A Model for Forecasting Emergency Hospital Admissions: Effect of Environmental Variables.

Abstract

This study modeled patterns and trends in emergency hospital admissions at a hospital in Madrid, Spain. The purpose was to quantify qualitative associations that have been detected between such admissions and a number of environmental variables. The following data were used: unscheduled daily emergency hospital admissions, Madrid air pollution data, and meteorological data. Timeseries analysis was performed, with Box-Jenkins modeling. A multivariate model was constructed, incorporating the different causes of admissions and the respective environmental variables, Statistically significant associations were found between hospital admissions and other variables, indicating relationships with temperature, relative humidity, and mean daily tropospheric ozone concentrations. Whereas the effect of heat on admissions was short term, that of cold was in evidence from the second week. The association with ozone showed a seven-day lag and basically manifested itself as an influence on admissions for circulatory disease.

Introduction

Numerous studies have pointed to a link between air pollution and mortality, not only for cases of episodic pollution (Firket, 1936; Ministry of Health [United Kingdom], 1954; Shrenk, Heimann, & Clayton, 1949), but also for levels below those specified by the air quality guidelines of the World Health Organization (WHO) (Momas et al., 1993; Seaton, MacNee, Donaldson, & Godden, 1995; WHO, 1987). Some authors have suggested that air pollution plays a role akin to that of a trigger mechanism, accelerating the death of already gravely ill individuals, and that pollution must thus be regarded as being of relatively minor importance (Spix et al., 1993). In recent years, however, a series of epidemiological studies have set out to establish the relationship between air pollution and emergency admissions (Schouten, Vonk, & Graaf, 1996; Schwartz, 1995; Vigotti, Rossi, Bisanti, Zanobetti, & Schwartz, 1996). Within the confines of the limitations to which such studies are said to be subject, the results show air pollut ion taking on a different dimension and an altogether greater transcendence.

Once the qualitative aspect of an association has been detected, the logical next step is to quantify the association. This step entails obtaining models that can diagnose and forecast all-cause emergency admissions, taking into account the possible effects of diverse atmospheric variables. The initial premise would be that, while not all admissions can be assumed to be due to such causes, there inevitably will be a proportion of causes that can be explained by reference to environmental factors.

With respect to air pollution, Madrid, Spain, is not very different from comparable European cities, Nevertheless, its extreme climate and the characteristics of its population (13.6 percent are over 65 years of age) lend special interest to the study of the relationship between mortality and environmental variables (Alberdi, Diaz, Montero, & Miron 1998; Montero, Miron Diaz, & Alberdi, 1997), particularly photochemical pollutants like ozone, which has registered an upward trend in recent years (Diaz, Alberdi, Montero, & Miron 1998).

This study modeled emergency-admission patterns and trends in a Madrid hospital to quantify the effect of these environmental variables. Timely prediction of the precise variables that form a model would enable health authorities to use the model to detect variations in the number of hospital admissions well in advance. Interventions could thus be programmed in such a way as to ensure optimal management and allocation of hospital health care resources.

Methods

The series of daily emergency hospital admissions examined for this study covered the period of January 1, 1994, to September 30, 1996 (1,004 days). This data series, supplied by the Gregorio Maranon University Teaching Hospital, included all unscheduled admissions excluding traumas and births. Causes of admission were defined according to the International Classification of Diseases, 9th Revision (ICD-9-MC) (Commission on Professional and Hospital Activities, 1978). Thus, admissions were grouped into the following categories: total organic-disease admissions (ICD-9: 1-799), circulatory-disease admissions (ICD-9: 390-459), and admissions due to diseases of the respiratory system (ICD-9: 460-519). Data on environmental variables took the form of daily mean air pollution values, in micrograms per cubic meter ([micro]g/[m.sup.3]), for sulfur dioxide ([SO.sub.2]), total suspended particles (TSP), nitrogen oxides ([NO.sub.x]), nitrogen dioxide ([NO.sub.2]), and ozone ([O.sub.3]), as furnished by the 24 monitoring stations making up Madrid's Municipal Automatic Air Pollution Monitoring Grid. Daily mean readings were recorded at all 24 of the grid's monitoring stations for all pollutants except ozone, which was monitored at only five stations. For ozone, the 24-hour mean was used, as that value had been shown by other studies to register the greatest association with mortality in Madrid (Diaz, Alberdi, Montero, & Miron, 1998). The following meteorological variables also were considered: daily maximum temperature (TMax), daily minimum temperature (TMin), and daily mean temperature (TMean), with TMean obtained on the basis of readings ([degrees]C) taken at 7 a.m., 1 p.m., 6 p.m., and 12 a.m. Relative humidity (RH) at 7 am, also was taken into account. Because of its convenient location in the vicinity of the Gregorio Maranon University Teaching Hospital, the Madrid Retiro Observatory was chosen as the meteorological observatory of reference.

