Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Air pollution and patients with implanted cardiac defibrillators : an epidemiological analysis and assessment… Rich, Kira 2002

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_2003-0292.pdf [ 17.2MB ]
Metadata
JSON: 831-1.0091086.json
JSON-LD: 831-1.0091086-ld.json
RDF/XML (Pretty): 831-1.0091086-rdf.xml
RDF/JSON: 831-1.0091086-rdf.json
Turtle: 831-1.0091086-turtle.txt
N-Triples: 831-1.0091086-rdf-ntriples.txt
Original Record: 831-1.0091086-source.json
Full Text
831-1.0091086-fulltext.txt
Citation
831-1.0091086.ris

Full Text

AIR POLLUTION A N D PATIENTS WITH IMPLANTED CARDIAC DEFIBRILLATORS: A N EPIDEMIOLOGICAL ANALYSIS A N D ASSESSMENT EXPOSURE by Kira Rich B.Sc, McGill University, 1999 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF M A S T E R OF SCIENCE in THE F A C U L T Y OF G R A D U A T E STUDIES School of Occupational and Environmental Hygiene We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH C O L U M B I A August 2002 © Kira Rich, 2002 UBC Rare Books and Special Collections - Thesis Authorisation Form Page 1 of 1 In p r e s e n t i n g t h i s t h e s i s i n p a r t i a l f u l f i l m e n t of the requirements f o r an advanced degree at the U n i v e r s i t y of B r i t i s h Columbia, I agree that the L i b r a r y s h a l l make i t f r e e l y a v a i l a b l e f o r reference and study. I f u r t h e r agree that permission f o r extensive copying of t h i s t h e s i s f o r s c h o l a r l y purposes may be granted by the head of my department or by h i s or her r e p r e s e n t a t i v e s . I t i s understood that copying or p u b l i c a t i o n of t h i s t h e s i s f o r f i n a n c i a l gain s h a l l not be allowed without my w r i t t e n permission. Department of The U n i v e r s i t y of B r i t i s h Columbia Vancouver, Canada Date A-pri I IM- |Q3 http ://www. library, ubc. ca/ spcoll/thesauth. html 4/14/2003 ABSTRACT This research extends previous investigations of adverse cardiovascular effects of particulate air pollutants to patients with implanted cardiac defibrillators (ICDs). Case crossover analyses were conducted for the February 14 to December 31, 2001 study period using data from ICDs worn by 34 Vancouver residents. Pollutant concentrations for days when ICD-detected arrhythmias were observed were compared to control day (± 7 days from ICD event days) concentrations. Although in general results were statistically non-significant, consistent trends suggested weak associations between summertime combustion-source primary air pollutants and cardiac arrhythmia. Odds ratios (OR) were higher in summer (7 of 9 >1) than in winter (1 of 9 >1) and ORs were highest for lag 0. For local combustion-source pollutants elemental and organic carbon, carbon monoxide and sulfur dioxide, ORs were above 1 at all lags (0-3 days) in summer. For summer and winter periods combined, results failed to indicate consistent associations between air pollution and ICD-detected arrhythmia. In the second component of the thesis, 19 patients with ICDs were monitored for exposure between May 15-August 31, 2001. For each subject, personal exposures to PM2.5 (particulate matter <2.5pm in aerodynamic diameter) mass concentration, filter optical absorbance (a surrogate for elemental carbon, a marker of traffic-related particles), and sulfate (a marker for regional ambient source particulate matter) were measured for 7 randomly selected 24-hour periods (>8 days apart). Supplementary exposure-related activity data was collected using time activity diaries. Ambient measurements of PM2.5, sulfate and absorbance were made at a single ambient monitoring location. Relationships between personal exposures and ambient concentrations were assessed to evaluate use of ambient measurements for exposure assessment. Median ambient and personal PM2.5 concentrations were 6.4ug/m3 and 13.5ug/m3, respectively, while median absorbance values were 10.8x10"5m"' and 7.2x10"5m"'. Median ambient and personal sulfate concentrations were 1.2ug/m3 and 1.15ug/m3. Median correlations (r) for individual personal versus ambient regressions for PM2.5, absorbance, and sulfate were 0.37, 0.50, and 0.85, respectively. The higher values for absorbance and sulfate are consistent with these components being associated with local and regional outdoor sources, respectively, whereas the lower correlation for PM2.5 reflects the impact of indoor sources on exposure. TABLE OF CONTENTS ABSTRACT ii TABLE OF CONTENTS iii LIST OF TABLES ix LIST OF FIGURES xi ACKNOWLEDGEMENTS xiv CHAPTER 1. INTRODUCTION 1 1.1 EPIDEMIOLOGICAL ANALYSIS 4 1.1.1 Implanted Cardiac Defibrillators 4 1.1.2 ICD population 5 1.1.3 Case crossover analysis 6 1.2 EXPOSURE ASSESSMENT 7 1.2.1 Particulate matter 7 1.2.2 Ambient concentrations 8 1.2.3 Particle measurement 8 1.2.4 Standards 8 1.2.5 Personal exposure monitoring 9 1.2.6 Relationships between personal and ambient concentrations 9 1.2.7 Source characterization 10 1.2.8 Sulfate 10 1.2.9 Absorbance 10 1.2.10 Study design 12 1.2.10.1 Hypotheses 12 1.2.10.2 Objectives 12 CHAPTER 2. METHODS 13 2.1 EPIDEMIOLOGICAL ANALYSIS 13 2.1.1 Study period 13 2.1.2 ICD Patient information 13 2.1.2.1 Data collection 13 2.1.3 Ambient pollutant and meteorological measurement 14 2.1.3.1 GVRD monitoring network 14 2.1.3.1.1 PM2.5 15 2.1.3.1.2 P M 1 0 15 2.1.3.1.3 Sulfate 15 2.1.3.1.4 Elemental and organic carbon 15 2.1.3.1.5 Carbon monoxide 15 2.1.3.1.6 Nitrogen dioxide 15 2.1.3.1.7 Sulfur dioxide 16 iii 2.1.3.1.8 Ozone 16 2.1.3.2 Meteorological data 16 2.1.4 Data Analysis 16 2.1.4.1 Case crossover analysis 16 2.1.4.1.1 Design of analysis 17 2.1.4.1.2 Grouping of events 17 2.1.4.1.3 Missing air pollutant data 17 2.1.4.1.4 Data preparation 18 2.1.4.1.5 Model variables 18 2.1.4.1.6 Lag periods 19 2.1.4.1.7 Season 20 2.1.4.2 Sensitivity Analyses 20 2.1.4.2.1 Meteorology 20 2.1.4.2.2 Grouping of ICD events 20 2.1.4.2.3 Inappropriate events 21 2.2 EXPOSURE ASSESSMENT 22 2.2.1 Study period 22 2.2.2 Personal exposure measurements 22 2.2.2.1 Study population - eligibility and recruitment 22 2.2.2.2 Subject identification and sampling schedules 23 2.2.3 Personal sampling equipment and forms 23 2.2.3.1 Personal exposure sampling for airborne particulate matter 23 2.2.3.2 Information about sources of pollutants 25 2.2.3.2.1 Time activity diaries 25 2.2.3.2.2 Dwelling information form 25 2.2.3.3 Lab preparation of personal samplers 26 2.2.4 Laboratory Analysis of personal filters 27 2.2.4.1 Gravimetric analysis 27 2.2.4.2 Quality control filters 27 2.2.4.3 Field and lab blanks 27 2.2.4.4 Maintenance of sampling equipment 28 2.2.5 Optical reflectance analysis 28 2.2.5.1 Reflectometer 28 2.2.5.2 Calibration 28 2.2.5.3 Linearity checks 28 2.2.5.4 Conversion to absorbance 29 2.2.6 Sulfate Analysis 29 2.2.6.1 Preparation 29 2.2.6.2 Procedure 29 2.2.7 Ambient Air pollutant measurements 29 2.2.7.1 Carbon measurements 30 2.2.8 Data Analysis 30 2.2.8.1 Particle sampling - data quality and descriptive statistics 30 2.2.8.1.1 Extreme values 30 2.2.8.1.2 Distribution of exposure variables 31 iv 2.2.8.2 Absorbance and various carbon measurements 31 2.2.8.3 Relationships between personal and ambient concentrations of the same exposure variable 31 2.2.8.4 Interrelationships between personal and ambient concentrations of different PM2.5 components 32 2.2.8.5 Composition of personal PM2.5 32 2.2.8.6 Predictors of personal exposures 32 2.2.8.6.1 Distance from monitoring station 32 2.2.8.6.2 Time activity diaries and dwelling information 32 2.2.8.7 Determinants of exposure modelling 33 2.2.8.7.1 Data preparation 33 2.2.8.7.2 Procedure for running models 33 2.2.8.7.3 Model fit 34 2.2.8.7.4 Sensitivity analyses 34 CHAPTER 3. RESULTS 35 3.1 EPIDEMIOLOGICAL ANALYSIS 35 3.1.1 Population Characteristics 35 3.1.2 Case Crossover Analysis 35 3.1.2.1 ICD discharges 35 3.1.2.2 Inappropriate ICD discharges 35 3.1.2.3 Model design 36 3.1.2.4 Model variables 37 3.1.2.5 Available data 37 3.1.2.6 Ambient air pollutants and meteorological variables 38 3.1.2.7 Model results 40 3.1.2.8 Stratification by season 42 3.1.3 Sensitivity Analyses 47 3.1.3.1 Sensitivity Analysis I: Meteorology 47 3.1.3.2 Sensitivity Analysis II: All Events 49 3.1.3.3 Sensitivity Analysis III: Appropriate events only 49 3.2 EXPOSURE ASSESSMENT 52 3.2.1 Study Subjects 52 3.2.1.1 Recruitment 52 3.2.1.2 Population characteristics 52 3.2.1.3 Compliance 53 3.2.2 Data Collection 53 3.2.2.1 Exclusion criteria 53 3.2.2.2 Sample duration 54 3.2.2.3 Quality control filters 54 3.2.3 Personal exposures to PM2.5, absorbance and sulfate 54 3.2.3.1 Exposure distributions 54 3.2.3.2 Personal P M 2 5 55 v 3.2.3.3 Personal absorbance 55 3.2.3.4. Personal sulfate 56 3.2.3.4.1 Data quality for sulfate analyses 57 3.2.4 Ambient Exposures 58 3.2.4.1 Missing ambient samples 58 3.2.4.2 Ambient P M 2 5 58 3.2.4.3 Ambient absorbance 58 3.2.4.4 Ambient sulfate 59 3.2.4.5 Distribution of ambient exposures 59 3.2.5 Absorbance and Various Carbon Measurements 60 3.2.5.1 Absorbance and Elemental Carbon 60 3.2.5.2 Absorbance and other carbon components 61 3.2.5.3 Laser Integrating Plate Method (LIPM) and carbon measures 63 3.2.6 Relationships between personal and ambient concentrations of the same exposure variable 65 3.2.6.1 All subjects combined 65 3.2.6.2 Individual subjects 67 3.2.6.3 Interrelationships between personal and ambient concentrations of different components of PM2.5 69 3.2.6.4 Composition of Personal PM2.5 71 3.2.7 Predictors of personal exposure 72 3.2.7.1 Distance from monitoring station ....72 3.2.7.2 Time Activity Data and Dwelling Information 72 3.2.7.2.1 Time Activity Diaries 72 3.2.7.2.2 Dwelling Information Form 74 3.2.7.3 Multiple Regression 75 3.2.7.3.1 Description of Models 75 3.2.7.3.2 Missing ambient pollutant data 76 3.2.7.3.3 Excluded variables 76 3.2.7.3.4 Categorical variables 76 3.2.7.3.5 Continuous variables 77 3.2.7.3.6 Intercorrelation of independent variables 78 3.2.7.3.7 Procedure for running models 79 3.2.7.3.8 OUTDOORS variable 79 3.2.7.3.9 Model results 80 3.2.7.3.10 Model fit 81 3.2.7.4 Sensitivity Anlyses 81 CHAPTER 4. DISCUSSION 83 4.1 EPIDEMIOLOGICAL ANALYSIS 83 4.1.1 Case crossover 83 4.1.2 Lag periods 84 4.1.3 Previous studies 84 4.1.4 Season 85 4.1.5 Sample size 86 vi 4.1.6 Strengths 86 4.1.7 Limitations 87 4.1.8 Suggestions for further research 90 4.2 EXPOSURE ASSESSMENT 91 4.2.1 Personal monitoring 91 4.2.2 Personal and ambient exposures 91 4.2.3 Absorbance as an exposure metric 91 4.2.4 Relationships between personal and ambient concentration 92 4.2.5 Personal to ambient concentration ratios 93 4.2.6 Interrelationships between exposure variables 93 4.2.7 PM25composition 94 4.2.8 Distance to ambient monitoring station 95 4.2.9 Multiple Regression 95 4.2.10 Strengths and limitations 96 4.2.11 Conclusions 97 REFERENCES 98 APPENDICES 106 APPENDIX 2.1a-b Recruitment letters 106 APPENDIX 2.2 Consent form 108 APPENDIX 2.3 Flow Log 110 APPENDIX 2.4 Time Activity Diary I l l APPENDIX 2.5 Dwelling Information Form 113 APPENDIX 3.1a-i Distribution of ambient pollutants during study period 115 APPENDIX 3.1j-m Distribution of meteorological variables during study period 124 APPENDIX 3.2 Results of case crossover analyses after stratification by season 129 APPENDIX 3.3 Results of case crossover analysis sensitivity analysis 1 131 APPENDIX 3.4 Results of case crossover analysis sensitivity analysis II 132 APPENDIX 3.5 Results of case crossover analysis sensitivity analysis III 133 APPENDIX 3.6a Quality control filter warning and control limits -personal filter KR-W1 134 APPENDIX 3.6b Quality control fdter warning and control limits -personal filter KR-W2 135 APPENDIX 3.6c Quality control fdter warning and control limits -personal fdter KR-W3 136 APPENDIX 3.7 Histograms showing distributions of untransformed and natural log-transformed personal exposure variables 137 APPENDIX 3.8 Comparison of sulfate analyses of 11 randomly selected samples repeated at Environment Canada and SOEH laboratories 138 APPENDIX 3.9 Histograms showing distributions of untransformed and natural log-transformed ambient exposure variables 139 APPENDIX 3.10 Comparisons of LIPM BC and two carbon fractions at the Slocan monitoring station 140 APPENDIX 3.1 la-c Distribution of ambient pollutants during exposure assessment 141 APPENDIX 3.12a-e Boxplots of continuous vs. categorical variables 144 APPENDIX 3.13 Crosstabulations between all categorical variables included in multiple regression 154 V l l l LIST OF TABLES Table 3 a. Inappropriate ICD discharges during the study period 36 Table 3b. Cases and controls included in case crossover analyses 38 Table 3c. Concentrations of ambient air pollutants and meteorological variables considered in case crossover analyses observed during the study period 39 Table 3d. Pearson correlations between all modeled air pollutants and meteorological variables 39 Table 3e. Results of initial case crossover analyses 40 Table 3f. Cases and controls included in case crossover analyses stratified by season 42 Table 3g. 75th percentile of pollutant concentrations for case and control periods used in season-stratified analyses 43 Table 3h. Warning (mean ± 2 SD) and control (mean ± 3 SD) limits for Quality Control Filters 54 Table 3i. Distribution of Personal PM2.5 Concentrations 55 Table 3j. Distribution of Personal Absorbance Values 56 Table 3k. Distribution of Personal Sulfate Concentrations 57 Table 31. Ambient PM2.5, absorbance and sulfate 59 Table 3m. Pearson correlations for individual regressions of personal versus ambient PM2.5, absorbance and sulfate 67 Table 3n. Correlation coefficients for univariate regressions between all exposure variables for all subjects combined 69 Table 3o. Correlation coefficients for univariate regressions between all exposure variables, by individual subject 71 Table 3p. Results of univariate regressions of personal to ambient Pearson r values against distance to ambient monitoring station... 72 Table 3q. Summary statistics of time activity information 73 Table 3r. Summary statistics of dwelling information 74 ix Table 3s. Dwelling variables not considered for inclusion in models 76 Table 3t. All independent variables introduced to models 77 Table 3u. Results of univariate regressions between dependent and independent variables considered for inclusion in models 78 Table 3v. Pearson correlations between all continuous variables considered for inclusion in models 78 Table 3w. Hypothesized direction of relationship with personal exposure variables 79 Table 3x. Results of P M 2 5 model run separately using INDOORS and OUTDOORS variables 80 Table 3y. Results of final mixed effects linear regression models 80 Table 3z. Equations of final mixed effects models 81 Table 3aa. R2 values predicted vs. observed values of dependent variables 81 Table 3ab. Variables to fall out of sensitivity analyses 82 Table 4a. Air pollutants considered in Boston and Vancouver studies of ICD patients 84 APPENDIX 3.2 Results of case crossover analyses after stratification by season 129 APPENDIX 3.3 Results of case crossover analysis sensitivity analysis 1 131 APPENDIX 3.4 Results of case crossover analysis sensitivity analysis II 132 APPENDIX 3.5 Results of case crossover analysis sensitivity analysis III 133 APPENDIX 3.8 Comparison of sulfate analyses of 11 randomly selected samples repeated at Environment Canada and SOEH laboratories 138 APPENDIX 3.13 Crosstabulations between all categorical variables included in multiple regression 154 LIST OF FIGURES Figure 1.1. Diagram of discharge event showing electrocardiogram during cardiac arrhythmia, shock, and regular heart rhythm 5 Figure 2.1. Map of GVRD ambient pollutant monitoring stations 14 Figure 2.2. Schematic of ambidirectional case crossover design 17 Figure 2.3. Proposed induction periods separating exposure to increased air pollution from ICD discharge 19 Figure 2.4. Ambidirectional case crossover design showing lag periods of 0, 1,2 and 3 days 20 Figure 2.5. Two study subjects during sampling sessions 24 Figure 2.6. PEM sampler assembly 26 Figure 3.1. ICD discharges over the study duration for each patient in the study sample ....36 Figure 3.2. All ICD event days considered in analyses 37 Figure 3.3. Results of initial case crossover analyses including discharges from all months 41 Figure 3.4a. Odds Ratios showing 95% confidence intervals for case crossover analyses stratified by season (PM2.5, PM10, SO4) 44 Figure 3.4b. Odds Ratios showing 95% confidence intervals for case crossover analyses stratified by season (EC, OC, CO) 45 Figure 3.4c. Odds Ratios showing 95% confidence intervals for case crossover analyses stratified by season (N0 2, 0 3 , S02) 46 Figure 3.5. Results of sensitivity analyses investigating the effect of the inclusion of meteorology 48 Figure 3.6. Odds Ratios showing 95% confidence intervals for case crossover analyses of pollutants using all events and the first event of a group, respectively 50 Figure 3.7. Odds Ratios showing 95% confidence intervals for case crossover analyses before and after the exclusion of ICD events considered inappropriate 51 Figure 3.8 Map of Study Area located in Vancouver, Canada 53 xi Figure 3.9. Comparison of sulfate analyses at Environment Canada and SOEH laboratories 58 Figure 3.10a. Relationship of absorbance and EC at South Burnaby monitoring station for the period 05/15/01 to 08/31/01 60 Figure 3.10b. Relationship of absorbance and EC at Slocan monitoring station for the period 08/13/01 to 08/31/01 61 Figure 3.10c. Relationship of absorbance and CC at South Burnaby monitoring station for the period 05/15/01 to 08/31/01 62 Figure 3.10d. Relationship of absorbance and CC at Slocan monitoring station for the period 08/13/01 to 08/31/01 62 Figure 3.10e. Relationship of absorbance and E C at South Burnaby for the period 05/15/01 to 08/31/01 63 Figure 3.1 Of. Relationship of absorbance and E C at Slocan for the period 08/13/01 to 08/31/01 63 Figure 3.10g. Relationship of absorbance and a black carbon indicator at Slocan for the period 08/13/01 to 08/31/01 64 Figure 3.10h. Relationship of LIPM BC and EC at Slocan for the period 08/13/01 to 08/31/01 : 64 Figure 3.11a. Ambient and personal PM2.5 concentrations during the study period 65 Figure 3.1 lb. Ambient and personal absorbance values during the study period 66 Figure 3.11c. Ambient and personal sulfate values during the study period 66 Figure 3.12. Histograms of Pearson r values for individual personakambient regressions...68 Figure 3.13. Univariate regressions between ambient PM2.5 and all measured personal exposure variables 70 Figure 3.14. Location of subjects during all sampling sessions 73 Figure 3.15. Summary of exposure to tobacco smoke for all samples 74 Figure 3.16. Distribution of results for the attached garage and range hood dwelling variables 75 Figure 4.1. Observed interrelationships between personal and ambient exposure variables 93 xn APPENDIX 3.1a-i Distribution of ambient pollutants during study period 115 APPENDIX 3.1j-m Distribution of meteorological variables during study period 124 APPENDIX 3.6a Quality control filter warning and control limits -personal filter KR-W1 134 APPENDIX 3.6b Quality control filter warning and control limits -personal filter KR-W2 135 APPENDIX 3.6c Quality control filter warning and control limits -personal filter KR-W3 136 APPENDIX 3.7 Histograms showing distributions of untransformed and natural log-transformed personal exposure variables 137 APPENDIX 3.9 Histograms showing distributions of untransformed and natural log-transformed ambient exposure variables 139 APPENDIX 3.10 Comparisons of LIPM BC and two carbon fractions at the Slocan monitoring station 140 APPENDIX 3.11 Distribution of ambient pollutants during exposure assessment 141 APPENDIX 3.12a-e Boxplots of continuous vs. categorical variables 144 xm ACKNOWLEDGEMENTS Only as this thesis came to a close did it occur to me that I had learned an incredible amount during the course of its preparation. This project as well as the benefit it has provided me would not have been possible without the participation of the 19 patients who generously volunteered so much of their time to the cause. Tt was a joy to get to know them. I would like to thank my family as well as Sanj for providing me with support and encouragement from start to finish. I would also like to thank the students from adjacent cubicles who, over the course of many late nights and bad meals, have become great friends. Although I spent what seems like a remarkable amount of time at SOEH, I was lucky to have almost constant company. I am extremely grateful to all of the SOEH professors as well as my committee members for always being approachable and willing to help. Finally, I would like to thank my supervisor Mike Brauer for his excellent guidance throughout this project. Thank you for always expecting more of me than I thought I could accomplish, for trusting I could do what I set out to, and for being honest in every situation. It was a pleasure to work with you. xiv 1.0 INTRODUCTION Chapter 1 Historically, studies of air pollution and health have focused on severe episodes of air pollution such as those observed in London in 1952. Early on, studies revealed that a large proportion of deaths and morbidity were due to cardiorespiratory ailments. Cardiovascular disease is an important area of focus for research on the effects of air pollution since it accounts for a large proportion of daily mortality (Levy et al, 2001). Ensuing studies have revealed that cardiovascular and respiratory morbidity and mortality also occurred at lower levels of air pollution. In the past decade, a body of epidemiologic evidence showing increased cardiovascular mortality and morbidity associated with exposure to current levels of air pollution has been established. Associations have been found with cardiovascular deaths as a whole as well as deaths from myocardial infarction and ventricular fibrillation (Dockery, 2001). Studies have revealed effects of air pollution exposure on both acute and long-term heart health (Herbert et al, 2000). Some of these findings are from large cohort studies. Dockery et al (1993), for example, reported that air pollution in six U.S. cities was positively associated with lung cancer and cardiopulmonary disease but not with death from other causes considered together. Pope et al (1995) identified an association between exposure to particulate air pollution at levels found in US cities and increased cardiopulmonary mortality. Most recently, Pope et al (2002) presented the results of a prospective mortality study that enrolled approximately 1.2 million American adults in 1982, concluding that long-term exposure to combustion-related fine particulate air pollution increases risk of all-cause, cardiopulmonary, and lung cancer mortality. Many other studies, most of them of the time-series design, have revealed that overall daily mortality increases as the concentration of particles in the air rises (Schwartz, 1994). Epidemiologic studies revealing associations between ambient particulate matter and cardiovascular effects have prompted interest in biological mechanisms that may mediate these associations. The biological plausibility of an effect of air pollution on cardiac health is supported by associations with cardiopulmonary health effects in epidemiological studies and by the fact that non-cardiopulmonary health effects have not often been associated with air pollution (Wilson and Spengler, 1996). The observation of stronger associations between air pollution and mortality in patients with heart disease than in the general population reinforces the evidence that a harmful effect of air pollution is mediated by cardiovascular mechanisms (Kwon et al, 2001). However, biological mechanisms that may mediate associations between air pollution exposure and cardiac effects have not been firmly established. It has been hypothesized that pulmonary inflammation due to deposition of ultrafine particles leads to changes in the blood including increased plasma viscosity, which increases the likelihood of myocardial infarction in those at risk (Seaton et al, 1995). Increases in plasma viscosity, which is determined largely by plasma fibrinogen concentration, and in C-reactive protein, an index of inflammation, have been observed in 1 randomly selected healthy adults in association with episodes of high particulate air pollution (Peters et al, 1997). Other studies have focused on a number of symptoms hypothesized to signify altered autonomic nervous system function. Particulate air pollution exposure has been associated with indicators of autonomic control of the heart including increased heart rate, decreased heart rate variability (HRV), and increased cardiac arrhythmias (Dockery, 2001). Pulse rate changes have been observed in association with exposure of elderly patients to PM (Pope et al, 1999). Systolic blood pressure has also been seen to increase in association with ambient particles (Ibald-Mulli et al, 2001). Exposure to PM2.5 was associated with decreased HRV in young workers in a recent study (Magari et al, 2001). Acceleration of heart rate and decreased HRV have also been documented in elderly persons exposed to PM. Associations between PM2.5 and decreased HRV were revealed by Liao et al (1999) and by Gold et al (2000), while Pope et al (1999a) observed decreased HRV in association with PM10. Since low HRV, a marker of poor autonomic control, is associated with higher risk of myocardial infarction and sudden cardiac death, these findings suggest a possible link between PM and cardiovascular disease mortality (Liao et al, 1999). Dozens of time series studies have revealed relationships between air pollutant exposure and cardiac morbidity and mortality and have produced consistent results in a variety of settings. However, these studies have tended to be subject to a number of shortcomings. First, many time series studies of air pollution and health effects have been ecologic in design, and have lacked individual level risk information such as smoking status or occupation. It may be especially important to include such individual level information in air pollution studies, since relative risks tend to be low across populations. Although fixed characteristics can be controlled for in time series studies, omitted factors that vary in time and influence risk can cause confounding. Further, if health outcomes are autocorrelated in time, it may be inappropriate to treat them independently (Vedal, 1997). Another limitation of previous time series studies is that disease outcomes have often been non-specific, limiting the interpretation of results. In addition, studies may have been confounded by copollutants and other putative disease risk factors such as meteorological variables (Levy et al, 2001). Further, while clinical studies have addressed the lack of specificity of outcomes by investigating detailed outcome measures in small populations, these studies have also had shortcomings. Because they tend to be labour intensive, these studies focus on small populations and lack continuous outcome information. Holter monitoring studies, for example, investigate outcomes for short periods of minutes to days, and the results may not be generalizeable to longer periods. The hypothesis that episodes of arrhythmia are associated with transient increases in air pollution is consistent with epidemiologic observations of increased cardiovascular mortality and morbidity on days of increased air pollution. It is also supported by studies of animals (Watkinson et al, 1998) and humans revealing heart rhythm disturbances in response to controlled exposure to air pollutants. Brauer et al (2001) observed associations between personal exposure to particles and increased supraventricular 2 ectopic heartbeats in patients with chronic obstructive pulmonary disease. Cardiac arrhythmia is of particular interest because' it mediates sudden cardiac death (Wannamethee et al, 1995). Most sudden cardiac deaths are arrhythmically mediated, generally secondary to ventricular tachycardia (rapid heart rate), ventricular fibrillation (irregular heart rate), or both (Peters and Gold, 2001). The excess relative risk of death associated with air pollution due to heart failure and arrhythmia have been revealed to be high relative to that for other cardiovascular causes of death (Hoek et al, 2001). Since risk estimates for health effects associated with air pollution exposure have traditionally been low at urban air pollutant concentrations, it is important to maximize both sensitivity and specificity of outcome information. It has been hypothesized that a population of cardiac patients wearing implanted cardiac defibrillators (ICDs) would be an ideal resource for an epidemiological analysis of air pollution and cardiac arrhythmia. Not only do they provide an objective record of arrhythmias in time, but they represent a population likely to be susceptible to cardiac effects of air pollution. The study of a susceptible population is thought to increase the potential for observation of any effect. Arrhythmias treated by ICDs are symptomatic of a variety of cardiac disorders including cardiomyopathies, coronary artery disease, and valve defects. Patients are fitted with ICDs to treat life-threatening cardiac arrhythmias. The study of ICD patients provides a convenient method of identification of cardiac rhythm disturbances. Only two other studies of ICD shock and air pollution have been described in the literature. Set in Boston, the first study of this kind investigated 100 ICD patients, 33 of whom experienced one or more ICD shocks during the study period. This study reported a statistically significant association between ICD shock and ambient NO2 concentration 2 days before the event (Peters et al, 2000). The other study, a case crossover analysis also set in Boston presently published in abstract form only, investigated 320 ICD patients, of which 121 experienced ICD shock during the 5 year study period. The authors present preliminary results indicating significant associations between ICD shock and a 9.2 ug/m3 increase in PM2.5 in the hour before the event as well as a 5 ppb increase in S0 2 in the hour before the event. They also report a nearly significant association between ICD shock and a 20 ppb increase in O 3 in the preceding 5 hours (Rich DQ et al, 2002). 3 1.1. EPIDEMIOLOGICAL ANALYSIS The first part of this study was an epidemiological analysis of the association between air pollution and cardiac arrhythmia in patients with ICDs. A case crossover approach was used to analyze the data. 1.1.1 Implanted Cardiac Defibrillators ICDs are small devices implanted in the chests of cardiac patients with a history of cardiac arrhythmia. An ICD consists of a sensor, a lead to the heart, and a battery. The sensor monitors the heart continuously, detecting rhythm changes. The heart rhythm is reset when arrhythmia reaches a preset threshold considered to be life-threatening. Determination of a suitable therapy for an arrhythmia is based on heart rate as well as duration of rhythm disturbance. When rate and duration of arrhythmia reach the predetermined threshold, the ICD delivers either antitachycardia pacing or shocks to halt the rhythm disturbance. Antitachycardia pacing consists of short bursts of pacing impulses at rates 10-20% higher than the tachycardia, which can stop most episodes of arrhythmia. Shocks terminate the arrhythmia by resetting the heart rhythm completely. The timing and nature of all arrhythmias and corresponding therapies are stored in the device until it is reset or the battery wears out after several years (Peters and Gold, 2001). ICDs are designed to treat ventricular tachycardia and fibrillation, since these are the most dangerous cardiac arrhythmias. Slower ventricular tachycardias are treated with antitachycardia pacing or low-energy shocks when pacing fails, while more severe ventricular tachycardias and ventricular fibrillation are treated with high-energy shocks. Maximum shock energies range from 25J - 42J in modern ICDs. Inappropriate shocks may be delivered when non-life threatening supraventricular arrhythmias (such as atrial fibrillation or sinus arrhythmia) are detected inappropriately as ventricular tachycardia. Modern ICD devices are increasingly sensitive to the nature of arrhythmias, decreasing the incidence of inappropriate treatments. In modern ICDs, stored information about heart rhythm as well as pacing and shock therapies delivered can be retrieved non-invasively by telemetry. The downloaded information reveals all therapies given, and indicates whether a shock has been delivered inappropriately in response to a non-life threatening supraventricular arrhythmia. The arrhythmia and shock information is downloaded upon regular visits to the cardiologist. (Glikson and Friedman, 2001) An example of the printout associated with an arrhythmia treated by ICD shock, or shock, is presented in Figure 1.1 below. The example follows the electrical activity of the heart in time through a period of ventricular fibrillation followed by a 26 J shock delivered by the ICD. A slower, more regular heart rhythm is observed following the shock therapy. 