METHODS

A time series “panel” study design was used¹¹ with two 8 week monitoring periods representing winter (13 January–10 March 1997) and summer (19 May–14 July 1997) conditions. Subjects were recruited from five primary schools in two urban locations—Birmingham, a major city in the English Midlands, and Sandwell, a large urban area contiguous with Birmingham—which were analysed separately. The schools selected were close to major motorways and congested city arterial roads. Study approval was obtained from the East Birmingham research and ethics committee.

Panel recruitment and monitoring

The panel comprised 9 year old children enrolled during September 1996. After giving written consent, parents completed a questionnaire detailing their child's respiratory symptoms, atopic illnesses, and relevant housing factors. Subjects were divided into subgroups on the basis of reported wheezing in the previous 12 months in the absence of a respiratory tract infection. Baseline spirometric tests and skin testing to common allergens (cat, dog, grass, trees, house dust mite, and cockroach) were performed. Children were considered atopic on the basis of at least one positive skin test (mean weal diameter at least 3 mm greater than negative control).

Children and teachers were instructed in peak expiratory flow (PEF) measurement and all subjects were issued with a PEF meter (Vitalograph Ltd, UK). The highest of three PEF readings at morning registration (08.45–09.00 hours) and at the end of the school day (15.30–15.45 hours) were recorded. At weekends parents were asked to ensure a reading at breakfast time and in mid-afternoon. Subjects were required to record medication taken each day and to respond to the following five questions:

• “Did you cough today?”

• “Were you ill today?”

• “Were you short of breath (SOB) today?”

• “Did you wake up last night with a cough or wheeze?”

• “Did you wheeze today?”

Diary cards were issued and collected weekly and inspected immediately for errors or omissions. Data were entered onto a spreadsheet by optical character recognition scanning (DRS Infotech, Milton Keynes, UK). Prior to analysis, PEF recordings were corrected for non-linear errors of the meters using an equation derived from the response of a sample of meters to a servo-controlled pump.¹²

Environmental monitoring

The five schools were near existing background urban air pollution monitoring stations, either as part of the National Automatic Urban and Rural Network or provided by local government. These stations measure nitrogen oxides (NO_x, NO, NO₂), sulphur dioxide (SO₂), ozone, carbon monoxide (CO), and PM₁₀ as hourly averages. Measurements of PM₁₀ and PM_2.5 used TEOM instruments equipped with 2.5 μm cyclone inserts operated at 50°C.¹³ At two sites denuders enabled measurement of ammonia (NH₃) and acid gases (SO₂, hydrochloric acid (HCl), and HNO₃), chloride (Cl⁻), NO₃⁻, and NH₄⁺.^14,15 Measurements of SO₄²⁻, NO₃⁻, and Cl⁻ were made by standard ion chromatographic methods, and of aerosol strong acid (H⁺) by aqueous extraction and determination of pH according to the method of Koutrakis and coworkers.¹⁶ The samples were collected daily and stored under refrigeration for batch analysis.

The University of Birmingham Weather Service and the Pollen Research Unit, Worcester provided daily maximum, mean and minimum temperatures, mean relative humidity, barometric pressure, wind speed, and summer pollen counts.

Statistical analysis

Analysis of data followed the methodology developed for the PEACE study.¹⁷ Subjects who failed to record data on more than 22 days in each period (40%) were excluded because of inadequate data collection. The first two days of all PEF records were ignored to reduce potential training effects, and each subject's PEF record was transformed into daily deviations from their individual mean PEF for morning and afternoon separately. PEF data were analysed as the daily mean of individual deviations (ΔPEF), weighted according to the number of reporting children that day. Symptom data were analysed as the daily proportion of subjects reporting each symptom (% prevalent symptoms) or reporting a new episode of each symptom (% incident symptoms). For each day the denominator was defined as the number of children recording both morning and afternoon PEF.

