Including viral infection data supports an association between particulate pollution and respiratory admissions

Transcription

1 Article Environmental Hazards Including viral infection data supports an association between particulate pollution and respiratory admissions Kyoko Fukuda Geohealth Laboratory, Department of Geography, University of Canterbury, New Zealand. Phil N. Hider Department of Public Health & General Practice, School of Medicine and Health Sciences, University of Otago, New Zealand Michael J. Epton Canterbury Respiratory Research Group, Christchurch Hospital, New Zealand Lance C. Jennings Canterbury Health Laboratories, Christchurch and Pathology Department, University of Otago, New Zealand Simon P. Kingham Geohealth Laboratory, Department of Geography, University of Canterbury, New Zealand Particulate air pollution has long been associated with increased respiratory morbidity, including hospital admission. 1-7 A statistical time series approach based on generalised additive models (GAMs) remains the standard method to investigate associations between air pollution, climate and hospital admissions, since it was introduced by Schwartz. 1 However, a number of potential problems with using default GAMs setting in S-Plus have been identified, such as the choice of various smoothers and the selection of the degrees of freedom for the analyses. 3,6 In practice, these differences in model specification tend to shift relative risk estimates only slightly and did not alter any key findings of studies to date using the statistical software such as S-Plus. 8,9 While the previous investigation in McGowan et al. 2 used default GAMs settings in S-Plus to determine the degrees of freedom, objectives of this study are to refine and revise the previously conducted air pollution and health for Christchurch, New Zealand study in McGowan et al. 2 according to the recommendations of Peng and Dominici. 6 The influence of respiratory pathogens introduces an additional degree of complexity in predictive models of air pollution and admissions. In particular, the periodicity of epidemics including respiratory syncytial virus (RSV) and influenza have been associated with increased admissions especially for young and elderly people worldwide. 10 Respiratory syncytial virus (RSV) is the most common virus pathogen that causes acute lower respiratory tract infection in infants, young children and the elderly, but can infect all age-groups, causing illness indistinguishable from influenza. 14,15 In fact, recent investigation Abstract Objective: To refine and revise previous air pollution, climate and health time series analysis in Christchurch, New Zealand, introducing viral identification data (positive identification count and outbreak, defined as two of more positive tests). Method: The effects on daily respiratory admissions for five years ( ) of air pollution (PM 10 ), climate and virology (incorporating actual counts and outbreaks of influenza A and B (INF), para influenza virus type 3 (PIV) and respiratory syncytial virus (RSV) were examined using generalised additive models (GAMs), which are one of semiparametric models. Results were also compared with a model that included climate and air pollution parameters but without the inclusion of virology data. The data were analysed aggregately and then stratified by age group and season. Results: Different virology data detected various association levels. The highest estimates were a 3.93% (CI: ) and a 3.88% (CI: ) rise in respiratory admissions for a rise of 10 µg/m 3 annual PM 10 with outbreak and actual counts of PIV respectively for 0-19 years old with a three-day lag. Conclusion: Refining a statistical model with the addition of virology data gives a similar estimation of the association between PM 10 levels and respiratory admissions to previous research. Use of the indicator of an outbreak of viral infection appears to be similar to actual count of viruses detected. Key words: Air pollution, climate, GAMs, outbreak, virology. Aust NZ J Public Health. 2011;163-9 doi: /j x Submitted: January 2010 Revision requested: March 2010 Accepted: August 2010 Correspondence to: Kyoko Fukuda, Geohealth Laboratory, Department of Geography, University of Canterbury, Private Bag 4800, Christchurch, New Zealand; kyoko.fukuda@canterbury.ac.nz 2011 vol. 35 no. 2 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH The Authors. ANZJPH 2010 Public Health Association of Australia