4 Figure 1.1. Diagram of shock event showing electrocardiogram during cardiac arrhythmia, antitachycardia pacing, shock, and regular heart rhythm 1.1.2 ICD population This is a unique population in that detailed, time-specific individual-level information is automatically collected over a relatively long time period for ICD patients. It differs from previous studies linking air pollution to cardiovascular morbidity and mortality in that studies have traditionally lacked specific information about outcomes. It differs again from other studies that have collected specific health outcome information in that the ICD acts as a passive data recorder. Therefore data collection is objective and large amounts of data may be collected with ease. Further, because downloads of ICD shocks occur retrospectively and with routine visits to the doctor, subjects will not change their behaviour as a result of participation in the study. Another advantage over some experimental studies is that health effects of realistic pollutant mixtures are investigated for this population. While some chamber studies have investigated the effects of concentrated mixtures of ambient pollutants (Brook et al, 2002), other laboratory-based studies that have used controlled human exposures to investigate acute health outcomes have focussed on particular components of the air pollution mix that may not approximate reality. 1.1.3 Case crossover analysis The case crossover approach to the analysis of event data was designed to investigate rare events associated with transient increases in risk. According to this approach, each case 'crosses over' between periods of different risk, and levels of risk are compared during case and control periods. Each individual acts as both case and control. Case periods are periods during which events are observed, and control periods are situated at times during which no case is observed, often at fixed interval(s) from cases. The case crossover analysis offers a number of advantages for the study of acute health effects associated with air pollution. The design avoids control-selection bias since cases act as their own controls, increasing efficiency. The use of cases as their own controls also circumvents 5 all confounding associated with fixed subject characteristics. Further, numerous control days can be compared to each case day, increasing the power of studies with small sample sizes. The use of ambidirectional control periods can also minimize bias associated with persistent trends in risk factors, while situation of cases and controls on the same weekday can control for potential day of the week effects. If case and control periods are selected effectively, biases due to autocorrelation of risk factors in time and seasonal trends may also be avoided. 6 1.2. EXPOSURE ASSESSMENT While the focus of the first part of the this study was on epidemiological relationships and proposed mechanisms of action between air pollution and heart health, the focus of the second part was to investigate questions regarding human exposure to particles of ambient origin. Specifically, the second part of this study was an assessment of exposure using personal monitoring of a subset of the ICD patient population from which subjects in the epidemiological analysis were drawn. These patients were monitored for personal exposure to PM2.5, PM2.5 filter absorbance and particle sulfate. 1.2.1 Particulate Matter Particulate matter is defined as all particles found in the air including dust, dirt, soot, and secondary acidic and organic aerosols (CCME, 2002). Unlike other air pollutants, particulate matter is not a specific chemical entity but a mixture of particles of different sizes, compositions and properties. Particle size is expressed in terms of aerodynamic diameter, defined as the diameter of a unit density sphere that has the same settling velocity (Hinds, 1982). PM2.5 refers to particulate matter with aerodynamic diameter less than 2.5 um. This fraction contains particles known as ultrafme (0.01-0.1 um) and fine (0.1-2.5 um). Although it contains both fine and ultrafme particles, PM2.5 is often referred to simply as the f i ne fraction. Particles in the fine fraction have long atmospheric residence times of days to weeks, and can be transported over long ranges of 100s to 1000s of kilometers (Wilson and Spengler, 1996). Studies have revealed that ambient PM2.5 is distributed relatively uniformly over space, suggesting that concentrations measured at a single monitoring site are able to characterize particle concentrations over urban areas well (Burton et al, 1996). In fact, a personal monitoring study conducted in Vancouver revealed that the use of PM2.5 measurements from five ambient monitoring stations did not improve estimates of exposure over measurements from the closest station, suggesting that the spatial variability of PM2.5 concentrations was minimal (Ebelt et al, 2000). The same may not be true for specific components of PM2.5, however. For example, a study conducted in three areas of Europe revealed that spatial contrasts were larger for the absorption coefficient of PM2.5, an indicator of traffic-source pollution, than for PM2.5 itself (Hoek et al, 2002). For airborne particles, the respiratory system is considered the point of contact, and the heart and lung are the target organs of concern (U.S. EPA, 1999). Behaviour of particles in the human respiratory tract is determined by size, shape, density and reactivity. A large component of PM2.5 is inhaled to the alveolar region of the lung. As evidence of this, Churg and Brauer (1997) observed that 96% of particles retained in autopsy tissue lay in the PM2.5 size range. PM2.5 is also of interest because sources of these particles are mainly anthropogenic. It is formed during combustion processes including those associated with motor vehicles, power plants and wood burning. This is important both because anthropogenic processes may increase the toxicity of particles and because, from 7 a regulatory perspective, particles generated by humans may be most easily controlled. The main chemical constituents of fine particulate matter include sulfate, nitrate, ammonium ion, elemental and organic carbon, and a variety of trace metals (Wilson and Spengler, 1996). 1.2.2 Ambient concentrations Ambient concentrations of PM2.5 vary considerably across geographic locations. Differences in local sources impact both the concentration and composition of particles. Since fine particles are produced primarily through combustion, concentrations are higher in urban areas and industrial regions. Ambient PM2.5 concentrations in Vancouver are low compared to other urban areas. A study of PM2.5 concentrations determined at 14 urban monitoring stations across Canada between 1986 and 1994 revealed that Vancouver levels were slightly above the average of all sites (15.6 vs 14.1 ug/m3). They were lower, however, than concentrations observed in Montreal (15.9-20.9 ug/m3), Toronto (16.8 ug/m3), and Windsor (16.8-18.1 ug/m3). Further, all cities with lower average PM2.5 concentrations were less populous than Vancouver (Brook et al, 1997). A report comparing Vancouver to 10 western U.S. cities of comparable population revealed that, for the period 1994-1998 inclusive, annual average Vancouver PM10 concentrations were lower than those of any other city. While the average PM10 concentration was 14.0 ug/m3 in Vancouver, the average of mean PM10 concentrations in the 10 other cities considered was 24.5 ug/m3 (range 18.6 - 41.2) (Brauer et al, 2000). 1.2.3 Particle measurement Consistent with the importance of particle aerodynamic diameter to particle behaviour and properties as well as effects on human health, sampling methods are designed to select particles according to size. The standard technique for determination of particle concentration involves gravimetric analysis of filters onto which particles collect while a measured volume of air is drawn through a size selective sampling head. This results in a size-specific particle concentration, conventionally reported in ug/m3. 1.2.4 Standards Regulations of particulate matter are based on mass as a function of aerodynamic size. Traditionally, investigations of the effects of particulate matter on health focused on all inhalable particles, known as Total Suspended Particulate (TSP), and PM10, or particles with aerodynamic diameter <10 um. In the past two decades, epidemiological studies have slowly changed their focus to PM2.5, as studies have revealed that PM2.5 or a component thereof may be responsible for the majority of observed health effects (Schwartz et al, 1996). In Canada, the Canadian Environmental Protection Act (CEPA) directs the regulation of environmental contaminants. Ambient particulate matter concentrations have traditionally been regulated by National Ambient Air Quality Objectives (NAAQO) set in accordance with the CEPA. The NAAQO include an objective concentration of TSP with no distinction according to particle size. In June of 2000, concerns about health and environmental risks of fine particles prompted the introduction of a Canada Wide Standard (CWS) for PM2.5. This 30 ug/m3 standard 8 represents a 24-hour average target concentration to be attained by the year 2010 (Health Canada, 2002). The United States has National Ambient Air Quality Standards regulated by the Environmental Protection Agency (EPA). The EPA has been regulating criteria air pollutants since the 1970 Clean Air Act (CAA) was passed. Until recently, EPA regulations considered only TSP and PMio. In 1997, however, standards were added to regulate PM2.5. The current standard requires that the 3-year average of the annual arithmetic mean of 24-hour PM2.5 concentrations not exceed 15 ug/m3 (U.S. EPA, 1999). 1.2.5 Personal exposure monitoring Personal particle exposure studies have been conducted on both healthy and health-compromised individuals. Personal exposure monitoring is an important tool for exploring relationships between individual exposures and ambient pollutant concentrations. An understanding of these relationships is necessary to validate epidemiological studies of air pollution and health, which rely on ambient monitors to estimate personal exposures. Also, since regulatory policy is most useful for limiting exposure to particles originating outdoors, it is important to understand the contribution of particles originating outside to personal exposure and health impacts. 1.2.6 Relationships between personal and ambient concentrations Relationships between personal and ambient concentrations of a number of pollutants have been elucidated in previous studies to reveal whether or not the use of ambient concentrations as estimates of personal exposure will result in exposure misclassification. A number of factors contribute to the potential for exposure misclassification. First, in areas where pollution sources are local, the distribution of ambient concentrations will vary spatially. The inability to capture local peaks may be particularly important if they contribute to health outcomes. Second, even when particles are distributed evenly outdoors, they may not penetrate indoors effectively or consistently for different building types. Third, indoor sources differ according to dwelling characteristics as well as the location and behaviour of residents. A 'personal cloud' associated with personal activity has been observed in many studies. These factors complicate the relationship between personal and ambient measures. While this is an important initial question, however, investigation of relationships between personal and ambient concentrations do not provide a complete understanding of relevant exposures. While poor personal to ambient correlations indicate that ambient and personal levels are quantitatively different, they do not indicate whether there are qualitative differences between ambient and personal pollutants that may affect exposure or health. In the case of personal exposure to PM2.5, studies have reported a range of results. While several have reported high longitudinal correlations between personal and ambient exposure to PM2.5, (Janssen et al, 1999; Williams et al, 2000) a number of other studies have revealed that ambient concentrations of PM2.5 do not approximate personal 9 exposures (Ebelt et al, 2000; Monn et al, 1997; Koistinen et al, 2001; Rodes et al, 2001). Ebelt et al (2000) revealed personal sulfate levels to be highly related to ambient PM2.5 in a Vancouver study of subjects with chronic obstructive pulmonary disease. The sulfate component of PM2.5 was more highly correlated with ambient PM2.5 than was personal PM2.5 itself, suggesting that sulfate may be a more appropriate exposure metric for personal exposure to ambient source PM2.5. This study will investigate the correlation of personal sulfate with ambient PM2.5 in a different population. Further, personal absorbance will be investigated as a possible exposure metric for ambient PM2.5 of local origin. 1.2.7 Source characterization Sulfate and elemental carbon are two of the predominant inorganic species of particulate matter. These two components of PM2.5 are investigated in this study for two major reasons. First, since each component has been linked to specific sources, measurement of personal concentrations of these components may reveal information regarding particle source. Second, it is important to determine whether personal exposure to either component is highly related to variation in ambient exposure to PM2.5, since such correlation would lend support to time-series epidemiological studies revealing associations between ambient PM2.5 and health outcomes. 1.2.8 Sulfate The average aerodynamic diameter of a sulfate particle is 0.48 um. Sulfate therefore exists almost exclusively as a component of PM2.5 mass. SO2 is the main precursor of sulfates, which exist as several atmospheric acid and neutralized sulfate species. S0 2 is emitted from the combustion of coal, gasoline, diesel and wood (Wilson and Spengler, 1996), and is oxidized to sulfate via sulfuric acid. Since SO2 is oxidized rather slowly, and sulfate particles are in the submicron size range, sulfate concentrations tend to be relatively constant over space (Suh et al, 1995). Submicron particles such as sulfate are also slow to deposit, are not subject to resuspension by personal activity, and have no major indoor sources. Because of the low spatial variability of ambient sulfate, and the lack of indoor sources, the correlation of personal and ambient concentrations tends to be high, making sulfate a good indicator of ambient source PM2.5. Further, particulate sulfate is found to be chemically and physically stable on filters, allowing accurate concentrations to be determined (Thatcher and Layton, 1995). A high indoor/outdoor ratio has repeatedly been observed for sulfate. As a result, this component of PM2.5 may be used as a source indicator of the proportion of PM2.5 originating outdoors. 1.2.9 Absorbance Absorbance is derived from optical reflectance, a measure of the light scattering properties of PM2.5 collected on a fdter. Absorbance is measured as a surrogate of elemental carbon (EC), since the two have been highly correlated in previous studies (Kinney et al, 2000). Elemental carbon, also known as black carbon (BC), is a component of PM2.5 associated with diesel exhaust particles. Although diesel exhaust 10 particles contain various types of carbon, elemental carbon is considered a superior marker of diesel particulate matter because the diesel engine is its major source (Birch and Cary, 1996). Diesel particles consist of a core of elemental carbon on which organic compounds are adsorbed. Typically more than 90% of the particle mass of diesel particulate is less than 2.5 um in diameter (Levelton Engineering Ltd. et al, 2000). Therefore, diesel exhaust particles exist almost entirely as a component of PM2.5. EC has also been found to have relatively consistent concentrations indoors and outdoors (Wheeler et al, 2002), indicating that this measure may also be used as a source indicator of PM2.5 originating outdoors. Whereas sulfate is an indicator of regional outdoor-source PM2.5, EC may be an indicator of outdoor-source PM2.5 of local origin. 11 1.2.10 Study design Due to the questions and issues described above, we designed a study to address the hypotheses and meet the objectives outlined below. 1.2.10.1 Hypotheses 1. Ambient air pollutant concentrations are associated with cardiac arrhythmia in patients with implanted cardiac defibrillators. 2. Using ambient measures of PM2.5 mass as surrogates for personal exposure results in greater misclassification of exposure for PM2.5 mass than for sulfate or elemental carbon. 1.2.10.2 Objectives a) To determine whether incidence of cardiac arrhythmia is associated in time with increases in ambient concentrations of specific pollutants. b) To take repeated measures of personal exposure to PM2.5, absorbance and sulfate in a population of patients with implanted cardiac defibrillators. c) To determine relationships between personal and ambient measures of PM2.5, absorbance and sulfate over the sampling period. d) To evaluate the utility of absorbance as a surrogate for elemental carbon. e) To determine the utility of time activity and location information and ambient concentrations as predictors of personal exposure to PM2.5, absorbance and sulfate. 12 2.0 METHODS Chapter 2 Overview There were two parts to this study. For both parts, the study population consisted of patients with implanted cardiac defibrillators (ICDs) seen at the only two referral ICD clinics in British Columbia 2.1 EPIDEMIOLOGICAL ANALYSIS The first part of the study was a retrospective epidemiological analysis of cardiac arrhythmia and ambient air pollution. Hourly concentrations of pollutants were obtained from Greater Vancouver Regional District (GVRD) and Environment Canada monitoring sites within the Lower Mainland of Vancouver, B.C., and hourly meteorological data was obtained from Environment Canada. The study population consisted of patients with ICDs seen at the two ICD clinics located in downtown Vancouver. Charts of all ICD patients seen at the St. Paul's Hospital Pacemaker Clinic and Dr. John Yeung's Burrard Pacemaker Clinic were reviewed. All patients residing within the GVRD who had experienced at least one ICD shock during the study period were included in epidemiological analyses. 2.1.1 Study period The study period was from February 14, 2000 to December 31, 2000, corresponding to the period for which ICD data were abstracted and for which enhanced air pollution monitoring data were available from Environment Canada. 2.1.2 ICD Patient information 2.1.2.1 Data collection ICD devices treat arrhythmia by delivering a shock that resets the rhythm of the heart. They also record all therapies that have been delivered by the ICD in an attempt to terminate ventricular tachycardia or fibrillation. Patient ICDs were manufactured by Guidant, Medtronic, and Intermedics. Because of individual physiological differences, ICDs were set to fire at different thresholds for each individual. ICD clinic records were manually abstracted to determine the age, gender, address, type of defibrillator, indication for implant, and date of implant for each patient. The dates on which patients experienced one or more ICD shocks as treatment for cardiac arrhythmia, and the most recent date that ICD records had been downloaded, were also recorded. The subject-specific duration of observation was bounded by the implant date on one side and the last date that ICD information was downloaded on the other. Shocks occurring within the first 3 days of implant were excluded from the dataset since these were usually induced to test the devices. Shocks were flagged if they were documented as being 'inappropriate' by a cardiologist, and the circumstances of the event were recorded. If more than one 13 event occurred for the same subject on the same calendar date, only one was retained in the dataset. 2.1.3 Ambient pollutant and meteorological measurements 2.1.3.1 GVRD monitoring network Ambient concentrations of specific air pollutants provided by the GVRD included hourly concentrations of ambient particulate matter of aerodynamic diameter less than 10 um (PMin), ozone (O3), nitrogen dioxide (NO2), carbon monoxide (CO) and sulfur dioxide (SO2) measured for the duration of the study. The pollutant monitoring network included 9 monitoring sites for P M 1 0 , 15 O 3 monitoring sites, 9 SO2 sites, 15 N 0 2 sites, and 12 CO sites. Sites consisted of a temperature-controlled shelter housing sampling equipment. Average hourly values of P M 1 0 , NO2, CO, and S0 2 were collapsed to 24-hour averages over all sites prior to analysis. For O 3 , 24-hour maximums were averaged over all sites. All pollutants were reported in parts per billion (ppb) except P M 1 0 , which was reported in ug/m3. For GVRD pollutants, missing values were filled in for each pollutant separately using an EM algorithm (Vedal et al, 2001). Missing values were not filled in for Environment Canada pollutants due to the fact that these measurements were conducted at a single monitoring station for a relatively short time period. Environment Canada provided hourly concentrations of PM 2 5, elemental carbon (EC), organic carbon (OC), and sulfate (SO4) averaged over 24-hour periods. Although these pollutants were monitored as part of an Environment Canada research initiative, measurements were conducted at a GVRD monitoring station located in South Burnaby, BC. Laboratory analyses were conducted in the Environment Canada laboratory located in Gloucester, Ontario. All pollutants were reported in ug/m3. Locations of monitoring stations are marked on Figure 2.1 below. 14 2.1.3.1.1 PM 2 . 5 Sampling for ambient PM2.5 was conducted with pre-weighed 47 mm Teflo® filters (2 um, Gelman) in a partisol sampler at a flow rate of 16.7 liters per minute. Samples were collected over 24-hour periods. Following sampling, PM2.5 mass was determined by post-weighing filters in a temperature and humidity controlled balance room using a microbalance sensitive to 0.1 ug. 2.1.3.1.2 PM10 PM10 concentrations were measured using a tapered element oscillating microbalance (TEOM) that determines concentrations by monitoring the frequency of oscillation of a tapered element, which decreases as particle mass accumulates on a filter mounted on its tip. Concentrations were measured every minute and averaged hourly. Hourly averages were collapsed to 24-hour averages and averaged over all monitoring sites. 2.1.3.1.3 Sulfate (SO4) Sulfate concentrations were determined from 47 mm Teflo filters sampled in a 2-stage filterpack. The filterpack was preceded in the sampling system by two 4-channel stainless steel denuders. This denuder/filterpack system was operated in a Versatile Air Pollutant Sampler (VAPS). The denuder/filterpack system was attached to a fine mass (PM2.5) channel and was run at 15 liters per minute. Samples were collected over 24 hour periods. Sulfate concentrations were determined by ion chromatography (Brook et al, 1997). 2.1.3.1.4 Elemental and Organic Carbon (EC and OC) Analysis for EC and OC was determined using a Sunset Labs thermal optical transmission instrument according to a modified NIOSH Method 5040. A temperature program was used which involved longer dwell times at each temperature step compared to NIOSH Method 5040. Samples were collected over 4, 8 or 12 hour periods and 24-hour averages were provided. 2.1.3.1.5 Carbon monoxide (CO) CO was measured using a continuous gas filter correlation analyzer that compares the infrared absorption spectrum of the sampled air to that of a high concentration sample of CO. Hourly average concentrations were collapsed to 24-hour averages that were averaged over all sampling sites. 2.1.3.1.6 Nitrogen dioxide (N02) NO2 concentrations were determined using a continuous chemiluminescence analyzer that establishes concentrations by measuring the light emission resulting from electronically excited NO2 molecules decaying to lower energy states. Specifically, the intensity of luminescence produced by the gas-phase reaction of nitric oxide and O 3 is 15 assessed. Hourly average concentrations were collapsed to 24-hour averages and averaged over all monitoring sites. 2.1.3.1.7 Sulfur dioxide (S02) Continuous SO2 concentrations were determined using a fluorescent SO2 analyzer that excites SO2 with pulsating ultraviolet light. The S0 2 molecules give off a characteristic decay radiation that is converted to a voltage proportional to the concentration of SO2 in the air sample. Hourly averages were collapsed to 24-hour averages and averaged over all monitoring sites. 2.1.3.1.8 Ozone (03) An ultraviolet photometric O 3 analyzer was used to determine continuous O 3 concentrations. The instrument established O 3 concentrations by measuring the attenuation of a specific wavelength of light due to 0 3 in an absorption cell. Hourly averages were collapsed to 24-hour maximums that were averaged over all monitoring sites. 2.1.3.2 Meteorological data Hourly meteorological data was provided by Environment Canada. Temperature, barometric pressure, rainfall, and wind speed were obtained from 4 monitoring stations, while relative humidity was obtained from 2 stations in the Vancouver area. Temperature, barometric pressure, wind speed and relative humidity values were averaged over 24-hour periods over all monitoring stations. These variables were reported in degrees Celsius (°C), kiloPascals (Kpa), kilometers per hour (km/h), and percent, respectively. Rainfall was reported as the percentage of hours of each 24-hour period that experienced any rainfall. Missing values were filled in as per the method used for air pollutants. 2.1.4 Data Analysis Excel 97 was used to create the initial database. SPSS Version 10.0 (©1999 SPSS Inc.) and Splus 2000 (©1988-2000 Mathsoft Inc.) were used for all statistical analyses in the epidemiological portion of the study. 2.1.4.1 Case crossover analysis A case crossover analysis is an adaptation of the traditional case-control study design wherein subjects act as their own controls. The case crossover design was introduced as a technique for assessing transient exposure-induced changes in the risk of a rare acute-onset health outcome (Maclure 1991). 16 2.1.4.1.1 Design of analysis In this analysis, levels of specific air pollutants measured at a period during or shortly before a subject experienced one or more ICD shocks (the so-called 'case' period) were contrasted with levels of the same pollutant during control periods when the subject did not experience a shock. The underlying hypothesis was that air pollution concentrations on or just preceding the date of ICD shock would be higher than those on control dates. The design was ambidirectional, with pollutant concentrations on each case date being compared to concentrations on two control dates 7 days before and after the case date, respectively. Both pre- and post-event control dates were included to increase the number of controls per case and to prevent persistent trends from biasing the analysis. If an omitted variable that had a positive relationship with ICD shock was increasing over time, for example, and control periods were always before case periods, the results would be artificially inflated by design. A seven day separation between case and control was chosen both because previous studies have demonstrated a one week interval to be sufficient to remove seasonal confounding (Sunyer et al, 2000), and because this design controls for potential day of the week effects. A schematic of the case crossover design utilized is presented in Figure 2.2 below. Figure 2.2. Schematic o f ambidirectional case crossover design control case control DAYS PRIOR TO DAY OF ICD DAYS AFTER ICD ICD SHOCK SHOCK SHOCK 2.1.4.1.2 Grouping of events ICD shocks occurring within a 72-hour period were considered as a single event in analyses. Events within 72 hours were collapsed because it was hypothesized that temporally clustered arrhythmias were initiated by the same air pollution event. If this was the case, analysis of all events would effectively double count events. Further, if air pollutant increases on or before the date of the first event initiated a series of events that occurred on the following day or days, and these days experienced low pollutant concentrations, the inclusion of events would dilute any true effect. Grouped events were attributed to the date of the first of the grouped events. 2.1.4.1.3 Missing air pollutant data Where air pollutant data was unavailable for GVRD pollutants, the design was modified to maximize the number of control dates considered. Specifically, if air pollutant data was unavailable for a proposed control date, the control was replaced by the date 7 days further from the event in order to preserve the control for day of the week. If air pollutant data was also unavailable for this date, it was replaced by the closest date on the same 17 side of the case date for which air pollutant data was available between 4 and 14 days of the case date and separated from any other event by at least 3 days. Replacing proposed control days with the closest days for which air pollutant data was available resulted in some comparisons of cases and controls on different weekdays. This was considered preferable to moving the control further than day 14 since the placement of cases and controls close to one another in time reduces the chance of seasonal effects or other omitted covariates from confounding the relationship. If air pollutant data was unavailable for a case date, the case was censored. If a case day was within 4 days of the beginning or end of the study period, the control situated outside of the study period was censored. 2.1.4.1.4 Data preparation Plots of air pollutant and meteorological variables over the duration of the study were analysed graphically prior to analysis to identify patterns. Descriptive statistics including mean and interquartile range of concentration were computed for each pollutant. Intercorrelations of each ambient pollution variable with all meteorological variables were computed to determine the degree of multicollinearity between variables to be considered for inclusion in models. 2.1.4.1.5 Model variables Single-pollutant models were run for 24-hour averages of nine separate pollutants (PM2.5, EC, OC, S0 4, PM10, CO, N0 2 , 0 3 , S02) measured from February 14 to December 31, 2000. Conditional logistic regression analyses were conducted using S-Plus. Analyses were stratified by individual. Since meteorological variables may covary in time with air pollutants, and an ambidirectional study design may not prevent all possible confounding by weather patterns, analyses were carried out both with and without average temperature, relative humidity, barometric pressure, rainfall and windspeed calculated as outlined above. These variables were included simultaneously as covariates, as has been reported in previous studies (Sunyer et al, 2000; Lee and Schwartz, 1999; Sheppard et al, 2001). To produce results relevant to the study region, and because the pollutants considered varied in magnitude as well as in unit of measurement, estimates and approximate 95% confidence intervals for odds ratios were calculated. Odds ratios were ratios of the odds of occurrence of ICD shock corresponding to increases in pollutant concentration equal to the interquartile range of the pollutant as measured during the study period. These were calculated as follows: OR = exp[coef*interquartile range] UCL = exp[(coef+1.96*SE)*interquartile range] LCL = exp[(coef-1.96*SE)*interquartile range] where OR=odds ratio, coef=coefficient, SE=standard error of the coefficient, UCL=upper 95% confidence limit, LCL=lower 95% confidence limit, interquartile range=range of values of the pollutant, as observed during the study period. 18 2.1.4.1.6 Lag periods While the purpose of this analysis was to test for acute effects of air pollution exposure on cardiac arrhythmia, the design was limited by a lack of understanding of the length of time it may take for air pollution exposure to translate into a cardiac rhythm disturbance. Because of this uncertainty about the length of the so-called 'induction period' separating exposure from outcome, the analysis was conducted assuming induction periods of 0, 1, 2, and 3 days, respectively. Apart from same-day pollutant concentrations, the analyses tested for effects of air pollution on ICD shock that are manifested 1, 2, and 3 days following exposure. A schematic of proposed induction periods is presented in Figure 2.3 below. Figure 2.3. Proposed induction periods separating exposure to increased air pollution from ICD shock Day 0 Day 1 1 day induction period Day 2 2 day induction period 3 day induction period Day 3 = point exposure to air pollutant induction period effect period To test the assumption of each hypothesized induction period, initial analyses were repeated for 'lags' of 1, 2, and 3 days separately. Case days were shifted backward 1, 2, and 3 days, respectively, and control dates were situated 7 days on either side of the resulting case dates. A 0 day lag refers to the condition when the same 24-hour period during which an ICD event occurs is considered as the case day. A 1 day lag refers to the condition when the 24-hour period directly preceding the 24-hour period during which ICD an event occurs is considered as the case day. A schematic of the placement of cases and controls for the initial analysis (lag 0) and for lags of 1, 2, and 3 days is presented in Figure 2.4 below. 