assumption that high-volume data were more comparable throughout the basin, we used season-and station-specific linear regressions to adjust continuous data to conform with high-volume data. We defined seasons to begin 1 January (winter), 1 April (spring), 1 July (summer), and 1 October (autumn).
Geographic differences were investigated across six regions defined by continuous PM0o monitoring stations (Figure 1). A region consisted of all zip codes that had a majority of their area closest to its station, except that some western coastal zip codes, which were separated from their closest station (region 1) by mountains, were assigned to region 2 to better represent their air quality. Admitted patients were assigned to regions by their residence zip codes. We excluded the 6.7% with zip codes missing or outside the South Coast Air Basin from regional analyses. We determined pollutant gas concentrations and temperatures for each region by averaging across all monitoring stations within it. Missing air monitoring data (4.4% for PM1O, smaller percentages for other variables) were replaced using analysis of variance with maximum likelihood estimation of missing values. For PM10, data from all days in the same season and all stations in the basin were used in estimation; for other variables, only stations in the same region were used. Statistical analyses. We used BMDP software (SPSS Inc., Chicago) and SAS software (SAS Institute, Inc., Cary, NC) for statistical analyses. Descriptive statistics and correlation patterns were examined regionally and seasonally for admission counts and atmospheric variables. Further descriptive analyses were performed to contrast weather and pollution characteristics between days with unusually high and unusually low observed morbidity relative to predicted values accounting for cyclical and secular trends. Predicted values were from regressions with indicator variables for the day of the week PM1, + gas + tsmperature monitor M Temperature + gas monitor * Gas monitor A Weather Irain, barometric pressure) monitor and for weekday holidays, with longer term variation modeled by fitting cubic splines to successive 28-day intervals of data. We then compared weather and pollution statistics between days with high admissions (residual > 85th percentile) and low admissions (residual < 15th percentile), as well as the immediately preceding days.
Time-series analytical approaches induded a) ordinary least squares (OLS) regression with admission count and atmospheric data filtered by the Shumway 19-day weighted moving average procedure (13), with or without an autoregressive component; b) regression of log-transformed daily admission counts using polynomial distributed lag models (14,15); and c) Poisson regression with allowance for overdispersion and autocorrelation, adapted from the analytical strategy of the Partide Epidemiology Evaluation Project (16) with modifications. In principle, daily counts are Poisson distributed and require approach c; however, given the generally large counts with filtering or smoothing, distributions were reasonably near normal so that other approaches were also feasible (17). The different approaches yielded similar conclusions when considering cardiovascular diseases. Polynomial-distributed lag models showed the largest significant effects consistently at lag 0, and effects beyond lag 1 were nearly always nonsignificant. Therefore, we adopted Poisson regression as the primary analytical tool. Predictors of daily admission counts included basis variables of a cubic-spline smooth on time (which accounted for secular trends and seasonal variation); indicator variables for the day of the week and for weekday holidays; indicator variables for hot days (maximum temperature > 85th percentile for entire study period), cold days (minimum temperature < 15th percentile), and rain days (> 0.01 inches at the Los Angeles International Airport); continuous atmospheric variables (one or more pollutant concentrations, barometric pressure, and mean temperature); and an autoregressive term-the residual admission count at lag 1, determined in a preliminary regression including all other predictors. Seasonal variation was more complex for pulmonary diseases than for others, probably because the timing and intensity of winter infectious disease outbreaks varied from year to year. Thus, cubic splines were determined at 28-day intervals when smoothing pulmonary disease counts, and at 4-month intervals otherwise.