Putative associations between environmental variables and health outcomes were initially explored using bivariate correlation (Pearson's correlation coefficient, r) and considered lags of 0–3 days and a 7 day prior mean. Final results were calculated using a linear regression approach for ΔPEF and logistic regression for symptoms. Final effect estimates were then derived from β (regression) coefficients in models accounting for relevant confounding factors. For each outcome, terms correcting for trend, weather elements, autocorrelation within the model residuals, and a dummy variable indicating schooldays (versus weekends and holidays) were included. For ΔPEF, linear and square root trend terms were considered for inclusion to adjust for lung growth and potential training effects. For symptoms, linear, quadratic and higher order polynomial trend terms were considered.¹⁷ Weather and summer pollen variables were considered for inclusion at the lag resulting in the strongest association to the health outcome. The a priori hypothesis required linear trend, a temperature term, and a term correcting for first order autocorrelation to be included. Other trend and weather terms were included on the basis of examining the residual variance and goodness of fit (r²) statistics. Terms correcting for higher order autocorrelation were included on the basis of visually examining the autocorrelation and partial autocorrelation functions of the model residuals. For comparability, both Birmingham and Sandwell models were required to contain the same terms, albeit at different lags for weather variables. Pooled results for the entire panel could then be derived for ΔPEF and symptom outcomes by combing the effect estimate from each location after weighting by the inverse of its variance.

In addition to considering each panel as a whole, subgroups based on atopic status and a history of recent wheezing were analysed with the main pollutants of interest using the models identified for the whole panel to determine whether these children were at increased risk of adverse health effects.¹⁸ In addition to examining effect sizes in these groups, comparisons were also made with results from the remaining panel members.

RESULTS

Panel recruitment

Consent was obtained in 162 of 264 eligible children (table 1). Four children were lost to the panel before the summer period. The subgroup with a history of recent wheezing comprised 39 subjects; sleep disturbance was reported in 13 of these (33%) and wheezing severe enough to limit speech in five (13%). No significant differences in the proportion of children with a history of recent wheezing were found between schools or sexes.

No differences in baseline spirometric subgroups based on a history of recent wheezing were recorded. Children with a such a history were significantly more likely to be labelled atopic than those without recent wheezing and were more likely to show specific reactions to house dust mite allergen, tree pollen, and cat dander (p<0.05).

Data collection from subjects

Morning PEF records were available for a median of 140/162 subjects in winter (87% response, range 106–159) and 126/158 in summer (79% response, range 93–142). In general, symptoms were more frequently recorded in the winter than the summer (fig 1), the most frequently reported symptom being cough. Inadequate data collection resulted in the exclusion of 14 and 20 children, respectively, from the winter and summer periods. Direct questioning determined whether a child possessed anti-asthma medication, but individual dosages were poorly recorded so these data were discarded.

Environmental results

Wintertime pollutant levels were unexceptional and PM₁₀ exceeded 40 μg/m³ on only four occasions (table 2 and fig 1). Such peaks were associated with increased oxides of nitrogen and SO₂. A modest increase in ozone was seen in March. Poor weather meant that summer pollutant levels were generally low. PM₁₀ exceeded 40 μg/m³ only in the final week (fig 1), associated with increased oxides of nitrogen. Acid and anion concentrations were especially low as conditions for their formation were unfavourable. Aerosol strong acidity (H⁺) levels were detectable on only 17 winter and 9 summer days and were therefore not analysed.

Multiple cross correlations were seen between pollutants for both the winter and summer periods. Very strong positive correlations between winter time PM₁₀ and PM_2.5 (r=0.93) were accompanied by similar relationships to NO₂ and, in the negative direction, to ozone. These associations were especially strong for PM_2.5 (r=0.88 and r=−0.83, respectively). Weaker associations were noted with SO₂. Most individual acid and anion species also showed a strong degree of positive correlation with each other and the particulate mass measures (PM_2.5 more than PM₁₀). However, Cl⁻, HCl, HNO₃ and, to a lesser extent, H⁺ were poorly associated with each other (except Cl⁻ and HCl) and other pollutants. A similar pattern of association was noted in the summer, although the values of r were generally lower. However, in the summer, levels of HNO₃ were correlated with those for the particulate mass measures (PM₁₀, r=0.77; PM_2.5, r=0.81) and NO₂ (r=0.65).