19 Figure 2.4. Ambidirectional case crossover design showing lag periods of 0, 1,2 and 3 days CONTROL C A S E CONTROL co es O 0 < < _3r_ _ 3 i 1 L LAG 1 i r I LLAGO L 1 r ^ I I i i i J 10 D A Y S PRIOR T O ICD S H O C K D A Y O F ICD S H O C K D A Y S A F T E R ICD S H O C K 2.1.4.1.7 Season Analyses were repeated to assess the importance of season, considering summer (May to September) and winter (October to Apri l ) events separately. Because the dataset did not include air pollutant data for January 1 to February 13, winter events represent October to December and February 14 to Apr i l only. Odds ratios and confidence intervals were expressed for the interquartile range of each pollutant, as observed during the study period. 2.1.4.2 Sensitivity Analyses Three sensitivity analyses were conducted as outlined below. A l l were conducted with the same 5 meteorological variables included simultaneously as covariates. A l l comparisons were conducted assuming a lag of 0 days. 2.1.4.2.1 Meteorology Odds ratios associated with each air pollutant variable were compared for models repeated with and without meteorological variables to evaluate the effect of inclusion of weather variables. This was conducted since the case crossover design has the potential to also control for meteorological variation i f meteorology doesn't differ within the period separating case and control days. 2.1.4.2.2 Grouping of I C D events In a second set of sensitivity analyses, all ICD events were considered separately to assess the effect of grouping all events occurring within a 72-hour period of one another. The initial analysis was repeated with all events included. Controls were placed 7 days on either side of each case day, as in the initial analysis. The results of the analysis of all events were then compared to those of the analysis of grouped events. 20 2.1.4.2.3 Inappropriate events A third set of analyses were run following the exclusion of ICD events identified by a cardiologist as being inappropriate. These events were identified by perusal of patient charts. These analyses were conducted in an effort to reveal how misclassification bias may have affected the analyses. The analyses were run exactly as were the initial analyses except that events known to be inappropriate were excluded. 21 2.2 EXPOSURE ASSESSMENT The second part of this research was a personal exposure monitoring study conducted from May 15 to August 31, 2001. Environment Canada provided ambient concentrations of PM2.5, SO4, and EC for this same period. The target sample size was 25 participants, and the study population consisted of 20 patients with ICDs, a subset of the study population considered in the epidemiological analysis. Either seven or eight sampling sessions were attempted for each subject. Each sampling session consisted of a 24-hour measurement of personal PM2.5 exposure. Time activity logs were completed by the subjects during each sampling period and a one-time dwelling characteristics questionnaire was also administered. The optical reflectance of each filter was determined using a reflectometer and sulfate concentrations were measured by ion chromatography. 2.2.1 Study period The duration of the exposure assessment portion of the study was from May 15 to August 31,2001. 2.2.2 Personal exposure measurements 2.2.2.1 Study population - eligibility and recruitment Before subject recruitment, ethical approval was obtained from the Clinical Research Ethics Board of the University of British Columbia. All participants received an honorarium of $250. For this part of the study, patients were eligible if they resided in the Lower Mainland of Vancouver and were neither current smokers nor living with current smokers. Priority was given to subjects with a history of ICD shocks and who lived within an accessible region including Vancouver, Bumaby, Richmond, North or West Vancouver, Surrey, Delta, Whiterock, or Langley. The location of all study subject dwellings is presented in Figure 3.8. Participants for the study were recruited through the St. Paul's Hospital Pacemaker Clinic and the pacemaker clinic of Dr. John Yeung with the cooperation of cardiologists and technicians at each of the clinics. Addresses of eligible candidates were provided by the two clinics, and study recruitment letters were sent to 47 randomly selected patients in a first round of recruitment. Copies of recruitment letters are included in Appendices 2.1a and 2.1b. Letters were sent to an additional 19 patients in a second round initiated because the target sample size was not attained after the initial round. Candidates were then contacted by telephone by UBC technicians. If the candidate confirmed receipt of the recruitment letter, the study was briefly described. If the candidate was interested in participating, a 30-45 minute introductory meeting was scheduled. During the introductory meeting, sampling equipment and procedures were demonstrated and questions were answered. An informed consent form containing information about the 22 study was provided, and candidates wishing to participate signed up for the study at the end of the introductory meeting. A copy of the consent form is included in Appendix 2.2. 2.2.2.2 Subject identification and sampling schedules A unique two-digit identification (ID) number was assigned to each participant. This was used throughout the study. Initially, sampling schedules were randomized using the random number generator function in Excel 98, by assigning a number to each possible 24-hour sampling date over the course of the 16-week study period. If any of the 7 dates provided by the random number generator were less than 8 days apart, they were adjusted forward or backward to ensure separation by at least 8 days. Sampling sessions were not scheduled on weekends and were not scheduled if 4 visits had already been scheduled for that day. Participants were not scheduled on the same day as the same other participant more than twice. Proposed sampling schedules were brought to each subject's home on their first sampling day. Subjects were informed that the schedule was tentative and sessions could be rescheduled if necessary. If a patient rescheduled a sampling date, the above-noted constraints were maintained. 2.2.3 Personal sampling equipment and forms 2.2.3.1 Personal exposure sampling for airborne particulate matter Personal exposures to PM2.5 were measured with personal PM2.5 impactors (PEM, MSP Corp.) loaded with 37mm 2um pore size Gelman Teflon filters (Teflo, R2PJ037, Gelman Sciences) and connected to 20 cm aluminum inlets. Air was pulled through the filter at 4.0 L/min using a flow-controlled battery operated pump (Aircheck Sampler 224-PCXR4, SKC Inc, or Gilman). To provide sufficient power for the 24-hour sample taken on each sampling day, secondary battery packs were connected to the pumps in series. Flows were measured at the beginning and end of each 24-hour sampling period with precision rotameters (Matheson, 603) calibrated with a frictionless piston meter (Bios Corp.). A flow log associated with each subject was used to record the date, time and pump flow at the beginning and end of each sampling session. A sample flow log is attached in Appendix 2.3. A precision sound level meter was used to document noise levels of sampling pumps. Observed noise levels ranged from 48 to 61 dBA. To aid in carrying the pump and to reduce noise levels, pumps were placed inside small bags or boxes lined with foam insulation and outfitted with a shoulder strap. This insulation reduced noise levels to between 39 and 51 dbA. The PEM sampler, attached to the pump by surgical latex tubing and positioned at one end of the aluminum inlet, was kept outside of the bag. The aluminum inlet served to protect the sampler from direct contact with clothing and other surfaces. Participants were requested to wear the monitor over their shoulder with the aluminum inlet oriented vertically, as close as possible to the breathing zone. The inlet 23 was fitted with velcro, which served to secure the sampler to the shoulder strap. Figure 2.5 shows two study subjects wearing the personal sampling equipment. Figure 2.5. Two study subjects during sampling sessions 24 Subjects were requested to wear the exposure sampler and pump whenever possible during the sampling session. They were allowed to place the monitor beside them while sitting in one room, but were requested to take the monitor with them if they moved from one room to another. Subjects were asked to keep the monitor beside them while sleeping. If the noise produced by the pump prevented them from sleeping, they were permitted to cover the pump with blankets to help muffle the sound provided that the sampler and inlet were not covered. Sampling sessions started and ended between the hours of 8:00 am and 11:00 am, keeping the sampling duration as close to 24 hours as possible. 2.2.3.2. Information about sources of pollutants 2.2.3.2.1 Time activity diaries To help determine potential sources of PM2.5 exposure, subjects were asked to complete a time activity diary for each of the seven sampling sessions. A copy of this diary is included in Appendix 2.4. The time activity diary was used to record information that might affect exposure to pollutants throughout the course of the sampling session. Subjects were asked to highlight boxes corresponding to their activity and location for the majority of each 30-minute period. Locations included 'indoors,' 'outdoors,' and in 'transit.' Outdoors was further defined as 'near' home if the individual's time was spent within a 3-block radius of the home. Transit was defined as 'car,' 'bus,' 'walk,' or 'other,' and subjects were asked to indicate whether or not they were on a busy road and, if so, the number of minutes spent in that location. Subjects were also asked to indicate whether their physical activity level was low, medium, or high. The three activity levels were described as being equivalent to running, walking, or sitting, respectively. Individuals were asked to highlight the 'cooking' or 'tobacco smoke' section for periods during which they were cooking or were in the same room as someone cooking or smoking. Finally, individuals were asked to indicate if windows were open in the room in which they were sitting, and to indicate any period during which the sampler was not in the room with them. This data was used to calculate the percentage of time spent in different locations or on different activities. Subjects were encouraged not to alter their normal behaviour on sampling days. 2.2.3.2.2 Dwelling information form A dwelling questionnaire was administered on a one-time basis to collect information about the individual's place of residence and location within the city. A copy of this form is included in Appendix 2.5. Subjects were required to estimate the percentage of their home covered by carpet. The length, width and height of every room in each subject's dwelling were ascertained with an electronic distance measuring tool (Sonin Inc.). The volume of each residence was then calculated. Addresses collected on the dwelling questionnaires were used to geolocate each residence using Arc View GIS 3.2 with Arcview Spatial Analyst and ArcView 3D Analyst. A digital cartographic Vancouver street network file from the 1996 census (publicly available as gsnf933r.e00) was used to locate each subject's street address. Arc View's Import71 utility was used to 25 convert the street network file from Arclnfo export format (.eOO) to an ArcView feature data theme. The 'Locate Address' function was then used to locate and mark dwellings. The distance in kilometers separating each dwelling from a major road (defined as a road with 4 or more lanes) and from the South Bumaby Monitoring Station (where ambient air pollutant measurements were taken) was calculated from the resulting ArcView map. 2.2.3.3 Lab preparation of personal samplers Samplers were loaded with pre-weighed filters in the lab before being taken to the field. Before loading, impactor plates were saturated with mineral oil. Once all sampler components were assembled, an adhesive label was used to identify the filter in each sampler. Samplers were leak-checked by fitting an adapter to seal the inlet of the sampler. The sampler was connected to a rotameter, which was in turn connected to a vacuum pump that pulled air through the sampler. If the rotameter indicated less than 10% of the target flow rate of 4 L/min (0.4 L/min), the sampler was considered sealed. If the rotameter indicated more than 0.4 L/min or the flow fluctuated, samplers were dismantled, and o-rings and filters were inspected for possible damage and replaced if necessary. A diagram of sampler assembly is presented in Figure 2.6 below. Figure 2.6. PEM sampler assembly (adapted from SKC Inc. www.skcinc.com. accessed August, 2002). nozzle cap Tuiano » d -Porous Vain less sSeel impaction ri Impaction ring suppori Stanleys s » d screen At the end of each sampling session, samplers were returned to the laboratory and disassembled. Filters were transferred back to their original petri plates using clean forceps, and the label from each sampler was transferred to the corresponding petri plate. The plates were then returned to the balance room where they were stored under temperature and humidity controlled conditions until the filters were reweighed. 2 6 2.2.4 Laboratory Analysis of Personal Filters 2.2.4.1 Gravimetric Analysis Before weighing unused filters, fdter packages were unsealed and allowed to equilibrate in a temperature and humidity controlled balance room for 48 hours or longer. The conditions of the balance room were monitored throughout the duration of the study. A temperature of (22°C + 1°C) and a relative humidity level of (45.5% + 12.5%) were maintained during weighing of filters. Prior to sampling, filters were pre-weighed in triplicate using a microbalance (Sartorius M3P; 1 ug resolution, ± 2 ug sensitivity). Static charge was removed with a radioactive neutralizer prior to weighing. Filters were reweighed if three consecutive weights of the same filter were not within 10 ug of one another. Pre-weighed filters were placed onto individual petri plates, and each plate was labeled with a 3-digit identification number. All filters were stored in the balance room, except those necessary for. the current week of sample collection, which were moved to the laboratory at the beginning of the week. Each week, between 8 and 14 filters were loaded into PEM samplers for sample collection. After sample collection, filters were unloaded from the samplers and returned to their original petri plates. The fdters were then returned to the balance room and allowed to re-equilibrate for a minimum of 48 hours before being post-weighed in triplicate using the same microbalance and procedure as was used for the pre-weights. Pre-weight was subtracted from postweight to determine PM2.5 sample mass. 2.2.4.2 Quality Control Filters In order to check the accuracy of filter weighing, three unused filters were used as quality control filters throughout the study. These three filters were weighed at the beginning of each weighing session. The weight of each was compared to the mean, warning (mean ± 2SD) and control (mean + 3SD) limits of all previous weighings of the same filter. Charts were evaluated for trends and for the presence of repeated weighings outside of limits. The mean, warning and control limit were plotted on a quality control chart to confirm that weights remained within limits. 2.2.4.3 Field and lab blanks One filter was set aside as a field blank each week, representing approximately 10% of the total filters used. Field blanks were prepared by loading filters into PEM samplers in the same manner as filters used for sample collection. All samplers were transported to the field in individual plastic bags. The field blank remained in its plastic bag for the duration of the 24-hour sampling period. Following this, it was taken to the lab, disassembled, and returned to the balance room for re-equilibration and post-weighing in the same manner as the sample filters. One filter was also set aside as a lab blank each week, representing another 10% of the total filters used. This pre-weighed filter was moved from the balance room to the lab at 27 the beginning of each week, and remained in its petri dish for the duration of the week, when it was moved back to the balance room to be post-weighed. 2.2.4.4 Maintenance of sampling equipment PEM samplers were cleaned following each use. Daily cleanings consisted of removing visible particles from the impactor plate with a razor blade and wiping down other sampler components with distilled, deionized water. On a weekly basis, impactor and backing plates were sonicated for 15 minutes in soapy water and then rinsed three times with distilled, deionized water. The outer casings of the samplers were soaked in soapy water for 15 minutes, scrubbed with a plastic brush, and rinsed three times with distilled, deionized water. All components were placed on Kimwipes and allowed to air dry. 2.2.5 Optical Reflectance Analysis 2.2.5.1 Reflectometer All filters were analyzed for optical reflectance using the EEL Smokescreen Stain Reflectometer Model 43D. Filters were analyzed following the manufacturer protocol, using a 5-point method wherein reflectance was taken as the average of measurements taken at the center and at 4 surrounding points. According to this protocol, the instrument was calibrated for each new production batch of filters, and a linearity check using standards provided by the manufacturer was conducted at the beginning of each measurement session. 2.2.5.2 Calibration The instrument was calibrated for each new batch of fdters using 5 unused so-called control fdters from the new batch. The reflectometer was placed over a randomly selected unused filter and the readout was adjusted to 100.0. The average reflectance of the other 4 filters was measured using the 5-point method and the standard deviation of the 5 average reflectance values (one value from each filter) was determined. If the SD was <0.5, the filter with the median reflectance was used as a primary control filter used for recalibration of the instrument during sample filter reflectance measurement for any filters originating from the same batch. 2.2.5.3 Linearity checks Prior to each measurement session, the reflectometer was placed over a white standard and the sensitivity control was adjusted to 100.0. The reflectometer was then placed over a second grey standard, and was required to read within the specified range of 34 ± 1.5. The instrument was placed over the center of the primary control filter and the readout adjusted to 100.0 immediately following the linearity check and following every fifth sample filter thereafter. 28 2.2.5.4 Conversion to absorbance Reflectance was converted to absorbance units prior to analysis. This conversion was conducted according to the following formula: a = ' A ' ln \Rf] 2V Rs where a =absorbance in 10"5m"', A=area of filter in m2, V=volume sampled in m3, i?/=average reflectance of the field blank filters, i?5=reflectance of the sample filter as a percentage of 100.0 2.2.6 Sulfate Analysis Sulfate analysis was conducted on all personal PM2.5 filter samples. Since the analysis involved destruction of fdters, it was conducted after a preliminary examination of the results of analyses for PM2.5 and absorbance. 2.2.6.1 Preparation All equipment was cleaned with soap and then thoroughly rinsed with distilled, deionized water and wiped with ethanol before commencing. Ethanol was also used to clean forceps, razor blades and glass plates between the handling of different fdters. 2.2.6.2 Procedure Each filter was moved from its petri plate onto a glass plate using forceps. A razor blade was used to make six to eight cuts into the plastic rim of each filter, allowing the fdters to fold easily. Each filter was then placed into a clean, labeled, plastic screw-top container. A micropipette was used to add lOOul ethanol to each container, ensuring that the filter was completely saturated. Five mL of distilled, deionized water was then added and the container was sealed. Containers were sonicated for 15 minutes to bring the sulfate into solution. Approximately 0.5 mL of the liquid was then transferred into a disposable poly vial via a fdter that removed large particles. The vial was capped and loaded into an autosampling device. From here, samples were analyzed for sulfate concentration using a Dionex DX-300 ion chromatograph with suppressed conductivity detection. A set of six standards and blanks were analyzed at the beginning and end of each batch to generate a batch-specific calibration curve. One standard was also inserted after every tenth sample to ensure comparable results were maintained throughout. 2.2.7 Ambient Air Pollutant Measurements All measurements of ambient pollution during the study period were provided by Environment Canada. They were collected at the GVRD monitoring station located in South Bumaby. All analyses of ambient filters were conducted in the Environment 29 Canada laboratory located in Gloucester, Ontario. Pollutants considered in the exposure assessment that were also considered in the epidemiological analysis included PM2.5, EC, OC, and sulfate. Methods of data collection and analysis for these ambient pollutants are outlined in the epidemiological analysis section above. Other pollutant measurements considered in the exposure assessment included carbonate carbon (CC), total carbon (TC), black carbon (BC), and another elemental carbon measure known as E C (elemental plus organic carbon). 2.2.7.1 Carbon measurements Carbonate carbon (CC), total carbon (TC) or elemental plus organic carbon, and E C , or elemental plus carbonate carbon, were determined using a Sunset Labs thermal optical transmission instrument (Sunset Laboratory Inc., Seattle, WA) according to NIOSH method 5040. Black carbon (BC) levels were determined using the laser integrating plate method (LIPM) at the Environment Canada laboratory. 2.2.8 Data Analysis 2.2.8.1 Particle sampling - data quality and descriptive statistics Excel 97 was used to create the initial database for this portion of the study. SPSS 10.0 and S-Plus 2000 were used to conduct all statistical analyses. Personal exposure sample particle data were reviewed to verify correct times and flows. Samples were excluded from further analysis if the tubing was found to be disconnected from the pump at the end of the sampling period, if the sampling duration was less than 18.0 hours, or if either pre- or post-sampling flow was more than 10% higher or lower than the target of 4.0L/min (3.6-4.4L/min). For PM2.5 and sulfate measurements, mean field blank levels were subtracted from all samples prior to calculating concentrations. Limits of detection, defined as the mean plus 3 standard deviations of blank concentrations, divided by the mean sample volume, were calculated for each component. Descriptive statistics including number of samples and range of concentration as well as arithmetic and geometric mean and standard deviation of concentration were presented for PM2.5, absorbance and sulfate. 2.2.8.1.1 Extreme values For samples with personal exposure values outside of the mean ± 2SD of the distribution of all subjects, laboratory analyses were repeated and spreadsheets were checked to ensure that the values were accurate. These values were also compared with time activity information and field notes to determine whether any activities may have contaminated the samples. Extreme values that could neither be explained by subject behaviour and/or location nor by laboratory error, and that were within a reasonable range for an urban setting, were flagged but included in analyses. Where relevant, analyses were repeated with and without extreme values. 30 2.2.8.1.2 Distribution of exposure variables Distributions of all exposure variables were examined graphically to determine whether variables were normally distributed. Variables that were noticeably skewed were natural log-transformed prior to further analysis. 2.2.8.2 Absorbance and various carbon measurements Optical absorbance was measured in this study because it is thought to represent elemental carbon, a component of local source traffic emissions. Previous studies have indicated high correlations between elemental carbon and absorbance (Kinney et al, 2000; Janssen et al, 2000). Absorbance has been associated with health impacts in epidemiologic studies (Prescott et al, 1998), and has been shown to be a good indicator of EC (Gotschi, 2002). EC, in turn, is an indicator of diesel exhaust emissions. Diesel exhaust particles, which have a mass median diameter of 0.05-1.Oum and are therefore a component of PM2.5, contain EC as one of their major constituents (Cohen and Nikula, 1999). Filter absorbance was compared to carbon measurements to determine the utility of this method of measurement in assessing air concentrations of different carbon components in this location. Optical absorbance is a relatively cheap and simple method, and is non-destructive, so may be conducted in combination with other analyses. These benefits provide incentive to identify the adequacy of this measure as a surrogate for EC. Ambient absorbance values were regressed against EC concentrations determined from colocated samples at two different Vancouver-area monitoring stations to assess the utility of absorbance as a surrogate for elemental carbon (EC). Absorbance was also regressed against carbonate carbon (CC), elemental plus carbonate carbon (EC), and black carbon (BC) to test the ability of absorbance to predict these other carbon measures. 2.2.8.3 Relationships between personal and ambient concentrations of the same exposure variable Personal exposures were compared to ambient concentrations of PM2.5 and component pollutants for all subjects combined using linear regression. Regression analyses were also conducted for individual subjects. Individual ranges and median correlation coefficients are presented. Scatter plots of personal versus ambient levels of each pollutant were examined, and extreme values were compared with time activity information and field notes to determine whether or not their removal was justified. Distributions of personal and ambient concentrations of each air pollutant measured were compared using boxplots to identify differences. Ratios of personal:ambient concentrations of PM2.5, absorbance and sulfate were calculated to further understand differences between personal and ambient measures. 31 2.2.8.4 Interrelationships between personal and ambient concentrations of different PM2.5 components Univariate regressions were conducted for all subjects combined between all personal and ambient exposure variables to identify relationships between the measured components. Regressions were also conducted for each individual subject. Ranges and median correlations were presented. 2.2.8.5 Composition of personal PM2.5 The ambient contribution to personal PM2.5 exposure was estimated using the so-called sulfate method (Sarnat et al, 2000). Specifically, for each sample, the ratio of personal to ambient sulfate was multiplied by the ambient PM2.5 concentration to provide an estimate of the amount of personal PM2.5 contributed by outdoor sources. This value was then divided by the personal PM2.5 concentration and multiplied by 100% to reveal the percentage of personal PM2.5 predicted to have originated from ambient sources. The equation for this calculation is as follows, with all pollutant concentrations in ug/m3: pPM-2.5a '^*aPM2.S aSQ4 pPMl.S *100% where p P M 2 . 5 a is the percentage of personal PM2 .5 contributed by ambient sources, pSOA is the personal sulfate concentration, aS04 is the ambient sulfate concentration, aPMl.5 is the ambient PM2 .5 concentration, and pPMl.S is the personal PM2 .5 concentration. To further characterize the composition of PM2.5, mass percent sulfate and EC were calculated for both personal and ambient samples. Mass percent was calculated by dividing the sample-specific mass of each component by the PM2.5 mass and multiplying the resulting fraction by 100%. EC mass was estimated from absorbance using the equation of the absorbance:EC relationship. 2.2.8.6 Predictors of personal exposures 2.2.8.6.1 Distance from monitoring station To determine if distance from monitoring station impacted relationships between personal and ambient exposures, Pearson correlations of individual personahambient univariate regressions were regressed against distance to the monitoring station. 2.2.8.6.2 Time activity diaries and dwelling information Time activity diary and dwelling questionnaire information was used to identify activities or housing characteristics predictive of personal exposures. For each time activity 32 variable, the number of 30-minute periods highlighted was tallied and divided by the total sample duration to determine the proportion of time associated with that location or activity. Means were presented in tables and summarized in pie charts. Descriptive statistics for all subjects combined were presented for dwelling information variables. 2.2.8.7 Determinants of exposure modeling 2.2.8.7.1 Data preparation Personal PM2.5, personal sulfate, and personal absorbance were used as the dependent variables in three separate multiple regression analyses. Data from time activity logs and dwelling questionnaires was collapsed into several categorical independent variables. Variables retained from time activity logs and dwelling information forms were included as independent variables in regression models. Variables with insufficient inter-subject variability were excluded from further analysis. Descriptive statistics (counts for categorical data, and means, ranges, and standard deviations for continuous variables) were calculated for all variables. Exposure data that was skewed and approximately lognormal was transformed (base e) prior to analysis. Variables were included only if they were logical predictors of the dependent variable. Pearson correlations between all independent variables were calculated to ensure that all pairs included in models had r<0.6. Strongly correlated variables were logically related to one another in all cases. For two intercorrelated variables, the variable thought to reasonably include the effect of the other variable was retained for analysis. Boxplots were reviewed to identify correlations between categorical and continuous variables. Cross tabulations were used to determine if categorical variables were correlated. Univariate regressions were conducted between all remaining independent and dependent variables. All variables with p<0.25 and R2>0.01 in univariate modeling were initially introduced into models. Since models had personal exposures as dependent variables, variables thought to represent sources of exposure were required to have positive coefficients in univariate analyses. 2.2.8.7.2 Procedure for running models A manual backward stepwise regression procedure was followed for the creation of three multiple regression models. First, if the sign on the coefficient of a particular variable was the opposite of that hypothesized, the variable was removed from the model. Second, variables with p>0.30 were eliminated stepwise until all variables in the model had p<0.30. A mixed effects linear modeling technique was used, with individual subject ID as the single random effect. Subject ID was included as a grouping variable to account for within-subject correlation. In order to preserve sample size, missing ambient pollutant values were estimated for the purposes of the multiple regression by averaging the value for the closest available previous date and the closest available date following the missing value. 33 2.2.8.7.3 Model fit For each model, coefficients and p values were reported for each independent variable that was retained in the model, and the equation of each model was presented. Overall model fit was assessed by examination of the R 2 values of univariate regressions of predicted values (generated by the model) against dependent variable values. R 2 values were approximate estimates of the amounts of variability explained by different combinations of independent variables, since they did not adjust for the different weights apportioned to different points due to the inclusion of a random grouping variable in the models. 2.2.8.7.4 Sensitivity analyses Three sensitivity analyses were conducted for the mixed effects models. Exposure to tobacco smoke is a well-known confounder in studies of personal exposure to particles. In a recent personal monitoring study, mean personal PM2.5 exposure concentrations of smokers were almost double those of participants exposed to tobacco smoke and three times those of participants not exposed to tobacco smoke (Koistinen KJ et al, 2001). Furthermore, exposure of study subjects to tobacco smoke was a violation of the initial inclusion criteria. Sensitivity analyses were therefore conducted to evaluate models following exclusion of two subjects who reported exposure to tobacco smoke during every sampling session. Similarly, models were rerun after the exclusion of all person-days with exposure to tobacco smoke. In a final sensitivity analysis, models were run following the removal of outliers in the dependent or independent variable. Outliers were defined as concentrations larger than the mean + 2SD of the distribution. These models were rerun to determine whether or not the inclusion of unusually high exposures changed model results. The variables that no longer met criteria for inclusion in the model in sensitivity analyses were reported. 