Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Occupational noise exposure and the risk of diabetes, rheumatoid arthritis, and cardiovascular disease Song, Chaojie 2013

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

Item Metadata

Download

Media
24-ubc_2013_spring_song_chaojie.pdf [ 2.91MB ]
Metadata
JSON: 24-1.0071919.json
JSON-LD: 24-1.0071919-ld.json
RDF/XML (Pretty): 24-1.0071919-rdf.xml
RDF/JSON: 24-1.0071919-rdf.json
Turtle: 24-1.0071919-turtle.txt
N-Triples: 24-1.0071919-rdf-ntriples.txt
Original Record: 24-1.0071919-source.json
Full Text
24-1.0071919-fulltext.txt
Citation
24-1.0071919.ris

Full Text

OCCUPATIONAL NOISE EXPOSURE AND THE RISK OF DIABETES, RHEUMATOID ARTHRITIS, AND CARDIOVASCULAR DISEASE  by Chaojie Song  B.M., Guangzhou Medical University, 2010  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  MASTER OF SCIENCE in THE FACULTY OF GRADUATE STUDIES (Occupational and Environmental Hygiene)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  April 2013  © Chaojie Song, 2013  Abstract  Occupational noise exposure is associated with an increased risk of cardiovascular disease. A biological model explains this association through a general stress reaction and the dysregulation of otherwise normal psychoneuralhormonal pathways. Given this evidence and the hypothesized link between stress and metabolic disorders or immunologic response (i.e. inflammation) we hypothesized that risk of diabetes and arthritis are elevated in those exposed to noise at work.  Cases and controls were drawn from the National Population Health Survey (NPHS), a longitudinal survey conducted by Statistics Canada. Eligible subjects were aged 18 to 55 in the first cycle (1994/1995) and who did not have the health outcome of interest up to and including the fourth cycle (2000/2001). Cases were determined from self reports in the fifth (2002/2003) through eighth cycle (2008/2009); each case was matched to five controls who were disease free at the time when the case was diagnosed. Subjects’ noise exposure at work was assessed using cumulative exposure. Noise levels by various occupations/industries were derived from a job exposure matrix built from WorkSafeBC’s noise exposure dataset and exposure duration was estimated according to the self-reported work status in the NPHS. Conditional logistic regression was used to estimate the odds ratios of developing the health outcomes of interest; models were adjusted for life style factors (physical activity, smoking, drinking), socio-economic factors (education, family income), and health status factors (hypertension, obesity).  In the current study the adjusted diabetes ORs (95% CI) for medium and high exposure groups (using low exposure group as reference) were 0.93 (0.61 – 1.41) and 1.04 (0.67 – 1.59), respectively. The corresponding ORs (95% CI) for rheumatoid arthritis were 1.01 (0.56 – 1.82) and 1.07 (0.57 – 2.01). For cardiovascular disease, the ORs (95% CI) were 0.87 (0.59 – 1.28) and 0.85 (0.57 – 1.27).  We did not find evidence of an increased risk of diabetes, rheumatoid arthritis, or cardiovascular disease in those exposed to occupational noise. Possible biases in this study may explain the lack of an observed association including misclassification bias – particularly of exposure - and the ii  healthy worker effect. Future studies are warranted and should focus on improving exposure assessment.  iii  Table of Contents  Abstract ..........................................................................................................................................ii Table of Contents.......................................................................................................................... iv List of Tables...............................................................................................................................viii List of Figures ............................................................................................................................... ix Acknowledgements ........................................................................................................................ x Dedication...................................................................................................................................... xi Chapter 1: Introduction ............................................................................................................. 1 1.1  Noise in the Workplace .............................................................................................................1  1.2  Adverse Health Effects of Noise Exposure ...............................................................................1  1.3  Biological Basis of the Noise-Disease Association...................................................................2  1.3.1 Noise and Stress ....................................................................................................................2 1.3.2 Stress Response .....................................................................................................................4 1.3.3 The Noise-CVD Biological Model ........................................................................................6 1.3.4 A More General Biological Model ........................................................................................7  1.4  1.3.4.1  Biological Basis for a Noise and Diabetes Association.................................................7  1.3.4.2  Biological Basis for a Noise and Rheumatoid Arthritis Association ............................8  Diabetes .....................................................................................................................................8  1.4.1 Epidemiological Studies of Noise and Diabetes....................................................................9 1.5  Rheumatoid Arthritis ...............................................................................................................11  1.5.1 Epidemiological Studies of Noise and Rheumatoid Arthritis..............................................11 1.6  Noise Exposure Assessment in Epidemiological Studies........................................................12  1.7  Research Objectives ................................................................................................................14  Chapter 2: Methods.................................................................................................................. 