Journal of Epidemiology and Community Health 1996;50(Supp 1):S59-S62 S59 J Epidemiol Community Health: first published as 10.1136/jech.50.Suppl_1.s59 on 1 April 1996. Downloaded from Asthma and ambient air pollution in Helsinki

Antti Ponka, Mikko Virtanen

Abstract increase the frequency of asthma attacks are Objective - To study whether ambient air not known with certainty. Nor is there adequate levels of sulphur dioxide (SO2), nitrogen information on the synergistic effects between dioxide (NO2), total suspended par- various pollutants and other factors such as ticulates (TSP), and ozone (03) affect the cold, the relative importance of the different number ofhospital admissions for asthma. causative factors, or the differences in sensi- Design - The associations between the tivity between the different age groups. The daily number of admissions and air pol- results of many studies have been impaired by lutants were analysed with Poisson re- small size of samples, confounding socio- gression, taking into account potential economic factors, insufficiently documented confounding factors by using the stand- levels of pollutants, and the biases in collecting ardised protocol of the APHEA project. information about illness. Setting - Helsinki, Finland, 1987-89. During recent years longitudinal studies have Patients and measurements - Patients ad- been used in these investigations because the mitted to hospitals through emergency populations studied serve as their own controls, rooms because of asthma (n = 2421). The eliminating a potential source of bias in cross daily mean concentration was 13-25 pg/m3 sectional studies.`811-14 Our study investigated for SO2, 33-41 pg/m' for NO2, 19-41 pg/m3 in daily time series the relationship of daily for 03, and 58-109 pg/m3 for TSP during hospital admissions to ambient air SO2, NO2, the different seasons. Values are means for 0,, and TSP concentrations, temperature, and various stations. The daily mean tem- relative humidity in Helsinki as part of a Euro- perature during the three year period was pean multicentre project. In particular, we in- + 5 4°C (range -37-27°C). vestigated whether the numbers of hospital Main results - Positive associations with admissions for asthma in different age groups admissions were observed for O3 levels in increase on days with high levels of pollutants. all children under 14 years, and for SO2 The methodology used was uniform for all levels in 15-64 year olds and among those A centres participating in the EEC APHEA older than 64. Significant associations project. 14 were also seen between admissions for di- gestive tract diseases (the control) and O3 levels. This suggests that the modelling, Methods which proved to be problematic, was un- or AIR POLLUTANTS AND METEOROLOGICAL satisfactory, it may be a statistical co- VARIABLES incidence. A rise in temperature was The pollution parameters used were 24 hour associated with a low number of ad- means for SO2, NO2, and TPS, an 8 hour mean http://jech.bmj.com/ missions for asthma among 0-14 year olds for 03 and 1 hour maximum values for SO2, and among 15-64 year olds, whereas NO2, and O3. The means of different stations humidity did not have a significant effect were used for all variables except 03, which on the number of admissions. was measured at one station only. Conclusions - It is possible than even low The city of Helsinki covers a relatively small level pollution may increase hospital ad- area, 185 km2. The main sources of air pol- missions for asthma. However, definitive lutants are energy production by coal fired and on September 30, 2021 by guest. Protected copyright. conclusions will be justified only after oil fired power plants, road traffic and, to a meta-analysis comprising several stud- small extent, industrialisation. Air pollution ies. The methodology was seen to have a measurements are conducted in Helsinki by strong effect on the results and stand- district municipal authorities. During the study ardised methods, possibly differing from period 1987-89, SO2 was measured hourly those for the study concerning mortality, by coulometric instruments at four automatic are needed to investigate the association monitoring stations, NO2 by chemilumin- between morbidity and ambient air pol- escence at two stations, and O3 by ultraviolet lutants. absorption at one station. The instruments are calibrated on a monthly basis for S02 and NO2, (J Epidemiol Comm Health 1996;50:(Supp 1)S59-S62) and bimonthly for O3, and as the results are followed continuously, any disturbance in their Helsinki City of the Environment, functioning is noticed immediately. Total sus- Sturenkatu 25, Several studies have shown an association be- pended particulates are collected by high vol- 00510 Helsinki tween ambient air pollutants and hospital ad- ume samplers at six stations, at four of them A Ponka missions or emergency room visits of patients every second day and at two every third day. M Virtanen with chronic obstructive pulmonary diseases At one weather station, measurements of