Following analysis of the components in the data series--that is, hospital-admission variables considered as dependent variables and environmental variables considered as independent variables--functional relationships were established between them via scatter-plot diagrams, with the respective correlation coefficients being analyzed to determine which afforded the best fit. Using this diagram, any emergency-hospital-admission/environmental-factor pattern could be analyzed. In the case of temperature, for example, a V-shaped distribution was observed, indicating the existence of admission peaks related to low and high temperatures, a pattern that rendered it necessary to divide the data series into two segments with reference to the comfort temperature, or the temperature associated with minimum admissions (Montero, Miron Diaz & Alberdi, 1997). Cold-temperature (TCold) and hot-temperature (THot) series were therefore created, defined as follows:

TCold = TComfort - TA if TA [less than] TComfort,

and

where TA = ambient temperature.

THot = TA - Tcomfort if TA [greater than] Tcomfort,

A similar pattern was in evidence for ozone. Consequently, high-ozone ([O.sub.3]-high) and low-ozone ([O.sub.3]-low) series were defined with reference to the ozone concentration associated with minimum admissions (Diaz, Alberdi, Montero, & Mirn 1998). The derivation of these series is further discussed in the Results section of this paper.

Series-specific analyses were performed for the whole year and for both winter (December to March) and summer (June to September). For analysis of the deterministic components in the data series, the pertinent basic descriptive statistics were ascertained. Trend and periodicities were analyzed with the relevant frequency spectra yielded by the Fast Fourier Transform method (Anderson, 1971).

Univariate AutoRegressive Integrated Moving Average (ARIMA) modeling (Box & Jenkins, 1976) was used to ascertain the nondeterministic components of the series-- that is, the autoregressive (AR) part and the moving average (MA). Using the Box-Ljung Portmanteau test, the authors selected these models where their partial autocorrelation functions (PACFs) and simple autocorrelation functions (ACFs) indicated white-noise structure.

Box-Jenkins prewhitening was performed to eliminate analogous periodicities and autocorrelations, as between the mortality and temperature series (Box & Jenkins, 1976). Next, the cross-correlation functions (CCFs) between their sets of residuals were calculated. This step established the lags at which significant relationships between the variables occurred. Lagged variables were created for all of the environmental variables listed above.

Once the hospital-admission/environmental-variable association had been established, ARIMA models of the hospital admissions variables were constructed, with the environmental variables included as exogenous inputs, in order to eliminate the effects of possible colinearities existing between environmental variables (Box & Jenkins, 1976). The authors evaluated goodness of fit by running the Box-Lujng Portmanteau test on the PACE and ACF residuals.

In addition to the environmental variables considered, other variables, such as day of the week and annual and six-month circular functions, were introduced to act as controls for possible confounding factors. Sensitivity analyses were performed on all models to test stability Statistical analysis of data was performed with the SPSS computer software package for Windows (Release 6.1).

Results

Figure 1 depicts the time trend in emergency hospital admissions for the period 1994-1996. A seasonal pattern is in evidence, marked by a maximum peak in winter and peaks of lower intensity in summer. As appears in the graphs in Figure 2 through Figure 5, air pollutants also showed a clear seasonal pattern, registering maximum values in winter. The single exception was ozone; because this pollutant is secondary in nature and sunlight is an essential prerequisite for its formation, ozone reached its peak during periods of maximum sunshine (Figure 6).

Table 1 provides descriptive statistics for the variables, including causes of emergency hospital admissions (total organic disease admissions as well as cause-specific admissions) and the various pollution and meteorological variables employed. Also shown are the periodicities and trends detected by spectral analysis. The daily mean for organic-disease admissions during the study period was 60, with a maximum of 108 and a minimum of 23.

By way of example, Figure 7 shows the spectrum for organic-disease hospital admissions with the corresponding 99 percent significance level. Periodicities that centered on the low frequencies were in evidence, pointing to an annual cycle. This pattern also was found for circulatory- and respiratory-disease admissions. At high frequencies, periodicities were likewise observed at seven and 3.5 days for organic-and circulatory-disease admissions.