2 Fukuda et al. Article on the Emergency Department admissions in New Zealand and Australia has reported that respiratory related illness and viral illness is found to be one of most frequent diagnoses. 16 Despite the fact that the association of influenza viruses and air pollutants in relation to health have been well investigated, few epidemiologic studies to assess the short-term effects between air pollutants and influenza in relation to human health have been undertaken. For example, Wong et al. 17 investigated the interaction between air pollution and influenza viruses in Hong Kong using GAMs. They found that influenza activity modifies the effects of air pollution, in particular ozone (O 3 ), on mortality and hospitalisation of respiratory diseases. However, other common respiratory viruses are rarely investigated. Hence, another objective of this study is to introduce the use of the actual numbers for respiratory viruses detected (virology counts) and the occurrence of outbreaks of influenza type A and B (INF), para influenza virus type 3 (PIV) and respiratory syncytial virus (RSV) to the model for New Zealand study for the first time (whereas McGowan et al. 2 used only climate and air pollution). Methods Study area The population of Christchurch, New Zealand, is approximately 369,000 people over an area of about 450 km 2 on the Canterbury Plains. 20 In Christchurch there is a serious winter air pollution problem related to the burning of wood for home heating, combined with poor air dispersion, 21,22 which has been linked to a range of measures of ill health. 4,23,24 In this study, five years of daily measurements of climate, air pollution and virology data from 1 October 1999 to 30 September 2004 were investigated for annual trends and then for cool seasons (from April to September) separately. Data were analysed across all age groups aggregately and then by the following age strata: 0-19, and 65 years and over. Air pollution and climate data Environment Canterbury provided daily mean PM 10 (TEOM at 40 C), temperature (measured at 1 m above the ground level in C) and relative humidity (in %) measurements, collected from a monitoring site, located in the centre of Christchurch City (Coles Place, St Albans). Hospital admission and virology data Daily counts of acute respiratory hospital admissions from the public hospital, Christchurch Hospital in the study area for the same period as the climate data were provided by the New Zealand Health Information Service. Over the study period all acute medical admissions to Christchurch Hospital were collated. Discharge diagnoses of admission records for all age groups with the following primary diagnoses based on the International Classification of Diseases, 9th Revision (ICD9) classification: pneumonia ( ), acute respiratory infections ( ), chronic lung diseases ( , ), asthma (493), ischaemic heart disease ( ), dysrhythmia (427) and heart failure (428) were investigated. Also, information on appendicitis ( ) admissions was included as a control group. Subjects were stratified into the following age groups: 0-19, and 65 years and over. Respiratory virus surveillance in Canterbury was carried out by the Virology Section of the Canterbury Health Laboratories on hospital inpatients. During the winter months, a network of sentinel general practices carried out influenza surveillance. From May through September, eight sentinel surveillance general practices recorded the number of patients seen daily meeting a case definition for influenza-like-illness and collect respiratory samples from the first patient meeting the case definition seen on Monday, Tuesday and Wednesday of each week. 25 These nasopharyngeal swab samples were forwarded to Canterbury Health Laboratories for viral diagnosis. In addition, swabs collected as part of routine assessments of patients in primary care, emergency departments and hospital admissions were also collected in the same dataset. Influenza and other viruses were detected by antigen detection using immunofluorescence assays and virus isolation in a range of cell cultures. 