16 2.1  Background Information about the National Population Health Survey .................................16  2.1.1 Target Population ................................................................................................................17 2.1.2 Sampling..............................................................................................................................17 2.1.3 Data Collection ....................................................................................................................17 2.2  Background Information about the Noise Exposure Data from WorkSafeBC ........................18  2.3  Study Design ...........................................................................................................................19 iv  2.4  Case and Control Definition ....................................................................................................20  2.5  Exposure Assessment ..............................................................................................................24  2.5.1 Building the Job Exposure Matrix.......................................................................................25 2.5.2 Estimating Exposure Duration ............................................................................................26 2.5.3 Assigning Noise Exposure Level ........................................................................................28 2.5.4 Calculating the Cumulative Noise Exposure .......................................................................30 2.5.5 Categorizing Exposure Groups............................................................................................31 2.6  Covariates ................................................................................................................................31  2.6.1 Income and Education .........................................................................................................31 2.6.2 Obesity.................................................................................................................................32 2.6.3 Physical Activity .................................................................................................................33 2.6.4 Smoking...............................................................................................................................34 2.6.5 Drinking...............................................................................................................................34 2.6.6 High Blood Pressure ............................................................................................................35 2.6.7 Covariates for the Three Health Outcomes of Interest ........................................................35 2.7  Statistical Analysis ..................................................................................................................35  2.7.1 Descriptive Statistics ...........................................................................................................35 2.7.2 Bivariate Analysis ...............................................................................................................36 2.7.3 Multiple Logistic Regression Analysis ...............................................................................36 2.7.4 Selection of Predictors.........................................................................................................36 2.8  Dealing with Missing Values ..................................................................................................37  2.8.1 Missing Health Outcome Measurement ..............................................................................38 2.8.2 Missing Exposure Assessment ............................................................................................38 2.8.3 Missing Covariates ..............................................................................................................39  Chapter 3: Results .................................................................................................................... 41 3.1  Job Exposure Matrix ...............................................................................................................41  3.1.1 Noise Exposure by Occupations ..........................................................................................42 3.1.2 Noise Exposure by Industries ..............................................................................................43 3.1.3 Grab Sample vs. Personal Sample .......................................................................................44 3.1.4 Temporal Trend ...................................................................................................................47 3.1.5 Quality of the Assigned NOC-S Codes ...............................................................................49 v  3.2  National Population Health Survey .........................................................................................50  3.2.1 Diabetes ...............................................................................................................................50 3.2.1.1  Characteristics of Cases and Controls .........................................................................50  3.2.1.2  Collinearity among Covariates ....................................................................................50  3.2.1.3  Association between Noise Exposure and Covariates .................................................50  3.2.1.4  Associations between Noise Exposure, Other Covariates and Diabetes .....................51  3.2.2 Rheumatoid Arthritis ...........................................................................................................56 3.2.2.1  Characteristics of Cases and Controls .........................................................................56  3.2.2.2  Collinearity among Covariates ....................................................................................56  3.2.2.3  Association between Noise Exposure and Covariates .................................................56  3.2.2.4  Association between Noise Exposure and Rheumatoid Arthritis ................................57  3.2.3 Cardiovascular Disease .......................................................................................................62 3.2.3.1  Characteristics of Cases and Controls .........................................................................62  3.2.3.2  Collinearity among Covariates ....................................................................................62  3.2.3.3  Association between Noise Exposure and Covariates .................................................62  3.2.3.4  Association between Noise Exposure and Cardiovascular Disease ............................63  Chapter 4: Discussion .............................................................................................................. 