For air pollutants, the frequency spectrum showed an annual seasonality for all pollutants except [NO.sub.2], which registered periodicities only at high frequencies (seven and 3.5 days). With respect to TSP, mention should be made of a slight downward trend that was detected over the spectrum.

The meteorological variables--TMax, Tmin, and relative humidity--registered an annual cycle. Relative humidity also showed a three-day periodicity.

Scatter-plot diagrams for air pollutants revealed that nitrogen oxides and particulates had a linear relationship with hospital admissions; Figure 8 gives the diagram for nitrogen dioxide as an example. In contrast, the relationship of [SO.sub.2] to hospital admissions was logarithmic. The relationships existed across the board for any concentration of these pollutants--that is, there were no threshold values below which the associations were not in evidence. Ozone had a quadratic relationship with hospital admissions; as shown in Figure 9, minimum admissions occurred when ozone was at a concentration of 45 [micro]g/[m.sup.3], and that value then served as the basis for defining high and low ozone. Among the temperature variables that were analyzed, Tmax was the most highly correlated with hospital admissions; minimum admissions were registered at 33[degrees]C.

When the authors analyzed cross-correlation functions between total and cause-specific admissions on the one hand and the different pollutants on the other--and controlled for the effect of temperature--the most significant result was that in the winter, [SO.sub.2] and TSP were associated with total hospital admissions at a lag of zero days. In the summer, however, the association held for particulates alone and at a lag of five days. When admissions were stratified by cause, TSP registered a yearlong association, at lags of zero and seven days, for circulatory-disease admissions; no significant association was observed for respiratory-disease admissions, With respect to daily mean ozone, a statistically significant association was found for all admission causes analyzed, with an immediate effect being registered for organic-disease admissions at a lag of zero. The cross-correlation functions for [NO.sub.x] registered a pattern similar to that recorded for [SO.sub.2] (in essence, a lag of zero).

ARIMA models, supplied with the necessary variables to control for possible confounding factors such as seasonality and day of the week, were likewise constructed to ascertain the influence of environmental variables. Table 2 shows the significance of the associations registered, the lags at which these occurred, and the values of the estimators for the different ARIMA models having exogenous variables.

These findings showed year-round organic-cause admissions to be linked to [SO.sub.2] and THot at a lag of zero days; high ozone at a lag of seven days; and TCold at a lag of 10 days. For the winter period, the association held solely for [SO.sub.2] and TCold; the summertime pattern proved identical to the year-round pattern.

Separate analysis of individual pollutants in relation to cause-specific admissions showed circulatory-disease admissions to have a short-term relationship with [SO.sub.2], high ozone, and relative humidity For respiratory-disease admissions, a relationship with Tcold registered in the long term, and a statistically significant association with ozone registered for values above 45 [micro]g/[m.sup.3] at a lag of zero in the summer. By taking estimator values for the individual pollutants into account, the authors then could determine the respective weights of the pollutants in any increase in emergency hospital admissions. [SO.sub.2] accounted for 2.3 percent of the increase when atmospheric concentrations rose 25 [micro]g/[m.sup.3] above the mean, the proportion climbing to 3.2 percent in the wintertime. The increase attributable to ozone was 18 percent when concentrations rose by 25 [micro]g/[m.sup.3] above 45 [micro]g/[m.sup.3]. With respect to temperature, there was an increase of 1.7 percent in daily all -cause admissions for every degree over 33[degrees]C; this effect was immediate.

Discussion and Conclusion

Numerous studies have reported the existence of statistically significant associations between air pollution and morbidity and mortality (Alberdi, Diaz, Montero & Miron, 1998; Firket, 1936; Ministry of Health United Kingdom, 1954; Montero, Miron, Diaz, & Alberdi, 1997; Seaton, MacNee, Donaldson, & Godden, 1995; Schwartz, 1995; Shrenk, Heimann, & Clayton, 1949). In the last few years, analysis has also turned to the relationship between hospital admissions and pollution, focusing fundamentally on admissions attributable to the different respiratory diseases (Bates & Sizto, 1983; Dab et al., 1996). Many such studies calculate the relative risk for admissions in exposed versus unexposed individuals. In many cases, total emergency admissions are analyzed, as in the study undertaken by Bates and Sizto (1983), which described an association between admissions and [SO.sub.2] levels, or in a Los Angeles-based study that linked incidence of overall emergency admissions to wintertime particulate levels (Goldsmith, Grif fith, Detels, Beeser, & Neumann, 1983). The most important studies have, however, centered on respiratory disease, implicating [O.sub.3], [SO.sub.2], and TSP as the causes underlying increases in the number of emergency admissions due to asthma and chronic obstructive pulmonary disease (Bates, Baker-Anderson, & Sitzo, 1990; Dab et al., 1996; Ponce de Leon, Anderson, Bland, Strachan, & Bower, 1996; Schwartz, 1994; Sunyer et al., 1993; Vigotti, Rossi, Bisanti, Zanobetti, & Schwartz, 1996). In some cases, an association has also been established between pollutant levels and cardiovascular admissions; an example is the link with [S0.sub.2] found by Sweeney (1982).