34 3.0 RESULTS Chapter 3 3.1 EPIDEMIOLOGICAL ANALYSIS 3.1.1 Population characteristics The study sample assessed in the epidemiological analysis was comprised of 34 patients living in the lower mainland of Vancouver and attending the St. Paul's Hospital and Burrard Pacemaker clinics who experienced ICD shocks during the February 14 to December 31, 2000 study period. Seven (20%) were female patients ranging in age from 22 to 79 years (mean: 53) and 27 (80%) were male patients ranging in age from 15 to 85 (mean: 62). 3.1.2 Case crossover analysis 3.1.2.1 ICD shocks All days with ICD shocks recorded for each ICD patient during the study period are shown in Figure 3.1. Date of entry into the study and date of end of observation are marked for each patient. An event day was defined as any day on which one or more ICD shock occurred. Of the 34 patients, 12 experienced one event day during the study period. Six patients experienced 2 event days, 10 experienced between 3 and 7 event days, and 5 experienced between 8 and 11 event days. The total person time was 8201 person-days during which a total of 128 person days with ICD shocks were recorded. Nine of the 34 patients were followed for the entire study period, 18 entered at the beginning of the study period but were not followed to the end, 4 were not followed at the outset but were followed to the end of the study period, and 3 entered the study after the beginning and exited before the end. The average length of observation was 241 days, and the average rate of ICD shock across all subjects was 9 event days per year (median: 3.6; SD: 15). While some patients with frequent ICD shocks experienced events interspersed over time, others experienced very clustered event days. 3.1.2.2 Inappropriate ICD shocks Because the objective of the case crossover analysis was to identify associations between air pollutant exposure and cardiac arrhythmia, it was important to distinguish ICD shocks triggered by ventricular tachycardia or fibrillation from 'false alarms', or shocks unrelated to life-threatening cardiac arrhythmias. Failure to distinguish the different etiologies of ICD shock would result in misclassification of events, diluting any true association. Of the 128 recorded event days, 7 were documented as inappropriate by a cardiologist. These event days, occurring in 5 patients, are noted on Figure 3.1. The circumstances of these events, as documented in patient charts, which included sinus (normal) heart rhythms as well as external factors, are listed in Table 3 a below. 35 SUBJECT SHOCK DATE(S) C IRCUMSTANCES of E V E N T 4 02/19/00 Patient mugged 02/20/00 7 07/27/00 Patient fixing a T V antenna in the rain 17 07/31/00 Inappropriate event (triggered by sinus rhythms) 27 08/21/00 Events related to arm fracture and missed medications 09/22/00 31 03/10/00 Inappropriate event (triggered by sinus rhythms) Figure 3.1. ICD shocks over the study duration for each patient in the study sample o LU —i CO Z> CO ALL ICD E V E N T DAYS R E C O R D E D DURING STUDY PERIOD 1 — e — 0 -—o— V •+• 1 V T i w V o + n V A  I o V A + i \J m\ V A T i w W o /s T r\ O O A + © KT 1 o \J U KJ V /s + T 4. o O T -L—O-«YX>G1D OO i r\ f-k o o O , A T  O <J \J *-» \J T o V /-» T o. Q /s 1 C J w V H D O E> O OO " T i w T 1 U \J o nt\ V T i V + o O O o g\ u v -f-CD GDDO -V O O O i o S\ O V T i o o o o n o / \ T KJ*J \J \J KJ KJ V \j \J V >s O ( D O u /S V + V - t o - ee -e— O 1 1 02/09/00 03/30/00 05/19/00 07/08/00 08/27/00 DATE 10/16/00 12/05/00 o ICD discharge -(-Entry into study ©End of observation •'Inappropriate'discharge 3.1.2.3 Model design An ambidirectional case crossover approach was used to compare air pollutant concentrations on dates of ICD shock with concentrations of the same pollutant on days when ICD shock did not occur. Events were analyzed using conditional logistic regression. In the initial analysis, cases were compared to 2 control dates, one 7 days before and the other 7 days after the case date. Figure 3.2 shows all ICD events considered in the analysis after grouping of event days. Collapsing of all event days within a 72-hour period of another event day resulted in a 36 total of 98 event days and an average ICD event day rate of 6.4 per year (median: 3.2, SD: 9). After grouping event days, events were less clustered for each subject. Figure 3.2. All ICD event days considered in analyses ICD EVENT DAYS CONSIDERED IN A N A L Y S E S CJ L U ~3 CD Z> to 1 C O O O CO o o o o o + eoco&© oo o o GO OQSO CO O O CO ©O © O o o + 0 o o oo o o oo© oo o o 0 o I o o o ie~ o o o 02/09/00 03/30/00 05/19/00 07/08/00 08/27/00 DATE o o o — © -10/16/00 12/05/00 o ICD Discharge + Entry into study © End of observation • 'Inappropriate' discharge 3.1.2.4 Model variables One-pollutant models were run for nine separate pollutants (PM2.5, EC, OC, SO4, PM10, CO, NO2, O 3 , SO2) measured at monitoring stations in the Vancouver area from February 14 to December 31, 2000. Since meteorological variables may covary in time with air pollutants, analyses were carried out both with and without weather variables including temperature, relative humidity, barometric pressure, rainfall and windspeed as covariates. In addition to same day (lag 0) analyses, models were fit for lag periods of 1, 2, and 3 days, respectively. 3.1.2.5 Available data Since missing values of routinely monitored air pollutants including P M ] 0 , CO, N0 2 , O 3 , and S0 2 were filled in according to an EM algorithm described previously, data was available for these pollutants for all days in the study period. Of the 322-day study period, air pollutant data was unavailable for 55, 44, 45 and 18 days for PM2.5, EC, OC, and SO4, respectively. Missing values of these pollutants were not filled in since, in contrast to the five above pollutants that were monitored at multiple sites spanning the 37 study region, measurements were taken at only one monitoring site. Further, missing values often occurred on sequential days, making it more difficult to estimate true values. Case days for which air pollutant data was unavailable were censored and proposed control days for which air pollutant data were unavailable were replaced as described in the Methods section. Because each air pollutant distribution was missing different days, the number of cases censored due to unavailable air pollutant data differed for one-pollutant models. Similarly, the number of controls that were censored because of unavailable air pollutant data or because they fell outside of the study period differed for each model. A summary of cases and controls included in each of the 9 models is presented in Table 3b below. The average number of controls per case was slightly less than the target of 2. Table 3b. Cases and controls included in case crossover analyses MODEL POLLUTANT Cases Controls Average # Controls per Case Cases Controls ANALYZED Analyzed Analyzed Censored Censored* 1 PM 2 . 5 77 149 1.94. 21 5 2 EC 78 148 1.90 20 8 3 OC 77 150 1.95 21. 4 4 S0 4 94 183 1.95 4 5 5-9 PM, 0 , CO, N 0 2 , 0 3 , so 2 98 191 1.95 0 5 ""Controls associated with censored cases are not reported 3.1.2.6 Ambient air pollutants and meteorological variables Analyses were conducted both with and without stratification for season. All models included 5 meteorological variables as covariates. Concentrations of ambient air pollutants and weather variables observed during the study period are presented in Table 3c. Plots of the distribution of each pollutant and meteorological variable over the study period are presented in Appendix 3.1. Seasonal patterns are apparent in both air pollutants and meteorological variables. 38 Table 3 c. Concentrations of ambient air pollutants and meteorological variables considered in case crossover analyses observed during the study period (February 14-December 31, 2000). Statistics are based on average values for all pollutants except Q 3 , for which they are based on daily maximums. n Mean 25th percentile 75,h percentile Interquartile range PM2.5 (ug/m3) 265 8.2 4.4 9.6 5.2 EC (ug/m3) 279 0.8 0.4 0.8 0.4 OC (ug/m3) 275 4.5 1.6 3.9 2.2 S04(ug/m3) 302 1.3 0.7 1.7 0.9 PM10(ug/m3) 322 13.3 9.3 16.7 7.4 CO (ppb) 322 553.8 439.4 602.1 162.7 N0 2 (ppb) 322 16.5 13.6 19.0 5.4 Oj (ppb) 322 27.5 20.4 33.8 13.4 S0 2 (ppb) 322 2.6 1.6 3.1 1.6 Temperature (°C) 322 10.5 6.4 14.5 8.1 Relative Humidity (%) 322 77.2 72.3 83.8 11.5 Barometric Pressure (Kpa) 322 101.8 101.5 102.2 0.8 Rainfall (% hours per 24 hour period with any rain) 322 13.2 0 21.9 21.9 Windspeed (km/h) 322 6.3 5.3 7.1 1.8 Since intercorrelation of pollutants and meteorological variables is important to the interpretation of models, Pearson correlations between meteorological variables and all air pollutants were calculated to identify the degree of multicollinearity of variables. Correlations are presented in Table 3d. Table 3d. Pearson correlations between all modeled air pollutants and meteorological variables POLLUTANT PM 2 5 PM 1 0 EC OC SO4 CO N0 2 0 3 so 2 PM2.5 1.00 PM,„ 0.47 1.00 H V EC 0.17 0.30 1.00 .TJTAr OC 0.15 0.28 0.92 1.00 .TJTAr S0 4 0.16 0.30 0.10 0.11 1.00 I—I hJ CO 0.40 0.57 0.24 0.17 0.07 1.00 o e-N0 2 0.41 0.63 0.24 0.18 0.08 0.68 1.00 Oj -0.14 -0.10 -0.10 -0.06 -0.02 -0.56 -0.13 1.00 so 2 0.33 0.71 0.22 0.18 0.14 0.75 0.67 -0.24 1.00 < Temperature -0.09 0.17 0.08 0.12 0.20 -0.39 -0.29 0.38 0.01 ROLOGIC RIABLE Relative Humidity 0.01 -0.23 0.03 <0.01 -0.01 0.15 -0.01 -0.47 -0.17 ROLOGIC RIABLE Barometric Pressure 0.11 0.23 0.05 0.04 0.03 0.29 0.14 -0.22 0.35 o < w > f- Rainfall -0.14 -0.51 -0.14 -0.17 -0.19 -0.18 -0.10 -0.11 -0.42 ME' Windspeed -0.33 -0.53 -0.25 -0.25 -0.16 -0.56 -0.47 0.38 -0.56 The highest correlation was observed between E C and OC. Other pairs of variables with intercorrelations >0.5 included P M ] 0 and C O , PM10 and N 0 2 , P M ) 0 and S 0 2 , CO and 0 3 , C O and S 0 2 , C O and N 0 2 as well as N O 2 and SO2. Intercorrelations of pollutants with 39 meteorological variables revealed r values >0.5 for PMio with rainfall as well as PMio, SO2 and CO with windspeed. 3.1.2.7 Model results Results of the initial case crossover analysis including events in all months are presented in Table 3e and Figure 3.3 below. T a b l e 3 e . R e s u l t s o f i n i t i a l c a s e c r o s s o v e r a n a l y s e s . O d d s r a t i o s a n d c o n f i d e n c e i n t e r v a l s a r e p r e s e n t e d p e r i n t e r q u a r t i l e r a n g e o f e a c h p o l l u t a n t , a s o b s e r v e d d u r i n g t h e s t u d y p e r i o d . M e t e o r o l o g i c a l v a r i a b l e s w e r e P O L L U T A N T L A G L O W E R 9 5 % C I O D D S R A T I O U P P E R 9 5 % C I 0 0 . 8 5 0.98 1.14 P M 2 . 5 ( u g / m 3 ) 1 0 . 5 5 0.80 1.18 2 0 . 4 7 0.75 1.18 3 0 . 3 5 0.61 1.05 0 0 . 8 5 1.05 1.30 E C ( u g / m 3 ) 1. 0 . 9 2 1.08 1.27 2 0 . 9 4 1.04 1.14 3 0 . 8 9 0.99 1.10 0 0 . 9 5 1.09 1.24 O C ( u g / m 3 ) 1 0 . 9 5 1.07 1.20 2 0 . 9 6 1.03 1.12 3 0 . 9 5 1.01 1.09 0 0 . 6 8 0.91 1.21 S 0 4 ( u g / m 3 ) 1 0 . 4 6 0.72 1.12 2 0 . 5 7 0.89 1.40 3 0 . 7 5 0.92 1.13 0 0 . 4 9 0.86 1.50 P M 1 0 ( u g / m 3 ) 1 0 . 3 6 0.62 1.07 2 0 . 4 2 0.69 1.13 3 0 . 3 6 0.63 1.11 0 0 . 6 0 0.91 1.38 C O ( p p b ) 1 0 . 5 3 0.79 1.18 2 0 . 6 0 0.84 1.19 3 0 . 5 6 0.84 1.27. 0 0 . 4 9 0.88 1 .60 N 0 2 ( p p b ) 1 0 . 4 4 0.73 1.20 2 * 0 . 3 0 0.52 0 . 9 0 3 0 . 3 2 0.56 1.00 0 0 . 5 8 1.15 2 . 2 9 0 3 ( p p b ) 1 0 . 7 7 1.66 3 . 5 8 2 0 . 4 9 0.99 2 . 0 1 3 0 . 4 7 0.95 1.92 0 0 . 5 3 0.86 1.40 S 0 2 ( p p b ) 1 0 . 5 7 0.93 1.51 2 0 . 4 4 0.71 1.13 3 0 . 4 2 0.66 1.05 * p < 0 . 0 5 40 3 °" Li- ra CM T-< CM o Z (O •* M O » (O < tM O o o o o o o i i v a s a a o o n v y s a a o o n v y s a a o 41 One significant effect (p<0.05) was seen for N0 2 at a 2 day lag in this initial analysis. This effect was in the direction opposite of what was expected. Odds ratios were consistently below 1.0 for P M 2 5 and PM10 as well as S0 4, CO, N0 2 , and S0 2. Odds ratios for pollutants with odds ratios below 1.0 were the opposite of what was expected, signifying that an increase in air pollutant concentration was associated with a decrease in ICD shock. A high odds ratio (OR=1.66) was observed for O3 at a 1 day lag. Odds ratios were above 1.0 at all lags for EC and OC. For PM2.5, EC and OC, odds ratios appeared to decrease as lag period increased. Effects of lags on other pollutants did not reveal a consistent pattern. 3.1.2.8 Stratification by season Analyses were also conducted after stratification by season to allow for the tendency of pollutants and meteorological variables to covary with seasonal patterns. The dataset was divided into events occurring in winter (October to April) and those occurring in summer (May to September). Cases and controls included in these analyses are outlined in Table 3f below. There were fewer cases analysed in the summer period than in the winter. Summer pollutant data provided for a slightly higher average controlxase ratio than did winter data. Table 3f. Cases and controls included in case crossover analyses stratified by season. Counts are presented MODEL POLLUTANT ANALYZED SEASON Cases Analysed Controls Analysed Average # Controls per Case 1 P M 2 5 summer 30 60 2.00 winter 47 88 1.87 2 EC summer 30 60 2.00 winter 48 88 1.83 3 OC summer 30 60 2.00 winter 47 89 1.89 4 S 0 4 summer 30 60 2.00 winter 64 120 1.88 5-9 P M 1 0 , CO, N 0 2 , o3, so2 summer 30 60 2.00 winter 68 131 1.93 Median ambient pollutant concentrations for case and control periods in summer and winter, respectively, are presented in Table 3g below. These are presented to indicate true differences that are not revealed by odds ratios. 42 Table 3g. 75' percentile of pollutant concentrations for case and control periods used in season-stratified analyses. Values are presented for lags of 0,1,2, and 3 days. WINTER SUMMER CASE CONTROL CASE CONTROL EC (ugm-3) lagO 0.77 1.03 0.51 0.56 lag 1 0.81 0.94 0.87 0.64 lag 2 0.91 1.22 0.71 0.71 lag 3 0.89 0.90 0.53 0.73 OC (ugm"3) lagO 3.63 5.27 4.18 2.59 lagl 4.36 4.26 4.36 2.84 lag 2 4.35 4.67 3.96 3.41 lag 3 3.64 4.79 3.17 4.10 Sulfate (ugm"3) lagO 1.47 1.50 1.92 1.86 lagl 1.47 1.47 1.62 1.89 lag 2 1.38 1.40 1.64 1.94 lag 3 1.59 1.33 1.72 2.19 PM2.5(ugm-3) lagO 10.56 11.72 7.12 7.60 lag 1 12.64 10.63 8.09 7.92 lag 2 11.32 10.64 7.94 8.59 lag 3 10.04 11.69 7.13 9.51 PM,o (ugm"3) lagO 14.81 16.80 16.17 15.26 lag 1 16.35 15.90 15.29 16.94 lag 2 16.90 16.97 16.70 17.34 lag 3 17.73 15.90 13.94 18.41 CO (ppb) lagO 601.32 662.14 531.87 499.26 lag 1 649.70 625.91 564.87 505.05 lag 2 679.72 633.36 566.67 522.53 lag 3 677.95 658.31 547.58 573.25 N 0 2 (ppb) lagO 19.83 20.27 15.80 16.18 lag 1 20.09 20.32 17.06 16.95 lag 2 20.17 20.60 16.72 17.55 lag 3 20.98 20.78 16.16 18.82 0 3 (ppb) lagO 35.00 34.21 36.51 36.15 lagl 36.82 36.96 36.71 37.43 lag 2 33.78 35.71 33.19 38.09 lag 3 36.93 37.36 33.29 40.32 S0 2 (ppb) lagO 2.65 3.49 3.15 2.75 lag 1 2.89 3.06 3.55 3.01 lag 2 3.00 3.02 3.03 3.02 lag 3 3.24 3.32 2.86 3.70 Results of analyses stratified by season are presented in Figures 3.4a-c below. Raw values used to create these figures are presented in Appendix 3.2. For the summer period, significant (p<0.05) results were seen for PMio (lags 2 and 3) and N0 2 (lag 2). None of these results were in the direction expected (OR>1.0). Odds ratios were below 1.0 at all lags for PM2.5 and SO4. Odds ratios were above 1.0 for all other pollutants at a 0 day lag in summer. With the exception of PM2.5, odds ratios for a 0 day lag were higher for summer events than for winter events. SO2 and CO (lag 0), as well as EC (lag 2) and OC (lags 2 and 3) were nearly significant in summer in the hypothesized direction. 43 Figure 3.4a. Odds Ratios showing 95% confidence intervals for case crossover analyses stratified by season. Meteorological variables were included in the models. Data are presented for lags of 0,1,2 and 3 days. PM 2.5 O I-< CC co Q Q O < CC co Q Q O < CC CO Q Q O 2.5 2.0 1.5 1.0 0.5 0.0 S U M M E R 5.0 4.0 3.0 2.0 1.0 0.0 2.5 2.0 1.5 1.0 0.5 0.0 WINTER 0 1 2 3 LAG S U M M E R PM 1 ( 0 1 2 3 LAG WINTER 0 1 2 3 LAG S U M M E R SO. 0 1 2 3 LAG WINTER 0 1 2 3 LAG 0 1 2 3 LAG 44 Figure 3.4b. Odds Ratios showing 95% confidence intervals for case crossover analyses stratified by season. Meteorological variables were included in the models. Data are presented for lags of 0,1,2 and 3 days. E C < CC CO Q Q O < CC CO Q Q O < CC CO Q Q O 2.5 2.0 1.5 1.0 0.5 0.0 S U M M E R 2.5 2.0 1.5 1.0 0.5 0.0 5.0 4.0 3.0 2.0 1.0 0.0 WINTFR 0 1 2 3 LAG S U M M E R O C 0 1 2 3 LAG WINTER I I I * 1 1 -1-0 1 2 3 0 1 2 3 LAG LAG C O S U M M E R WINTER 1 2 3 LAG 0 1 2 3 LAG 45 Figure 3.4c. Odds Ratios showing 95% confidence intervals for case crossover analyses stratified by season. Meteorological variables were included in the models. Data are presented for lags of 0,1,2 and 3 days. Two values are not shown for 0 3 (upper 95% confidence limits = 11.1 and 19.9 for lags 2 and 3, respectively) < cr co Q Q O < CO Q Q O 10 8 6 4 2 0 10 8 6 4 2 0 NO, S U M M E R WINTER T T T I T I L i i I i i I i 0 1 2 3 0 1 2 3 LAG LAG S U M M E R WINTER i — r 1 2 LAG 1 2 LAG SO, < a: CO Q Q O 10 8 6 4 2 0 S U M M E R WINTER I T I I I I * 1 2 LAG 2 LAG 46 For the winter period, EC (lags 1-3) and OC (lag 3) produced odds ratios significantly below 1.0. The result for OC (lag 2) was almost significant. O 3 (lag 3) was significant in the hypothesized direction. Season appeared to have a potentially important effect on odds ratios as well as on confidence intervals. Confidence intervals varied according to season for several pollutants, but no consistent pattern was seen. For most pollutants, confidence intervals were larger in summer than in winter. This was expected since the number of ICD shocks observed was considerably smaller in summer than in winter. Interestingly, for EC and OC as well as CO and SO2, odds ratios were >1.0 for all lags in the summer and <1.0 for all lags in the winter. For other pollutants, no pattern was observed between summer and winter. For PM2.5, the effect of lag on odds ratio was similar from summer to winter, but odds ratios were slightly lower in summer. Odds ratios increased with increasing lag time in winter for SO4 and to a much greater extent for 0 3 . 3.1.3 S ensiti vity Analyses 3.1.3.1 Sensitivity Analysis I: Meteorology In addition to the initial case crossover analyses, three sensitivity analyses were conducted. The first tested the effect of the inclusion of weather variables. Meteorological variables were included as covariates in initial analyses to adjust for possible temporal trends associated with weather. Results of analyses with and without meteorological variables are presented in Figure 3.5 below. Tables of results are presented in Appendix 3.3. 47 CO 0) UJ O c o o o o o o z CO E i2 "cu E o c a) (O T f w o o o ci d ci oiivy saao onvy saao O O O C O T t C N O C O C D ^ r C M O onvy saao 48 Odds ratios and confidence intervals were compared to determine the effect of the inclusion of weather variables on the air pollutant variable. Examination of the odds ratios of pollutants in models with and without meteorology reveals that the inclusion of weather variables as covariates had only small effects on coefficients at a lag of 0 days. In all cases, odds ratios >1.0 for models including meteorology remained >1.0 after meteorological variables were taken out. Similarly, any meteorology-adjusted model with an odds ratio <1.0 retained an odds ratio <1.0 following the removal of weather variables. However, all confidence intervals were smaller in models without meteorological variables. 3.1.3.2 Sensitivity Analysis II: All events To reveal the effect of including all events within a 72-hour period of one another in the same strata, another sensitivity analysis was conducted using all shocks recorded during the study period. The results of this analysis are presented in Figure 3.6 below. Raw results are presented in Appendix 3.4. Grouping events had a very small effect on results, and did not change the direction of any relationship at a lag of 0 days. Odds ratios and confidence intervals were altered slightly, but no consistent pattern was observed. 3.1.3.3 Sensitivity Analysis III: Appropriate events only The final sensitivity analyses were conducted with the exclusion of inappropriate events. Upon comparison of these results to the initial analysis, no change in significance was observed for any pollutant. Only minor changes in odds ratios and confidence intervals distinguished the results of this analysis from the original, and no patterns in these changes were observed at a lag of 0 days. The results of this sensitivity analysis are presented in Figure 3.7. Raw results are included in Appendix 3.5. 49 c CU > CO i_ cu o X I CZ CD CD W C CU > CD ro ro c CO ZJ OJ c ro 13 "5 Q . H— o CO cu CO >* ro c ro i_ cu CO > CD O X J O CO O T— OS E o he CL cu CO CZ ro o X J i cu o X I *— CO o ro c erv cu i we nee les cu •o ro H— C ro o > o ro o L O ' r a a> o ro o c i o 3 o sh C O CO oi Ratio 'ely Ratio u Q_ CO cu X I Q . X I CO o OJ C D co no cu i cn C iqu ro u. c> c CU > CU ^ o I o > CD CZ cu > cu c <D > 03 O CO CO ^ CM O T-1 O O CZJ CJ o onvy saao I D ^ - C N O C O C O - l - C N O oiivy saao cz 03 > 03 CZ 03 > 03 o o c 03 > 03 O CO CO <t CM o O O O O O O CO CO t CN o ^ - " 0 0 0 0 0 onvy saao onvy saao c cu > cu » o £ LU cz 03 > 03 CZ 03 > 03 CN o c cu > cu cu > cu C O - t N O C O C D ^ C M O onvy saao CO 1 o c > 03 CZ cu > cu O LO O LO CO CN CN ^ p i n o T— C3 C3 onvy saao ^ C N p c q c q ^ c N p onvy saao O CO CO 'C? CN o ^ 0 0 0 0 0 cz CD > cu a> > cu O LO O LO CO CN CN T -O LO o f- o d onvy saao onvd saao 50 o o o CO o o u C N o o Q. IO f (N O CO CO o o onvy saao onvy saao O O ( D T f N q o o o o o oi±vy saao ra T3 I ro o co ro £ o I o co B cu c CO CD . CO Ira £ ™ <•> ^ ro o ro co S o Q 8 | CD ^ O O , 3 CU 0 o > o tn 43 ro cn ra c > o o i l l I ro (1) "D O B • ra " & =3 Q. . co ro C O C D ^ N O C O C D ^ f W O onvu saao in C N t M O C O O ^ I M O T - ' T ^ ^ O O O C D O onvd saao o o C D C N O CO CD f CM O o o o o o OLLVH saao o LU Q. a. ro I f l t C N O C O C O ^ C M l onvy saao L O O m O L O O l O O c o r o c N i c N ^ ^ d d onvy saao 1 o Q. Q. ro o LO o cn o co c ri p i ci r ^ 6 I ouvy saao 51 3.2 EXPOSURE ASSESSMENT 3.2.1 Study subjects 3.2.1.1 Recruitment A separate subset of the ICD patient population from which patients considered in the epidemiological analysis was drawn were the focus of the exposure analysis. Not all patients included in the exposure analysis had experienced ICD shock. Fourteen patients from the SPH Clinic and 20 patients from the Burrard Clinic were contacted by letter and then by telephone in a first round of recruiting. After telephone calls, 6 patients from the Burrard Clinic and 7 patients from the SPH Clinic had decided not to participate in the study. Two patients from the Burrard Clinic stated that they were too busy, 1 cited health reasons, and 3 cited other reasons for declining to participate. Four patients from the SPH Clinic were going away, 2 were too busy, and 1 declined to participate for health reasons. After introductory meetings, a further 4 patients from the Burrard Clinic and 4 patients from the SPH clinic chose not to participate. Two patients from each clinic declined because the pump was too loud, 3 declined for other reasons, and one patient declined because he was preparing to have surgery. One subject did not meet the study eligibility criteria since his wife was a smoker. In a second round of recruiting initiated because the target sample size had not been met, 17 patients from the Burrard Clinic were contacted by letter and followed up by telephone. After telephone calls, 10 had decided not to participate in the study. One individual was too busy to take part, 3 were going away, 1 was unable to participate due to brain injury, 1 declined because he feared that the study might have a negative impact on his health, and 4 were simply not interested. In total, 20 individuals agreed to participate in the study, yielding a 39% (20/51) participation rate. Three of these individuals were from the SPH Clinic (21% participation rate) and 17 were from the Burrard Clinic (46% participation rate). One subject was unable to complete the minimum number of sampling sessions due to poor compliance. This individual was excluded from all data analyses, leaving the total number of subjects at 19, slightly lower than the original target of 20-25 participants. 3.2.1.2 Population characteristics The exposure assessment study population consisted of 16 male subjects aged 33 to 74 years (mean: 63) and 3 female subjects aged 32 to 45 years (mean 40). All subjects had an ICD and resided in Vancouver, North Vancouver, Burnaby, Richmond, Surrey, Whiterock, or Langley. Locations of the homes of each subject and the South Burnaby monitoring station at which measurements of ambient pollutants were taken are shown in Figure 3.8 below. Seven of the 19 subjects (37%) either worked or attended university regularly. All subjects were current non-smokers and did not live with smokers with the exception of one subject who lived with a smoker and one subject who worked with a smoker. For these subjects, the smokers agreed to smoke outside or not at all on sampling days. For this reason, these subjects were allowed to remain in the study despite not meeting the initial study criteria. 52 Figure 3.8 Map of Study Area located in Vancouver, Canada 3.2.1.3 Compliance For the 19 ICD subjects, 140 personal samples were collected between May 14 and August 31, 2001. Each subject included in the study underwent at least 7 sampling sessions, yielding a 100% compliance rate. However, some personal exposure samples were invalid, resulting in less than 7 samples for some individuals. In these cases, if subjects agreed to provide additional samples and it was possible to maintain the original criteria of >8 days between sampling sessions, one extra sample was scheduled. Originally scheduled sample days were changed to accommodate 5 individuals, again keeping consecutive samples at least 8 days apart. 3.2.2 Data collection 3.2.2.1 Exclusion criteria Fifteen samples were excluded from analyses because pre- or post-sampling flow was outside of 4.0 L/min + 10% (4 samples), because sample duration was less than 75% of the target of 24 hours (6 samples), or because post-sample flow or sample volume could 53 not be determined as a result of the tubing having become disconnected or sampler o-ring having become loose (5 samples). A further 2 samples were deleted due to laboratory error, leaving 123 (88%) valid samples, or 5-7 valid personal exposure samples for each subject on 58 different weekdays. The mean personal sample flow rate was 3.9 L/min (SD: 0.1; range 3.6-4.4 L/min). Eighty-two percent of the samples had flows of 4.0 L/min+5% (3.8-4.2 L/min). The mean personal sample duration was 23:49 hours (SD: 42 min; range 19:52-25:22 hours) and the mean personal sample volume was 5.61 m3 (SD: 0.3; range 4.7-6.5 m3). 3.2.2.2 Sample duration Personal samples were started and stopped between 8:00 am and 11:00 am, with the aim of 24-hour sample durations. Mean start and stop times for personal measurements were 9:25 am (SD 55 min; range 7:39-11:15 am) and 9:18 am, respectively. Average ambient start and stop times were compared to personal start and stop times for each day. The mean percentage of personal sample duration that overlapped ambient sample duration was 94%. Pump-reported times were compared to changes in wristwatch-recorded time to ensure that pumps were timing sample duration accurately. The average absolute difference between wristwatch time and pump time was 1.4 minutes. This comparison was used to validate the used of wristwatch-feported times when pump timers were not functional. 3.2.2.3 Quality control filters Three quality control filters were weighed repeatedly throughout the study to ensure that laboratory conditions and weighing procedures remained consistent over time. Graphs of consecutive weighings over time, displaying the mean, warning (mean ± 2 SD) and control (mean ± 3 SD) limits are attached in Appendix 3.6. The quality control filters showed no systematic change over time. Warning and control limits of quality control fdters are presented in Table 3h below. Table 3 h . Warning (mean ± 2 SD) and control (mean ± 3 SD) limits for Quality Control Filters Quality Control Filter Warning Limit (ug) Control Limit (ug) 1 ± 16.3 ±24.4 2 ±31.3 ±46.9 3 ± 17.5 ±26.3 3.2.3 Personal exposures to PM2 . 5 , absorbance and sulfate 3.2.3.1 Exposure distributions Distributions of personal exposure variables were examined prior to analyses since assumptions about exposure distributions were important for some statistical analyses that were conducted. Upon examination of graphical representations of the distributions of personal PM2.5, absorbance and sulfate values, it was established that these variables were not normally distributed. Both natural log-transformed and untransformed results 54 are presented in Tables 3i-k below. Histograms of personal exposure variable distributions are presented in Appendix 3.7. 3.2.3.2 Personal PM 2 . 5 Throughout the study, 9% of all personal filters were lab blanks (n=15) and 9% were field blanks (n=15). PM2.5 mass differences were calculated for all lab and field blank filters. The mean mass differences were -1.6 ug (SD: 5.9 ug) for lab blanks and 12.6 ug (SD: 24.7 ug) for field blanks. The mean mass increase on all sample filters was 113.4 ug, corresponding to a mean sample concentration of 17.9ug/m . The mean mass increase on field blanks, calculated as the post-sampling weight minus the pre-sampling weight, was subtracted from all personal samples before data analysis. The limit of detection (LOD) was defined as the mean plus 3 standard deviations of all lab blanks divided by the mean sample volume of 5.61 m3. This resulted in a personal PM2.5 sample LOD of 2.9 ug/m3. Two personal samples were below the LOD. These samples were flagged but retained in the data analysis. Table 3i. Distribution of Personal PM2 .5 Concentrations Subject n PM2.5 Concentrations (ug/m3) Range AM SD GM GSD 1 6 18.3-42.7 29.7 10.1 28.3 1.4 2 7 9.6-42.0 24.2 11.9 21.5 1.7 3 6 4.2-70.3 25.6 23.2 18.4 2.5 4 7 ' 3.5-30.9 14.2 9.0 11.7 2.0 5 7 6.2-23.0 14.3 5.3 13.4 1.5 6 7 5.1 -29.7 14.9 7.4 13.4 1.7 7 6 0.9-25.5 15.1 9.8 10.1 3.6 8 7 7.1 -25.7 11.0 6.6 9.9 1.6 9 7 8.4-16.5 11.8 3.4 11.4 1.3 10 7 4.0- 19.7 12.2 5.8 10.8 1.8 11 6 3.9-16.1 9.2 4.1 8.4 1.6 12 6 7.0-52.8 20.2 17.3 15.6 2.1 13 7 8.9 - 24.3 16.5 6.0 15.5 1.5 14 6 5.0-45.2 13.9 15.4 10.0 2.2 15 7 11.1 -41.7 27.0 11.2 24.8 1.6 16 5 0.2-7.1 4.5 2.8 2.9 4.2 17 7 11.0-37.9 19.3 10.4 17.4 1.6 18 5 17.1 - 174.4 51.3 68.8 31.2 2.6 19 7 6.8-20.6 11.6 4.5 11.0 1.4 ALL 123 0.2-174.4 17.9 18.1 13.5 2.2 AM=arithmetic mean, SD=standard deviation, GM=geometric mean, GSD=geometric standard deviation 3.2.3.3 Personal absorbance For personal absorbance, field blank values were factored into a reference value used in the calculation of absorbance. Therefore it was not necessary to correct for nonzero field blank values. Assuming the mean sample volume of 5.61 m3, the mean absorbance values were 0.61 (SD: 0.79) for lab blanks and 0.18 (SD: 0.38) for field blanks. The mean sample absorbance was 9.3 absorbance units. The LOD for absorbance was 55 defined as 3 times the SD of field blank absorbance. For personal samples, this LOD corresponded to 1.15 absorbance units. No personal samples were below the LOD. Table 3j. Distribution of Personal Absorbance Values Subject n Absorbance (10_5m"') Range AM SD GM GSD 1 6 6.6 - 69.0 20.7 23.9 14.4 2.3 2 7 4.0- 12.4 7.1 2.9 6.6 1.5 3 6 11.7-28.0 17.0 5.8 16.3 1.4 4 7 1.9-5.5 3.2 1.4 3.0 1.5 5 7 6.3-46.3 16.8 13.3 14.0 1.8 6 7 2.7-10.5 6.6 2.5 6.2 1.5 7 6 4.6-19.3 12.9 5.1 11.8 1.7 8 7 6.7-15.2 9.1 2.9 8.8 1.3 9 7 3.7-11.4 7.8 2.9 7.3 1.5 10 7 1.4-14.3 7.2 4.1 5.9 2.1 11 6 2.8-13.3 7.7 3.6 6.9 1.7 12 6 2.2-6.6 4.1 1.9 3.7 1.6 13 7 2.5-37.1 13.1 13.8 8.4 2.7 14 6 2.4 - 9.2 5.9 2.3 5.4 1.6 15 7 6.0-16.4 9.1 3.8 8.6 1.4 16 5 1.8-4.7 3.4 1.3 3.1 1.5 17 7 8.6-16.4 12.3 2.4 12.1 1.2 18 5 1.9-10.0 5.1 3.3 4.2 2.0 19 7 3.8-10.0 6.9 2.4 6.5 1.4 ALL 123 1.4-69.0 9.3 8.4 7.3 2.0 AM=arithmetic mean, SD=standard deviation, GM=geometric mean, GSD=geometric standard deviation 3.2.3.4. Personal sulfate For ion chromatographic determination of sulfate levels, nonzero sulfate values for control standards repeated in different analytical runs were not significantly different according to paired sample t tests. Therefore it was not necessary to correct for run-specific changes. Assuming the mean sample volume of 5.61 m3, the mean sulfate concentration was 0.12 ug/m3 (SD: 0.04) for lab blanks and 0.12 ug/m3 (SD: 0.04) for field blanks. The mean sulfate concentration on all sample filters was 1.28 ug/m3 (SD: 0.92). The mean field blank sulfate mass increase was 0.67 ug. This value was subtracted from all personal sulfate samples prior to data analysis. The concentration LOD for personal sulfate samples was defined as 3 standard deviations of all field blank concentrations assuming the mean personal sample volume of 5.61 m3. This value was 0.12 ug/m3. No personal sulfate samples were below the LOD. 56 Table 3k. Distribution of Personal Sulfate Concentrations Subject n Sulfate Concentrations (ug/m3) Range AM SD GM GSD 1 6 0.3-2.0 0.8 0.6 0.7 2.1 2 7 0.5-2.5 1.0 0.7 0.9 1.8 3 6 0.6-1.7 0.9 0.4 0.9 1.5 4 7 0.3-2.2 1.4 0.7 1.2 2.1 5 7 0.5-2.1 1.2 0.6 1.1 1.7 6 7 0.4-2.5 1.3 0.7 1.1 1.8 7 6 0.6-2.3 1.0 0.7 0.9 1.8 8 7 0.4-1.8 1.2 0.6 1.0 1.8 9 7 0.8-2.5 1.4 0.7 1.3 1.5 10 7 0.3-2.2 1.0 0.7 0.8 2.2 11 6 0.4-1.2 0.9 0.3 0.8 1.5 12 6 0.3-1.9 1.0 0.6 0.8 2.1 13 7 0.3-2.5 1.1 0.7 0.9 1.9 14 6 0.5-0.8 0.7 0.1 0.7 1.2 15 7 0.7-2.2 1.6 0.6 1.5 1.6 16 5 0.2-1.0 0.5 0.3 0.5 1.8 17 7 0.7-8.6 2.2 2.9 1.5 2.3 18 5 0.4-2.7 1.2 1.0 1.0 2.2 19 7 0.4-2.3 1.1 0.7 0.9 1.9 A L L 123 0.2-8.6 1.2 0.9 0.9 1.9 AM=arithmetic mean, SD=standard deviation, GM=geometric mean, GSD=geometric standard deviation Personal exposure to sulfate appeared relatively consistent between subjects, with arithmetic means ranging from 0.5-2.2 ug/m3, or about fourfold. Both PM2.5 and absorbance, on the other hand, were more variable, with arithmetic mean exposures ranging about sixfold, from 4.5-29.7 ug/m3 and 3.2-20.7 absorbance units, respectively. Coefficients of variation for all samples were 1.0 and 0.9 for PM2.5 and absorbance, respectively, as compared to 0.75 for sulfate. 