tem- Correspondence to: including asthma, or asthma alone. The con- Dr A Ponka. perature and relative humidity are recorded centrations at which ambient air pollutants hourly. The total number of stations is eight; S60 Pdnka, Virtanen J Epidemiol Community Health: first published as 10.1136/jech.50.Suppl_1.s59 on 1 April 1996. Downloaded from four of them are situated in the centre of the Many attempts were made to reduce the town and four in the suburbs, at heights of number of uninterpretable results. Firstly, uni- 2-10 m. The sites of these stations have been form selection criteria for transformations were selected to represent the actual exposure of the selected for different age groups, and the effects population to ambient air pollutants. ofthe sinusoidals and differences between years were studied. The time plots showed that the seasonal patterns changed from year to year. THE MORBIDITY DATA Also, it seemed clear from both the plots and The data on morbidity comprise details on all the exploratory STL (Seasonal and trend de- patients admitted to hospitals for asthma via composition using Loess) 6 that the weekly pat- emergency wards in Helsinki in 1987-89. The terns change with time. These patterns were age groups 0-14 years, 15-65 years, and >64 modelled in many different ways, but no clear years were analysed separately. The diagnosis answer was found. of asthma is based on the Ninth International The number of sinusoidal terms was ori- Classification of Diseases (ICD-9 code 493) is- ginally restricted to six, but up to 19 were tried. sued by the World Health Organization. The Different sinusoidal terms for each year were number of admissions for diseases of the di- also tried, without success. Better adjustments gestive system (ICD 9 codes 520-579) was were found by using different trends and espe- studied similarly and used as a control. The cially different weekly dummies for each year. data concerning hospital admissions for ex- Although the STL decomposition suggested acerbations of asthma were obtained from the that the trend would differ between the days, register kept ofall episodes ofillnesses requiring this was found difficult to model using standard admission to hospital. The register contains regression methodology. On including more information on the dates of hospital stay, and sinusoidal terms, the results became more reas- on the diagnoses and ages of the patients. The onable, and the opposite was also true. This register, its comprehensive coverage, diagnostic could mean that there were some exceptional criteria, and quality assessment have been de- periods (heatwaves, etc) that the model was scribed in detail earlier.915 The population of not able to take into account. However, the Helsinki was 488 604 in 1987,491 148 in 1988, task of finding such periods proved impossible. and 491 777 in 1989. Another set of difficulties was encountered when determining the proper transformations for the pollutants. The data were rather kurtotic MISSING VALUES - that is, most of the values did not differ much The study period comprised 1096 days. In- from the mean, whereas there were some very formation on the concentration of TSP was extreme values. These extremes tended to dom- lacking for 375 days because of the sampling inate the fit if no transformations were used. strategy, and so were values for NO2 and SO2 In addition, the time plots seemed rather non- for 13 and four days, respectively. However, stationary. Logarithmic transformation was the frequencies of admissions during the days therefore used. In future, generalised additive with missing observations were similar to those models'7 in addition to the STL seem to be an on the other days. Thus, the effect of the attractive way of avoiding such problems. missing observations is probably small. The Lags of 0, 1, and 2 days and cumulative lags missing values were imputed from regression of up to 3 days were tried for the effects of all analysis. pollutants. For those of 03, cumulative lags up http://jech.bmj.com/ to 5 days were tried. The lags for untransformed values of temperature and humidity were MODELLING chosen from lags of 0, 1, and 2 days, and from The variables were included in the regression the averages of lag 0 and 1, and of lag 0, 1, analysis in the following order: (1) long term and 2 days. The lags for influenza epidemics trend, (2) season, (3) epidemics, (4) day of were chosen from lag 0 to 15 days. week, (5) holidays, and (6) temperature and on September 30, 2021 by guest. Protected copyright. relative humidity. The final analysis in- vestigating the association between average Results levels of pollutants measured at different sta- NUMBER OF ADMISSIONS tions and admission for asthma was made by There were altogether 2421 admissions to hos- Poisson regression analysis. pital for asthma during the three year period- Long term trends have been controlled both that is, an average of 2-21 admissions daily separately and jointly for each year. In addition, (table 1). yearly "dummies" have been used. Separate