The results yielded by this study proved very similar to those of studies that have linked air pollution and mortality in Madrid, even though the study periods and populations were different. The functional relationships detected, the response times, and even the estimator values were very similar, thus underscoring the fact that air pollution, rather than acting as a factor that merely triggers deaths destined to occur within a matter of days, might indeed be an initial link in such a chain of events.

If these results are deemed of interest, it is perhaps because the principal contribution that this study sets out to make is not merely to establish a pollutant-hospital admission relationship. The main thrust is the attempt to model the time trend in these admissions, based not only on the history of the data series (the ARIMA part of the models) but also on various statistically significant environmental variables. The latter basis endows univariate models with the possibility of forecasting episodic situations that are linked to environmental variables and that no series-related history could otherwise reproduce unless a comparable event had previously taken place. Hence, the importance of this study does not lie simply in the establishment of a relationship between pollution and hospital admissions. Rather, it lies in the quantification of this relationship; by establishing models capable of diagnosing and forecasting the trend over time with a mean error of 15 percent, this methodology offers a useful tool for hospital management purposes (Figure 10).

Moreover, a hospital-admission-forecasting model of this nature opens the way to epidemiological surveillance of pollution, enabling preventive measures to be taken and specific purpose-designed actions to be implemented in the case of health care risks associated with pollution and episodes of extreme temperatures. Currently, the model is being implemented at the Hospital Gregorio Maranon de Madrid.

Acknowledgements: This study was funded by Health Sciences Research Project Grant No. 08.7/0007/1999 2 from the Madrid Regional Education and Culture Authority.

Corresponding Author: Dr. Julio Diaz Jimenez, Centro Universitario de Salud Publica, C/ General Oraa 39, E-28006 Madrid, Spain. E-mail: [less than][e-mail][greater than].

REFERENCES

Alberdi, J.C., Diaz, J., Montero, J.C., & Miron, I.J. (1998). Daily mortality in Madrid community 1986-1992: Relationship with meteorological variables. European journal of Epidemiology, 14,571-578.

Anderson, T.W (1971). The statistical analysis of time series. New York: John Wiley.

Bates, D.V., & Sizto, R. (1983). Relationship between air pollutant levels and hospital admissions in Southern Ontario. Canadian Journal of Public Health, 74, 117-122.

Bates, D.V., Baker-Anderson, M., & Sitzo, R. (1990). Asthma attack periodicity: A study of hospital emergency visits in Vancouver. Environmental Research, 51(1), 51-70.

Box, G.E.P., & Jenkins, G.M. (1976). Time series analysis Forecasting and control. San Francisco: Holden-Day.

Commission on Professional and Hospital Activities (1978). The international classification of diseases (9th Rev. Clinical Modification). Ann Arbor, MI: Author.

Dab, W, Medina, S., Quenel, P, Le Moullec, Y., Le Tertre, A., Thelot, B., Monteil, C., Lameloise, P., Pirard, P., Momas, I., Ferry R:, & Festy, B. (1996). Short term respiratory health effects of ambient air pollution: Results of the APHEA project in Paris Journal of Epidemiology of Community Health, 50(Suppl 1), 42-46.

Diaz, J., Alberdi, J.C., Montero, J.C., & Miron, I.j. (1998). Efectos a corto plazo de la contaminacion atmosferica sobre la mortalidad diaria en Madrid (Espana) de 1990 a 1992: Un analisis de series temporales. Informacion Tecnologica, 9(1), 33-42.

Firket, J. (1936). Fog along the Meuse Valley. Transactions of t Faraday Society, 32, 1992-1997.