68 4.1  Summary of Key Findings.......................................................................................................68  4.1.1 Diabetes ...............................................................................................................................68 4.1.2 Rheumatoid Arthritis ...........................................................................................................70 4.1.3 Cardiovascular Disease as “Sensitivity Analysis”...............................................................71 4.2  Potential Impact of Bias ..........................................................................................................72  4.2.1 Misclassification in Health Outcome and Covariates..........................................................73 4.2.2 Biases in Exposure Assessment...........................................................................................75 4.2.2.1  Sampling Stage ............................................................................................................75  4.2.2.2  Coding Stage ...............................................................................................................76  4.2.2.3  JEM Building Stage.....................................................................................................76  4.2.2.4  Linking Health and Exposure Stage ............................................................................79  4.2.2.5  Estimating Cumulative Noise Exposure Stage ............................................................80  4.2.3 Hearing Protectors ...............................................................................................................82 4.2.4 Healthy Worker Effect ........................................................................................................82 vi  4.3  Strengths and Limitations ........................................................................................................83  4.3.1 Strengths ..............................................................................................................................83 4.3.2 Limitations...........................................................................................................................84 4.4  Future Perspectives..................................................................................................................85  4.5  Conclusions .............................................................................................................................87  Bibliography................................................................................................................................. 88 Appendices ................................................................................................................................. 101 Appendix A: ......................................................................................................................................101 Appendix B: ......................................................................................................................................103  vii  List of Tables  Table 2. 1  Sample of the Noise Job Exposure Matrix .............................................................. 27  Table 2. 2  Summary of the Composition of the Job Exposure Matrix ..................................... 27  Table 2. 3  List of Covariates for Different Health Outcomes .................................................. 37  Table 2. 4  Procedures of Dealing with Missing Values in Exposure Assessment .................... 40  Table 3. 1  Noise Exposure Levels by Occupations (1-digit NOC-S Level)............................. 42  Table 3. 2  Noise Exposure Levels by Industries (2-digit NAICS Level) ................................. 43  Table 3. 3  Comparison between Grab Sample and Personal Sample ....................................... 44  Table 3. 4  Comparison between Grab Sample and Personal Sample by Occupations ............. 45  Table 3. 5  Comparison between Grab Sample and Personal Sample by Industries ................. 46  Table 3. 6  Characteristics of Diabetes Cases and Controls ...................................................... 52  Table 3. 7  Correlation Matrix for Covariates ........................................................................... 53  Table 3. 8  Associations between Exposure Groups and Covariates ......................................... 54  Table 3. 9  Summary of Results for Diabetes from Single Logistic Regression and the Final  Multivariate Logistic Regression .................................................................................................. 55 Table 3. 10  Characteristics of Rheumatoid Arthritis Cases and Controls ................................ 58  Table 3. 11  Correlation Matrix for Covariates ......................................................................... 59  Table 3. 12  Associations between Exposure Groups and Covariates ....................................... 60  Table 3. 13  Summary of Results for Rheumatoid Arthritis from Single Logistic Regression  and the Final Multivariate Logistic Regression ............................................................................ 61 Table 3. 14  Characteristics of CVD Cases and Controls .......................................................... 64  Table 3. 15  Correlation Matrix for Covariates ......................................................................... 65  Table 3. 16  Associations between Exposure Groups and Covariates ....................................... 66  Table 3. 17  Summary of Results for Cardiovascular Disease from Single Logistic Regression  and the Final Multivariate Logistic Regression ............................................................................ 67 Table B. 1  Model Selection Process for Diabetes .................................................................. 103  Table B. 2  Model Selection Process for Rheumatoid Arthritis .............................................. 106  Table B. 3  Model Selection Process for Cardiovascular Disease ........................................... 108 viii  List of Figures  Figure 1. 1  Noise and Cardiovascular Disease Reaction Model ................................................ 4  Figure 2. 