3.2.3.4.1 Data quality for sulfate analyses Personal sulfate concentrations were determined at the School of Occupational and Environmental Hygiene (SOEH) laboratory located at the University of British Columbia. Since ion chromatographic laboratory analysis of ambient and personal sulfate concentrations was performed at different laboratories, a subset of randomly selected personal filters were analyzed for sulfate at both sites to confirm that there were no systematic difference in concentrations. The mean SOEH - Environment Canada percent concentration difference was -0.15% (SD: 8.9%; range -8.0 - 21.7%), corresponding to a concentration difference of 0.12 ug/m3. Systematic differences were not observed. Concentrations reported by each lab are presented in Figure 3.9 below. Raw data is presented in Appendix 3.8 57 Figure 3.9. Comparison of sulfate analyses at Environment Canada and SOEH laboratories Sulfate Analyses at 2 Laboratories 3.00 1 2 3 • Environment Canada • SOEH 4 5 6 7 Sample ID 8 9 10 11 3.2.4 Ambient exposures 3.2.4.1 Missing ambient samples Ambient PM2.5, absorbance and sulfate samples were collected on each of the 61 days on which personal samples were collected. Of these, 9 PM25, 10 absorbance, and 4 sulfate samples were deleted due to invalid field or laboratory methods as determined by Environment Canada. Thus 160 (87%) of the 183 ambient samples were successfully collected. Ambient PM2 5 concentrations for the South Burnaby monitoring site were available for 49 of 58 days (84%) for which personal exposures were successfully measured. Ambient sulfate concentrations measured at the South Burnaby monitoring station were available for 54 of 58 days (93%), and ambient absorbance was available for 48 of the 58 days with personal measurements (83%). 3.2.4.2 Ambient PM2.5 For ambient PM2.5 samples, the mass increase measured on field blanks ranged from 1.2 to 3.0 ug (mean 1.8 ug). The average mass increase on sample filters was 84.0 ug (SD: 39.3), corresponding to a mean sample concentration of 7.0 ug/m3. A week-specific average nonzero field blank value was subtracted from each sample prior to analysis. The LOD, calculated as the mean plus 3 times the standard deviation of these field blank values, was 3.6 ug, corresponding to a concentration LOD of 0.3 ug/m3. No ambient PM2 5 samples were below the LOD. 3.2.4.3 Ambient absorbance As in the calculation of personal absorbance, field blank values were factored into a reference value used in the computation of ambient absorbance. The LOD for ambient absorbance was defined as 3 times the SD of field blank absorbance. Assuming the mean sample volume of 12.0 m 3 for ambient samples, the LOD corresponded to 1.32 58 absorbance units. The average ambient absorbance value was 5.6 absorbance units. Four ambient samples were below the LOD. These samples were flagged but retained in the data analysis. 3.2.4.4 Ambient sulfate Ambient sulfate samples were analyzed by ion chromatography at the Environment Canada laboratory in Ontario. According to the procedures of this laboratory, sulfate concentrations were not corrected for field blank values and an LOD was not calculated, since field blank values approximated zero. The average mass increase on sample filters was 15.6 ug (SD: 8.5), corresponding to an average concentration of 1.2 ug/m3. 3.2.4.5 Distribution of ambient exposures Distributions of ambient exposure variables were also considered prior to statistical analyses. Histograms of ambient exposure variables are presented in Appendix 3.10. Ambient sulfate concentrations approximated a normal distribution, while ambient PM2.5 and absorbance values were not normally distributed. Log transformed and non-log transformed results are presented in Table 31. Data are presented separately for days on which personal measurements were taken and for all days within the study period, since ambient samples but not personal samples were attempted on all days within the study period. Distributions of pollutant values for days with personal measurements are most relevant since these values are used in further analyses. Summary statistics are presented for all days within the study period to demonstrate that personal measurements were not taken on days with systematically different pollutant concentrations than those of days on which samples were not taken. This is particularly important since personal samples were collected on weekdays, which may experience relatively higher emissions related to vehicle traffic. Histograms of ambient exposure variable distributions for days on which personal measurements were taken are included in Appendix 3.9 Table 31. Ambient P M 2 . 5 , absorbance and sulfate DAYS WITH PERSONAL MEASUREMENTS PM2.5(ug/m3) Absorbance (xl0"5m"') Sulfate (ug/m3) n 49 48 54 AM 6.7 5.6 1.3 SD 2.9 2.4 0.7 GM 6.1 5.1 1.1 GSD 1.6 1.5 2.3 Range 1.6-13.6 2.1-12.1 0.0-3.1 ALL DAYS WITHIN STUDY PERIOD PM25(ug/m3) Absorbance (xl0"5m"') Sulfate (ug/m3) n 93 94 103 AM 6.7 5.1 1.6 SD 2.9 2.6 3.2 GM 6.0 4.5 1.2 GSD 1.6 1.7 2.1 Range 1.6-16.0 1.0-12.7 0.0-33.3 AM=arithmetic mean, SD=standard deviation, GM=geometric mean, GSD=geometric standard deviation 59 The comparability of the two PM2.5 distributions confirmed that PM2.5 concentrations observed on sampling days were likely to be generalizable to the entire study period. While mean sulfate values were comparable for both distributions, however, the range in values differed considerably between days with personal measurements and all days. Examination of individual values revealed that this was due to one day with exceptionally high sulfate levels. 3.2.5 Absorbance and Various Carbon Measurements 3.2.5.1 Absorbance and Elemental Carbon (EC) An investigation of relationships between absorbance and other carbon components was considered useful to the determination of the utility of absorbance as a surrogate of various carbon measures. Absorbance values were regressed against concentrations of EC and other carbon measurements in co-located samples to evaluate their relationships. The linear regression of ambient EC versus ambient absorbance measured at the South Burnaby monitoring station during the study period had an R value of 0.77. This regression is shown in Figure 3.10a. Figure 3.10a. Relationship of absorbance and EC at South Burnaby monitoring station for the period 05/15/01 to 08/31/01 SOUTH BURNABY: Absorbance v Thermal Method EC y = 11 9 6 x + 0.91 0 0 .2 0.4 0 .6 0.8 1 1.2 1.4 Elemental Carbon (ugm-3) This relationship agrees well with the relationship observed between absorbance and EC for the Slocan monitoring station, another measurement site in the Vancouver area, for a subset of the study period (R2=0.71). This regression is shown in Figure 3.10b. 60 Figure 3.10b. Relationship of absorbance and EC at Slocan monitoring station for the period 08/13/01 to 08/31/01 S L O C A N : Absorbance v Thermal Method E C y = 14.53X + 2 98 o J , _ 0 0.2 0.4 0.6 0.8 1 1.2 Elemental Carbon (ugm3) The intercept and slope of each line were examined to identify differences in baseline absorbance measurements at either site and to confirm that similar increases in EC at either station are matched by similar increases in absorbance. The equations displayed on each of Figures 3.10a and 3.10b indicate similar slopes, at approximately 12 and 14.5 for the South Burnaby and Slocan sites, respectively. Comparison of the lines reveals an intercept near 1.0 absorbance units for the South Burnaby station as compared to an intercept of just under 3.0 for the Slocan station, indicating that baseline absorbance was slightly higher at Slocan than at South Burnaby. 3.2.5.2 Absorbance and other carbon components To evaluate the ability of absorbance to act as a surrogate for other carbon components, univariate regressions were conducted between absorbance and other carbon components measured at two Vancouver-area monitoring stations during the study period. The other components considered were CC (carbonate carbon) and E C (elemental plus carbonate carbon). The relationships of absorbance with CC are presented in Figures 3.10c-d. These relationships were weaker and less consistent between sites than was the absorbance:EC relationship. 61 Figure 3.10c. Relationship of absorbance and C C at South Burnaby monitoring station for the period 05/15/01 to 08/31/01 S O U T H B U R N A B Y : Abso rbance v Thermal Method C C 4 .44X + 2.31 E IA b cu o c ro A L. o V) < Carbonate Carbon (ugm ) Figure 3. 08/31/01 Od. Relationship of absorbance and C C at Slocan monitoring station for the period 08/13/01 to S L O C A N : Absorbance v Thermal Method C C y= 5.27x + 4.69 E b 0) o c ro n o in < 0.4 0.6 0.8 1 1.2 Carbonate Carbon (ugm' 3) 1.4 1.6 The relationship between absorbance and E C (EC + CC) was not as strong as that for EC alone, but the relationship was consistent between sites. These relationships are demonstrated in Figures 3.10e and 3.1 Of below. Relationships were better at the South Burnaby than at the Slocan site for all three carbon components. 6 2 Figure 3.10e. Relationship of absorbance and E C at South Burnaby for the period 05/15/01 to 08/31/01 o c re Si o CO A < S O U T H B U R N A B Y : Absorbance v Thermal Method E C y = 3.84X+ 1.36 R 2 = 0.60 Elemental plus Carbonate Carbon (ugm ) Figure3.10f. Relationship of absorbance and E C at Slocan for the period 08/13/01 to 08/31/01 CD O c re A i— o in < S L O C A N : Absorbance v Thermal Method E C y = 6.54x+ 1.72 Elemental plus Carbonate Carbon (ugm ) 3.2.5.3 Laser Integrating Plate Method (LIPM) and carbon measures To investigate the further application of the optical absorbance method as an indicator of different carbon measures, absorbance was also compared to an indicator of black carbon (BC) taken at the Slocan monitoring station in August of 2001. BC was measured using the indirect LIPM, a method of measurement of relative filter transmittance. The relationship is presented in Figure 3.10g below. 63 Figure 3.10g. Relationship of absorbance and a black carbon indicator at Slocan for the period 08/13/01 to 08/31/01 Absorbance v LIPM B C y = oo ix + 099 R 2 = 0.78 0 500 1000 1500 2000 2500 Black Carbon (10'9ngm3) 2 The R for the absorbance:LIPM BC relationship was 0.78. Although the relationship of optical absorbance and LIPM BC was good, and in the range of that for optical absorbance with elemental carbon (R2=0.77 and 0.71 for South Burnaby and Slocan, respectively), the relationship between LIPM BC and EC was not similarly strong. This relationship (R2=0.36) is presented in Figure 3.1 Oh below. Regressions of LIPM BC against CC and E C produced R 2 values of 0.24 and 0.43, respectively. These plots are presented in Appendix 3.10. Figure 3.10h. Relationship of L I P M BC and EC at Slocan for the period 08/13/01 to 08/31/01 Thermal Method E C vs LIPM B C y = o.oox + 0.06 R 2 = 0.36 0 500 1000 150Q 2000 2500 Black Carbon (10 ngni*) 64 3.2.6 Relationships between personal and ambient concentrations of the same exposure variable 3.2.6.1 All subjects combined The causality of associations between ambient pollution concentrations and health outcomes hinges on the presumption of a relationship between ambient concentration and personal exposure. Personal and ambient PM2.5 concentrations are compared both to test the validity of this assumption and to gain information regarding possible sources of exposure to particles. Figure 3.1 la shows the distributions of 24-hour averaged ambient and personal PM2.5 concentrations for all samples and subjects combined. This reveals that the median as well as the range of concentrations of personal P M 2 5 were considerably higher than those observed for ambient PM2.5. Similarly, it shows that the high values observed in personal samples were not seen in ambient measurements. The median daily ratio of personal to ambient PM25 was 3.2. This ratio was 1.4 and 0.8 for absorbance and sulfate, respectively. Boxplots of ambient and personal concentrations of absorbance and sulfate for all samples combined are included in Figures 3.11b and 3.1 lc. Plots reveal that overall distributions of personal and ambient sulfate concentrations are more similar than are overall distributions personal and ambient concentrations of PM2.5 or absorbance. Figure 3.11a. Ambient and personal P M 2 5 concentrations during the study period. Boxplots show 2 5 t h and 75 t h percentiles as well as medians. Concentrations are in ug/m\ Outliers are not shown. 50 40 CO 30 N= 93 123 ambient personal 65 Figure 3.11b. Ambient and personal absorbance values during the study period. Boxplots show 25' and 75th percentiles as well as medians. Concentrations are in 10"5m"'. Outliers are not shown. 2.0T 0 N= 94 123 ambient personal Figure 3.11c. Ambient and personal sulfate values during the study period. Boxplots show 25th and 751 percentiles as well as medians. Concentrations are in ug/mJ. Outliers are not shown. 66 3.2.6.2 Individual subjects To determine the degree of misclassification of exposure associated with the use of ambient PM2.5 measurements to define exposure, individual univariate regressions of personal versus ambient levels of PM2.5, absorbance and sulfate were conducted for each subject. Pearson correlations for these regressions are presented in Table 3m. Table 3m. Pearson correlations for individual regressions of personal versus ambient PM2.5, absorbance and sulfate. Variables are untransformed. Subject PI VI2.5 Absorbance Sul fate N r n r n r 1 5 -0.20 5 -0.07 6 0.96 2 5 -0.62 5 0.99 7 0.97 3 5 -0.81 4 0.81 6 0.97 4 6 0.42 6 -0.36 6 0.93 5 5 -0.93 5 0.19 6 0.81 6 5 -0.73 5 0.08 7 0.96 7 5 -0.92 4 0.67 6 0.82 8 7 -0.14 7 0.51 6 0.94 9 7 0.65 7 0.06 5 0.64 10 5 -0.42 5 0.20 6 0.96 11 6 -0.20 6 0.39 6 0.37 12 4 0.25 4 0.66 6 0.71 13 4 -0.58 4 0.66 7 0.95 14 5 0.11 5 0.11 6 0.31 15 5 -0.76 4 0.87 6 0.57 16 5 0.71 7 -0.77 5 0.82 17 7 0.35 7 0.28 7 0.20 18 4 0.75 4 0.90 5 0.91 19 5 -0.06 5 0.55 6 0.94 Mean 5.26 -0.16 5.21 0.35 6.05 0.78 Median 5 -0.20 5 0.39 6 0.91 Range 4-7 -0.93 - 0.75 4-7 -0.77-0.99 5-7 0.20-0.97 Histograms of Pearson correlations for individual personahambient relationships for each pollutant are presented in Figure 3.12. 67 Figure 3.12. Histograms of Pearson r values for individual personal:ambient regressions. PM2.5 12 10 8 6 -1.00 -.75 -.50 -.25 0.00 .25 .50 .75 1.00 PersonatAmbient r values ABSORBANCE -1.00 -.75 -.50 -.25 0.00 .25 .50 .75 1.00 Personal:Ambient r values SULFATE 12 10 8 6 4 2 0 -1.00 -.75 -.50 -.25 0.00 .25 .50 .75 1.00 Personal:Ambient r values 68 Personal to ambient correlations are both stronger and more similar between individuals for sulfate than for PM2.5. Whereas all subject-specific Pearson r values are positive for sulfate, those for PM2.5 demonstrate a wide range from -0.93 to 0.75. These differences indicate that changes in ambient sulfate coincide with changes in personal levels of the same pollutant more consistently than do changes in ambient PM2.5. Time activity and dwelling information was reviewed for individuals with low personahambient sulfate correlations in order to identify specific exposure-related activity or location information that could account for the poor correlation between ambient and personal exposures. Upon examination of time activity diaries, no reported behaviours or locations were identified as being associated with uniquely high or low exposures to sulfate. Similarly, time activity information was reviewed for individuals with particularly high or low personahambient PM2.5 and absorbance correlations, but recorded variables did not appear to explain the observed patterns. 3.2.6.3 Interrelationships between personal and ambient concentrations of different components of PM2.5 As mentioned above, the elucidation of relationships between personal and ambient levels of the same component of the air pollutant mix are important in reducing the misclassification of exposure that results from equating ambient concentrations with personal exposures. A related strategy involves the investigation of relationships between personal and ambient measures of different pollutants to identify ambient parameters that may act as surrogates for personal exposures. Regressions of personal and ambient levels of all measured air pollutants were conducted for all subjects together. To identify relationships between the different components of PM2.5, univariate regressions between all combinations of measured personal and ambient exposure variables were also conducted for all subjects together. Results of these regressions are presented in Table 3n. Regressions of each personal exposure variable with ambient PM2.5 are presented in Figure 3.13 below. Table 3n. Correlation coefficients for univariate regressions between all exposure variables for all subjects combined PEARSON r VALUES FOR REGRESSIONS FOR ALL SUBJECTS COMBINED Personal Exposure Variable n lnPM 2 5 InAbsorbance InSulfate Ambient Exposure Variable lnPM 2 5 100 0.30** 0.12 0.56** InAbsorbance 98 0.03 0.16 0.32** Sulfate 115 -0.04 0.06 0.70** **p<0.05 *p<0.10 69 Figure 3.13. Univariate regressions between ambient P M 2 5 and all measured personal exposure variables. Regressions were conducted for all subjects combined. 61 PM £ 5 „ ro 1 < E cn _c LO CN C L ro c o w <D Q. Rsq = 0.0904 .5 1.0 1.5 2.0 2.5 In ambient PM2.5 (lnugmA-3) 3 0 ABSORBANCE Rsq = 0 0154 In ambient PM2.5 (ugmA-3) In regression analyses of all subjects together, personal and ambient sulfate are relatively well correlated (r=0.7; p<0.05). Personal sulfate also has significant relationships with ambient PM2.5 and absorbance. The only other significant relationship observed occurred 70 between personal and ambient PM2.5. To account for repeated measures on one subject, regressions were also conducted by subject. Summaries of these regression analyses are presented in Table 3o. Table 3o. Correlation coefficients for univariate regressions between all exposure variables, by individual subject. Variables are untransformed. PEARSON r VALUES FOR INDIVIDUAL REGRESSIONS Personal Exposure Variable nt PM2.5 Absorbance Sulfate median range median range median range median Range Ambient Exposure Variable PM2.5 5 4 to 7 -0.20 -0.93 to 0.75 0.63 -0.56 to 0.98 0.67 -0.09 to 0.93 Ambient Exposure Variable Absorbance 5 4 to 7 -0.16 -0.87 to 0.98 0.39 -0.77 to 0.99 0.22 -0.57 to 0.92 Ambient Exposure Variable Sulfate 6 5 to 7 -0.06 -0.83 to 0.56 0.06 -0.78 to 0.90 0.91 0.20 to 0.97 tsample sizes reported are per subject Individual regressions revealed different trends than did pooled regressions. Because they account for intersubject variability, they are considered more representative of true relationships. While the reported ranges in r values indicated that correlations were inconsistent between individuals for all three pollutants, reported median r values revealed that personal absorbance and sulfate tended to correlate well with ambient PM2.5. When all subjects were considered individually, ambient absorbance and sulfate did not appear to be predictive of any personal measures other than absorbance and sulfate, respectively. 3.2.6.4 Composition of Personal PM2.5 Sulfate is considered to have no primary indoor sources, and the ratio of indoor to outdoor sulfate was close to unity, consistent with previous research demonstrating that sulfate penetrates indoors as well as PM2.5. This rationalizes the use of the personahambient sulfate ratio to estimate the contribution of ambient PM2.5 to personal PM2.5 exposure (Samat et al, 2000). Median ratios were calculated for each subject according to the formula presented in the Methods section. Over all subjects, the median portion of personal PM2.5 exposure contributed by PM2.5 of ambient origin was 38%. To further characterize personal PM2.5, the mass percentage of personal PM2.5 composed of sulfate was calculated. Expressed as a percentage of personal PM2.5, the mean mass percent sulfate for all personal samples was 10.4% (median: 7.9%; SD: 13.2). For ambient PM2.5, the mean mass percent sulfate was 21.2%. Since EC may be an indicator of specific sources of exposure to particles, it is useful to estimate the relative amount of EC found in personal and ambient samples. The South Bumaby absorbance :EC relationship presented above was used to estimate personal EC mass from personal absorbance values. The result was an estimated average personal EC mass of 0.7 ug/m3 (median: 0.5; SD: 0.7) as compared to the average ambient EC mass of 0.34 ug/m3 (median: 0.3; SD: 0.2). Expressed as a percentage of PM2.5, these values are 6.4% (median: 4.3; SD: 11.1) and 5.4% (median: 5.4; SD: 2.5) mass EC, respectively. 71 3.2.7 Predictors of personal exposure 3.2.7.1 Distance from monitoring station It was hypothesized that personal exposures would be better approximated by ambient concentrations for subjects living closest to the South Burnaby monitoring station. Individual personal to ambient correlation was regressed against distance from the ambient monitoring station to determine whether this played a role. Results of separate regressions for PM2.5, absorbance and sulfate are presented in Table 3p. Table 3p. Results of univariate regressions of personal to ambient Pearson r values against distance to POLLUTANT R r P PM2.5 0.02 0.14 0.56 Absorbance 0.16 -0.40 0.09 Sulfate 0.23 0.48 0.04 No relationship was seen between personal to ambient PM2.5 correlation and distance. A small negative relationship was observed between Pearson r values for absorbance and distance from South Burnaby, indicating that the ability of ambient measurements to predict personal exposures deteriorated as distance from ambient monitors increased. For sulfate, a small positive relationship was seen, indicating that distance from South Burnaby coincided with improved correlation between personal and ambient sulfate. 3.2.7.2 Time Activity Data and Dwelling Information 3.2.7.2.1 Time Activity Diaries Each subject was provided with a time activity diary at the beginning of each sampling session. Subjects periodically recorded information about their activity and location for the majority of each 30-minute period of the sampling session. Amounts of time spent indoors, outdoors or in transit were recorded along with more specific information about location. Subjects also recorded the length of time that they were exposed to cooking or tobacco smoke. Diaries were collected at the end of each sampling session, at which time subjects were requested to fill in any boxes that had been left blank. This data was entered into spreadsheets and used to calculate the proportion of time that subjects spent in different locations or on different activities. Since sampling durations were not all exactly 24 hours, proportions were calculated using session-specific sampling duration as the denominator. If the sample duration was within 30 minutes of 1440 minutes (24 hours), time activity information was entered for all 48 30-minute periods. If the sampling duration was less than 1410 minutes, 30-minute intervals during which sampling did not occur were omitted from analyses. One time activity log was collected in association with each successful sample, for a total of 123 logs. Time activity variables provide an indication of within-subject variability related to behaviours and locations thought to be associated with exposure to particles. Means of total time (minutes) as well as of the proportion of the total duration of each 72 sample spent in each location are presented. Values add to more than 1.0 in some cases, since not all categories are mutually exclusive. Subjects spent most of their time indoors (mean 0.84) with approximately three quarters of this time being with windows open (mean 0.74). Figure 3.14 below summarizes subject location over all subjects and sampling sessions. Figure 3.14. Location of subjects during all sampling sessions, expressed as mean proportions (values add to >1 because subjects may report being outdoors and in transit) Summary of reported time spent indoors, outdoors, and in transit for all samples 0 1 1 • indoors at home 0. • indoors at work • indoors other • outdoors • in transit A summary of the variables considered for analysis is presented in Table 3q below. Table 3q. Summary statistics of time activity information TOTAL MINUTES Mean SD Median MIN MAX time at home 1060 235 1080 540 1440 time at work 78 167 0 0 540 time at other indoor location 56 102 0 0 420 time with windows open 870 435 930 0 1440 time exposed to tobacco smoke 28 88 0 0 660 time cooking or near cooking 52 55 40 0 300 TOTAL time indoors 1194 159 1200 690 1440 OUTDOORS TOTAL time outdoors 137 125 90 0 510 time in transit 151 119 150 0 570 AS A PROPORTION OF SAMPLE DURATION Mean SD Median MIN MAX time at home 0.74 0.16 0.75 0.38 1.03 time at work 0.05 0.12 0.00 0.00 0.38 time at other indoor location 0.04 0.07 0.00 0.00 0.31 time with windows openf 0.74 0.37 0.80 0.00 1.50 time exposed to tobacco smoke 0.02 0.08 0.00 0.00 0.65 time cooking or near cooking 0.04 0.04 0.03 0.00 0.21 TOTAL time indoors 0.84 0.11 0.86 0.48 1.10 OUTDOORS TOTAL time outdoors 0.10 0.09 0.06 0.00 0.36 time in transit 0.11 0.08 0.10 0.00 0.40 t As a proportion of time indoors 73 Reported exposure to tobacco smoke is summarized in Figure 3.15 below. No tobacco smoke exposure was reported for almost 80% of sampling days. The remainder of samples were split fairly evenly between 30-60 and >90 minute exposures per 24-hour sampling session. Figure 3.15. Summary of exposure to tobacco smoke for all samples Reported exposure to tobacco smoke on sampling days • 0 mins • 30-60 mins • 90-120 mins • >120 mins 3.2.7.2.2 Dwelling Information Form A one-time dwelling information questionnaire collected from each participant provided particle exposure-related information about each subject's residence. Subjects fdled out and returned the questionnaires on or before the end of their final sampling session. Subject-reported information included the type and frequency of use (times per year) of air filtration and heating systems, the extent of carpet coverage as a percentage of floors in the home, and frequency of use of kitchen range hoods (always/sometimes/never) and attached garages (yes/no). Results of frequency of use of heating systems are not presented here because heating systems were not in use during the study period. Air filtration systems were not used by any of the subjects. Addresses collected on dwelling questionnaires were used to geolocate subject residences on an ArcView map. The distance of each dwelling from any major road (>4 lanes) and from the South Burnaby monitoring station was recorded. Dwelling information results were analyzed to identify between-subject variability in characteristics of subject residences thought to be associated with exposure to particles. Results are summarized in Table 3r and Figure 3.16 below. Table 3r. Summary statistics of dwelling information Mean SD Median MIN MAX Distance from major road (km) 0.7 0.9 0.4 0.01 4.0 Distance from ambient monitoring station (km) 14.6 7.7 13.1 1.7 30.6 Volume of dwelling (m3) 322 200 274 67 991 Carpet coverage (%) 77 23 80 20 100 74 Figure 3.16. Distribution of results for the attached garage and range hood dwelling variables. Results presented as percentages of all study subjects 16% • no attached garage • attached garage in use • attached garage not in use 11% 16% • range hood used sometimes • range hood used always The range hood variable provides relevant information since kitchen range hoods remove cooking smoke that may contain substantial amounts of particulate matter. Subjects that didn't have a kitchen range hood were included in the 'no range hood' category, while those reporting irregular use of a kitchen range hood were included in the 'range hood used sometimes' category, and subjects reporting that they always used a range hood were included in the 'range hood used always' category. As revealed in Figure 3.16 above, the majority of subjects (84%) reported some use of a kitchen range hood. The garage variable is another potentially important measure since it may allow particles released in vehicle exhaust to enter the subject's residence. Subjects without an attached garage are included in the 'no attached garage' category, while those with an attached garage are divided between 'attached garage in use' if they reported using the garage for an automobile, or 'attached garage not in use' if they reported no use of the garage for an automobile. Figure 3.16 demonstrates the frequency of use of an attached garage among study subjects. Sixty eight percent reported no used of an attached garage. 3.2.7.3 Multiple Regression Multiple regression analyses were conducted to determine the amount of variability in personal exposure measurements that could be explained by ambient levels of the same pollutant as well as by time activity and dwelling variables collected during the sampling period. 3.2.7.3.1 Description of Models Three separate mixed effects multiple linear regression models were conducted to predict personal P M 2 . 5 , absorbance and sulfate concentrations. Dependent variables were PERSONAL P M 2 . 5 , PERSONAL ABSORBANCE, and PERSONAL SULFATE for Models 1, 2 and 3, respectively. The single random effect was subject ID in all models. 75 3.2.7.3.2 Missing ambient pollutant data Ambient pollutant data was missing for 8, 10 and 4 dates for PM2.5, absorbance, and sulfate, respectively. These values were filled in according to the procedure outlined in the Methods section. Distributions of ambient pollutants during the study period are presented in Appendix 3.11. 