Goldsmith, J.R., Griffith, H.L., Detels, R., Beeser, S., & Neumann, L. (1983). Emergency room admissions, meteorologic variables and air pollutants: A path analysis. American Journal of Epidemiology 5, 759-788.

Ministry of Health (United Kingdom). (1954). Mortality and morbidity during the London fog of December 1952 (Reports on Public Health and Medical Subjects, 95). London: Her Majesty's Station Office.

Momas, I., Pirard, P., Quenel, P., Medina, S., Le Moullec, Y., Dab, W., Ferry, R., & Festy, B. (1993). Pollution atmospherique urbaine et mortalite: Une synthese des etudes epidemiologiques publiees entre 1980 et 1991. Revue Epidemiologique et Sante Publique, 41(1), 30-43.

Montero, J.C., Miron, I.J., Diaz, J., & Alberdi, J.C. (1997). Influencia de variables atmosfericas sobre la mortalidad por enfermedades respiratorias y cardiovasculates en los mayores de 65 anos de la Comunidad de Madrid. Gaceta Sanitaria, 11,164-170.

Ponce de Leon, A., Anderson, H.R., Bland, J.M., Strachan, D.P., & Bower, J. (1996). Effects of air pollution on daily hospital admissions for .respiratory disease in London between 1987-1988 and 1991-1992. Journal of Epidemiology and Community Health, 33(Suppl. 1), 63-70.

Schouten, J.P., Vonk, J.M., & Graaf, A. (1996). Short term effects of air pollution on emergency hospital admissions for respiratory disease: Results of the APHEA project in two major cities in the Netherlands, 1997-89. Journal of Epidemiology of Community Health, 50(Suppl 1), 22-29.

Schwartz, J. (1994). Air pollution and hospital admissions for the elderly in Birmingham, Alabama. American Journal of Epidemiology, 139, 589-598.

Schwartz, J. (1995). Air pollution and hospital admissions for respiratory disease. Epidemiology Resources, 7(1), 20-28.

Shrenk, H.H., Heimann, H., & Clayton, G.D. (1949), Air pollution in Donora, PA: Epidemiology of the unusual smog episode of October 1948. Preliminary report. Public health bulletin, 306. Washington, DC: U.S. Public Health Service.

Seaton, A., MacNee, W., Donaldson, K., & Godden, D. (1995). Particulate air pollution and acute health effects. Lancet, 345, 176-178.

Spix, C., Heinrich, J., Dockery, D., Schwartz, J., Volsch G., Schwinkowski, K., Collen, C., & Wichmannn, H.E. (1993). Air pollution and mortality in Erfurt, East Germany, 1980-1989. Environmental Health Perspectives, 5, 518-525.

Sunyer, J., Saez, M., Murillo, C., Castellsague, J., Martinez; J., & Anto, J.M. (1993). Air pollution and emergency room admissions for chronic obstructive pulmonary diseases. American Journal of Epidemiology, 137, 701-5.

Sweeney, J.C. (1982). Air pollution and morbidity in Dublin. Irish Geography, 15(1), 1-10.

Vigotti, M.A., Rossi, G., Bisanti, L., Zanobetti, A., & Schwartz, J. (1996). Short term effects of urban air pollution on respiratory health in Milan, Italy, 1980-1989. Journal of Epidemiology and Community Health, 50(Suppl. 1), 71-75.

World Health Organization. (1987). Air quality guidelines for Europe. (W.H.O. Regional Publications, European Series No 23). Copenhagen: W.H.O. Regional Office for Europe.

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

TABLE 1

Statistics for Hospital-Admission and Environmental Variables


Variable                                Mean  Standard Deviation

Organic disease (daily admissions)      59.9                12.7
Circulatory disease (daily admissions)   9.8                 3.5
Respiratory disease (daily admissions)   7.6                 3.7
TMax ([degrees]C)                       20.9                 8.6
Thin ([degrees]C)                       10.7                 6.5
[SO.sub.2] ([micro]g/[m.sup.3])         26.1                18.5
[O.sub.3] ([micro]g/[m.sup.3])          26.2                14.3
TSP ([micro]g/[m.sup.3])                39.1                13.8
[NO.sub.2] ([micro]g/[m.sup.3])         65.0                17.4
[NO.sub.x] ([micro]g/[m.sup.3])          160                80.8
Relative humidity (%)                   64.6                17.8