26 The virus dataset made available consisted of the time and date the sample arrived in the laboratory, date of patient birth and sex. During the studied period, more than 10,000 respiratory samples were collected and laboratory identification counts of influenza type A and B (INF), parainfluenza type 3 virus (PIV) and respiratory syncytial Virus (RSV) per day were extracted. Approximately 27,000 people, 7% of the Christchurch urban area, were included in the sentinel GP practices that were broadly representative of the local population. Because the surveillance and routine diagnostic virology samples were indistinguishable on the data set, an outbreak index for each virus was created, which followed the definition of outbreak that is when there are two or more cases within a 24 hour period (usually linked to a common source 27 ). Hence, the virology outbreak index of 1 indicates when there are two or more positive swabs for a particular virus per day, otherwise the index recorded 0. The virology outbreak index for each INF, PIV and RSV was created from all age groups to indicate an outbreak event for the entire studied population, whereas actual virology counts of INF, PIV and RSV (all age groups) were investigated by dividing them into specific age groups as defined above. Each age group was then incorporated separately into the model to predict a matched admission age group. Results for actual virology counts are reported as INF, PIV and RSV, and outbreak results are shown as INF outbreak, PIV outbreak and RSV outbreak. Statistical method We estimated the log-relative risk of PM 10 to measure the percentage (with 95% confidence intervals, CI) increase in hospital admission per unit increase in PM 10 (10 µ/m 3 ) using semiparametric GAMs assuming a Poisson distribution, with smoothing splines to control for seasonal patterns. A single actual virology counts or outbreak of either INF, PIV or RSV was incorporated respectively to the model along with temperature and 164 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 2011 vol. 35 no. 2

3 Environmental Hazards Particulate pollution and respiratory admissions relative humidity measurements and dummy holiday variables (day of the week, weekend and national holidays). The statistical software package R version with the gam and splines packages was used in the analysis. 6 This study reported the estimated parameter of PM 10 with the smallest mean squared error by iteratively running the model with various df per year of data. The minimum of 2 df per year was chosen to consider the long-term increase in hospital admission counts for seasonal structure. In order to decide how many df should be tested for the GAMs in this study, the following steps were taken. Firstly, the optimal df were calculated to best predict the PM 10 time series as outcomes using the bruto function in the mda package. Secondly, the optimal df for predicting the hospital admission count time series was calculated (details in Peng and Dominici 6 ). We found that optimum df vary over different age groups but its ranges were between 5 to 9 df for annual and 3 to 9 df for cool seasons. Note that the optimum df were found to be the same over variations in virology data. From here, the 2 to 10 df per year was iteratively tested on each set of data (each age group applied with a single actual virology counts and outbreak) to identify which df provide the estimated parameter of PM 10 with the smallest standard error. The maximum of 10 df was chosen as to include the potentially maximum optimum df per year. The minimum of two df per year was chosen to consider the long-term increase in hospital admission counts for seasonal structure. From the experiment of running the model iteratively using 2 to 10, all sets of data recorded the smallest standard error with two df. Then, small number of df, such as selected from the experiment, was chosen for this analysis, because the investigation aims to