Results
Seasonal air quality and hospital admissions. Table 1 presents seasonal pollution, weather, and hospital admission statistics for the entire basin for 1992-1995. Overall means ± SDs VOLUME 108 1 NUMBER 5 May 2000 * Environmental Health Perspectives were 1.5 ± 0.8 ppm for CO, 3.4 ± 1.3 parts per hundred million (pphm) for NO2, 45 ± 18 pg/m3 for PM1O, and 2.4 + 1.2 pphm for 03. We determined basinwide means from the six regional values by weighting each region according to its proportion of cardiovascular plus pulmonary admissions (considered to reflect its proportion of the population at risk). These levels reflect a > 80% reduction of CO since the 1960s (18), and more modest reductions in the other pollutants. Year-round means and SDs of daily admissions were 428 ± 76 for cardiovascular, 207 ± 54 for pulmonary, 74 ± 14 for cerebrovascular, and 244 ± 39 for abdominal diseases. Seasonal means for abdominal diseases (not tabulated) varied < 3%. All disease categories showed marked variation by day of the week, consistent across seasons. Relative to Monday admissions, Sunday admissions averaged 64% for cardiovascular, 70% for pulmonary, 76% for cerebrovascular, and 67% for abdominal diseases. Table 2 shows pairwise correlations of basinwide average daily pollutant concentrations, mean temperature, and barometric pressure within each season. NO2 showed high positive correlations (r . 0.8) with CO in all seasons, and correlations nearly as high with PM O. 0 3 was positively correlated with all three other pollutants only in the spring, and most strongly with PM O0 03 showed a weaker positive relationship to PM1O in the summer and a negative relationship to CO and NO2 in the winter. Higher mean temperatures were associated with higher pollutant levels in all seasons, with the exception of CO in the autumn. Barometric pressure showed varying relationships with pollutants. Expressing the data as residuals from cubicspline smoothing brought about no marked changes in these correlations, except that in autumn the positive relationship between 03 and temperature became nonsignificant. Regional measurements and basinwide averages correlated strongly for every pollutant in every season (r > 0.7), except for PM1O in the summer (r = 0.5-0.6 for some regions). Different regions' measurements of a given pollutant also correlated positively in all seasons. Southern coastal region 4 and eastern inland region 6 contrasted most sharply, with r-values between 0.3 and 0.7. In light of this generally similar behavior of air quality in different regions, time-series analyses focused on the entire basin, and regional comparisons were limited to regions 4 and 6. was highest in the summer and autumn, particularly in the eastern inland region 6, but seasonal variation was less for PMIO than for CO. 03 was highest in the spring and summer, particularly in inland regions 3 and 6. Contrast of atmospheric conditions between days with high and low admission counts. Table 3 summarizes significant (p < 0.05) differences in basinwide weather and pollution statistics between days with unusually high and unusually low admission counts (residuals from cubic-spline smoothing) in a particular broad disease category. Highadmission days (and/or immediately preceding days) tended to have relatively warm dry weather. Primary pollution (CO and NO2) was significantly elevated on winter, spring, or autumn days with high cardiovascular admissions; spring and summer days with high pulmonary admissions; and spring and autumn days with high cerebrovascular admissions.
Elevated PMIO tended to accompany elevated primary pollutants on days with high cardiovascular or pulmonary admissions; PMIO also was associated with high abdominal admissions in the spring. 03 was increased (along with the other pollutants) on days with high pulmonary admissions in spring and summer, the seasons of the highest mean 03 concentrations (Table 1). By contrast, 03 was decreased on days with high cardiovascular admissions in the winter, when 03 was generally low and negatively correlated with the other pollutants. Table 4 shows mean weather and pollution conditions on days of high and low cardiovascular admissions in the winter and summer for the contrasting southern coastal region 4 and eastern inland region 6. In the summer, pollution (except for CO) and heat were markedly greater in region 6, but there were no clear pollution or temperature differences between highand low-admission days in either region. In the winter, CO was markedly higher in region 4, and other regional differences were modest. In region 4, winter high-admission days had significantly higher temperature, barometric pressure, CO, NO2, and PM1O, and significantly lower probabiliy of rain, than low-admission days. In region 6, these tendencies were less obvious, but CO and NO2 were significantly elevated on the days preceding high-admission days. In similar analyses of pulmonary diseases (not tabulated), we found only a few significant associations with high admissions: high same-day PMIO in region 4 in the winter, Analyses of admission counts in broad disease categories.   