Particulate levels were similar between the two locations, particularly in the case of PM_2.5, consistent with its long atmospheric lifetime (winter PM_2.5, r=0.80). Measured components of the particulate matter (SO₄²⁻, NO₃⁻, NH₄⁺, Cl⁻) were reasonably correlated between the two sites (r=0.63−0.93), while the gaseous acid and anions (HCl, HNO₃, NH₃) showed a lower degree of correlation (r=0.12−0.73).

Identification of regression models

For ΔPEF, square root trend terms were included as this improved model fit for the Birmingham panel. Similarly, quadratic trend terms were included in symptom models as model fit was generally improved for both panels. Minimum temperature was included in all models, although variations were seen in the lag chosen. The closest association between wintertime ΔPEF, prevalent symptoms, and minimum temperature was seen for the 7 day prior mean. In contrast, incident symptoms were most clearly associated with the same day's minimum temperature (except Birmingham illness and wheeze models which included the 7 day prior mean). The majority of summer outcomes were most closely associated with the same or previous day's minimum temperature, except incident and prevalent illness for which models also included the 7 day mean. Additionally, inclusion of relative humidity improved model fit for winter ΔPEF and prevalent symptoms, but not other meteorological variables or summer pollen count.

In general, first order autocorrelation terms were adequate to remove autoregressive effects from the model residuals, but winter SOB and summer morning ΔPEF required additional higher order terms.

Pollutant effects on health outcomes

Statistically significant associations between pollutants and ΔPEF or respiratory symptoms were seen in both winter (tables 3 and 4) and summer (tables 5 and 6). However, there were no consistent changes, either adverse or otherwise, in any symptom or lung function index when the total panel was considered. Results for incident symptoms and the acid and anion species HCl, Cl⁻, NH₄⁺ and NH₃ are not shown for brevity. No pattern in the nature of the pollutants or the lag of greatest measured effect were noted and, in particular, there were no consistent responses to ozone or particles as PM₁₀ or PM_2.5.

Analysis of subgroups based on atopic status or history of recent wheezing

The results do not indicate that children with atopy or a history of recent wheezing are more susceptible to the short term respiratory health effects of air pollutants. Few statistically significant results were noted and the results are presented for winter morning ΔPEF only (table 7). However, no pattern between pollutants, their lags, or apparent direction of measured effect are evident for any of the health outcomes studied, nor is there any apparent consistency between the two locations.

DISCUSSION

This study provides little evidence for a relationship between the measured pollutants and daily changes in health outcomes after correction for the confounding effects of weather, trends in the data, and autocorrelation. In particular, there is no suggestion that PM_2.5 provides a better index of short term health effects than PM₁₀, and that individual acid or anion species were more closely associated with such effects than particulate mass measures. In epidemiological studies involving multiple comparisons it is important not to overemphasise individual “significant” results, but rather to attempt to identify clear consistent patterns. In this study no such consistency in pollutant, lag, or location was noted.

The identification of particulate health effects where aerosol strong acidity is very low³ has focused attention away from acid species generally, and these compounds have not been regarded as important in the UK since the Clean Air Act. In this study, aerosol strong acidity was virtually undetectable and no relationships with acid or basic species, gas or aerosol, were found, although concentrations were low. Sulphate has been regarded by some authors as a surrogate for the mass concentration of fine particulates¹⁹ and, in situations of high aerosol strong acidity, has been found to be more strongly related to respiratory admissions and some symptoms than PM measures.^9,20 In contrast, low levels of aerosol strong acidity were reported in a wintertime Dutch panel of children which identified small effects of PM₁₀, NO₂, SO₄²⁻, NO₃⁻, and HNO₃ (but not SO₂) on lung function,¹⁰ although only short lags were considered. The effects of SO₂ and SO₄²⁻ on respiratory symptoms were also reported.