provide a more precise estimate of the air pollution coefficient, even though it can involve a weak adjustment for unmeasured confounding. Additionally, the plot of estimated percentages over various df showed that stable estimated percentages but extremely low values (can be even negative) were observed after eight or nine df. We reported the estimated percentages at current PM 10 exposure (lag 0). Additionally, we calculated up to 14-day PM 10 exposure lag. Two estimated percentages were reported; the maximum estimated percentages taken from the shorter lag effects (from 1 to 6-day lag) and longer lag effects (from 7 to 14-day lag), to examine variations in PM 10 exposure lag effects on the admission count. We also report results using only climate measurements (temperature and humidity) with the pollution data, and results are compared with the model using climate measurements and each virology count. Results Summary statistics of daily air pollution, climate, respiratory hospital admission and virology measurements for each age group in Christchurch ( ) are shown in Table 1. The data contained a very small proportion of missing data (1-2%) of PM 10 and relative humidity measurements due to equipment maintenance requirements, which were considered to be insignificant. Cool seasons recorded generally higher PM 10 levels, e.g. median of PM 10 is µg/m 3 in cool seasons whereas the median annual PM 10 is µg/m 3 (Table 1). More than half the total respiratory admission counts (n=22,392) were recorded in cool seasons (n=13,882). Median daily hospital admission counts, recoded from Christchurch Hospital, for all age groups during cool seasons was 15, compared with 11 over the entire year (Table 1). A summary of estimated percentage increase (with 95% confident intervals) in respiratory admissions for various age groups in relation to an increase in 10 μg/m 3 of PM 10 is shown for the entire year (Table 2) and separately for the cool seasons (Table 3). All results were reported for the estimated percentage at current (lag 0) and a maximum estimated percentage selected from the shorter (1-6 lag days) and longer (7-14 lag days) PM 10 lag effects. Similar estimated percentage increases were observed between actual virology counts and outbreaks. Likewise the results do not significantly vary between models using climate data only, and those using a combination of both climate and virology data. Higher estimated percentages were generally observed at PM 10 exposure with varying time lags compared with no lag. Increases were usually slightly higher with shorter rather than longer lag periods especially in the cool seasons. Discussion This is the first Australasian study that has investigated the association between respiratory hospital admissions, PM 10 with the inclusion of virology data. It has refined and revised a previous air pollution, climate and respiratory hospital admission study from a 10-year period ( ) 2 by applying the model to five years of more recent data ( ) and adding results from virological analysis. Other studies have examined the effect of PM 10 on the respiratory admission rate in the short term (lag less than three days). 28,29 This study and McGowan et al. 2 observed the maximum estimates (up to six days) were at a two-day lag; for example, there was a 2.62% ( ) increase in admissions with RSV and 2.43% ( ) with INF (Table 2), and McGowan et al. 2 recorded a 2.28% ( ) rise in admissions although virology data were not included. Note that a percentage increase value in McGowan et al. 2 was converted to describe an increase in 10 μg/ m 3 of PM 10. Similarly, for the longer lag periods (7-14 days) the results in this study are also consistent with those reported by McGowan et al. 2 The overlapping confidence intervals for the percentage increase in respiratory admissions in relation to PM 10 between this and the McGowan study suggest that estimates in general may not significantly differ despite the different statistical approaches used to investigate the study populations. McGowan et al. 2 used the default S-PLUS setting in the GAMs whereas our method employs a minimum standard error during the selection of the degrees of freedom in the model. Our results do not show any significant impact on the estimation of the association between particulate pollution and hospital admissions when virology data are included vol. 35 no. 2 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 165