year-round and winter analyses, and also to NO2 or CO in autumn. Cerebrovasculardisease admissions were significantly related only to CO or NO2 only in the spring. Abdominal-disease admissions were significantly related only to NO2, and only in the year-round analysis. 03 showed either negative or nonsignificant positive relationships with cardiovascular, pulmonary, cerebrovascular, and abdominal disease admissions in year-round and singleseason analyses. The same was true in OLS regressions of Shumway-filtered data (not tabulated). Alternative analyses intended to give additional weight to high-03 conditions, by expressing exposure in terms of daily maxima or in exceedances of an assumed threshold, or by restricting the analysis to the three high-03 inland regions, still showed no significant positive associations. In Poisson models excluding mean temperature and barometric pressure, daily mean 0 showed significant positive associations wit1 pulmonary diseases in the spring and year-round. Table 6 illustrates the sensitivity of results to the choice of regression procedure and model for the cardiovascular disease/CO relationship in the winter and the pulmonary disease/03 relationship in the spring. Across a broad range of models with and without weather and other pollutants as predictors, 'Highand low-admission days (> 85th and < 15th percentiles, respectively) are determined by residuals from regressions accounting for temporal effects (see text). A significant increase in atmospheric variables on high-admission days relative to low-admission days land/or the immediately preceding days) is indicated by +; a significant decrease by-. the change in atmospheric measurement from the preceding day was significantly more positive on high-than lowadmission days, alfthough values measured on high and low days were not significantly different 'For pulmonary diseases in winter, high-admission days' increases in NO2, PM,,, and high temperature approached significance (p < 0.10).
%or abdominal diseases in winter, high-admission days' increases in CO approached significance (p < 0.10).  Table 3 for definition of highand low-admission days. bMeasured at the Los Angeles International Airport, closer to region 4 than region 6. Other measurements made within the indicated region. *Significant (p < 0.05) differences. **Although this difference did not reach significance, the difference between days immediately preceding highand low-admission days was significant (p < 0.05). VOLUME 108 NUMBER 5 May 2000 * Environmental Health Perspectives B[, the estimated winter CO effect was always significant and was reasonably consistent in size. None of the other pollutants' effects was significant when included in a model with CO. The spring 0 effect on pulmonary admissions was significant when 03 was the only atmospheric factor in the model, predicting a 1.5% increase in admissions for a 1-pphm increase in daily mean 03 concentration. However, the 03 effect was nonsignificant if the model included weather and/or other pollutant variables. Interpretation of these findings is complicated by collinearity and by possibly different characteristics of exposure measurement error for different pollutants (19). Nevertheless, it seems dear that in the winter, CO was the analyzed atmospheric factor that was most closely linked with excess cardiovascular morbidity. In the spring, 03 was the pollutant most closely associated with excess pulmonary morbidity; however, morbidity was still more closely associated with warm temperatures, and all four pollutants tended to rise with temperature, making interpretation difficult.
Because diabetes mellitus is an important risk factor for cardiovascular disease, we reanalyzed cardiovascular admissions separately for diabetics (all of those with ICD code 250 entered among four additional diagnoses in the record; approximately 20% of all patients) and for others, using the autoregressive Poisson model. In year-round analyses, the slope ± SE was 0.039 ± 0.006 for diabetics as compared to 0.031 ± 0.004 for others.
Year-round analyses of NO2 and PMIO effects showed similar modest slope increases for diabetics, as did single-season analyses. None of the slope differences between diabetics and others was statistically significant.