terms were tested for each year. Sinusoidal terms were introduced into the model up to Table 1 Number of emergency hospital admissions for the 6th order. The combination with the best asthma in Helsinki, 1987-89 adjusted R2 was selected, including both sine Diagnosis and No Admissions per day and cosine terms. In addition, different terms age group No (SD) for each year were tested, but found not to be significant. Six dummy variables were formed Asthma, all ages 2421 2-21 (1 91) Asthma, 0-14y 961 0-88 (1-26) for the days of the week and one for holidays. Asthma, 15-64y 813 0-74 (1-08) For each Christmas, a separate exponentially Asthma, >64y 647 0-59 (0 82) Control* 9017 8-23 (3 63) decaying term - that is, exp - [days from Christ- * mas] - was used. Emergency admissions for digestive system diseases. Asthma and ambient air pollution in Helsinki S61 J Epidemiol Community Health: first published as 10.1136/jech.50.Suppl_1.s59 on 1 April 1996. Downloaded from Table 2 Mean values ofpollutants, temperature, and relative humidity by season in RELATION OF AIR POLLUTANTS TO ADMISSIONS Helsinki, 1987-89 A summary of the significant associations is December- March- J7une- September- presented in table 3. Admissions for asthma February May August November among 0-14 year olds were related to the 8 h S02 (,ug/m') 26 22 13 15 03 levels, and those among 15-64 year olds NO, (pg/M3) 38 44 39 34 03 (pg/Mr) 20 30 23 15 and >64 year olds to the 24 h SO2 levels. TSP (pg/M3) 64 116 62 64 Admissions for digestive tract disorders, Temperature ('C) -4-3 +4-0 +15-8 +5-8 Relative humidity (%) 89 79 75 86 which were used as a control, were significantly associated with 8 h 03 levels, but not with other pollutants, temperature, or humidity. Table 3 Association between the number of admissions and levels ofpollutants with various lags in the Poisson regression analysis Discussion Disease and Pollutant Lag Parameter (SE) p age group (d) estimate The modelling was problematic. In the first attempts, over correction for cyclicity and un- Asthma 0,, 8h* 0 0-1600 (0-0778) 0 040 0- to 14-y der correction for autocorrelation were made. Asthma S02, 24h 2 0-2176 (0-1081) 0-44 Therefore, the number of sinusoidal terms was 15- to 64-y Asthma 52, 24h 0-3 0-3086 (0-1545) 0-046 restricted from nine to six (but up to 19 were 15- to 64-y tried). Various transformations of pollution Asthma S02, 24h 2 0-2412 (0 0956) 0-012 >64 y variables were tried; and finally logarithmic Digestive tract 0,, 8h* 0 0-0641 (0 0282) 0-023 values were used. diseases Digestive tract 03, 8h* 0-4 0-1152 (0-0417) 0-006 The difficulties of the analysis suggest that diseases the method of analysis may be unsatisfactory. * One station only The model was first used for analysing assoc- iations between air pollution and mortality.'4 It is possible that the patterns of cyclical variation POLLUTANTS AND METEOROLOGICAL VARIABLES found in data for mortality and hospital ad- Concentrations of most pollutants in the am- mission may be different, a phenomenon that bient air were relatively low. The mean con- might be explained by the epidemiology as- centrations of SO2, NO2, and 03 during the sociated with these outcome variables. three year period, presented as arithmetic The results obtained here differ from those means of the values measured at the respective of an earlier analysis of the same data with a stations, were 19 ,ug/m3, 39 pg/M3, and 22 ,ug/m3. different methodology.9 In the earlier analysis, The mean concentration of TSP was 76 jg/M3, linear regression was used and the effect of a high value that resulted from the met- cold was controlled with a less sophisticated eorological conditions and the erosions of street method. With these methods, significant assoc- surfaces caused by the use of studded tyres iations were observed between hospital ad- during the winter, as well as the use of sand missions for asthma and the levels of SO2, NO2, on the streets to treat icy surfaces. Mean values 03 and TSP.9 of pollutants by season are given in table 2. In the present study, not only average values ofpollutants measured at different stations, but also the measurements at the station with the RELATION OF CONFOUNDERS AND COFACTORS highest values were used first in the analysis. http://jech.bmj.com/ TO ADMISSIONS The results of these analyses were in part il- The time plots of residuals of the regression logical, showing that these measurements do analyses were checked after inclusion of in- not sufficiently reflect the exposure ofthe popu- tercept, trend, season, sinusoidal terms, in- lation to pollutants. fluenza epidemics, day of week, holidays, and Hospital admissions for digestive tract dis- temperature and relative humidity. orders were significantly associated with the In the final Poisson models, short term cycles 8 h 03 levels at the only station. There are two