3.2.7.3.3 Excluded variables Some information that was collected as dwelling information was not considered for inclusion in models. Variables were not considered if they were thought to be incorporated in other variables, if they had insufficient variability to contribute to models, or if they were not relevant during the study period. Definitions of these variables, as well as the reasons for their exclusion, are included in Table 3s. In addition to these dwelling variables, personal activity level, a categorical variable collected on time activity diaries, was excluded because it was not considered relevant to analyses. Table 3s. Dwelling variables not considered for inclusion in models VARIABLE CATEGORIES REASON FOR EXCLUSION TYPE OF DWELLING Single house/townhouse/ apartment building/other Correlated with home volume STREET CANYON (RATIO OF DISTANCE FROM DWELLING TO STREET AXIS<1.5) Yes/no Correlated with home volume FLOOR NUMBER - Correlated with home volume CORNER UNIT? Yes/no Insufficient variability SIDE OF BUILDING North/East/South/West Insufficient variability AREA OF HOME - Correlated with home volume NUMBER OF ROOMS IN HOME - Correlated with home volume TYPE OF VENTILATION Natural only/System Insufficient variability AIR CONDITIONING Yes/No Insufficient variability HEATING SYSTEM Electrical/Gas/Forced Air/ Furnace/Hot Water/ Radiator/Other Not relevant during study period (summer) FIREPLACE Yes/No Not relevant during study period (summer) INDEPENDENT AIR FILTER Yes/No Insufficient variability NUMBER OF WINDOWS OPEN - Incorporated in windows variable on time activity diary 3.2.7.3.4 Categorical variables Variables indicating use of attached garage and proximity to major roads were collected as categorical variables. In addition, continuous variables with particularly skewed or irregular distributions were collapsed into categories and introduced to models as categorical variables. These changes were made to accord less weight to extreme values unlikely to have an important effect. For example, while exposure to cooking is thought to increase the potential for exposure to particles, it is unlikely that living in a home with an estimated 60% carpet will result in 2 times as much particle exposure as will living in 76 a home with an estimated 30% carpet. Variables converted to categories are evident in Table 3t. Table 3t. All independent variables introduced to models CONTINUOUS DESCRIPTION AMBIENT P M 2 5 Natural log-transformed 24-hour averaged PM2.s concentration in lnug/m3 AMBIENT ABSORBANCE Natural log-transformed 24-hour averaged absorbance value in lnl0"5m"' AMBIENT SULFATE 24-hour averaged SO4 concentration in ug/m3 TRANSIT Subject-reported time in transit as a proportion of sample duration INDOORS Subject-reported time indoors as a proportion of sample duration OUTDOORS Subject-reported time outdoors as a proportion of sample duration VOLUME Volume of subject dwelling in m3 [ABLES WINDOWS Subject-reported time indoors with windows open as a proportion of sample duration [ABLES CATEGORICAL # o f c a t e g o r i e s c a t e g o r y DESCRIPTION VAR INSTANT FOR URATI ON OF VDYPERIOD MAJOR ROAD 2 1 Subject dwelling is situated <150 m from any road with > 4 lanes H Z Ed INSTANT FOR URATI ON OF VDYPERIOD 0 Subject dwelling is situated >150 m from any road with > 4 lanes 'END! INSTANT FOR URATI ON OF VDYPERIOD GARAGE 2 1 Subject reports regular use of attached garage 'END! INSTANT FOR URATI ON OF VDYPERIOD 0 Subject reports no use of attached garage PL-W y o ^ CARPET 2 1 >80% of subject dwelling carpeted Q 0 <80% of subject dwelling carpeted 2 Subject reported >120 minutes of cooking or being near cooking :NTFOR EA ING SESSIO COOKING 3 1 Subject reported <120 minutes of cooking or being near cooking :NTFOR EA ING SESSIO 0 Subject reported neither cooking nor being near cooking in one sampling session DIFFERE SAMPL* ETS 2 1 Subject reported some exposure to tobacco smoke DIFFERE SAMPL* 0 Subject reported no exposure to tobacco smoke 3.2.7.3.5 Continuous variables Independent continuous variables were required to meet a number of criteria before introduction into models. It was required that univariate regression of each continuous independent variable against the dependent variable return an R2>0.01 and p<0.25. All independent variables meeting these criteria were initially introduced into models. Results of these univariate regressions are presented in Table 3u below. Variables that did not meet the criteria for inclusion are indicated in bold. 77 Table 3u. Results of univariate regressions between dependent and independent variables considered for inclusion in models D E P E N D E N T V A R I A B L E P E R S O N A L PM2.5 P E R S O N A L A B S O R B A N C E P E R S O N A L S U L F A T E INDEPENDENT V A R I A B L E R 2 P INDEPENDENT V A R I A B L E R 2 P INDEPENDENT V A R I A B L E R 2 P AMBIENT PM2.5 0.09 O.01 AMBIENT A B S O R B A N C E 0.03 0.12 AMBIENT SULFATE 0.49 <0.01 TRANSIT 0.04 0.03 TRANSIT 0.10 O.01 T R A N S I T <0.01 0.60 INDOORS 0.05 0.02 INDOORS 0.08 .01 INDOORS 0.02 0.13 OUTDOORS 0.01 0.20 O U T D O O R S <0.01 0.38 OUTDOORS 0.03 0.05 W I N D O W S <0.01 0.96 WINDOWS 0.03 0.05 WINDOWS 0.12 <0.01 VOLUME 0.02 0.12 VOLUME 0.01 0.18 VOLUME 0.01 0.18 3.2.7.3.6 Intercorrelation of independent variables Before introduction to the models, it was ascertained that all continuous variables had Pearson intercorrelations of I rl <0.5. A correlation matrix showing Pearson correlations for all relevant combinations of independent continuous variables is shown in Table 3v. Table 3v. Pearson correlations between all continuous variables considered for inclusion in models T R A N S I T I N D O O R S O U T D O O R S V O L U M E W I N D O W S T R A N S I T 1.00 -0.48 -0.03 0.05 0.30 I N D O O R S -0.48 1.00 -0.69 -0.19 -0.28 O U T D O O R S -0.03 -0.69 LOO 0.20 0.07 V O L U M E 0.05 -0.19 0.20 1.00 -0.05 W I N D O W S 0.30 -0.28 0.07 -0.05 1.00 A M B I E N T P M 2 5 0.23 -0.32 0.31 -0.08 0.32 A M B I E N T 0.11 -0.18 0.27 -0.09 0.31 A B S O R B A N C E A M B I E N T S U L F A T E 0.18 -0.18 0.17 -0.05 0.28 Boxplots were created to determine relationships between continuous and categorical independent variables. Separate boxplots of categorical variables and all dependent and independent continuous variables are presented in Appendix 3.12. Categorical variables MAJOR ROAD, GARAGE, CARPET, COOKING, and ETS did not appear to have strong correlations with any continuous variables. However, some trends were identified from the plots. Dwellings within 150 meters of a major road were observed to be smaller in volume than dwellings >150 meters from a major road. Time spent cooking also appeared to have inverse relationships with both home volume and time spent outdoors. Crosstabulations were also reviewed to determine whether or not categorical variables were intercorrelated with one another. Crosstabulations are included in Appendix 3.13. It was revealed that GARAGE categories were correlated with both MAJOR ROAD and ETS categories. Specifically, all subjects reporting exposure to tobacco smoke reported use of an attached garage, and all subjects reporting use of an attached garage lived >150 meters from a major road. Despite these findings, all variables were introduced into 78 models on the basis that they were believed strongly to be associated with particle exposure. 3.2.7.3.7 Procedure for running models A manual backward stepwise multiple regression approach was employed for each model. Subject was included as a random grouping variable in all models with the intercept alone as random. All variables meeting the initial criteria for inclusion were introduced to the models. One variable was removed after each run of a model until all remaining variables had coefficients with the hypothesized sign and p values <0.3. After each run, any variable returning a coefficient with a sign opposite to that hypothesized was removed from the model first, and the model was run again. Variables predicted to have a positive linear relationship with the dependent variable were expected to return positive coefficients. The directions of hypothesized relationships are presented in Table 3w. When all variables returned coefficients with the hypothesized signs, variables with p>0.3 were removed one at a time, starting with the variable with the highest p value. Table 3w. Hypothesized direction of relationship with personal exposure variables INDEPENDENT VARIABLE SIGN EXPECTED ON COEFFICIENT RATIONALE AMBIENT PM2.5 + Some penetration indoors expected AMBIENT ABSORBANCE + Some penetration indoors expected AMBIENT SULFATE + Some penetration indoors expected TRANSIT + Expected to increase exposure to vehicle emissions INDOORS - Expected to decrease exposure to ambient particles OUTDOORS + Expected to increase exposure to ambient particles VOLUME - Expected to increase dispersion of particles WINDOWS + Expected to increase exposure to ambient particles MAJOR ROAD + Expected to increase exposure to vehicle emissions GARAGE + Expected to increase exposure to vehicle emissions CARPET + Expected to increase exposure to indoor source particles COOKING + Expected to increase exposure to particles from combustion ETS + Expected to increase exposure to particles from combustion 3.2.7.3.8 OUTDOORS variable The OUTDOORS and INDOORS variables did not add exactly to 1.0 because time activity data was quantified to the nearest V2 hour, while sample duration was quantified to the nearest minute. As shown in Table 3v, however, the correlation between the two variables had an absolute value >0.5, and therefore the variables could not be included in the same model. The PM2.5 model was therefore attempted separately with INDOORS and OUTDOORS forced in as independent variables to determine whether the distinction had an important impact. It was established that the inclusion of either variable produced 79 comparable coefficients. The results of the two models are presented in Table 3x below. The O U T D O O R S variable was used in all models thereafter. Table 3x. Results of PM2.5 model run separately using INDOORS and OUTDOORS variables. Results are presented for initial models, prior to removal of any variables. COOKING 1 and COOKING 2 refer to (0 minutes >cooking <120 minutes) and (cooking >120 minutes) respectively. Using OUTDOORS Using INDOORS Variable Coefficient P Coefficient P OUTDOORS 0.94 0.34. -INDOORS - - -0.93 0.29 AMBIENT PM2.5 0.31 0.07 0.32 0.05 MAJOR ROAD -0.18 0.59 -0.15 0.63 COOKING 1 -0.18 0.36 -0.19 0.32 COOKING 2 -0.31 0.36 -0.33 0.33 ETS -0.32 0.17 -0.34 0.14 TRANSIT 2.21 0.05 1.62 0.21 VOLUME -0.001 0.14 -0.001 0.13 GARAGE 0.13 0.70 0.11 0.75 3.2.7.3.9 Model results The final results of each of the three models are presented in Table 3y below. Table 3y. Results of final mixed effects linear regression models MODEL DEPENDENT INTERCEPT INDEPENDENT coefficient p value VARIABLE p value VARIABLE 1.89 <0.0001 AMBIENT PM2.5 0.38 0.02** 1 PERSONAL P M 2 . 5 TRANSIT 2.14 0.04** VOLUME -0.0006 0.28 AMBIENT 0.20 0.15 PERSONAL ABSORBANCE 0.71 0.01 ABSORBANCE 2 TRANSIT 1.25 0.11 WINDOWS 0.41 0.02** GARAGE 0.69 0.01** PERSONAL SULFATE -0.95 <0.0001 AMBIENT SULFATE 0.60 <0001** 3 OUTDOORS 0.84 0.08* WINDOWS 0.21 0.08* VOLUME -0.0005 0.12 * * p < 0 . 0 5 * p < 0 . 1 These results are summarized in the equation of each final model in Table 3z below. 80 Table 3z. Equations of final mixed effects models MODEL EQUATION 1 lnpPM2i= 1.89* + 038(lnaPM2J)* + 2.14(TRANSIT)* - 6.00 x 10"4(VOLUME) 2 Inpabs = 0.71 + 0.20(lnaabs) + 1.25(TRANSIT) + 0.41 (WINDOWS)* + 0.69(GARAGE)* 3 lnpS04 = -0.85* + 0.60(aSO4)* + 0.84(OUTDOORS)* + 0.21 (WINDOWS)* -4.65x10'4(VOLUME) *p<0.05 lnpPM2! = natural log-transformed personal P M 2 5 (lnugm"3); lnaPM,j = natural log-transformed ambient PM2.5(Inugm'3) Inpabs = natural log-transformed personal absorbance (lnl0"5m"'); Inaabs - natural log-transformed ambient absorbance (lnl0"5m"') lnpS04 = natural log-transformed personal sulfate (lnugm'3); aS04 = ambient sulfate (ugm3) In each case, the ambient exposure variable corresponding to the personal exposure variable predicted remained in the model. It is clear from the highly significant p value associated with AMBIENT SULFATE that this is a better predictor of its personal counterpart than is AMBIENT ABSORBANCE. AMBIENT P M 2 5 also returned a significant p value (<0.05). Coefficients could not be compared directly between models since a one-unit change in the dependent variable is likely to have a different importance for each of the three personal exposure variables. The TRANSIT variable remained in models for both PM2.5 and absorbance, WINDOWS remained in models for both absorbance and sulfate, and VOLUME remained in models for both PM2.5 and sulfate. Additionally, GARAGE remained in the absorbance model, while OUTDOORS remained in the sulfate model. For personal PM2.5, the most important predictor appeared to be TRANSIT, while GARAGE seemed to be the most important predictor of personal absorbance. For personal sulfate, AMBIENT SULFATE was followed in importance by OUTDOORS as a predictor. 3.2.7.3.10 Model fit R2 values of univariate regressions of predicted values against each dependent variable are presented in Table 3aa below. Table 3aa. R 2 values predicted vs. observed values of dependent variables MODEL I DEPENDENT VARIABLE I R2 1 PERSONAL PM 2 . 5 0A6 2 PERSONAL ABSORBANCE 063 3 PERSONAL SULFATE 0.65 These values indicate that variability in personal exposure to sulfate and absorbance was better explained by mixed effects modeling than was variability in personal exposure to PM2.5. They imply that more than half of the variability in personal absorbance and sulfate may be explained by independent variables. 3.2.7.4 Sensitivity Analyses Three sensitivity analyses were conducted for each model. The first two focused on exposure to tobacco smoke. These were conducted because tobacco smoke is considered an important source of PM2.5, and such exposures may have interfered with predictions of pollutant exposure if they were not incorporated into the models. The contribution of tobacco smoke to particulate exposure is a potential confounder considered in many 81 personal exposure studies. In the first sensitivity analysis, models were rerun after removal of the two subjects who reported some exposure to tobacco smoke during every sampling session. In the second sensitivity analysis, models were rerun after exclusion of all person days on which any exposure to tobacco smoke was reported. In the third, models were repeated after removing outliers associated with the dependent and independent exposure variable. A summary of the sensitivity analyses conducted and the variables that fell out of each model are presented in Table 3ab. Table 3ab. Variables to fall out of sensitivity analyses P M 2 5 ABSORBANCE S U L F A T E I: Subjects with daily ETS exposure removed TRANSIT TRANSIT -II: Person days with ETS exposure removed - TRANSIT -III: Outliers removed - - OUTDOORS The sensitivity analyses indicated that TRANSIT may be confounded by exposure to tobacco smoke, since this variable fell out of both PM2.5 and absorbance models when the dataset was restricted according to ETS exposure. For the sulfate model, the removal of one extreme personal sulfate value resulted in the OUTDOORS variable falling out of the model, indicating that this variable was not as important as the coefficient in the original model indicated. 8 2 4.0 DISCUSSION Chapter 4 4.1 EPIDEMIOLOGICAL ANALYSIS 4.1.1 Case crossover This study revealed that in the population investigated, ambient air pollutant concentrations had small associations with ICD shocks to treat cardiac arrhythmia. Although in general there were no statistically significant results, consistent trends indicated associations between air pollutants and ICD shocks. Odds ratios (OR) were consistently higher in summer (7 of 9 >1) than in winter (1 of 9 >1), indicating a possible seasonal effect. Further, the highest ORs were consistently observed at a lag of 0 days, implying a short delay between exposure and outcome. For local combustion-source pollutants EC, OC, CO and SO2, ORs were above 1 at all lags (0-3 days) in summer. For summer and winter combined, results failed to indicate consistent associations between air pollution and ICD shock. Exceptions were slight trends for EC and OC, for which ORs were above 1 at all lags, a protective effect for NO2 at a lag of 2 days and high but non-significant odds ratios for O 3 . Taken together, these findings suggest a weak association between summertime combustion-source primary air pollutants and cardiac arrhythmia. Previous studies of humans (Peters et al, 1999; Gold et al, 2000) and animals (Watkinson et al, 1998; Wellenius et al, 2002) have demonstrated changes in heart rhythm in association with exposure to air pollution. The case crossover analysis compared air pollutant concentrations during or preceding ICD shock treatments for cardiac arrhythmia to air pollution at other times to assess the association between air pollution and cardiac rhythm disturbance. The results were not statistically significant with one exception. This significant result for NO2 at a lag of 2 days was in the opposite direction to that hypothesized, and was likely due to chance. Although N 0 2 concentrations in Vancouver are low relative to other urban places, it is unlikely that they are truly protective against ICD shock. High odds ratios were observed for O 3 in the hypothesized direction, with odds ratios of 1.15 and 1.66 for lags of 0 and 1 days, respectively. The observation of high odds ratios for O 3 and low odds ratios for other pollutants reflected the negative correlation observed between O 3 and all other pollutants. Although O 3 and CO were relatively well correlated (r=-0.56), the effect on ICD shock is more likely to have resulted from increased concentrations of the irritant gas O 3 than from decreases in CO concentrations. This finding for O 3 is consistent with a Holter monitoring study reporting changes in heart rate and heart rate variability with increases in O 3 . (Gold et al, 2000) 83 4.1.2 Lag periods Observation of effects for different lag periods revealed an interesting trend for some pollutants. The tendency for odds ratios to decrease as lag increased for EC, OC and PM2.5 may reveal an acute effect of combustion-source pollution on ICD shock. It is not surprising that EC and OC revealed a similar trend since these two pollutants were so highly correlated (r=0.92) during the study period. Their intercorrelation makes it difficult to draw conclusions about which pollutant or combination may be responsible for observed effects. Although nonsignificant odds ratios make the trends difficult to interpret, decreasing odds ratios with lag time may indicate that the induction time between exposure and outcome was within one or two days. A similar trend of odds ratios with lag period was observed in a study of the effect of PM2.5 exposure on onset of myocardial infarction (Peters et al, 2001). In this study, a positive association was observed with elevated concentrations between 24 and 48 hours before the onset of symptoms. Odds ratios increased from a 0 to a 1 day lag, and decreased steadily through a lag of 5 days. Studies of the effect of PM2.5 on heart rate variability in humans have revealed induction periods of hours rather than days (Magari, 2002). The short induction period is also consistent with preliminary findings of another case crossover study of air pollution and ICD shock (Rich DQ et al, 2002). This study revealed cardiac response to acute air pollution episodes in the previous 1-5 hours. A study of the lag structure between particulate air pollution and cardiovascular deaths in U.S. cities concluded that same day exposures were likely the main contributor to deaths due to myocardial infarction (Braga et al, 2001). 4.1.3 Previous studies Besides the Rich DQ et al (2002) study for which preliminary results have been presented, only one other study of the effects of air pollution on ICD shock was located. This study, carried out between January 1995 and December 1997, revealed that N0 2 concentrations 2 days before significantly increased the odds of experiencing an ICD shock (Peters et al, 2000). The air pollutant concentrations observed during this study, which was conducted in South Boston, are presented in Table 4a. Table 4 a . Air pollutants considered in Boston and Vancouver studies of ICD patients Boston study Vancouver study A M IQR A M IQR P M 1 0 (ugm3) 19.3 11.6 13.3 7.4 P M 2 . 5 (ugm3) 12.7 8.5 8.2 5.2 C O (ppm) 580 230 553.8 162.7 0 3 (ppm) 18.6 14.0 27.5 13.4 S 0 2 (ppm) 7.0 6.0 2.6 1.6 N 0 2 (ppm) 23.0 11.0 16.5 5.4 AM=arithmetic mean, IQR=interquartile range. Note that 0 3 values are calculated from daily maxima in the Vancouver study and daily averages in the Boston study. In general, it is likely that the low air pollutant concentrations in Vancouver as compared to Boston and other urban areas contributed to the absence of strong findings in this study. 84 4.1.4 Season The effect of season on odds ratios produced striking results. While seasonal differences did not affect odds ratios systematically for particulate matter or some of the gaseous pollutants considered, odds ratios increased considerably in summer relative to winter for pollutants including EC, OC, CO and SO2. All were nearly significant in the hypothesized direction in summer. Because EC and OC as well as CO and SO2 were highly correlated, it is not surprising that these pollutants displayed similar results. It is unclear for each pair of pollutants whether either or both were responsible for effects seen. Regardless of which pollutants were responsible, if concentrations were higher for cases in summer, the seasonal differences observed might point to the existence of a pollutant concentration threshold above which exposures led to arrhythmia and below which they had no effect. Comparison of the 75th percentile of these pollutant concentrations for cases in summer and in winter, however, did not reveal concentrations of pollutants to be higher for cases in summer. In fact, case concentrations were similar in summer and winter for EC, OC and SO2, and case concentrations were actually lower for cases in summer than in winter for CO. These observations indicate that differences lay in the control periods. Examination of control concentrations reveals that summer controls indeed experienced lower pollutant concentrations than did winter controls. Further, despite the observation of summer odds ratios >1.0 and winter odds ratios <1.0, the 75th percentiles of summer case concentrations were lower than winter control concentrations for these four pollutants. There are many potential explanations for this. The fact that cases coincided with lower pollutant concentrations in summer than in winter may indicate that air pollutants acted in concert with other factors to induce arrhythmia in summer. An alternative explanation is that an omitted protective variable that covaried with pollution acted in winter but not in summer to cancel out the effect of increased pollutant concentrations, explaining the observation of protective effects in winter at higher pollutant concentrations. Either way, it is possible that trends seen in odds ratios for these air pollutants reflect the action of factors that covaried with the pollutants but influenced heart health independently. In summer, odds ratios were >1.0 at a 0 day lag for 7 of 9 pollutants. This may reflect a short delay in the effect of air pollution on arrhythmia. The multicollinearity of pollutants makes it difficult to determine which pollutant or pollutants might have been responsible for this effect. However, the consistently positive association at a 0 day lag in summer but not winter indicates that the effect is sensitive to season. In addition, since high odds ratios observed in winter in the hypothesized direction for O 3 were consistent with those seen in the initial analysis, the all-year effect could have been driven by a winter association. Seasonal differences may also be influenced by subject behaviour. Although subjects are thought to spend the majority of their time indoors, relative amounts of time spent indoors and outdoors are likely to vary with seasonal patterns from summer to winter. 85 This could systematically alter exposure to specific air pollution components, with a decreased relative exposure to ambient source pollutants in winter, when more time is spent indoors. It is possible that the strong associations between O 3 and ICD shock observed in summer relative to winter were confounded by temperature. Although models were adjusted for ambient temperature, it is probable that subjects were exposed primarily to indoor temperatures in winter. Although indoor temperatures were not documented, it is likely that in winter, indoor temperatures were consistently higher than ambient temperatures. It is thus likely that adjustment for ambient temperatures resulted in a more accurate model of actual exposure in summer than in winter. This differential adjustment could have confounded seasonal associations between O 3 and ICD shock, particularly in view of the positive association between ambient temperature and O 3 . 4.1.5 Sample size The patient population considered in the case crossover analysis was smaller than was originally expected. The smaller sample size as well as the shorter duration of this study may account in part for the non-significant results. This decreased the power of the study and widened confidence intervals associated with estimated odds ratios. The Peters et al (2000) study followed 100 ICD patients between 1995 and 1997, although only 33 patients experienced ICD shock. The Rich DQ et al (2002) study of ICD shocks and air pollution, which reported significant results for PM2.5, SO2, and O 3 , followed 320 ICD patients between 1995 and 2000, of which 121 experienced ICD shock. The above-mentioned study which revealed a significant association between PM2.5 air pollution concentrations and myocardial infarction involved 772 patients (Peters et al, 2001). 4.1.6 Strengths The case crossover design had several important strengths for this analysis. It had continuous, individual, detailed outcome information for a relatively long period. Further, because each subject acted as his own control, separate controls were unnecessary. A side benefit was that control selection bias was eliminated, since cases acted as their own controls (Maclure, 1991). Self matching of cases provided ideal controls, since fixed individual characteristics did not need to be considered. The case crossover analysis is also touted for its ability to control for time trends and seasonal patterns by design (Sunyer et al, 2001). The fact that the inclusion of meteorological variables had only small effects on pollutant coefficients indicates that the design of this analysis controlled for seasonal patterns effectively. Cases and controls were located close in time to one another, so results were unlikely to be affected by unmeasured time trends in omitted variables. The use of ambidirectional control periods similarly reduced the chance that time trends in the exposure variable would bias results. Another strength of the case crossover approach was that it increased power by focusing on cases and excluding noise. The use of multiple controls per case also increased 86 power, allowing more stable results to be obtained than would otherwise have been possible with such a small sample size. This particular study had a number of other unique strengths that set it apart from previous work in the field. The use of ICDs as an objective reporter of ICD shock minimized the information bias that occurs with self-reporting in other studies, for example (Schlesselman, 1981). Further, ICDs are able to record all events occurring over extended time periods, providing a large dataset relative to more labour intensive studies that rely on Holter monitoring for health outcome data. Abstraction of ICD shock data from the only two ICD referral clinics in BC provided as large and representative study population as was possible in this region. The study population was restricted to ICD patients living in the Lower Mainland to improve the relevance of measurements taken at ambient monitors located in and around Vancouver. The identification of the study population from two clinics serving the entire ICD patient population of the province of British Columbia improved the generalizability of results. Further, the population studied which included females and males ranging in age from 15 to 85, incorporated a broad variety of individual characteristics. 4.1.7 Limitations Despite the important strengths of the case crossover analysis in this context, it also suffers from several limitations. The case crossover analysis was designed to compare exposures during rare events to those during control periods. Rarity was necessary for two reasons. Events had to be rare enough that arrhythmias could be considered unrelated to one another. If they were not independent, it would be inappropriate to analyze them as separate events. While grouping of events within 72 hours diminished this problem, events were still relatively clustered for some subjects. Rarity was also necessary to minimize the overlap of each case's 14 day control-control period with the 14 day control-control periods associated with other events. Overlap between these periods increased the potential for autocorrelated pollutant concentrations to be inappropriately treated as independent. Because the temporal and mechanistic association between air pollution and cardiac arrhythmia is poorly understood, the number of days that must separate periods from one another in order to treat them as independent is unclear. If, for example, an event is influenced by an air pollution increase that is autocorrelated with air pollutant concentrations over several days on either side of the event, controls should be placed outside of the window of autocorrelation. If another case exists within that window for the same subject, cases may not be sufficiently rare to run a case crossover analysis. Although ICD shocks were separated by relatively long time periods for some patients in the study population, shocks were clustered for others. The clustering of these shock events prevented control periods from being applied to each case in exactly the same manner, and increased the potential for autocorrelation of events with other events as well as with controls. This makes the results more difficult to interpret. 87 Another weakness of the case crossover design in the context of this study was that it required the event studied to be completely transient, so that periods of increased risk could be compared to control periods of no assumed risk both before and after the effect. Because air pollution exposure has been shown to result in both acute (Pope et al, 1999; Gold et al, 2000) and long-term (Morgan et al, 1998; Schwartz, 1997; Wong et al, 1999) cardiac effects, it is unclear whether air pollution mediated cardiac arrhythmia may be treated as a completely transient effect. If, for example, a pollution-mediated ICD shock made a patient more vulnerable to cardiac effects of air pollution, the susceptibility of an individual could increase at the same time as air pollutant concentrations returned to normal. The next event would then coincide with lower air pollutant concentrations, diluting any effect. A related obstacle is that it is difficult to distinguish whether air pollution is solely responsible for the ICD shocks it may trigger, or if these events were inevitable anyway, and the air pollution just acted to elicit the events. Further, due to the multicollinearity of pollutants, it cannot be determined with certainty whether a pollutant is responsible for an observed effect, or whether it is acting as a surrogate for another pollutant or omitted correlated variable. Events that were caused by sinus rhythms or were known to be triggered by events unrelated to air pollution (eg. Patient being mugged) were considered 'inappropriate' in this study. There was no systematic method of identification of inappropriate events in patient charts. Therefore, although sensitivity analyses revealed that inclusion of inappropriate events had small effects on odds ratios, the value of the results of this sensitivity analysis are limited by the fact that inappropriate shocks identified may not represent the true magnitude of inappropriate events. This increased the potential for misclassification, which would have diluted any true effect. Uncertainty about the length of the so-called effect period, or the period of increased risk of disease onset following exposure (Maclure, 1991) also makes the results difficult to interpret. If an effect exists, and the effect period is truly within 3 days, one of the lag periods applied is likely to pick up the effect, but since the length of the effect period is uncertain, it is unclear which lag is most relevant. Similarly, if, as studies of cardiac arrhythmia and myocardial infarction with air pollution have indicated, the lag structure separating exposure from outcome is within 1 day, the use of 24-hour averaged pollutant concentrations may be inappropriate. The use of 24-hour averages has the potential to mask shorter-term peaks that could be responsible for triggering arrhythmias. This potential for misclassification of exposure would, again, attenuate any true effect. Another design-specific limitation was that control days were censored for ICD events occurring near both ends of the sampling period due to missing data. In these situations, ambidirectional sampling, the purpose of which was to avoid bias associated with time trends persisting from pre- to post-event periods, was not possible. Since both the beginning and the end of the study period occurred during the winter period, unidirectional sampling due to censored controls occurred only in winter. This resulted in a lower control:case ratio in winter, and created the potential for bias which could affect results of all-year as well as seasonally stratified analyses. If, for example, air pollutant concentrations were steadily increasing in the weeks surrounding the beginning 88 of the sampling period, censored controls would occur on days with systematically lower pollutant concentrations than controls included in analyses. This would result in artificially deflated odds ratios. A similar effect would occur if pollutant concentrations were steadily declining in the weeks surrounding the end of the sampling period. Unfortunately the true pollutant distribution could not be determined without data from outside of the sampling period. Patients with pre-existing heart conditions which increase their likelihood of experiencing arrhythmia are studied because their susceptibility increases the chance of a true effect being identified. The use of ICD shock as an indication of health outcome, however, had the limitation that it only treated events that were deemed life-threatening. Because of this, minor arrhythmias were not included in analyses, making any effect more difficult to identify. By using ICD shock as an indicator of cardiac arrhythmia, severe events were selected for. Among patients with ICDs, those actually experiencing shocks suffer the most severe arrhythmia, and may therefore be the most susceptible to an effect of air pollution. They provide a good population of study for identification of a rare effect. On the other hand, effects observed in individuals with severe arrhythmia may not be generalizable to all individuals experiencing cardiac arrhythmia. Holter monitoring studies have detected less severe cardiac effects in association with air pollutant concentrations including small changes in heart rate variability (Pope et al, 1999a) and supraventricular ectopic heartbeats (Brauer et al, 2001). Because continuous monitoring is labour-intensive and requires the participation of willing subjects, however, these studies have involved relatively few participants. Further, Holter monitoring is usually conducted over short time intervals on the order of hours or days, and results may not be generalizeable to longer periods. Subject behaviour may change in ways that affect personal exposures to air pollution in the week following an ICD shock, creating asymmetry in the bi-directional control strategy. Also, any unreported individual change in risk that does not covary with air pollution could dilute the observed effect. Conversely, any within individual change that does covary with air pollution could cause confounding. Another limitation of the case crossover design is that the adaptation used to maximize the number of controls could introduce a systematic bias. Proposed control days for which air pollution data were unavailable were replaced by days with air pollution data. If missing data was associated with air pollution levels, this could bias the results. Another potential limitation of this study is that conditional logistic regression analysis may not be valid for some case crossover designs. There is a potential for bias when the timing of each control period is dependent on the response, or the corresponding case period (Levy et al, 2001a; Lumley and Levy, 2000). The validity of analyzing these data by conditional logistic regression relies on an assumption that the exposure distribution is stable over time. Ideally, controls should be chosen independently and with equal probability from the entire study period. When controls are not chosen at random, the conditional likelihood, or the probability of the observed data configuration relative to all 89 other possible configurations, may be invalid. However, simulations conducted by Levy et al (2001a) to explore the impact of different referent selection strategies indicated that the approach employed in this study, using ambidirectional sampling 7 days from each case, would result in only a small bias. To completely remove this bias, a time-stratified case crossover design was suggested. In this approach, the time period is divided a priori into fixed strata and the remaining days in a stratum are used as controls for a case in that stratum. This design avoids seasonal confounding and provides the correct conditional likelihood. 