4 Fukuda et al. Article It is notable that the elderly were under-represented in swab counts for all the viruses and especially PIV and RSV (Table 1). These low counts may be important as the elderly are particularly at risk for admission to hospital for a respiratory illness, and our models may be not accurately capture the trends for this age group (Table 2, 3). It is possible that using low count observations may correlate with other variables, leading to a multicollinearity problem that occurs when an approximate linear relationship between low counts of the elderly swab counts and hospital admissions provide unreliable regression estimates. 30 Adverse effects of other pollutants such as NO 2, SO 2, CO, O 3, ultrafine particle (UFP) and PM 2.5 (PM<2.5 µm in aerodynamic diameter) on health have been detected worldwide Hagen et al. 35 reported that PM 10 may be an indicator of other pollutants such as gaseous air pollutants, e.g. NO 2, and such other pollutants are more important determinants of acute hospitalisation for Table 1: Summary statistics of daily air pollution, climate, virology and hospital admission data. Variables n Mean ± SD Min Q1 Med Q3 Max (a) Annual atmospheric and pollution data PM10 (µg/m 3 ) ± Temperature ( C) ± Relative humidity (%) ± (b) Cool seasons atmospheric and pollution data PM10 (µg/m3) ± Temperature ( C) ± Relative humidity (%) ± (c) Annual respiratory hospital admission counts All age group ± age group ± age group ± age group ± (d) Cool seasons respiratory hospital admission counts All age group ± age group ± age group ± age group ± (e) Annual respiratory syncytial virus counts (RSV) All age group ± age group ± age group ± age group ± (f) Cool seasons respiratory syncytial virus counts (RSV) All age group ± age group ± age group ± age group ± (g) Annual influenza A and B counts (INF) All age group ± age group ± age group ± age group ± (h) Cool seasons influenza A and B counts (INF) All age group ± age group ± age group ± age group ± (i) Annual para influenza virus type 3 counts (PIV) All age group ± age group ± age group ± age group ± (j) Cool seasons para influenza virus type 3 counts (PIV) All age group ± age group ± age group ± age group ± AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 2011 vol. 35 no. 2

5 Environmental Hazards Particulate pollution and respiratory admissions respiratory conditions than particulate mass for Europe Furthermore, diesel exhaust particles are known to particularly increase airway inflammation and exacerbate asthma while the inflammatory impacts of O 3 on respiratory disease have also been investigated. 38 Although this study examined PM 10, it has been reported that 90% of PM 10 in Christchurch is estimated to be within the PM 2.5 fraction, suggesting a close relationship between the two measures. 39 A Ministry of Environment report 40 has estimated that more than 100 deaths per year in Auckland may be attributable to exposure to ozone concentrations. However, the role of ozone was not investigated here given the large uncertainties in estimating area-level exposure. It has also been noted that Christchurch is unlikely to experience significantly elevated levels of ozone (see McGowan et al. 2 ). In recent years, the heterogeneity of air pollution, climate and health effects over different cities or multi-pollutant models have been investigated. Single site or single pollutant investigation can lead to imprecise or biased results, given that sensitivity of the effect estimates in different cities or combinations of pollutants can vary. 41 Thus, future investigation should attempt multiple pollutant analyses, e.g. PM 10 and NO 2, taken from multiple monitoring sites to examine the various effects on the admission pattern. Another area of considerable research interest is the potential use of virology detection data to serve as a prediction tool for hospital admission numbers. Many