and the effect of weekdays were seen in all the possible explanations for this illogical finding. on September 30, 2021 by guest. Protected copyright. groups studied: among emergency admissions Firstly, the modelling may be defective; if so, for asthma in the various age groups and among the association between 03 levels and ad- emergency admissions for digestive system dis- missions due to asthma among 0-14 year olds orders. should also to be interpreted with caution. A rise in temperature was associated with a Alternatively, the apparent correlation may be low number of admissions due to asthma a statistical coincidence. among those aged 0-14 years and among 15-64 The mean daily S02 values measured in year olds, whereas humidity did not have a Helsinki were lower than the maximum value significant effect on the number of admissions. recommended by the World Health Or- ganization - 125 pg/m3. 8 The daily guideline was not exceeded. However, a significant assoc- AUTOCORRELATION AND DESCRIPTIVE STATISTICS iation was seen between daily S02 con- The control of autocorrelation was checked centrations and the daily counts of emergency by using the residual diagnostic plots of the room admissions among the 15-64 year old Poisson regressions and with the periodo- patients and among those over 64 years old. grams. The autocorrelation and partial auto- However, this association was not observed correlation functions of the series included among children. The inconsistency of the find- approximate 95% confidence intervals defined ings may suggest that there is no real effect. as + 2/n-ag = + 0-06. The other possibility is that there is a relation S62 Pdnka, Virtanen J Epidemiol Community Health: first published as 10.1136/jech.50.Suppl_1.s59 on 1 April 1996. Downloaded from 5 Goldstein JF, Weinstein AL. Air pollution and asthma: between the number of patients admitted for Effect of exposure to short-term sulfur dioxide peaks. asthma and the presence of SO2 at lower con- Environ Res 1986;40:332-45. 6 Bates DV, Sitzo R. Hospital admissions and air pollutants centrations than previously believed harmful or in Southern Ontario: the summer acid haze effect. Environ than are listed in the guidelines given in many Res 1987;43:317-31. 7 Bates DV, Baker-Anderson M, Sizto R. Asthma attack peri- countries and by the World Health Organiza- odicity: a study of hospital emergency visits in Vancouver. tion. The previous study from Helsinki sug- Environ Res 1990;51:51-70. 8 Pope CA, III. Respiratory hospital admissions associated gested the same among patients with chronic with PM,, pollution in Utah, Salt Lake, and Cache valleys. bronchitis. 5 Arch Environ Health 1991;46:90-7. 9 Ponka A. Asthma and low level air pollution in Helsinki. Although our results, using the uniform Arch Environ Health 1991;46:262-9. methodology of the APHEA project, suggest 10 Sunyer J, Anto JM, Murillo C, Saez M. Effects of urban air pollution on emergency room admissions for chronic that even low level pollution may increase hos- obstructive pulmonary disease. Am J Epidemiol 1991;134: pital admissions for asthma, no definitive con- 277-86. 11 Schwartz J, Slater D, Larson TV, Pierson WE, Koenig JQ. clusion can be drawn from the results without Particulate air pollution and hospital emergency visits for meta-analysis. The effect of methodology was asthma in Seattle. Am Rev Respir Dis 1993;147:826-31. 12 Dockery DW, Schwartz J, Spengler JD. Air pollution and seen to have a strong effect on the results and daily mortality: Association with particulates and acid new standardised methods, possibly different aerosols. Environ Res 1992;59:362-73. 13 Schwartz J, Wypij D, Dockery D, Ware J, Zeger S, Spengler from those for the study on mortality, should J, Ferris B Jr. Daily diaries of respiratory symptoms and be developed to investigate the associations air pollution: Methodological issues and results. Environ Health Perspect 1991;90:181-7. between ambient air pollutants and morbidity. 14 Katsouyanni K, Zmirou D, Spix C et al. Short-term effects of air pollution on health: A European approach using epidemiologic time series data. Eur RespirJ 1995;8:1030- 8. 15 Ponka A, Virtanen M. Chronic bronchitis, emphysema, and 1 Logan WPD. Mortality in the London fog incident. Lancet low-level air pollution in Helsinki. Environ Res 1994;65: 1 953;i:336-8. 207-17. 2 Ministry ofHealth. Morbidity and mortality during the London 16 Cleveland RB, Cleveland WS, McRae JE, Terpenning I. fog. Reports on Public Health Medicine Subjects No 95. STL: A seasonal trend decomposition procedure based London: Her Majesty's Stationary Office, 1954. on Loess. Journal of Official Statistics 1990;6:3-73. 3 Cohen AA, Bromberg S, Buechley RW, Heiderscheit MA, 17 Hastie TJ, Tibshirani RJ. Generalized additive models. Lon- Shy CM. Asthma and air pollution from a coal-fueled don: Chapman & Hall, 1990. power plant. Am I Public Health 1972;62:1181-8. 18 World Health Organization. Air quality guidelines for Europe. 4 Whittemore AS, Korn EL. Asthma and air pollution in the WHO Regional Publications, European Series No. 23. Los Angeles area. Am Public Health 1980;70:687-96. Copenhagen: WHO, 1987. http://jech.bmj.com/ on September 30, 2021 by guest. Protected copyright.