Analyses ofcardiovascular disease admission counts by age, sex, and ethniity. Table   7 presents results from single-pollutant autoregressive Poisson models applied to cardiovascular admission counts in three age strata (30-64, 65-74, and 2 75 years of age) separately for men and women. Results for 03 (not tabulated) were never statistically significant. CO effects were near-significant for women 30-64 years of age and sig-nificant in all other age-sex groups in year-round analyses and in one or more seasonal analyses. Effect sizes increased with age similarly in both sexes, but age-related differences were not significant. NO2 effects showed a similar pattern of significance, but with less suggestion of age dependence. PMIO effects were also significant year-round and/or in one season for all groups except men > 75 years of age. years of age in four ethnic categories-white (non-Hispanic), black, Hispanic, and other. The other category indudes people of Asian-Pacific ancestry (the large majority), Native Americans, and others not dassifiable in the first three groups. 03 effects (not tabulated) were never significant. Regression coefficients, though not significantly different, suggested meaningful ethnic differences in exposure-response relationships. CO, NO2, and PM0O effects were significant in whites in yearround, winter, and autumn analyses. In blacks, CO and NO2 effects were significant year-round (also in the winter for CO) and were similar to these effects in whites. CO and NO2 effects in Hispanics were significant yearround but were smaller than these effects in whites and blacks. The remaining (other) category, with a relatively small number of admissions, showed consistently small and nonsignificant regression slopes.
Analyses ofadumission counts for more specifi diagnoses. Table 9 presents results in adults > 30 years of age from singlepollutant autoregressive Poisson models relating basinwide daily average pollutant concentrations with same-day admission counts. Occlusive strokes showed the most consistent positive relationships to pollution: significant associations with 03 in the summer only; and with CO, NO2, and PMIO year-round and in at least two single-season analyses. Asthma, COPD, and CHF were  .000 (0.008) 'Regression analyses used 24-hr average measurements of pollutants and same-day admission counts for patients > 30 years of age throughout the South Coast Air Basin. Example interpretation: the coefficient 0.038 relating winter cardiovascular admissions to CO indicates that admissions increase by a factor of aem., i.e., by 3.9%, with a 1-ppm increase in CO concentration, after allowing for the effects of time and weather on admission rates. *Significant in expected direction, p < 0.05. **Significant in the 'wrong' direction, p < 0.05. year-round and in one or more single-season coastal region 4 and eastern inland region 6. analyses. Myocardial infarction was associat- The two regions showed reasonably similar ed with CO and NO2, and arrhythmias with daily admission counts and similar positive CO, in year-round analyses only.

correlations of daily CO and PMIO levels,
We also analyzed asthma admissions for but markedly different concentration ranges patients 0-29 years of age. In year-round ( Figure 2). By OLS regression allowing for analyses, slopes ± SEs were 0.036 ± autocorrelation, a wintertime 1-ppm rise in 0.016/ppm CO, 0.024 ± 0.008/pphm NO2, CO predicted a 9-pg/m3 rise in PM10 in and 0.0011 ± 0.0006/pg/i3 PM -all sigregion 4, but a 25 pg/m3 rise in region 6. In nificant (p < 0.05) and appreciably larger single-pollutant autoregressive Poisson modthan the slopes in adults > 30 years of age. els, region 4 showed highly significant rela-0 effects were nonsignificant. Most of the tionships between PMIO and admissions, almitted patients in this youngest group year-round and in the winter, despite its low were children: the mean age was 7.
PM10. In region 6, despite its high PM1O, Relationships of cardiovascular disease regression slopes were significantly lower admissions to CO or PM10 in the two most than in region 4, and were not significantly contrasting regions. Table 10 presents different from zero. Admissions showed a comparative statistics for cardiovascular more plausible relationship with CO across   Table 5 for explanation of regression procedure and coefficients. bAnnual mean daily admission count ± SD, for patients . 30 years of age. C03 results not tabulated; none was significantly positive. *Year-round relationship significant, p < 0.05. the two regions, with highly significant positive slopes in region 4 and modestly lower non-significant slopes in region 6, consistent with its generally lower and less variable CO concentrations.