virology samples are taken in community in advance of a hospital admission. In recent years, various computational methodologies such as neural networks, 42 theoretical computer algorithms 5 and data mining techniques 7,43 have been applied to study the relationship between air pollution, climate and hospital admissions. Such computational methods can be used conjunction with climate and virology data and may offer a method that can better predict or understand hospital admissions. Table 2: Summary results of annual data. Annual Lag 0 Shorter lag (1-6 days) Longer lags (7-14 days) Increase % (95% CI) Increase % (95% CI) Lag Increase % (95% CI) Lag Temperature and humidity into GAMs All age groups 1.72 ( ) 2.99 ( ) ( ) years old 1.52 ( ) 3.78 ( ) ( ) years old 1.84 ( ) 3.81 ( ) ( ) years old 1.93 ( ) 3.25 ( ) ( ) 10 Temperature, humidity and virology into GAMs INF All age groups 1.05 ( ) 2.43 ( ) ( ) years old 0.94 ( ) 3.43 ( ) ( ) years old 1.25 ( ) 3.11 ( ) ( ) years old 1.51 ( ) 3.05 ( ) ( ) 10 INF outbreak All age groups 1.10 ( ) 2.49 ( ) ( ) years old 0.86 ( ) 3.19 ( ) ( ) years old 1.27 ( ) 3.32 ( ) ( ) years old 1.32 ( ) 2.59 ( ) ( ) 10 PIV All age groups 1.54 ( ) 3.11 ( ) ( ) years old 1.42 ( ) 3.88 ( ) ( ) years old 1.54 ( ) 3.82 ( ) ( ) years old 1.76 ( ) 3.31 ( ) ( ) 10 PIV outbreak All age groups 1.54 ( ) 3.06 ( ) ( ) years old 1.37 ( ) 3.93 ( ) ( ) years old 1.60 ( ) 3.77 ( ) ( ) years old 1.76 ( ) 3.30 ( ) ( ) 10 RSV All age groups 1.48 ( ) 2.62 ( ) ( ) years old 1.21 ( ) 2.87 ( ) ( ) years old 1.69 ( ) 3.82 ( ) ( ) years old 1.74 ( ) 3.30 ( ) ( ) 10 RSV outbreak All age groups 1.44 ( ) 2.59 ( ) ( ) years old 1.18 ( ) 2.81 ( ) ( ) years old 1.60 ( ) 3.73 ( ) ( ) years old 1.68 ( ) 3.11 ( ) ( ) 10 The highest estimated percentages within the same age group across various lag results were highlighted in bold. The underlined estimated percentages indicate that the highest estimated percentage within the same virology category over any age groups and any lags vol. 35 no. 2 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 167

6 Fukuda et al. Article In relation to such a prediction tool, the results from this study suggest that actual virology counts and outbreak virology data generally provide similar results. The use of an outbreak as an indication of actual virology counts, namely the identification of two cases or more cases, may be sufficient enough to represent the same outcome. Such a proxy measure suggests a more cost effective surveillance strategy in future that does not require the collection of respiratory samples from large numbers of patients for all age groups in order to obtain reasonable estimates. Acknowledgements This study was supported by a grant from 2008 Research Project Grant Round, Canterbury Medical Research Foundation. We thank the ANZJPH reviewers who provide us useful comments, to Environment Canterbury (Ms. Teresa Aberkane), New Zealand Health Information Service and Canterbury Health Laboratories for providing data, to Professor Ian Town for his support, and to P. Pearson for editorial work. References 1. Schwartz J. Nonparametric smoothing in the analysis of air pollution and respiratory illness. Canadian Journal of Statistics. 1994;22: McGowan JA, Hider PN, Chacko E, Town GI. Particulate air pollution and hospital admissions in Christchurch, New Zealand. Aust N Z J Public Health. 2002;26 (1): Dominici R, McDermott A, Zeger SL, Samet JM. On the use of Generalized Additive Models in time series studies of air pollution and health. Am J Epidemiol. 2002;156(3): Wilson JG, Kingham S, Sturman A, Pearce J. Intraurban particulate air pollution and restricted activity days in New Zealand school children. Epidemiology. 