4.1.8 Suggestions for further research As discussed above, this type of analysis would benefit from the increased power provided by a larger sample size. This might result in more conclusive findings. A longer study duration as well as a more complete set of air pollutant data would also improve this type of study. As was done in the Rich DQ et al (2002) study, it would be useful to have a cardiac electrophysiologist analyze electrograms to characterize events in order to minimize the misclassification error associated with inclusion of inappropriate events. This is one of the first studies of its kind, however, and although confidence intervals were wide as a result of the relatively small sample size, future research may build on the preliminary findings reported here. 90 4.2 EXPOSURE ASSESSMENT 4.2.1 Personal monitoring This analysis contributes data from a unique, susceptible population to the collection of studies of personal exposure to PM2.5. It both confirms previous reports and provides new information. The characteristics of the population studied provide an interesting context for interpretation of exposure analyses. Among subjects monitored for personal exposure, 37% worked or attended university. This range of characteristics and activities sets this study apart from previous personal exposure studies that have focused on susceptible populations with limited mobility (Williams et al, 2000; Rodes et al, 2001). Although the 39% response rate for the personal monitoring portion of this study was relatively low compared to some other kinds of studies, it was comparable to that for other personal monitoring studies (Stieb et al, 1998; Ebelt et al, 2000). The number of study subjects involved in the personal monitoring was also similar to that in other studies of this type (Adams et al, 2001; Janssen et al, 2000). Compliance was excellent, at 100%, and the only loss of samples was due to equipment failure and laboratory error. The number of samples lost to equipment failure was <10%, consistent with other studies of this kind (Janssen et al, 1998). Temporal overlap of personal measurements with ambient levels was high, at 94%, suggesting that comparisons of personal and ambient concentrations was appropriate. 4.2.2 Personal and ambient exposures Personal measurements were comparable to those seen in other studies. Overall, ambient PM2.5, absorbance and sulfate concentrations were low compared to those of other urban areas (Brook et al, 1997; Brauer et al, 2000). The extreme values observed in personal PM2.5 samples were not seen in ambient measurements. These findings suggest that a substantial amount of personal exposure to PM2.5 originates from local sources. Conversely, personal and ambient sulfate concentrations were very consistent. Personal absorbance values were relatively high compared to ambient absorbance. The fact that personal and ambient absorbance values were more similar than personal and ambient PM2.5 concentrations, but less similar than personal and ambient sulfate concentrations, indicates that absorbance may have local or indoor as well as ambient sources. 4.2.3 Absorbance as an exposure metric The relatively high R 2 values (0.77 and 0.71) for absorbance and EC at both monitoring sites were comparable to the R 2 values of approximately 0.80 reported in previous studies using absorbance as a surrogate for EC (Gotschi et al, 2002; Janssen et al, 2000). The R 2 values were lower, however, than the R 2 values of 0.95 and 0.92 reported by Kinney et al (2000) and Janssen et al (2001). However, these studies were conducted on sidewalks 91 and near motorways, respectively, where diesel emissions are likely to be higher. These trends reinforce the observation that absorbance "is a surrogate for EC emissions, which result primarily from motor vehicle traffic (Hies et al, 2000). In this study, absorbance was not well correlated with other carbon fractions, but was well correlated with measurements of black carbon (BC) as measured via the laser integrating plate method. This is consistent with the high absorbance:EC correlation, since BC is considered interchangeable with EC. 4.2.4 Relationships between personal and ambient concentration Subject-specific personal:ambient PM2.5 relationships displayed a large range of Pearson correlations. Individual r values ranged from -0.93 to 0.75, with a median of-0.20. By itself, this indicates that indoor and or local outdoor sources played an important role in personal exposures. Although other studies have conceded that ambient concentrations provide an inadequate metric for personal exposure to PM2.5, previously reported personal:ambient PM2.5 correlations associated with 24-hour personal exposure measurements have generally been better than those found in this study. Ebelt et al (2000), for example, reported a median individual personafambient PM2.5 r value of 0.48 in a study of 54 to 86 year old patients with chronic obstructive pulmonary disease in Vancouver, while a study of two panels of 50-84 year old subjects with cardiovascular disease reported medians of 0.76 and 0.79 (Janssen et al, 2000). A possible explanation for the particularly poor individual personahambient PM2.5 correlations revealed in this study is that this population was younger (age range 32-74) and tended to be more mobile than other populations with heart disorders. It has been hypothesized that healthier individuals experience more personal PM2.5 exposure through generation of a 'personal cloud' during indoor activities (Wallace, 2000; Monn, 1997). Older, sicker populations may have experienced personal exposures more consistent with ambient levels because they were less mobile. At least one study of children has also reported high personal to ambient PM2.5 correlations (Janssen et al, 1999). Although children would be expected to generate a larger personal cloud, they may have spent more time outdoors, decreasing exposure to indoor source particles. Unfortunately, proportions of time spent indoors and outdoors were not reported in the Janssen et al study. Personal to ambient absorbance relationships were somewhat stronger than those for PM2.5. A range of individual correlations were observed, however, consistent with the hypothesis that absorbance is an indicator of regionally variable traffic-source emissions. Personal to ambient sulfate relationships produced higher r values than either PM2.5 or absorbance. This finding is consistent with the observation in previous studies that sulfate is a regional source pollutant with virtually no residential sources (Wilson and Suh, 1997; Mage et al, 1999; Landis et al, 2001). The fact that personahambient relationships were weaker for absorbance than for sulfate indicates that indoor or local outdoor sources made a greater contribution to personal absorbance than to personal sulfate. The implication of these findings is that whereas ambient absorbance appeared to 92 be a very poor measure of personal exposure tx> PM2.5, which is dominated by indoor sources, ambient sulfate seemed to provide a good representation of personal exposure, since indoor sources were unimportant. 4.2.5 Personal to ambient concentration ratios Average personal to ambient ratios of each measured component also produced interesting results. The mean personal:ambient PM2.5 ratio was 3.2, as compared to 1.4 for absorbance and 0.8 for sulfate. Because this population spent a large proportion of their time indoors (mean=0.84), it is likely that indoor rather than local outdoor sources accounted for the higher PM2.5 ratio. The comparatively smaller ratio for absorbance than PM2.5 was consistent with ratios reported in a previous study (Gotschi, 2002). Relatively small absorbance and sulfate ratios were consistent with these components having fewer local or indoor sources. The low sulfate ratio again confirms that little or no sulfate originates indoors. Since coal combustion has decreased over time as a source of residential heat, ratios of less than 1.0 for sulfate have been consistently reported (Spengler et al, 1981). When compared to that for sulfate, the slightly higher absorbance ratio indicates that absorbance may have regional as well as indoor and localized outdoor sources. 4.2.6 Interrelationships between exposure variables Relationships between ambient PM2.5 and all personal exposure parameters were investigated. Observed relationships between personal and ambient exposure variables are presented schematically in Figure 4.1 below. Figure 4.1. Observed interrelationships between personal and ambient exposure variables A PM2.5 = ambient P M 2 5 P PM2.5 = personal P M 2 5 A Abs = ambient absorbance P Abs = personal absorbance A S0 4 = ambient sulfate P S0 4 = personal sulfate H E A L T H O U T C O M E S Strong relationship ^ No relationship ^ Some relationship 93 While correlations between ambient PM2.5 and health effects have been revealed, it is unclear whether these relationships are mediated by personal PM2.5 exposure. The fundamental problem is that simple measures of personal PM2.5 include PM2.5 of ambient and non-ambient origin. Epidemiological studies exist only for ambient PM2.5 and health, so the importance of non-ambient source PM2.5 is not known. This is a question that has been addressed by many studies over the past decade, with variable findings (Wilson et al, 2000; Samat et al, 2000). The results of this study suggest that personal PM2.5 exposure is not well correlated with ambient PM2.5, indicating that the use of ambient PM2.5 to predict personal PM2.5 exposure could lead to misclassification of exposure. However, misclassification of exposure is relevant to conclusions about health effects only if non-ambient PM2.5 has properties similar to those of ambient PM2.5. While it is clear from the results of this study that some properties of ambient PM2.5 distinguish it from indoor source PM2.5, it is unclear whether those properties are significant to health. As discussed above, since health effects have been associated with ambient PM2.5, it is useful from a regulatory perspective to identify personal exposures that are correlated with ambient particle concentrations. While indoor sulfate is considered a marker of regional combustion source pollution, absorbance, as described above, appears to be a surrogate for EC. Although the median individual correlation between ambient PM2.5 and personal absorbance was not extremely high (r=0.63), it was almost as high as that between ambient PM2.5 and personal sulfate (r=0.67), and it was higher than that between ambient and personal absorbance (r=0.39). Most interestingly, the median correlation between ambient PM2.5 and personal absorbance was much higher than that between ambient and personal PM2.5 (r=-0.20). This suggests that some personal exposure to EC results from ambient sources. These findings indicate that both personal sulfate and personal absorbance exposures may be driven by ambient PM2.5 concentrations. 4.2.7 PM2.5 composition The estimation of the proportion of personal PM2.5 contributed by ambient sources was based on the relationship between personal and ambient sulfate concentrations. This was a crude estimate since it assumed both that PM2.5 penetrated indoors as effectively as sulfate, and that no ambient source PM2.5 settled and was re-entrained indoors. According to this prediction, 38% of personal PM2.5 was from ambient sources. This is slightly lower than what has been reported in other studies, which have reported ambient contributions to PM2.5 near 50% in elderly populations (Meng et al, 2002; Sheldon et al, 2002). The lower proportion found in this study is consistent with the particularly important indoor 'personal cloud' hypothesized for this population. Calculated mean percent mass sulfate was twice as high for ambient as for personal PM2.5 (20% vs. 10%). Since sulfate is predicted to have no indoor sources, this finding is consistent with the large indoor contribution to personal PM2.5. The estimated mean percent mass EC, on the other hand, was much more similar between personal and ambient PM2.5 (6.4% vs. 5.4%). Since personal PM2.5 concentrations were larger than ambient PM2.5 levels, the personal percent mass corresponds to a larger amount of EC in 94 personal PM2.5 samples. Although the contribution of both indoor and outdoor exposures to personal PM2.5 make it difficult to determine sources of its components, it is possible that some indoor-source EC contributed to personal exposures. Since absorbance is an indicator of combustion-source exposures, activities such as cooking and candle burning as well as exposure to vehicle emissions may have contributed to personal exposures. 4.2.8 Distance to ambient monitoring station Regression of individual personal to ambient PM2.5 correlation r value against distance to the ambient monitoring station revealed no relationship. This was consistent with the domination of personal PM2.5 exposure by indoor sources. PM2.5 is spatially uniform and PM2.5 exposure is driven by indoor sources. The small negative relationship observed between individual personahambient absorbance Pearson r values and distance from the monitoring station suggests that personal absorbance levels are affected by spatial variability in ambient concentrations. 4.2.9 Multiple Regression Because many time activity variables were qualitative and precision was low, descriptive multiple regression models were run with the objective of identifying variables likely to play a role in the determination of personal exposures. Results were generally consistent with hypotheses about sources. Estimated R 2 values indicated that the PM2.5 model explained the least variability in personal exposure. This indicated that variability in PM2.5 exposure was well characterized neither by ambient source P M 2 5 nor by available time activity and dwelling information. This suggests that omitted variables may have been responsible for the majority of exposure to PM2.5. Ambient PM2.5 did remain in the model, however, indicating that personal exposures may result in part from ambient concentrations. VOLUME was the only other variable that remained in all sensitivity analyses for PM2.5. It is logical that this variable was a predictor of personal exposure to PM2.5 since home volume could influence both indoor and outdoor source PM2.5. Specifically, a larger volume allows for greater dispersion of particles generated indoors as well as particles entering from outdoors, resulting in lower personal exposures. Absorbance was predicted by variables representing vehicle emissions, consistent with the hypothesis that absorbance (EC) is an indicator of particles associated with local traffic sources. Results of multiple regression analyses for personal absorbance were compared to models of indoor absorbance presented by Gotschi et al (2002). In the Gotschi et al study, ambient absorbance explained most of the variability in indoor absorbance. While time activity and location variables explained most of the variability in personal absorbance in this study, however, ambient absorbance did not remain in the model as a significant (p<0.05) predictor. This again implies that ambient absorbance was an inadequate metric for personal exposure to absorbance, likely due to regional variation in vehicle source emissions. 95 In the sulfate multiple regression, ambient sulfate was the most predictive independent variable, as has been found in other studies (Ebelt et al, 2000). This was expected, and confirms the implications of findings previously discussed. The fact that the WINDOWS and VOLUME variables remained in sulfate models was also consistent with sulfate being an ambient source pollutant. Increased time with windows open could increase the penetration of ambient-source particles, while increased volume could allow for greater dispersion of ambient-source particles, decreasing personal exposure to sulfate. Sensitivity analyses revealed that some variables were confounded by tobacco smoke exposure and outlying data points. Specifically, the fact that the TRANSIT variable fell out of models for both PM2.5 and absorbance when subjects experiencing exposure to tobacco smoke every day were excluded indicates that increased personal exposures may have been the result of exposure to tobacco smoke or time in transit or both. Similarly, the fact that the OUTDOORS variable fell out of the sulfate model when one outlying data point was removed indicates that time spent outdoors was not as important a predictor of personal sulfate exposure as was indicated by the initial model. 4.2.10 Strengths and limitations Among personal exposure studies, this investigation has a number of strengths. It involved 24-hour personal samples, ideal for comparison with ambient concentrations that are often averaged over the same period. This was also the first personal exposure monitoring study to involve patients with ICDs. Assessment of this unique, susceptible population contributes useful information to the body of evidence suggesting effects of air pollution on heart health. Another important strength is the use of integrated sampling techniques, allowing some investigation of the composition of PM2.5. On the other hand, although it would preclude further analysis of samples, continuous monitoring has been suggested as a means of gaining other important information (Long, 2001). For example, activities have been shown to be very important to peaks in personal exposure (Quintana, 2001). Since this study used 24-hour averaged PM2.5 concentrations rather than continuous monitoring, time activity data could not be used to determine time-resolved relationships between activity and exposure. Comparison of time activity data on days of personal sampling with time activity on other weekdays in a study of personal exposure to particles in groups of 50-70 year old adults revealed that subjects change their behaviour due to the wearing of personal sampling equipment. Specifically, it was found that adults spent significantly less time outdoors and more time at home on days of personal sampling compared to other weekdays (Janssen et al, 1998). Although subjects were encouraged to maintain normal activities on sampling days, altered behaviour may have increased the contribution of indoor sources, decreasing the true correlation between personal and ambient PM2.5. 96 4.2.11 Conclusions As discussed above, there were a number of limitations associated with epidemiological research of particulate matter and health effects. In this study, the intercorrelation of risk factors provided a major obstacle to the identification of causal pathways linking exposure to health outcomes. Further, the potential for misclassification of exposure limited the interpretation of results. A susceptible population was investigated from both an epidemiological and a human exposure perspective. The exposure portion of this study attempted to address some of the many limitations faced in the case crossover analysis. With respect to personakambient sulfate correlations, the results of this study confirm the findings of previous studies. Ambient sulfate has been well-established as a good indicator of exposure, but the applicability of this exposure metric in epidemiological analyses is limited because indoor or local source PM may also contribute to health outcomes. Results for PM2.5 and absorbance were less consistent with those revealed by other studies, perhaps because this population was younger and more mobile than those studied elsewhere. Mixed effects models were conducted in an effort to understand what drove personal exposures, with some interesting results. Considering the results of all analyses, personal absorbance appears to be a good indicator of local source PM2.5. Ambient absorbance, on the other hand, may be a poor indicator of exposure due to the potential for misclassification. Ambient PM2.5 may remain the best exposure metric, since this is correlated with both personal absorbance and personal sulfate. The discrepancy between ambient and personal concentrations of particles underscores the need for an understanding of the factors that mediate the relationship between ambient PM2.5 and health. Even when relationships between ambient concentrations and personal exposures are predictable, questions remain regarding which aspects of particles are most important to health. These broad questions may be answered only through an integrated approach incorporating toxicological, human exposure and epidemiological studies. 97 REFERENCES Adams HS, Nieuwenjuijsen MJ, Colvile RN, McMullen MAS, Khandelwal P. (2001) Fine particle (PM2.5) personal exposure levels in transport microenvironments, London, UK. The Science of the Total Environment, 279, 29-44. Birch ME, Cary RA. (1996) Elemental carbon-based method for monitoring personal exposures to particulate diesel exhaust. Aerosol Science and Technology, 25, 221-41. Braga ALF, Zanobetti A, Schwartz J. (2001) The lag structure between particulate air pollution and respiratory and cardiovascular deaths in 10 US cities. Journal of Occupational and Environmental Medicine, 43, 927-33. Brauer M, Brumm J, Ebelt ST. (2000) Evaluation of ambient air pollution in the lower mainland of British Columbia: Public health impacts, spatial variability and temporal patterns. Final Report to the Administrative Council of Lower Mainland Medical Health Officers, July 17, 2000. Retrieved August 01, 2002 from: (http://www.southfraserhealth.com/ImagesAVhats Happening/PDFs/Air%20Quality%20 Report%20-%20Final.pdf) Brauer M, Ebelt ST, Fisher TV, Brumm J, Petkau J, Vedal S. (2001) Exposure of chronic obstructive pulmonary disease patients to particles: respiratory and cardiovascular health effects. Journal of Exposure Analysis and Environmental Epidemiology, 11(6), 1-10. Brook JR, Dann TF, Burnett RT. (1997) The relationship among TSP, PM, 0 , PM 2 . 5 , and inorganic constituents of atmospheric particulate matter at multiple Canadian locations. Journal of the Air & Waste Management Association, 47 (1), 2-19. Brook RD, Brook JR, Urch B, Vincent R, Rajagopalan S, Silverman F. (2002) Inhalation of fine particulate air pollution and ozone causes acute arterial vasoconstriction in healthy adults. Circulation, 105(13), 1534-6. Burton RM, Suh HH, Koutrakis P. (1996) Spatial variation in particulate concentrations within metropolitan Philadelphia. Environmental Science and Technology, 30(2), 400-7. Canadian Council of Ministers of the Environment (CCME) (2002) Retrieved July 28, 2002 from: http://www.rsc.ca/ozone_&_pm/Ozreport.pdf Churg A, Brauer M. (1997) Human lung parenchyma retains PM2.5. American Journal of Respiratory and Critical Care Medicine, 155, 2109-11. Cohen AJ, Nikula K. (1999) The Health Effects of Diesel Exhaust. In Air Pollution and Health. Eds. Holgate ST, Samet JM, Koren HS, Maynard RL. San Diego: Academic Press. 98 Dockery DW, Pope CA III, Xu X, Spengler JD, Ware JH, Fay ME et al. (1993) An association between air pollution and mortality in six U.S. cities. The New England Journal of Medicine, 329(4), 1753-9. Dockery DW. (2001) Epidemiological evidence of cardiovascular effects of air pollution. Environmental Health Perspectives, 109 (S4), 483-6. Ebelt ST, Petkau JA, Vedal S, Fisher TV, Brauer M. (2000) Exposure of chronic obstructive pulmonary disease patients to particulate matter: Relationships between personal and ambient air concentrations. Journal of the Air & Waste Management Association, 50, 1081-94. Glikson M, Friedman PA. (2001) The implantable cardioverter defibrillator. Lancet, 357, 1107-17. Gold DR, Litonjua A, Schwartz J, Lovett E, Larson A, Nearing B, Allen G, Verrier M, Cherry R, Verrier R. (2000) Ambient pollution and heart rate variability. Circulation, 101(11), 1267-73. Gotschi T, Oglesby L, Mathys P, Monn C, Manalis N, Koistinen K, Jantunen M, Hanninen O, Polanska L, Kunzli N. (2002) Comparison of black smoke and PM2.5 levels in indoor and outdoor environments of four European cities. Environmental Science and Technology, 36, 1191-7. Health Canada, 2002. Retrieved July 28, 2002 from: (http://www.hc-sc.gc.ca/hecs-sesc/air quality/cws.htm) Herbert R, Schechter C, Smith DA, Phillips R, Diamond J, Carroll S, Weiner J, Dahms TE, Landrigan PJ. (2000) Occupational coronary heart disease among bridge and tunnel officers. Archives of Environmental Health, 55(3), 152-63. Hies T, Treffeisen R, Sebald L, Reimer E. (2000) Spectral analysis of air pollutants. Part 1: Elemental carbon time series. Atmospheric Environment, 34, 3495-502. Hinds WH. (1982) Aerosol Technology: Properties, Behaviour and Measurement of Airborne Particles. New York: J. Riley. Hoek G, Brunekreef B, Fischer P, van Wijnen, J. (2001) The association between air pollution and heart failure, arrhythmia, embolism, thrombosis, and other cardiovascular causes of death in a time series study. Epidemiology, 12(3), 355-7. Hoek G, Meliefste K, Cyrys J, Lewne M, Bellander T, Brauer M, Fischer P, Gehring U, Heinrich J, van Vliet P, Brunekreef B. (2002) Spatial variability of fine particle concentrations in three European areas. Atmospheric Environment (in press). 99 Ibald-Mulli A, Stieber J, Wichmann H-E, Koenig W, Peters A. (2001) Effects of air pollution on blood pressure: A population based approach. American Journal of Public Health, 91(4), 571-7. Janssen NAH, Hoek G, Harssema H, Brunekreef B. (1998) Personal sampling of airborne particles: method performance and data quality. Journal of Exposure Analysis and Environmental Epidemiology, 8(1), 37-49. Janssen NAH, Hoek G, Harssema H, Brunekreef B. (1999) Personal exposure to fine particles in children correlates closely with ambient fine particles. Archives of Environmental Health, 54(2), 95-101. Janssen NAH, de Hartog JJ, Hoek G, Brunekreef B. (2000) Personal exposure to fine particulate matter in elderly subjects: Relation between personal, indoor, and outdoor concentrations. Journal of the Air & Waste Management Association, 50, 1133-43. Janssen NAH, van Vliet PHN, Aarts F, Harssema H, Brunekreef B. (2001) Assessment of exposure to traffic related air pollution of children attending schools near motorways. Atmospheric Environment, 35, 3875-84. Kinney PL, Aggarwal M, Northridge ME, Janssen NAH, Shepard P. (2000) Airborne concentrations of PM2.5 and diesel exhaust particles on Harlem sidewalks: A community-based pilot study. Environmental Health Perspectives, 108, 213-8. Koistinen KJ, Hanninen O, Rotko T, Edwards RD, Moschandreas D, Jantunen MJ. (2001) Behavioral and environmental determinants of personal exposures to PM2.5 in EXPOLIS - Helsinki, Finland. Atmospheric Environment, 35, 2473-81. Kwon H-J, Cho S-H, Nyberg F, Pershagen G. (2001) Effects of ambient air pollution on daily mortality in a cohort of patients with congestive heart failure. Epidemiology, 12, 413-9. Landis MS, Norris GA, Williams RW, Weinstein JP. (2001) Personal exposures to PM2.5 mass and trace elements in Baltimore, MD, U.S.A. Atmospheric Environment, 35, 6511-24. Lee J-T and Schwartz J. (1999) Reanalysis of the effects of air pollution on daily mortality in Seoul, Korea: A case-crossover design. Environmental Health Perspectives, 107(8), 633-6. Levelton Engineering Ltd., Alchemy Consulting Inc., Bates DV. (2000) Diesel particulate matter and associated environmental concerns, health risks and tradeoffs. Vancouver, BC. Levy D, Sheppard L, Checkoway H, Kaufman J, Lumley T, Koenig J, Siscovick D. (2001) A case-crossover analysis of particulate matter air pollution and out-of-hospital primary cardiac arrest. Epidemiology, 12(2), 193-9. 100 Levy D, Lumley T, Sheppard L, Kaufman J, Checkoway H. (2001a) Referent selection in case-crossover analyses of acute health effects of air pollution. Epidemiology, 12(2), 186-92. Liao D, Creason J, Shy C, Williams R, Watts R, Zweidinger R. (1999) Daily variation of particulate air pollution and poor cardiac autonomic control in the elderly. Environmental Health Perspectives, 107, 521-5. Long CM, Suh HH, Catalano PJ, Koutrakis P. (2001) Using time- and size-resolved particulate data to quantify indoor penetration and deposition behaviour. Environmental Science and Technology, 35, 2089-99. Lumley T, Levy D. (2000) Bias in the case-crossover design: implications for studies of air pollution. Environmetrics, 11, 689-704. Maclure M. (1991) The case-crossover design: A method for studying transient effects on the risk of acute events. American Journal of Epidemiology, 133(2), 144-53. Magari SR, Hauser R, Schwartz J, Williams PL, Smith TJ, Christiani DC. (2001) Association of heart rate variability with occupational and environmental exposure to particulate air pollution. Circulation, 104(9), 986-91. Magari SR, Schwartz J, Williams PL, Hauser R, Smith TJ, Christiani DC. (2002) The association between personal measurements of environmental exposure to particulates and heart rate variability. Epidemiology, 13(3), 305-10. Mage D, Wilson W, Hasselblad V, Grant L. (1999) Assessment of human exposure to ambient particulate matter. Journal of the Air & Waste Management Association, 49, 1280-91. Meng QY, Turpin BJ, Korn L, Lee JH, Giovanetti R, Kwon JM, Almokhtari S, Weisel CP, Shendell D, Jones J, Winer A, Colome S, Maberti S, Stock T, Morandi M, Spektor D. (2002) Influence of outdoor sources on indoor and personal fine particle concentrations: Analyses of RIOPA data. Epidemiology, 13(4), S205. Monn C, Fuchs A, Hogger D, Junker M, Kogelschatz D, Roth N, Wanner H-U. (1997) Particulate matter less than 10 um and fine particles less than 2.5 um: Relationships between indoor, outdoor and personal concentrations. The Science of the Total Environment 208, 15-21. Morgan G, Corbett S, Wlodarczyk J. (1998) Air pollution and hospital admissions in Sydney, Australia, 1990 to 1994. American Journal oj~ Public Health. 88, 1761-66. NIOSH method 5040 Retrieved July 05, 2002 from www.cdc.niosh.org. 101 Peters A, Perz S, Doring A, Stieber J, Koenig W, Wichmann HE. (1999) Increased heart rate during an air pollution episode. American Journal of Epidemiology, 150, 1094-8. Peters A, Liu E, Verrier RL, Schwartz J, Gold DR, Mittleman M, Baliff J, Oh JA, Allen G, Monahan K, Dockery DW. (2000) Air pollution and incidence of cardiac arrhythmia. Epidemiology, 11(1),11-17. Peters A, Doring A, Wichmann HE, Koenig W. (1997) Increased plasma viscosity during an air pollution episode: A link to mortality? Lancet, 349(9065), 1582-7. Peters A, Dockery DW, Muller JE, Mittleman MA. (2001) Increased particulate air pollution and the triggering of myocardial infarction. Circulation, 103(23),2810-2815. Peters RW, Gold MR (2001) Implantable cardiac defibrillators. Cardiac Arrhythmias, 85(2), 343-64. Pope CA III, Thun MJ, Namboodiri MM, Dockery DW, Evans JS, Speizer FE, Heath CW. (1995) Particulate air pollution as a predictor of mortality in a prospective study of US adults. American Journal of Respiratory and Critical Care Medicine, 151, 669-74. Pope CA III, Dockery DW, Kanner RE, Villegas GM, Schwartz J. (1999) Oxygen saturation, pulse rate, and particulate air pollution. American Journal of Respiratory and Critical Care Medicine, 159(2), 365-72. Pope CA III, Verrier RL, Lovett EG, Larson AC, Raizenne ME, Kanner RE, Schwartz J, Villegas M, Gold DR, Dockery DW. (1999a) Heart rate variability associated with particulate air pollution. American Heart Journal, 138, 890-9. Pope CA III, Burnett RT, Thun MJ, Calle EE, Krewski D, Ito K, Thurston GD. (2002) Lung cancer, cardiopulmonary mortality, and long-term exposure to fine particulate air pollution. Journal of the American Medical Association, 287(9), 1132-41. Prescott GJ, Cohen GR, Elton RA, Fowkes FGR, Agius RM. (1998) Urban air pollution and cardiopulmonary ill health: a 14.5 year time series study. Occupational and Environmental Medicine, 55, 697-704. Quintana PJ, Valenzia JR, Delfino RJ, Liu LJ. (2001) Monitoring of 1-min personal particulate matter exposures in relation to voice-recorded time-activity data. Environmental Research, 87(3), 199-213. Rich DQ, Schwartz J, Link M, Luttmann-Gibson H, Mittleman MA, Verrier RL, Gold D, Dockery DW. (2002) Case-crossover analysis of the effects of hourly air pollution on ICD-detected cardiac arrhythmias. Epidemiology, 13(4) S168. Rodes CE, Lawless PA, Evans GF, Sheldon LS, Williams RW, Vette AF et al. (2001) The relationships between personal PM exposures for elderly populations and indoor and 102 outdoor concentrations for three retirement center scenarios. Journal of Exposure Analysis and Environmental Epidemiology. 