Discussion
Limitations; recommendations for future research. Problems with this and other time-series studies indude exposure misdassification, response misclassification, and model misspecification. Exposure misclassification occurs when the monitored environmental factors are not the ones responsible for health effects, when monitoring errors are appreciable and differ by time and location, when monitoring station data poorly represent background air near patients' homes, when personal microenvironments differ from background, or when exposures that precipitate hospital admissions occur away from home. Future expansion of the monitoring program, at least for particulate pollution, should provide better background concentration estimates, allowing more powerful tests for regional differences in effects. New personal monitoring studies, designed to elucidate longitudinal relationships between background and personal exposures, might help to disentangle the effects of particulate pollutants and covarying gases (e.g., CO and NO2). Small panel studies have suggested that personal particulate exposures track background concentrations dosely in healthy older adults and children in The Netherlands (20,21), but not in older adults with COPD in Los Angeles (22). To our knowledge, no longitudinal studies of personal CO exposure have been reported. Response misclassification can result from errors in diagnosis or in medical record coding. Reviews suggest that 15-20% of assigned ICD codes are inaccurate (23,24).
Inaccuracies should increase random errors in specific disease counts and reduce the statistical significance of disease/pollution relationships, but should not introduce bias unless coding inaccuracies covary with pollution. Misdiagnoses are hard to evaluate, but are undoubtedly important, given the complexities of disease processes and the fuzzy boundaries between diagnoses. Wrong diagnoses or codes would likely shift patients to different specific disease counts within the same broad category, and thus should have relatively little effect on broad-category analyses. In any event, we have had only limited success in finding specific pollutant-disease relationships with mechanistic or public-health implications. Future studies focusing on precise diagnoses and accounting for other risk factors (e.g., additional diagnoses and particular demographic characteristics) might be more successful. VOLUME 108 NUMBER 5 May 2000 * Environmental Health Perspectives ,These results are for [1992][1993][1994] only. See footnote to Table 5 for explanation of regression procedure and coefficients. bAnnual mean daily admission count ± SD, for patients . 30 years of age. CO3 results not tabulated; none was significantly positive. *Year-round relationship significant, p < 0.05.
Limitations of our primary analytical to indusion or exclusion ofweather and other model indude the use of only one pollutant pollutant variables. If pollutants not in the at a time and possibly incomplete account-model affected admissions, the likely result ing for weather and temporal influences.
would be to overestimate effects of the mod-Because estimated CO effects were similar in eled pollutant and underestimate total effects various singleand multipollutant models of pollution (12). Thus, effects we associated that accounted for seasonal and weekly with CO might be at least partly due to cycles, more complete modeling of weather covarying gases (e.g., oxides of nitrogen) or to or temporal effects should not change the particulate substances. Similarly, incomplete conclusions concerning CO. By contrast, accounting for lagged effects would likely estimates of 03 effects were highly sensitive result in overestimated effects of very recent   Table 5 for explanation of regression procedure and coefficients. bAnnual mean daily admission count ± SD for patients 2 30 years of age. 0See text concerning asthma in patients < 30 years of age. *Year-round relationship significant, p < 0.05.  (14).

Conclusion
In general, our results from metropolitan Los Angeles appear consistent with reports from elsewhere (7)(8)(9)(10)(11)(12) that day-to-day increases in urban CO and/or PMIO and/or NO2 are associated with meaningful increases in cardiovascular illnesses. We found only a few equivocally positive relationships between cardiopulmonary morbidity and 03, in situations when other pollutants and heat stress increased along with 0 This is surprising, in light of severe 03 potlution in Los Angeles, obvious acute respiratory effects of 03 in animal and human exposure studies (2), and recent observations of 03-related hospital admissions in Toronto, Canada, where 03 levels are lower than in Los Angeles (12). 03 has been linked to mortality in Los Angeles (13), although PMIO might explain that association (25). On the other hand, a recent time-series study of asthma admissions in central and western Los Angeles (26) generally supports our findings, showing associations with PMIO but not with 03. The tendency of 03 concentrations to decrease indoors, where most people spend most of their time (27), might attenuate morbidity/03 relationships, but would not likely do so in Los Angeles more than in Toronto. In any event, our results suggest that the excess risk ofhospitalization in Los Angeles is greater on high-primary-pollution days than on high-0 days. The greatest risk of pollution-related hospital admissions apparently occurs on autumn/winter days with weak Santa Ana weather conditions, when air incursion from the desert approximately counterbalances that from the ocean, resulting in maximal atmospheric stagnation.
