Association of School Absence with Air Pollution in Areas around Arterial Roads

Association of school absence with air pollution from suspended particulate matter (SPM) and nitrogen dioxide was analyzed in areas around arterial roads for five years, from 1993 to 1997. The prevalence of absence was calculated using the data for school absence in two schools around arterial roads, one of which is near a crossing (Sch.A), and the other is adjacent to an arterial road (Sch.B). Although the results from annual correlation analyses did not indicate common findings for five years or in the two schools, the prevalence of absence correlated positively with SPM, nitrogen dioxide, or relative humidity, and negatively with atmospheric temperature. As the results from multiple regression analyses, atmospheric temperature in Sch.A was adopted as the optimum explanatory variable, whereas SPM and relative humidity were considered in Sch.B. Odds ratios for the prevalence of absence to SPM were elevated and were significant in Sch.B, when using a quintile method. The other odds ratios for the air pollutants were not significant, but exceeded 1. When the data were classified by day of the week, significant associations of the prevalence of absence were observed with atmospheric temperature in Sch.A and with SPM in Sch.B. The slope of the regressive equations by day of the week became steeper with the day in Sch.B. SPM was weakly associated with the prevalence of absence in Sch.A and was closely associated in Sch.B according to the optimum variables selected from the multiple regression analyses by day of the week.


INTRODUCTION
Exposure to high levels of particulate matter and nitrogen dioxide in ambient air is known to cause acute adverse effects on health of schoolchildren 1-5).School absence is one of the adverse effects, and it was reported that a positive relationship was found between daily school absence and air pollutants , particularly particulate matter, in longitudinal studies 7-11).However, school absence is likely caused more frequently by reasons other than air pollution in Japan.The most common reason for school absence is -infectious diseases such as influenza 12).Since the endemic season of such diseases is in December through March in Japan, it is very difficult to detect the effects of air pollution on school absence during this period.Therefore, the study design must be maturely considered to analyze an association of school absence with air pollution.
The following items were considered for study areas, study schools, study periods, air pollutants.statistical models and others for the longitudinal study design.1) High levels of air pollutants must be monitored for the suitable study area.
2) There are monitoring stations near the schools.3) The daily data for school absence can be obtained for the five full years of the decided preservation period of the data.4) The time of exposure to air pollutants and months to examine the data are limited to raise sensitivity.5) Some statistical models are used to analyze relationships.

Study Area
The area around arterial roads was considered for the study on health effects of particulate matter.The Yamatocho cross-Received August 14, 1999 ;accepted February 21 , 2000.Address for correspondence : Tokyo Metropolitan Research Laboratory of Public Health 3-24-1 Hyakunincho , Shinjuku-ku, Tokyo, 169-0073 Japan.lence of absence and the concentrations of particulate matter and nitrogen dioxide.The number of absences rapidly increased for those four months because of epidemic diseases such as influenza, or other reasons such as household circumstances, in every year.The absence data of the days following holidays were not also used, for the number of absences on those days is higher than on other days.
As the environmental factors, the following four factors were selected.Data for daily mean concentrations of SPM and nitrogen dioxide were obtained in the two stations for the five years from 1993 to 1997.Similarly data for meteorological factors such as atmospheric temperature and relative humidity were obtained in the station of Site B. Twelve-hour means from eight A.M. to eight P.M. were calculated for each of these data in consideration of outdoor behavior of children.

Statistical Analysis
Statistical analyses were separately conducted for the two schools/stations.The data of school absences in the day corresponded to air pollution or meteorological factors of the previous day when there is no time lag of days.Linear regression models were used to analyze the prevalence data, whereas explanatory variables were one or some of SPM, nitrogen dioxide, atmospheric temperature, and relative humidity.However, the two concentrations in both sites could not be directly compared as exposure indices.Sch.A is not adjacent to Site A, while Sch.B is adjacent to Site B. In this analysis, the statistical analyses were conducted by year, and time lag of days was examined up to three days, using correlation analyses.Logistic regression and Poisson regression models were also used for reference.Auto-correlation analyses between the school absence data were examined for a week, since there is probability of dependence on a day of the week for school absence.In addition, odds ratios from the multiple logistic model were examined by using the quintile method.The first quintile by frequency from lower concentrations of air pollutants was applied to the control group through the total study period.
In order to study the effects of a day of the week on the relationship between the prevalence of absence and air pollutants or meteorological factors, regression analyses were conducted by day of the week through the total study period.When the explanatory variables were plural, selection of the optimum variables was decided by AIC (Akaike's Information Criterion) statistics.The prevalence of absence was significantly correlated with SPM, nitrogen dioxide and atmospheric temperature in Sch.A (Table 2).SPM, nitrogen dioxide, and relative humidity were correlated positively, and atmospheric temperature was negatively in some significant cases.Atmospheric temperature showed significant correlation in three of the five years in Sch.A.However, results from correlation analyses by year could not find a certain tendency in the two schools for the five years.On the other hand, there was almost no difference among the three models for the correlation analysis.Therefore, the linear regression model was mainly selected in the following analyses.

A summary of the data was presented in
Besides, correlation coefficients between the two air pollutants were from 0.78 and 0.22 for the five years in the two sites.Though these correlations were all significant (p<0.001), the coefficients varied widely by year.Next, multiple regression analysis was conducted by year and selection of the optimum variables was performed by AIC sta-tistics (Table 3).Although a certain tendency could not be obtained, as the correlation analyses, atmospheric temperature in Sch.A and SPM and relative humidity in Sch.B were selected as the optimum variables in more than three of the five years and during the total period.In addition, multiple correlation analyses showed significant result in the four years from 1993 to 1996 in Sch.A, in the two years of 1995 and 1996 in Sch.B, and in the total period.Correlation coefficients calculated with time lag of days up to three days did not exceed the coefficients when there was no time lag of days.Furthermore, the auto-correlation coefficients by year for a week were not significant in all cases when the data for the school absence on air pollutants were found on Sunday.According to the results from the correlation analyses by day of the week (Table 5), SPM was significantly correlated with Thursday in Sch.A and in most days of the week in Sch.B, but nitrogen dioxide was not on any day of the week.As for the meteorological factors , atmospheric temperature, which was correlated negatively , was more closely related to the prevalence of absence than relative humidity.
When the regression equations in the prevalence of absence to the SPM concentration by day of the week were examined in Sch.B (Figure 3), the slopes of the equations became steeper with the day of the week.Results from multiple regression analyses by day of the week indicated a close relationship between the prevalence of absence and the SPM concentration particularly, in Sch.B (Table 6).

DISCUSSION
Analyzing the tendency of school absence by grade may be interesting in a survey with a large number of children, as the Figure 2. Odds ratios and the confidence intervals for the prevalence of absence to SPM and NO2 in using the quintile method (*:p<0.05).
Table1for the fiveyear study period, but the data for air pollutants were not included for the duration of summer vacation (from July 21*to August 31*) or holidays.Numbers of schoolchildren were between154 and 190 in Sch.A and between 380 and 458 in

Table 1 .
Summary of data for the prevalence of absence in Sch.A and Sch.B, and the environmental factors in Site A and Site B.

Table 2 .
Annual correlation analyses between the prevalence of absence and the environmental factors.

Table 3 .
Annual multiple regression analyses for the prevalence of absence with environmental factors.

Table 4 .
Mean values for the prevalence of absence and the environmental factors by day of the week.

Table 5 .
Correlation analyses by day of the week between the prevalence of absence and the environmental factors.

Table 6 .
Multiple regression analyses by day of the week for the prevalence of absence with the environmental factors.