Our findings are consistent with those from the PEACE study in which urban and rural panels of 6–12 year old children with a history of recent night time cough or wheeze were monitored across 14 centres for at least two winter months.⁷ Measures of fine particulate levels and acid species were omitted from this study and few statistically significant associations were found overall, despite a wide range of pollutant and climatic experiences. The authors identified a number of possibilities for these negative findings. Firstly, overall panel effect estimates could potentially be biased by a subgroup within the panel with a different pattern of response. In the PEACE study children with diagnosed asthma taking respiratory medication showed a positive relationship between PEF and pollutant levels, although bronchodilator use was not related to pollutants.¹⁸ In our study, subgroup analyses revealed no consistent effects of pollutants on PEF or respiratory symptoms in children with a previous history of wheezing or atopy. Other authors report contrasting findings; a study of Dutch adults demonstrated that airway “lability”, expressed as PEF variability or bronchial hyperresponsiveness (BHR), predicted susceptibility to pollutant effects²¹ and, in Dutch children, “allergy”, in addition to BHR, has been implicated in susceptibility to increased PM₁₀, NO₂, and SO₂.²² It has been suggested that, in susceptible subjects, pollutants may act as “potentiators”, increasing the effects of other factors such as allergens which could influence lung function.²³ However, such a relationship could potentially be reversed by the action of bronchodilating medication so that the inclusion of children with diagnosed asthma in an epidemiological study of air pollutants may obscure any real relationships or lead to the paradoxical result of high pollutant levels associated with better health outcomes. This could occur if those with asthma modify their behaviour on high pollutant days, either by the use of bronchodilating drugs³ or by staying indoors and reducing activity.

Selecting individuals with a history of recent wheezing on the basis of a questionnaire may also result in a heterogeneous group. In contrast to our symptomatic subgroup, selected on the basis of recent wheezing only, most centres in the PEACE study included children with nocturnal cough alone. Such children were in the majority in many panels and had a lower prevalence of atopy, higher MMEF, and lower day to day PEF variability²⁴ and could have responded differently to air pollutants.¹⁸

Unmeasured confounders which vary across a suitable time course—for example, respiratory tract infections—could also be important, although fever has proved to be a relatively insensitive surrogate marker for this.⁷ In our study the daily prevalence of school absence (or days' data missing if at weekends) said to be due to illness (excluding accidental injury) was used as an imperfect measure of such infections and was not found to be associated with daily pollutant levels. Data missing from an individual's record due to ill health could weaken the apparent effect of pollutants if the missed days included illness precipitated or exacerbated by air pollution. The validity of the health outcome measures where children record their own data is not known and misclassification arising in this area could weaken any apparent effects of pollutants. Invented, misread, or inadequate PEF values can occur and these have been shown to increase with time.²⁵ Less is known about the validity and accuracy of symptom responses, but these may reduce with waning interest. In addition, the low prevalence of symptom reporting, particularly in the summer, may also have reduced the sensitivity of our study.

The number of subjects enrolled in this study was greater than in some widely reported panel studies that have shown an association between PM₁₀ and decrements in PEF in this age group.^2–4 However, these studies have exclusively examined populations selected on the basis of existing respiratory symptoms or asthma. In addition, in all these cases pollutant levels were greater than seen in this study where, in general, pollutant levels were modest in winter, though not unrepresentative of the UK, and very low in summer. It may therefore be that this study had insufficient power to detect effects in an unselected cohort of children.

In conclusion, this study does not provide evidence for day to day respiratory health effects of pollutants, including particulates and individual acid and anion species, in a panel of UK inner city primary schoolchildren or subgroups with atopy or pre-existing wheeze. However, only short term effects were considered so these results do not preclude an effect of very fine particulates or acid species on longer term changes in lung function, symptoms, or lung development. Previous authors have suggested that short term pollutant effects occur without threshold. We believe that, if such effects exist, they are likely to be marginal at these observed pollutant concentrations. It is likely that the complexity of adequately dealing with both intraindividual and interindividual variability, in addition to quantitatively small associations between population average responses and pollutants, may be beyond the modelling approach adopted by the PEACE study. Further work should concentrate on more homogeneous groups thought to be at high risk of adverse effects and attempt to improve the validity of health status monitoring.