Variable                                Maximum  Minimum

Organic disease (daily admissions)          108    23
Circulatory disease (daily admissions)       25     2
Respiratory disease (daily admissions)       29     0
TMax ([degrees]C)                          40.9     0.4
Thin ([degrees]C)                          25.8    -7.6
[SO.sub.2] ([micro]g/[m.sup.3])             128     6
[O.sub.3] ([micro]g/[m.sup.3])               71     2
TSP ([micro]g/[m.sup.3])                    147    18
[NO.sub.2] ([micro]g/[m.sup.3])             143    26
[NO.sub.x] ([micro]g/[m.sup.3])             596    46
Relative humidity (%)                        99    28



Variable                                    Trend

Organic disease (daily admissions)            No
Circulatory disease (daily admissions)        No
Respiratory disease (daily admissions)        No
TMax ([degrees]C)                             No
Thin ([degrees]C)                             No
[SO.sub.2] ([micro]g/[m.sup.3])               No
[O.sub.3] ([micro]g/[m.sup.3])           Yes (upward)
TSP ([micro]g/[m.sup.3])                Yes (downward)
[NO.sub.2] ([micro]g/[m.sup.3])               No
[NO.sub.x] ([micro]g/[m.sup.3])               No
Relative humidity (%)                         No



Variable                                Periodicity

Organic disease (daily admissions)      Annual, 7 days, 3.5 days
Circulatory disease (daily admissions)  Annual, 7 days, 3.5 days
Respiratory disease (daily admissions)  Annual, 3.5 days
TMax ([degrees]C)                       Annual
Thin ([degrees]C)                       Annual
[SO.sub.2] ([micro]g/[m.sup.3])         Annual, 7 days, 3.5 days
[O.sub.3] ([micro]g/[m.sup.3])          Annual, 7 days (95%), 3.5 days
TSP ([micro]g/[m.sup.3])                Annual, 7 days, 3.5 days
[NO.sub.2] ([micro]g/[m.sup.3])         7 days, 3.5 days
[NO.sub.x] ([micro]g/[m.sup.3])         Annual, 7 days, 3.5 days
Relative humidity (%)                   Annual, 3 days
TABLE 2

Associations Between Hospital Admissions and Exogenous
Variables--Significant Coefficients and Lags (in Parentheses) for Nine
ARIMA Models


                   Organic Disease

Variable           Year             Winter          Summer

Log [SO.sub.2]     2.005 [*] (0)    2.72 (0)        --
TSP                --               --              --
[O.sub.3] (high)   0.43 [***] (7)   --              0.30 [*] (7)
[0.sub.3] (low)    --               --              --
[NO.sub.2]         --               --              --
[NO.sub.X]         --               --              --
THot               0.98 [**] (0)    --              1.17 [**] (0)
TCold              0.32 [***](10)   0.50 [**] (10)  0.42 [**] (3)
Relative humidity  --               --              --




                   Circulatory Disease

Variable           Year                 Winter          Summer

Log [SO.sub.2]     0.63 [*] (1)         --              --
TSP                --                   --              --
[O.sub.3] (high)   0.09 [*] (6)         --              0.10 [*] (6)
[0.sub.3] (low)    --                   --              --
[NO.sub.2]         --                   --              --
[NO.sub.X]         --                   --              --
THot               --                   --              --
TCold              --                   0.15 [***] (7)  0.13 [**] (3)
Relative humidity  0.03 [***] (7)       --              0.03 [*] (9)




                   Respiratory Disease

Variable           Year                 Winter

Log [SO.sub.2]     --                   --
TSP                --                   --
[O.sub.3] (high)   --                   --
[0.sub.3] (low)    --                   --
[NO.sub.2]         --                   --
[NO.sub.X]         --                   --
THot               --                   --
TCold              0.21 [***] (8,12)    0.32 [*] (10,12)
Relative humidity  --                   0.03 [*] (5)






Variable           Summer

Log [SO.sub.2]     --
TSP                --
[O.sub.3] (high)   0.07 [*] (0)
[0.sub.3] (low)    --
[NO.sub.2]         --
[NO.sub.X]         --
THot               --
TCold              0.19 [*] (11,3)
Relative humidity  --



[O.sub.3]-high = [O.sub.3] values higher than 45 [micro]g/[m.sup.3]

[0.sub.3]-low = [0.sub.3] values lower than 45 [micro]g/[m.sup.3]

THot = TMax values higher than 33[degree]C.

TCold = TMax values lower than 33[degree]C.

(*)p [less than] .05.

(**)p [less than] .01.

(***)p [less than] .001.