We could not distinguish clearly among CO-, NO2-and particle-associated effects. CO showed the strongest statistical relationships with most indices of morbidity even in the regions and seasons with the highest and widest ranging PM1O. NO2 tracked CO dosely enough that CO-associated effects might reasonably be attributed to NO2 and/or another oxide of nitrogen. Weaker statistical relationships of illness to PM10 might have resulted from less accurate exposure assessment even if PMIO were inherently more toxic (19). Too lirtle is known about the relationships between the ambient background and personal exposures to judge which pollutants are most subject to exposure misdassification. Even if PM10/morbidity associations were entirely explainable by CO/morbidity associations, some particulate species closely associated with CO might be the active agent(s). Alternatively, our findings might reflect separate effects of CO and some Environmental Health Perspectives * VOLUME 1081 NUMBER 5 May 2000 'Observed values minus values predicted by day of the week and cubic-spline smooth of longer term trends. 1From single-pollutant autoregressive Poisson models including temporal and weather effects. *Significant, p < 0.05. component of PM1O, as suggested by the Schwartz (9) Tucson, Arizona, study. One argument against CO effects per se is that typical ambient background CO concentrations are below normal bloodstream concentrations of metabolically produced CO (3). Even on most high-CO days in Los Angeles, inhaling the background concentration should reduce the blood's oxygen-carrying capacity by < 1%. However, CO concentrations near sources (e.g., heavy traffic) exceed background levels and may cause appreciable cardiovascular stress (9,10). If both are driven by atmospheric stagnation, these higher microenvironmental concentrations should track background levels. Thus, a low range of monitored background CO does not necessarily rule out an effect of CO on cardiovascular morbidity.
The observed association of all pulmonary diseases with PMIO or NO2 more than CO, and of all cardiovascular diseases with CO more than PM1O (Table 5), appears consistent with the well-known properties of CO as a circulatory toxicant without direct effects on the lungs, and of some particulate species as respiratory irritants. The association of occlusive strokes with all four tested pollutants appears consistent with the hypothesis of Seaton et al. (28) that urban pollution provokes alveolar inflammation, releasing mediators which increase blood coagulability. A previous finding of increased plasma viscosity during a primary pollution episode in Germany (2,9 also supports that hypothesis. By our data, we cannot test Seaton et al.'s (28) attribution of the inflammatory effect to ultrafine particles. We found possibly meaningful demographic differences in morbidity/pollution relationships, although none of them attained statistical significance. Persons . 65 years of age and diabetics showed somewhat increased cardiovascular disease effects as compared to others without those risk factors, but men did not appear to be more at risk than women of similar age. Persons younger than 30 years of age showed the largest pollution-related effects on asthma. Although air pollution health risks are believed to fall disproportionately on ethnic minorities (30), whites usually showed the largest pollution-related effects on cardiovascular disease. Effect sizes in blacks, the minority group generally at greatest risk for cardiovascular disease, were similar to those in whites, whereas effect sizes were generally smaller in Hispanics and undetectable in the other (predominantly Asian) ethnic category. Definitive interpretation would require evaluation of ethnic differences in exposure, susceptibility, and access to hospitals. On average, 03 exposures in the basin appear higher for whites than for blacks or Asian/Pacific Islanders (31). Differences in other exposures apparently have not been studied formally, but the high-primary-pollution regions 2 and 4 have high proportions of minority residents. Thus, smaller effects in some minorities (if real) probably are not explained by less exposure.
The relatively nonspecific pattern of diagnoses suggests that excess patients admitted to hospitals on high-pollution days in metropolitan Los Angeles are individuals with preexisting problems which make them highly vulnerable to any extra stresses on their oxygen delivery systems, including unfavorable changes in the air environment. If so, generalized efforts to preserve cardiopulmonary health should help to prevent (or at least to postpone) pollution-associated illnesses. Our findings suggest that control of primary pollutants is more important to public health than control of 0, which in any case depends on control ofprimary pollutants.