2005;16:S Fukuda K, Takaoka T. Analysis of Air Pollution (PM10) and Respiratory Morbidity Rate using K-Maximum Sub-array (2-D) Algorithm. Proceedings of the 22nd Annual 2007 ACM Symposium on Applied Computing; 2007 Mar 11-15; Seoul, Korea. New York: ACM; p Table 3: Summary results of cool season data. Cool seasons Lag 0 Shorter lag (1-6 days) Longer lags (7-14 days) Increase % (95% CI) Increase % (95% CI) Lag Increase % (95% CI) Lag Temperature and humidity into GAMs All age groups 1.02 ( ) 1.80 ( ) ( ) years old 0.39 ( ) 2.25 ( ) ( ) years old 1.89 ( ) 3.19 ( ) ( ) years old 1.36 ( ) 2.76 ( ) ( ) 10 Temperature, humidity and virology into GAMs INF All age groups 0.58 ( ) 1.49 ( ) ( ) years old 0.06 ( ) 2.13 ( ) ( ) years old 1.45 ( ) 2.65 ( ) ( ) years old 1.02 ( ) 2.61 ( ) ( ) 10 INF outbreak All age groups 0.62 ( ) 1.67 ( ) ( ) years old ( ) 1.97 ( ) ( ) years old 1.50 ( ) 2.92 ( ) ( ) years old 0.94 ( ) 2.40 ( ) ( ) 10 PIV All age groups 0.87 ( ) 1.99 ( ) ( ) years old 0.24 ( ) 2.41 ( ) ( ) years old 1.65 ( ) 3.19 ( ) ( ) years old 1.25 ( ) 2.81 ( ) ( ) 10 PIV outbreak All age groups 0.86 ( ) 1.97 ( ) ( ) years old 0.23 ( ) 2.42 ( ) ( ) years old 1.66 ( ) 3.12 ( ) ( ) years old 1.24 ( ) 2.81 ( ) ( ) 10 RSV All age groups 0.90 ( ) 1.83 ( ) ( ) years old 0.26 ( ) 1.80 ( ) ( ) years old 1.76 ( ) 3.17 ( ) ( ) years old 1.21 ( ) 2.79 ( ) ( ) 10 RSV outbreak All age groups 0.87 ( ) 1.82 ( ) ( ) years old 0.23 ( ) 1.79 ( ) ( ) years old 1.71 ( ) 3.18 ( ) ( ) years old 1.23 ( ) 2.78 ( ) ( ) 10 The highest estimated percentages within the same age group across various lag results were highlighted in bold. The underlined estimated percentages indicate that the highest estimated percentage within the same virology category over any age groups and any lags. 168 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 2011 vol. 35 no. 2

7 Environmental Hazards Particulate pollution and respiratory admissions 6. Peng RD, Dominici F. Statistical Methods for Environmental Epidemiology with R. A Case Study in Air Pollution and Health. Baltimore (MD): Springer; Fukuda K. Identifying Fewer Key Factors by Attribute Selection Methodologies to Understand the Hospital Admission Prediction Pattern with Ant Miner and C4.5. In: Velásquez JD, et al. editors. KES 2009, Part II, Lecture Notes in Computer Science, LNAI Berlin (DEU): Springer; p Wilson AM, Salloway JC, Wake CP, Kelly T. Air pollution and the demand for hospital services: a review. Environment International. 2004;30: Colburn KA, Johnson PR. Public health. Air pollution concerns not changed by S-Plus flaw. Science. 2003;229: Jordan RE, Hawker JI, Ayres JG, Tunnicliffe W, Adab P, Olowokure B, et al. A case-control study of elderly patients with acute respiratory illness: Effect of influenza vaccination on admission to hospital in winter. Vaccine. 2007;25: Wyde PR. Respiratory syncytial virus (RSV) disease and prospects for its control. Antiviral Res. 1998;39: Simoes EAF. Respiratory syncytial virus infection. Lancet. 1999;354: Noyola DE, Zuviri-González A, Castro-García JA, Ochoa-Zavala JR. Impact of respiratory syncytial virus on hospital admissions in children younger than 3 years of age. J infect. 2007;54: Zambon MC, Stockton JD, Clewley JP, Fleming DM. Contribution of influenzalike illness: an observational study. Lancet. 2001;358; Nicholson KG, McNally T, Silverman M, Simons P, Stockton JD, Zambonc MC. Rates of hospitalisation for influenza, respiratory syncytial virus and human metapneumovirus among infants and young children. Vaccine. 2006;24: Acworth J, Babl F, Borland M, Ngo P, Krieser D. Patterns of presentation to the Australian and New Zealand Paediatric Emergency Research Network. Emerg Med Australasia. 2009;21: Wong CM, Yang L, Thach TQ, Chau PYK, Chan KP, Thomas GN, et al. Modification by influenza on health effects of air pollution in Hong Kong. Environ Health Perspect. 2009;117(2): Touloumi G, Samoli E, Quenel P, Paldy A, Anderson RH, Zmirou D, et al. Short-term effects of air pollution on total and cardiovascular mortality: the confounding effect of influenza epidemics. Epidemiology. 2005;16: Katsouyanni K, Touloumi G, Samoli E, Gryparis A, Le TA, Monopolis Y, et al Confounding and effect modification in the short-term effects of ambient particles on total mortality: results from 29 European cities within the APHEA2 project. Epidemiology. 2001;12: Statistics New Zealand. Population Mobility of Urban/rural Profile Areas [report on the Internet]. Wellington (NZ): Government of New Zealand; 2006 [cited 2009 Dec 21]. Available from: Kossmann M, Sturman AP. The surface wind field during winter smog nights in Christchurch and coastal Canterbury, New Zealand. International Journal of Climatology. 