11, 103-15. Roorda-Knape MC, Janssen NAH, de Hartog JJ, van Vliet, PHN, Harssema H, Brunekreef B. (1998) Air pollution from traffic in city districts near major motorways. Atmospheric Environment, 32(11), 1921-30. Sarnat JA, Koutrakis P, Suh HH. (2000) Assessing the relationship between personal particulate and gaseous exposures of senior citizens living in Baltimore, MD. Journal of the Air & Waste Management Association, 50, 1184-98. Schlesselman JJ. (1981). Case-Control Studies. New York: Oxford University Press. Schwartz J. (1994) What are people dying of on high air pollution days? Environmental Research, 64, 26-35. Schwartz J. (1994a) Air pollution and daily mortality: a review and meta-analysis. Environmental Research, 64, 36-52. Schwartz J, Dockery DW, Neas LM. (1996) Is daily mortality associated specifically with fine particles? Journal of the Air & Waste Management Association, 46(10), 927-39. Schwartz J. (1997) Air pollution and hospital admissions for heart disease in eight U.S. counties. Epidemiology, 10, 17-22. Seaton A, MacNee W, Donaldson K, Godden D. (1995) Particulate air pollution and acute health effects. The Lancet, 345, 176-8. Sheldon L, Williams R, Suggs J, Evans G, Rea A, Vette A, Burke J, Croghan C, Leovic K, Creason J, Walsh D, Rodes C, Thomburg J, Lawless P, Ejire A, Sanders W. (2002) Summary findings from the U.S. EPA's particulate matter panel studies. Epidemiology, 13(4) S83. Sheppard L, Levy D, Checkoway H. (2001) Correcting for the effects of location and atmospheric conditions on air pollution exposures in a case-crossover study. Journal of Exposure Analysis and Environmental Epidemiology, 11, 86-96. Spengler JD, Dockery DW, Turner WA, Wolfson JM, Ferris BG. (1981) Long-term measurements of respirable sulfates and particles inside and outside homes. Atmospheric Environment, 15, 23-30. Stieb DM, Brook JR, Broder I, Judek S, Burnett RT, Beveridge RC. (1998) Personal exposure of adults with cardiorespiratory disease to particulate acid and sulfate in Saint John, New Brunswick, Canada. Applied Occupational and Environmental Hygiene, 13, 461-8. 103 Suh HH, Allen GA, Koutrakis P. (1995) Spatial variation in acidic sulfate and ammonia concentrations within metropolitan Philadelphia. Journal of the Air & Waste Management Association, 45(6), 442-52. Sunyer J, Schwartz J, Tobias A, Macfarlane D, Garcia J, Anto JM. (2000) Patients with Chronic Obstructive Pulmonary Disease are at increased risk of death associated with urban particle air pollution: A case-crossover analysis. American Journal of Epidemiology, 151(1), 50-56. Thatcher TL, Layton DW. (1995) Deposition, resuspension, and penetration of particles within a residence. Atmospheric Environment, 29(13), 1487-97. U.S. EPA, 1999 Retrieved July 28, 2002 from: (http://www.epa.gov/oar/oaqps/ozpmbro/current.htm) Vedal S. (1997) Ambient particles and health: Lines that divide. Journal of the Air & Waste Management Association, 47(5), 551-81. Vedal S, Brauer M, Petkau J, Zidek J, Brook J, Kerr C, Yeung J. Air pollution health effects in susceptible populations: refining the assessment of exposure. TSRI Final Report, May 31,2001. Wallace L. (2000) Correlations of personal exposure to particles with outdoor air measurements: A review of recent studies. Aerosol Science and Technology, 32(1), 15-25. Wannamethee G, Shaper AG, Macfarlane PW, Walker M. (1995) Arrhythmias, Pacing, Sudden Death: Risk Factors for Sudden Cardiac Death in Middle-Aged British Men. Ci'rcH/aftow,91(6),1749-1756. Watkinson WP, Campen MJ, Costa DL. (1998) Cardiac arrhythmia induction after exposure to Residual Oil Fly Ash particles in a rodent model of pulmonary hypertension. Toxicological Sciences, 41, 209-16. Wellenius GA, Saldiva PHN, Batalha JRF, Murthy GGK, Coull BA, Verrier RL, Godleski JJ. (2002) Electrocardiographic changes during exposure to residual oil fly ash (ROFA) particles in a rat model of myocardial infarction. Toxicological Sciences, 66(2), 327-335. Wheeler A, Suh H, Koutrakis P, Reid C, Wallace LA, Ryan PB. (2002) Analysis of components of particulate matter (PM2.5) for an exposure assessment study of two sensitive cohorts in Atlanta, GA. Epidemiology, 13(4), S84. Williams R, Suggs J, Zweidinger R, Evans G, Creason J, Kwok R, Rodes C, Lawless P, Sheldon L. (2000) The 1998 Baltimore particulate matter epidemiology-exposure study: Part 1. Comparison of ambient, residential outdoor, indoor and apartment particulate 104 matter monitoring. Journal of Exposure Analysis and Environmental Epidemiology 10, 518-22. Wilson R, Spengler J. (1996) Particles in our Air: Concentrations and Health Effects. Harvard School of Public Health: Harvard University Press. Wilson WE, Suh HH. (1997) Fine particles and coarse particles: Concentrations relationships relevant to epidemiological studies. Journal of the Air & Waste Management Association, 47, 1238-49. Wilson WE, Mage DT, Grant LD. (2000) Estimating separately personal exposure to ambient and nonambient particulate matter for epidemiology and risk assessment: Why and how. Journal of the Air & Waste Management Association, 50, 1167-83. Wong TW, Lau TS, Yu TS, Neller A, Wong SL, Tarn W, Pang SW. (1999) Air pollution and hospital admissions for respiratory and cardiovascular diseases in Hong Kong. Occupational and Environmental Medicine, 56, 679-683 105 Appendix 2.1a. Recruitment letter (Burrard Clinic) April 18, 2001 Dear, In cooperation with Dr. John Yeung's clinic, Dr. Sverre Vedal, Dr. Michael Brauer, Ms. Kira Rich and colleagues at the University of British Columbia are conducting a study to evaluate the exposure of individuals with implantable cardiac defibrillators (ICDs) to air pollutants. Recent studies suggest that patients with cardiac arrhythmia may have increased susceptibility to air pollutants. We hope to identify factors that predict exposure and evaluate the health impact of air pollutants on patients with ICDs. Measurements of air pollutants have traditionally been taken outdoors. There are important questions about the relationship between outdoor air pollution and the exposure of individuals to this pollution. This study is being conducted to examine the relationship of measured outdoor and indoor air pollution and how these contribute to personal exposures. Each participant in the study will be asked to wear a personal pollutant monitor for seven 24 hour periods randomly spaced over the months of May to August, 2001. The monitor weighs approximately 500 grams (1 pound) and is mounted in a camera bag. When relaxing in one place or in bed, the pump may be removed and set down. A research assistant will visit each participant's home to demonstrate how to attach and start the pump, and will return the next day to pick it up. Participants will also complete a 'time-activity' diary recording their activities and location during monitoring. Equipment similar to what is being used in this study has already been used in studies conducted in Atlanta, Los Angeles, Nashville and Boston. The monitoring is noninvasive and your participation in this study should not interfere with your daily activities. The study has been approved by the Behavioural Research Ethics Board of the University of British Columbia. All information will remain confidential and participation in the study will not affect the medical care you receive. Monetary compensation in the amount of $250 will be provided to each participant. Your participation in this study would be greatly appreciated. A member of our research staff will be contacting you shortly with further details about procedures involved in collecting data and to ask about your interest in participating in this project. We hope that you will agree to participate and we thank you for your time. Sincerely, Kira Rich M.Sc. graduate student School of Occupational and Environmental Hygiene University of British Columbia 106 Appendix 2.1b. Recruitment letter (St. Paul's Hospital Clinic) April 18, 2001 Dear, In cooperation with Dr. Charles Kerr and the St. Paul's Hospital Pacemaker Clinic, Dr. Sverre Vedal, Dr. Michael Brauer, Ms. Kira Rich and colleagues at the University of British Columbia are conducting a study to evaluate the exposure of individuals with implantable cardiac defibrillators (ICDs) to air pollutants. Recent studies suggest that patients with cardiac arrhythmia may have increased susceptibility to air pollutants. We hope to identify factors that predict exposure and evaluate the health impact of air pollutants on patients with ICDs. Measurements of air pollutants have traditionally been taken outdoors. There are important questions about the relationship between outdoor air pollution and the exposure of individuals to this pollution. This study is being conducted to examine the relationship of measured outdoor and indoor air pollution and how these contribute to personal exposures. Each participant in the study will be asked to wear a personal pollutant monitor for seven 24 hour periods randomly spaced over the months of May to August, 2001. The monitor weighs approximately 500 grams (1 pound) and is mounted in a camera bag. When relaxing in one place or in bed, the pump may be removed and set down. A research assistant will visit each participant's home to demonstrate how to attach and start the pump, and will return the next day to pick it up. Participants will also complete a 'time-activity' diary recording their activities and location during monitoring. Equipment similar to what is being used in this study has already been used in studies conducted in Atlanta, Los Angeles, Nashville and Boston. The monitoring is noninvasive and your participation in this study should not interfere with your daily activities. The study has been approved by the Behavioural Research Ethics Board of the University of British Columbia. All information will remain confidential and participation in the study will not affect the medical care you receive. Monetary compensation in the amount of $250 will be provided to each participant. Your participation in this study would be greatly appreciated. A member of our research staff will be contacting you shortly with further details about procedures involved in collecting data and to ask about your interest in participating in this project. We hope that you will agree to participate and we thank you for your time. Sincerely, Kira Rich M.Sc. graduate student School of Occupational and Environmental Hygiene University of British Columbia 107 Appendix 2.2. Consent form U N I V E R S I T Y O F B R I T I S H C O L U M B I A D E P A R T M E N T O F M E D I C I N E RESPIRATORY DIVISION 277S Heather Street Vancouver, BC Canada, V5Z33S Ph {.604)87S-«22 F«(604)875-469S CONSENT FORM (version date October 2,2000) Study title: Air pollution health effects in potentially susceptible populations: refining the assessment of exposure Investigators: Sverre Vedal, MD, MSc (principal investigator) Charles Kerr, MD Michael Brauer, ScD John Petkau, PhD Jim Zidek, PhD Study purpose: 1 understand that the purpose of this study is to gather information on personal exposure to air pollution to see if this improves the ability to detect effects of air pollution on people with heart conditions. Study procedures: 1. Diary: I will be asked to complete what is called a "time-activity" diary for 7 separate days spread out over a period of about 2 to 3 months. The diary asks me to write down where I am and what I am doing for each of the seven 24-hour periods. ; 2. Pollution monitor: For the same 7 days in which I am completing the diary, I will be asked to wear a small and quiet battery-powered pump when I am up and about during the day. When I am relaxing in one place or in bed, I may remove the pump and set it down to work without being attached to me. A research assistant will visit my home to show me how to attach and start the pump, and then will return the next day to pick up the monitor and the diary. Confidentiality: I understand that it is important to the researchers that the information they collect on me be kept confidential. To make sure this happens, they will not keep any information that identifies me in the computerized files. The identification sheet for the diary will be kept in locked storage. Only identification numbers will be used to keep my information separate from the information on other patients. Access to the computerized files will be protected by password. Participation: I understand that my participation in this study is completely voluntary. Whether I participate or 108 CD O _ i o o </) DC LU a. 0 0 o CN a CU tx cx < o o CM _ o CD DQ E 00 •a CD > o o c CD 3 > co Q -CD CQ co O CD CL C X <D LU ^ CO c o CO 1 CD CL co CD 4—» tz CD E c E 2 £ > -E LU O CD CD O -a Q. CD O § o V 3 M— ° s .!= O < 00 d -*—* c CD Q. O • • + 3 CD l i CD CD iZ. a. < Comment Actual flow (L/min) u. u. Watch time O Pump Count (min) Flow (RU) CC tt Date (m/d/y) Actual Flow (L/min) Pump Count at Start z Watch Time O Flow (RU) CC tt (0 0. tt Date (m/d/y) Filter ID KR-KR-KR-KR-KR-KR-KR-Sample Day CNI co m CD >>-1 1 0 TT) Is |5 o •c a n p a> « 5 co >- > >- >- >- > > >- >- > 2 z >- > z z >-a> o o to c o o 2 >-z > z z >- > z >- >- > :> 5 o i l X TJ X TJ X "D X TJ X TJ X X •o X TJ X TJ X TJ X TJ X •a X TJ X TJ o X TJ X TJ X TJ X TJ X TJ X TJ X TJ 2 CO > > > > > > > CD sz CD sz CD sz CD sz <D x: x. ca 03 ca x CD cn ZJ ca x ro cn ca ro co ZJ ca CO CO ca ro co ZJ ca ro co ZJ CD x ro co ca CO cn ZJ ca co co ZJ ca CD -*-* O x ro CO ca CD CD x: CD x: CD x: ro CO ZJ ca ro co ZJ CD X ro CO ca x ro co ca x CO CO ca CD x: j—< O x ro co ca <D x: O ca co ca CD x: -4—> O x ro co ca CD sz X ro co ZJ CD CD x: O x ro co ca co O ro U ro O ro O ro O ro O ro O ro O ro O ro O ro O ro O ro O ro O ro U ro O ro O ro O ro O ro O ro O CD 2 I E § o s o o O <S J= 1 •p — n > ST I ro 2r1 CO 1 CO 2r1 ro < K? 1 ro ro ST 1 ro ST1 ro < ro < S^1 ro < S T 1 ro < co 2T1 ro < >> I ro < 5T1 co < ro S T 1 ro < ro < f^1 co I ca S'1 ro < CD E H ro b CD 4—' r s > a 0 r < rs CD CO E i-r^ CN X 9 K TJ c pen rso ter: CL CD < CL LL 3 CO p O 5 § S i ca CD ca CD ro CD ca CD ro CD co CD ro CD ro CD ro CD ro CD ro CD ro CD ro <D ro CD ca CD ro CD ro CD ro CD co CD ca CD ca CD CD sz CD x: CD x: CD SZ CD SZ CD x: CD x: E 1 o CD E o X o CO cd ci o do o o X o o ai 1 o CO CO i o E o X o 00 a> 1 o o ai 1 o CD E o X < o o o o CO ai s o o X o CO o o o 1 o CD E o X o o o CO 1 o o X o CO o o 1 o CD E o X C L o o o CO o CD E o X o CO o o csi XL 1 o CD E o X o o o CO csi XL 1 o CD E o X o CO o o XL 1 o CD E o X o o cSi 1 o CO XL 1 o CD E o X o CO CN 1 o o XL 1 o CD E o X o o CO I o CO XL l_ I O CD E o X o CO CO I o o CO XL o 0 E o X o o o CO CO o XL I ol CD E o X o CO o o O x: 1 1 o CD E o X o o in • o CO X 1 I o CD E o X o CO in • o o in XL o 5 CD E o X o o CD o CO X 1 o o X o CO CD I o o CD Notes ring #-» 5 z z z z Z z z z z z z z z z z z z z z z Z z z z z Wea Samt >- >- >- > >- > > > >- > > > >- > >- >- > > idows •pen z z z z z z z z z z z z z z z z z Z z z Z z z z z u >- > >- > >- >- > > > > > > > > > >-Tobacco' Smoke >-z z z > z > z z > z > z > z z z > z :>-z > z >-z >-z >-z z >-z z > z > z > z >-z z > okinc z z z z z z z z z z z z z z z z z z z z z z z z z o <J >- > > > > > > > > >- > > > > >->. -*—• I X X X X X X X X X X X X X X X X X X X X X X X X > a> > •o •o TJ T3 TJ "D •D T3 TJ TD T3 T3 TD TJ Q) • IE !s < o ~j o _ l o _J o _ l o _ l o _ l o _ l o _ l o _J o _ l o _ l o _ l o _ l o _ l o _ l o _J o _J o _ l o _ i o _J o _J o _ l o _ l o _ l o _J o _ l minutes busy road z z z z z z z z z z z z z z z z z z z z z z z z z busy road >- > > >- > >- > > > >- >- > > > > Trans CD -C 0 sz 0) sz 0 x: 0 x: 0 x: 0 sz 0 x: 0 x: 0 0 sz 0 x: 0 x: 0 x: 0 x: 0 x: 0 x: 0 x: 0 x: 0 sz 0 x: 0 x: 0 x: 0 x: 0 x: Trans O O o o O O o o o o o o o o O O O O O O O o o o o o ^ JX. ^ J<: 1 Wal Wal Wal Wal Wal Wal Wal Wal Wal Wal Wal Wal Wal Wal Wal Wa! Wal Wal Wal Wal Wal Wal Wal Wal Wal Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus Bus CD o Car Car Car Car Car Car Car Car Car Car Car Car Car Car Car Car Car Car Car Car Car Car Car Car Car doors 5 E s P Home Away Away Away Away Away Away Away Away Away Away Away Away Away Away Away Away Away Away Away Away Away Away Away Away Away Outi A/ear Home Near Near Near Near Near Near Near Near Near Near Near Near Near Near Near Near Near Near Near Near Near Near Near Near Near CD 0 SZ CD X: 0 x: 0 sz 0 sz 0 sz 0 sz 0 x: 0 0 x: 0 X! 0 x: 0 x: 0 x: 0 x: 0 x: 0 x: 0 x: 0 x: 0 x: 0 x: 0 x: 0 x: 0 x: 0 x: ors O O O O O O O O. O O O O O O O O O O O O O O O O O O ors ^: opu Wo; Wor Wor Wor Wor Wor o Wor Wor Wor Wor Wor Wor Wor Wor Wor Woi o Woi Woi Woi Woi Woi Woi Woi Woi Home Home Home Home Home Home Home Home Home Home Home Home Home Home Home Home Home Home Home Home Home Home Home Home Home Home :30-10:00 PM 12:30-1:00 AM Time 30-8:00 PM 00-8:30 PM :30-9:00 PM :00-9:30 PM :30-10:00 PM 10:00-10:30 P 3:30-11:00 P 1:00-11:30 P 1:30-12:00 A 12:00-12:30 A 12:30-1:00 AM :00-1:30 AM :30-2:00 AM :00-2:30 AM :30-3:00 AM :00-3:30 AM :30-4:00 AM :00-4:30 AM :30-5:00 AM :00-5:30 AM :30-6:00 AM :00-6:30 AM :30-7:00 AM :00-7:30 AM :30-8:00 AM 00 CO CO CD 10:00-10:30 P 12:00-12:30 A 12:30-1:00 AM CM CM CO CO ^3- ^r in in CD CD Appendix 2.5. Dwelling Information Form Page 1 of 2 ID: Dwelling Information Data Entered: Date: 1. Address: POSTAL C O D E Latitude Longitude 2. Proximity (<50m) to major roads (major road = 4 lanes): On a major road? Yes No If not, how many meters away from major road? m 3. Type of building: single house townhouse apartment building other:_ Building location: Street canyon (street for which the ratio of the distance from the buildings to the axis of the street and the height of the building was less than 1.5 - this can be estimated by field worker) Yes No 4. Apartment location in building (if applicable) Floor number: Corner unit? Yes No Side of building: North South East West 5 . Size of home Area: Volume: 6. Number of rooms: 7. Estimate percentage of floor space covered with carpets (entire house) 8. Kitchen Range hood? Yes No If yes, is it used? Yes No If yes, how often? Always Sometimes Never 9. Type of Ventilation? Natural only System: 10. Air conditioning? Yes No If yes, what kind? How often used? 113 Page 2 of 2 11. Heating System? Electrical Gas Forced Air/Furnace Hot Water/Radiator Other: 12. Fireplace? Yes No If yes, how many? Type and number: Wood Gas How often is it used? 13. Independent air filter/cleaner? Yes No Type: Location: How often is it used? 14. Windows Description of windows that are opened (type, quantity, average use per summer day): Always: Sometimes: Never: 15. Attached garage Yes No 114 TJ O "C CD Q. > TJ 3 CA CO '(/> >. CO c re ro o "5> o o £ TJ Q. Q> CT _C 'k. 3 TJ C V E < S3 a, a, < 00/22/21-002/2 L/21 0002/2 VIZ 00/22/L U OO/SL/LL 0002/L L/C oo/re/OL I 00/H/OL 0002/0 L/f 00/^ 2/60 00/H/60 0002/6/fr 00/92/80 tj 00/91780 0002/8/9 00/92//0 00/9 L/ZO 0002/Z/9 00/92/90 00/91-/90 0002/9/9 00/Z2/90 00/ZL/90 0002/9/Z 00/Z2/frO OO/ZL/fO 0002/WZ 00/92/80 j 00/91-/80 I 0002/8/9 00/172/20 00/H/20 ( iuBn) uojiBJjueouoo ^z]fid 115 (£.ui6n) "OS 116 00/33/31-003/31 IZ V 0003/3 L/3 00/23/11 00/£L/LL 0003/1- i/e 00/fr3AH oo/t^ u/ou 0 0 0 3 / 0 m 00/^3/60 00/H/60 0003/6/fr =| 00/93/90 00/91780 0003/8/9 cu 00/93/ZO ra Q 00/9 VI LO 0003/Z/9 00/93/90 00/91790 | 0003/9/9 00/Z3/S0 00/ZL/90 0003/9/Z 00/ZZ/M) OO/U/vO 0003/WZ 00/93/eo i 00 /9 i/eo 0003/E/9 00/^3/30 00/H/30 (£ uiBn) 0 0 117 I 00/33/31-003/3 L/31 0003/31-/3 00/33/1-1 oo/zmi 0003/U/G 00/K/OI-0003/0 L/f g 00/^ 3/60 00/^ 1-/60 0003/6/17 00/93/80 00/9 L/80 0003/8/9 1 00/93/ZO 1 00/91./Z0 0003/Z/9 00/93/90 00/91-/90 0003/9/9 I 00/Z3/90 00/ZL/90 | 0003/9/Z 00/Z3/170 I 00/ZL/t70 I 0003/WZ I 00/93/80 00/9 i./eo I 0003/8/9 00/t73/30 oo/n/30 (£ w6n) 03 118 (£ w6n) 0L | / \ |d 119 (qdd) oo 120 00 T 3 a CL) PH OH TJ O ™ CD Q. >» TJ 3 Cfl .S2 "co >* re c re "re o o o E CD TJ 'a. cu u> c 3 TJ CD TJ X O a c Cl) CJ) o c CD E < (qdd) ZQN 121 ro E x CO E 3 O SZ Ui c '(/) 3 T J o *c CD Q . > T5 3 <-» (/> ( A "to >» CO c CO "ro u "5> o o E <o T J 'a. <D Ui C "i_ 3 T J 0) c o N o c a> ! Q E < (qdd) e 0 122 TJ O 'C Q) Q. >. TJ 3 (A W '35 >» (0 c CO "ro o o o CD TJ Q. 0) CD 3 TJ CD TJ "x O 3 (/> c .22 !5 E < s cu C L , a, (qdd) 3QS 123 TJ O ™ CU Q . >i TJ 3 CO (fl 'to ra c re "ro o 'cn o o 1 TJ 'a. c 3 TJ 2! 3 +-* re L . 0) Q. E cu T J c CD CU < (sms|eo saajBap) ajn)ejadiuai 124 TJ O O Q. >» TJ 3 + J (A _</> 'tfl >. ro c re re o 'cn o o E <a> TJ 'o. a> D) _c 3 TJ £ TJ E 3 X > ro c OH OH 000Z/H/2 (%) m 125 (Bd>|) ajnssajd 126 TJ O ™ o a. >. TJ 3 Ui CO t/5 > re c re "re o "5> o o I CD TJ '5. CD D) 3 TJ 1 C ' r e o o o o o o h - CO LO CO CNJ T — (%) l l B J u ! e j 3UJOS 6 u p u a u a d x a p o u a d jnou , f2 i p e a j.o s j n o H 127 C\J O 0 0 CD (L-U,W>I) p a a d s p u j M 128 Appendix 3.2. Results of case crossover analyses after stratification by season. Odds ratios are presented per interquartile range of each pollutant, as observed during the study period. Meteorological variables were included in all models. Results are presented for lags of 0,1,2 and 3 days. P O L L U T A N T A N A L Y S E D S E A S O N L A G L O W E R 95% CI O D D S R A T I O U P P E R 95% CI 0 0.643 0.908 1.283 summer 1 0.421 0.865 1.777 2 0.137 0.518 1.965 P M 2 5 3 0.140 0.471 1.577 0 0.817 1.026 1.289 winter 1 0.442 0.788 1.406 2 0.325 0.678 1.416 3 0.273 0.621 1.411 0 0.862 1.087 1.371 summer 1 0.816 1.253 1.924 2 0.965 1.101 1.256 E C 3 0.509 1.055 2.185 0 0.315 0.609 1.180 winter 1 0.392 0.470 0.563 2 0.165 0.353 0.755 3 0.364 0.421 0.486 0 0.931 1.073 1.237 summer 1 0.931 1.064 1.217 2 0.975 1.064 1.161 O C 3 0.984 1.049 1.118 0 0.380 0.746 1.466 winter 1 0.330 0.758 1.743 2 0.231 0.482 1.005 3 0.187 0.405 0.879 0 0.641 0.920 1.322 summer 1 0.134 0.410 1.258 2 0.170 0.442 1.144 S 0 4 3 0.258 0.597 1.378 0 0.594 0.905 1.378 winter 1 0.592 0.905 1.382 2 0.726 1.140 1.791 3 0.859 1.405 2.298 0 0.509 1.548 4.703 summer 1 0.111 0.384 1.328 2 0.061 0.219 0.788 P M 1 0 3 0.033 0.179 0.964 0 0.332 0.680 1.394 winter 1 0.307 0.632 1.299 2 0.456 0.838 1.541 3 0.423 0.826 1.614 0 0.957 2.026 4.290 summer 1 0.737 1.756 4.187 2 0.533 1.201 2.706 C O 3 0.554 1.534 4.251 0 0.342 0.657 1.264 winter 1 0.334 0.637 1.215 2 0.453 0.751 1.246 3 0.403 0.729 1.319 129 0 0.636 1.569 3.874 summer 1 0.287 0.668 1.552 2 0.135 0.356 0.938 N 0 2 3 0.189 0.516 1.408 0 0.325 0.663 1.350 winter 1 0.404 0.762 1.438 2 0.338 0.645 1.232 3 0.386 0.779 . 1.574 0 0.528 1.394 3.681 summer 1 0.386 1.239 3.972 2 0.261 0.665 1.694 0 3 3 0.269 0.703 1.839 0 0.223 0.773 2.677 winter 1 0.672 2.269 7.662 2 0.624 2.634 11.110 3 1.202 4.892 19.917 0 0.962 2.701 7.585 summer 1 0.438 1.200 3.288 2 0.383 0.998 2.603 s o 2 3 0.632 1.768 4.950 0 0.244 0.517 1.093 winter 1 0.324 0.751 1.740 2 0.277 0.561 1.137 3 0.377 0.707 1.328 130 APPENDIX 3.3 Results of case crossover analysis sensitivity analysis I. Odds ratios and confidence intervals are presented for all pollutants with and without meteorological variables as covariates for lags of With meteorological variables Without meteorological variables P O L L U T A N T L A G LOWER ODDS UPPER LOWER ODDS UPPER 95% CI RATIO 95% CI 95% CI RATIO 95% CI 0 0.85 0.98 1.14 0.85 0.98 1.13 P M 2 5 (ug/m3) 1 0.55 0.80 1.18 0.77 0.92 1.10 2 0.47 0.75 1.18 0.68 0.90 1.19 3 0.35 0.61 1.05 0.49 0.74 1.11 0 0.85 1.05 1.30 0.87 1.04 1.25 E C (ug/m3) 1 0.92 1.08 1.27 0.95 1.12 1.32 2 0.94 1.04 1.14 0.95 1.05 1.15 3 0.89 0.99 1.10 0.90 0.99 1.09 0 0.95 1.09 1.24 0.96 1.08 1.21 O C (ug/m3) 1 0.95 1.07 1.20 0.97 1.08 1.21 2 0.96 1.03 1.12 0.96 1.04 1.12 3 0.95 1.01 1.09 0.95 1.01 1.08 0 0.68 0.91 1.21 0.73 0.93 1.17 S0 4(ug/m 3) 1 0.46 0.72 1.12 0.73 0.92 1.16 2 0.57 0.89 1.40 0.61 0.87 1.24 3 0.75 0.92 1.13 0.72 0.90 1.11 0 0.49 0.86 1.50 0.63 0.92 1.35 PM l 0 (ug/m 3 ) 1 0.36 0.62 1.07 0.68 0.97 1.40 2 0.42 0.69 1.13 0.59 0.86 1.24 3 0.36 0.63 1.11 0.52 0.76 1.13 0 0.60 0.91 1.38 0.69 0.94 1.27 C O (ppb) 1 0.53 0.79 1.18 0.81 1.08 1.44 2 0.60 0.84 1.19 0.73 0.95 1.25 3 0.56 0.84 1.27 0.75 1.01 1.36 0 0.49 0.88 1.60 0.64 0.96 1.43 N 0 2 (ppb) 1 .44 0.73 1.20 0.73 1.06 1.53 2 0.30 0.52 0.90 0.54 0.80 1.18 3 0.32 0.56 1.00 0.59 0.87 1.28 0 0.58 1.15 2.29 0.69 1.13 1.86 O 3 (ppb) 1 0.77 1.66 3.58 0.56 0.89 1.41 2 0.49 0.99 2.01 0.57 0.92 1.49 3 0.47 0.95 1.92 0.42 0.67 1.15 0 0.53 0.86 1.40 0.66 0.93 1.30 S 0 2 (ppb) 1 0.57 0.93 1.51 0.82 1.13 1.57 2 0.44 0.71 1.13 0.62 0.87 1.22 3 0.42 0.66 1.05 0.60 0.82 1.11 131 Appendix 3.4. Results of case crossover analysis sensitivity analysis II. Odds ratios and confidence intervals are presented for interquartile ranges for analyses of all events and of the first of events occurring within 72 hours, respectively. Meteorological variables were included as covariates in all models. All E V E N T S CONSIDERED L O W E R 95% UPPER 95% P O L L U T A N T C O N F I D E N C E ODDS R A TIO C O N F I D E N C E LIMIT LIMIT P M 2 . 5 all events 0.837 0.971 1.127 1 st event 0.849 0.982 1.135 S 0 4 all events 0.603 0.866 1.243 1 st event 0.680 0.908 1.213 E C all events 0.852 1.048 1.289 1st event 0.854 1.053 1.297 O C all events 0.952 1.083 1.232 1st event 0.952 1.086 1.239 PM,o all events 0.470 0.782 1.303 1st event 0.488 0.856 1.501 C O all events 0.574 0.918 1.468 1 st event 0.605 0.912 1.376 N 0 2 all events 0.359 0.788 1.729 1 st event 0.491 0.884 1.591 0 3 all events 0.600 1.238 2.555 1 st event 0.577 1.148 2.285 so2 all events 0.662 0.902 1.228 1 st event 0.531 0.863 1.401 132 Appendix 3.5 Results of case crossover analysis sensitivity analysis III. Odds ratios and confidence intervals are presented for interquartile ranges for analyses with and without the exclusion of events considered inappropriate. Meteorological variables were included as covariates in all models. All results are POLLUTANT EVENTS CONSIDERED LOWER 95% CONFIDENCE LIMIT ODDS RATIO UPPER 95% CONFIDENCE LIMIT EC appropriate 0.858 1.058 1.304 all 0.854 1.053 1.297 OC appropriate 0.955 1.090 1.245 all 0.952 1.086 1.239 PM2.5 appropriate 0.859 0.991 1.142 all 0.849 0.982 1.135 S0 4 appropriate 0.621 0.972 1.523 all 0.680 0.908 1.213 PM.o appropriate 0.536 0.919 1.577 all 0.488 0.856 1.501 CO appropriate 0.496 0.828 1.383 all 0.605 0.912 1.376 N0 2 appropriate 0.365 0.856 2.012 all 0.491 0.884 1.591 0 3 appropriate 0.638 1.400 3.073 all 0.577 1.148 2.285 so 2 appropriate 0.659 0.907 1.250 all 0.531 0.863 1.401 133 00 • c o o •o c i 5 I ? Lu * p i i o c 00 5 03 CO T3 fl CD CL CL < CM O CN O O CO CN O O CM CM O CN O O O CN O o CD O CN O O 00 O CN O o r-o CM o ( B L U ) m B j a M 134 00 CD (A E CN c o o TJ c to cn c E £ re a: 5 * 11 i i u-o JS c ° o S2 O a> 00 Ui c cn 5 re 3 o v© -a a to OH OH o o LO od CT> o CO od O CD cci o 00 o og od ( B L U ) jiiBsaM o o od o oo CO od 135 00 CD CN CD C 00 cn '3 5 CD o o o o o O o O O LO CO CM O o > 00 s - s - N - CD CD CD c o c d C0 00 00 00 CO 00 00 o > c n c n O) O) 03 CD (Btu) juiBjaM 136 Appendix 3.7. Histograms showing distributions of untransformed and natural log-transformed personal exposure variables. PERSONAL P M . . 3.5 24.5 45.5 66.5 87.5 108.5129.5150.5171.5 Untransformed personal PM2.5 (ug/m3) 20 15 10 o n h,n XL 1 .7 1.2 1.8 2.4 3.0 3.6 4.1 4.7 5.3 In-transformed personal PM2.5 (Inug/m3) PERSONAL ABSORBANCE 2.4 12.1 21.8 31.4 41.1 50.8 60.4 70.1 Untransformed personal Absorbance (10A-5mA-1) •3 1.0 1.8 2.5 3.3 4.0 In-transformed personal Absorbance (ln10A-5mA-1) PERSONAL SULFATE •2 1.7 3.2 4.6 6.1 7.5 Untransformed personal Sulfate (ugmA-3) •0 .3 .6 .9 1.2 1.5 1.8 2.0 2.3 In-transformed personal Sulfate (lnugm"-3) 137 Appendix 3.8. Comparison of sulfate analyses of 11 randomly selected samples repeated at Environment Canada and SOEH laboratories. Concentrations are presented in ug/mL sulfate. Validation Sample ID Sulfate Concentration (ug/mL) Sulfate Concentration (ug/mL) Change in Sulfate Concentration % change in Sulfate Concentration 1 0.22 0.24 0.02 7.61 2 1.82 1.51 -0.31 -17.08 3 0.13 0.20 0.07 55.93 4 2.20 1.77 -0.43 -19.50 5 1.40 1.39 -0.01 -0.75 6 1.20 1.25 0.05 4.15 7 2.52 2.21 -0.31 -12.29 8 2.60 2.16 -0.44 -16.79 9 1.15 1.10 -0.05 -4.03 10 0.46 0.49 0.03 6.85 11 0.90 0.86 -0.04 -4.34 138 Appendix 3.9. Histograms showing distributions of untransformed and natural log-transformed ambient exposure variables. AMBIENT PM, < 1.5 3.0 4.5 6.0 7.5 9.0 10.5 12.0 13.5 Untransformed ambient PM2.5 (ugmA-3) .5 .8 1.0 1.3 1.5 1.8 2.0 2.3 2.5 In-transformed ambient PM2.5 (lnugmA-3) AMBIENT ABSORBANCE 0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0 12.0 Untransformed ambient Absorbance (10A-5mA-1) J 1.0 1.3 1.5 1.8 2.0 2.3 2.5 In-transformed ambient Absorbance (ln10A-5mA-1) AMBIENT SULFATE 0.00 .50 1.00 1.50 2.00 2.50 3.00 Untransformed ambient Sulfate (ugmA-3) -4.5 -3.5 -2.5 -1.5 -.5 .5 In-transformed ambient Sulfate {lnugmA-3) 139 Appendix 3.10. Comparisons of LIPM BC and two carbon fractions at the Slocan monitoring station. Figure 1. Relationship of LIPM BC and CC at Slocan for the period 08/13/01 to 08/31/01 2.5 E o> 3^ C o ra O « ra c o .Q i— ra O 1.5 0.5 LIPM BC v Thermal Method CC 500 1000 1500 Black Carbon (10"9ngm'3) y =0.0003x + 0.3091 R 2 = 0.2432 • • t • • <> • • • • • • • 2000 2500 140 o E o 3 CO I s 3 II Q- C > 5 1 j? w « to o 'to c ^ fl) ra 3 ra ™ 3 C CO O o 5=: CL C x re a> a> co E .E o != += •2 0 ' E iB £ £ I 3 W co <2 c 0) 5 o J3 II TJ d) *•• +-» O TJ ( £ . iuBn) 142 IU SOLU|) aoueqjosqv iuaiqiueii| 143 Appendix 3.12a. Boxplots of continuous vs. categorical variables: MAJOR ROAD Page 1 of 2 144 Appendix 3.12a. Boxplots of continuous vs. categorical variables: MAJOR ROAD P a g e 2 of 2 N = 89 34 NO Y E S <0.15km from major road 145 Appendix 3.12b. Boxplots of continuous vs. categorical variables: G A R A G E Appendix 3.12b. Boxplots of continuous vs. categorical variables: G A R A G E P a g e 2 o f 2 G A R A G E IN U S E 147 Appendix 3.12c. Boxplots of continuous vs. categorical variables: C A R P E T Page 1 of 2 >80% carpet s 0 % c a r p e , 148 Appendix 3.12c. Boxplots of continuous vs. categorical variables: C A R P E T Page 2 of 2 149 Appendix 3.12d. Boxplots of continuous vs. categorical variables: COOKING Page 1 of 2 >120mins > 120 mins C O O K I N G 70 12 <120mins >120mins C O O K I N G 70 12 <120mins >120mins C O O K I N G 0 < 120 mins C O O K I N G 120 mins C O O K I N G < 1 2 0 m i n s > 1 2 0 m i n s C O O K I N G C O O K I N G 70 12 < 1 2 0 m i n s > 1 2 0 m i n s C O O K I N G 150 Appendix 3.12d. Boxplots of continuous vs. categorical variables: COOKING Page 2 of 2 N- 41 70 12 N= 41 70 12 0 <120mins >120 mins 0 <120 mins >120 mins COOKING COOKING 1200 1000 800 LU 0 <120mins >120mins COOKING 151 Appendix 3.12e. Boxplots of continuous vs. categorical variables: ETS Page 1 of 2 -1J . . I .4] N = 98 25 N = 98 25 N O Y E S N O Y E S E T S E T S 152 Appendix 3.12e. Boxplots of continuous vs. categorical variables: ETS Page 2 of 2 o N = 98 25 NO YES ETS 153 Appendix 3.13. Crosstabulations between all categorical variables included in multiple regression. <0.15 km from major road * garage in use Crosstabulat ion garage in use Total Y E S NO <0.15 km from major road NO 37 52 89 Y E S 34 34 Total 37 86 123 <0.15 km from major road * >80% carpet Crosstabulation >80% carpet Total NO Y E S <0.15 km NO 20 69 89 from major Y E S 7 27 34 road Total 27 96 123 <0.15 km from major road * COOKING Crosstabulat ion COOKING Total 0 <120 mins >120 mins <0.15 km from NO 33 49 7 89 major road Y E S 8 21 5 34 Total 41 70 12 123 <0.15 km from major road * ETS Crosstabulat ion ETS Total NO Y E S <0.15 km from major road NO 69 20 89 Y E S 29 5 34 Total 98 25 123 garage in use * >80% carpet Crosstabulation >80% carpet Total NO Y E S garage in use Y E S 7 30 37 NO 20 66 86 Total 27 96 123 garage in use * COOKING Crosstabulation COOKING Total 0 <120 mins >120 mins garage in use Y E S 8 28 1 37 NO 33 42 11 86 Total 41 70 12 123 154 garage in use * ETS Crosstabulat ion ETS Total NO Y E S garage in use Y E S 37 37 NO 61 25 86 Total 98 25 123 >80% carpet * COOKING Crosstabulation COOKING Total 0 <120 mins >120 mins >80% carpet NO 10 16 1 27 Y E S 31 54 11 96 Total 41 70 12 123 >80% carpet * ETS Crosstabulat ion ETS Total NO Y E S >80% carpet NO 24 3 27 Y E S 74 22 96 Total 98 25 123 COOKING * ETS Crosstabulat ion ETS Total NO Y E S COOKING 0 33 8 41 <120 mins 55 15 70 >120 mins 10 2 12 Total 98 25 123 155 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0091086/manifest

Comment

Related Items