2004;24: McKendry IG, Sturman AP, Vergeiner J. Vertical profiles of particulate matter size distributions during winter domestic burning in Christchurch, New Zealand. Atmospheric Environment. 2004;38: Cavanagh JAE, Brown L, Trought K, Kingham S, Epton MJ. Elevated concentrations of 1-hydroxypyrene in schoolchildren during winter in Christchurch, New Zealand. Sci Total Environ. 2007;374: Fisher G, Kjellstrom T, Kingham S, Hales S, Shrestha R. Health and Air Pollution in New Zealand, Final Report [report on the Internet]. Wellington (NZ): Health Research Council, Ministry for the Environment and Ministry of Transport; 2007 [cited 2009 Jul 24]. Available from: Jennings L, Huang QS, Baker M, Bonne M, Galloway Y, Baker S. Influenza surveillance and immunisation in New Zealand New Zealand Public Health Report. 2001;8: Jennings LC, Anderson TP, Werno AM, Beynon KA, Murdoch DR. Viral Etiology of Lower Respiratory Tract Infections in Children Presenting to Hospital: Role of RCR and Demonstration of Multiple Infections. Pediatr Infect Dis J. 2004;23: Population and Environmental Health Group. Annual Summary of Outbreaks in New Zealand [report on the Internet]. Wellington (NZ): Institute of Environmental Science and Research; 2008 [cited 2009 Jul 23]. Available from: Moolgavkar SH, Luebeck EG, Anderson EL. Air pollution and hospital admissions for respiratory cases in Minneapolis St. Paul and Birmingham. Epidemiology. 1997;8: Luginaah IN, Fung KY, Gorey KM, Wills C. Association of ambient air pollution with respiratory hospitalization in a government designated area of concern : the case of Windsor, Ontario. Environ Health Perspect. 2005;113: Verbeek M. A Guide to Modern Economics. Hoboken (NJ): John Wiley & Son; Schwartz J. Air pollution and hospital admissions for respiratory disease. Epidemiology. 1996;7: Anderson HR, Bremner SA, Atkinson RW, Harrison RM, Walters S. Particulate matter ad daily mortality and hospital admissions in the west midlands conurbation of the United Kingdom: associations with fine and coarse particles, black smoke and sulphate. Occup Environ Med. 2001;58: Barnett GA, Williams GM, Schwarts J, Neller AH, Best TL, Petroeschevsky AL, et al. Air pollution and Child Respiratory Health. Am J Respir Crit Care Med. 2005;171: Breitner S, Stölzel M, Cyrys J, Pitz M, Wölke G, Kreyling W, et al. Short-Term Mortality Rates during a Decade of Improved Air Quality in Erfurt, Germany. Environ Health Perspect. 2009;117: Hagen JA, Nafstad P, Skrondal A, Bjørkly S, Magnus P. Associations between outdoor air pollutants and hospitalization for respiratory diseases. Epidemiology. 2000;11: Roemer W, Clench-Aas J, Englert N, Hoek G, Katsuyanni K, Pekkanen J, et al. Inhomogeneity in response to air pollution in European children (PEACE project). Occup Environ Med. 1999;56: Fusco D, Forastiere F, Michelozzi P, Spadea T, Ostro B, Arcà M, et al. Air pollution and hospital admissions for respiratory conditions in Rome, Italy. Eur Respir J. 2001;17: Bernstein J, Alexis N, Barnes C, Bernstein L, Nel A, Peden D, et al. Health effects of air pollution. J Allergy Clin Immunol. 2004:114; Kingham S, Durand M, Harrison, Cavanagh J, Epton M. Temporal variations in particulate exposure to wood smoke in a residential school environment. Atmospheric Environment. 2008:42; Ministry of Environment. Health Effects of CO, NO 2, SO 2, Ozone, Benzene and Benzo(a)pyrene in New Zealand. Air Quality Technical Report No.: 43. Wellington (NZ): Ministry of Enviornment; Simpson R, Williams G, Petroeschevsky A, Best T, Morgan G, Denison L, et al. The short-term effects of air pollution on daily mortality in four Australian cities. Aust N Z J Public Health. 2005;29 (3): Bibi H, Nutman A, Shoseyoc D, Shalom M, Peled R, Kivity S, et al. Prediction of emergency department visits for respiratory symptom using an artificial neural network. Chest. 2002;125(5): Fukuda K. Attribute Selection for Environmental Science Problems: Innovative Data Mining and Statistical Tools. Saarbrücken (DEU): VDM Verlag Dr. Müller Aktiengesellschaft & Co; vol. 35 no. 2 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 169