1  Flowchart of Selection Cases and Controls for Diabetes....................................... 22  Figure 2. 2  Flowchart of Selection Cases and Controls for Rheumatoid Arthritis................... 23  Figure 2. 3  Flowchart of Selection Cases and Controls for Cardiovascular Disease ............... 24  Figure 2. 4  Noise Assignment Procedures................................................................................ 30  Figure 3. 1  Histogram of Noise level ....................................................................................... 41  Figure 3. 2  Q-Q Plot of Noise Level ........................................................................................ 41  Figure 3. 3  Scatter Plot of Noise Exposure Level by Year ....................................................... 47  Figure 3. 4  Scatter Plots of Noise Level in Certain Occupations ............................................. 48  Figure 3. 5  Scatter Plots of Noise Level by Industries ............................................................. 49  Figure A. 1  Temporal Trend of Noise Level by Occupations ................................................ 101  Figure A. 2  Temporal Trend of Noise Levels by Industries ................................................... 102  ix  Acknowledgements  I would like to thank Dr. Hugh Davies, my thesis supervisor, who has supported me throughout my thesis with his patience and expertise. Hugh was a great mentor and I really learnt a lot from him. It was a great experience to work with him. I would also like to thank Dr. Mieke Koehoorn and Dr. Kay Teschke for their support and suggestions on my thesis work. I was fortunate to have them in the committee with knowledge and experience.  I would like to acknowledge the British Columbia Research Data Centre, where I accessed the National Population Health Survey data. I would also like to thank Karen King and Lee Grenon at the Research Data Centre for their guidance in using the sources in the centre.  I would like to thank Nichole Garzia, who offered me great help in applying for the access to the Research Data Centre, and Calvin Wu, who helped me with the final edits of my thesis.  I would also like to acknowledge WorkSafeBC and the Bridge Program for their funding support.  Finally, I’d like to thank my parents, who are always there for me even there is time and space between us. Their love and support are the motivation which continuously keeps me move forward. Mom, dad – thank you!  x  Dedication  Dedicated to my parents, the shelter and beacon of my life.  xi  Chapter 1: Introduction  1.1  Noise in the Workplace  Noise is one of the most common hazards in the workplace globally (Nelson et al., 2005). In the European Union, 28% of workers involved in a survey reported that they were exposed to occupational noise which was so loud that they needed to raise their voices to talk to others (above 85 to 90 dBA) during one quarter of their working hours (EASHW, 2000). Twenty two million (17.2%) workers in the United States of America are exposed to occupational noise at a hazardous level (Tak et al., 2009). The exposure situation is probably worse in developing countries due to lack of effective control methods (Morata et al., 1997; Shaikh, 1996; Mukherjee et al., 1995; Oleru et al., 1990).  The number of noise-exposed workers in British Columbia (BC) is not known, but since the number of the workers in Canada is estimated at about 10% of that in the USA, and in British Columbia about 10% of that in Canada, a crude estimate of the number of noise-exposed workers (85 dBA or over) in BC is 220,000. From 2011 WorkSafeBC statistics, we also know that 138,818 workers had audiometric test for a Hearing Conservation Program (WorkSafeBC, 2011). So the number of noise-exposed workers in BC can be estimated as within the range of 140,000 to 220.000.  1.2  Adverse Health Effects of Noise Exposure  Adverse health effects caused by exposure to noise include noise-induced hearing loss (NIHL), sleep difficulties, annoyance, psychological stress, elevated blood pressure, and cardiovascular diseases (Nelson et al., 2005; Ising and Kruppa, 2004). Among these, NIHL is regarded as the most well understood and has received the most attention. As one of the most common sources of noise, occupational exposure plays an important role in causing NIHL. It is estimated that about 16% of disabling hearing loss can be attributed to occupational noise exposure (Nelson et 1  al., 2005). Although much emphasis has been placed on NIHL, there is increasing evidence of other non-auditory health effects, in particular cardiovascular disease (Gopinath et al., 2011; Selander et al., 2009; Sbihi et al., 2008; Babisch et al., 2005; Davies et al., 2005). Noise is known to be a potent stressor and non-auditory effects are assumed to be stress-mediated. Noise exposure triggers a stress response that in turn leads to various health outcomes. Now interest is turning to other diseases that may be stress-related, such as diabetes and rheumatoid arthritis.  1.3  1.3.1  Biological Basis of the Noise-Disease Association  Noise and Stress  The perception of sound (or noise – i.e. unwanted or harmful sound) is a complex process in which two auditory pathways are involved, one from the inner ear to the auditory cortex, and the other from the inner ear to the reticular activating system which then connects to the limbic system as well as the autonomic nervous system and the neuroendocrine system (Westman and Walters, 1981; Cohen, 1977).  In the first pathway, nerve impulses excited by sound stimuli travel along the ascending nerve path from the receptor cells in the Corti to the auditory centers in the cortex. These impulses reach the temporal lobe and are perceived and interpreted consciously. A descending nerve path also exists, from the temporal cerebral cortex to the dorsal cochlear nucleus, exerting inhibitory and some minor excitatory effects.  In the second pathway, nerve impulses are also generated at the inner ear. However, the nerve fibers branch out and connect to the motor cell nuclei subserving reflexes within the brainstem and to the reticular activating system (RAS) in the midbrain. Since the RAS is also linked to the sympathetic-adrenal neuroendocrine (SAM) system and the hypothalamo-pituitary-adrenal neuroendocrine (HPA) system, the impulses arriving at the RAS can also be conveyed to these two systems. The SAM system regulates the secretion of catecholamines, adrenaline (epinephrine) and noradrenaline (norepinephrine) while the HPA system regulates the 2  corticosteroids. These two systems and the hormones regulated by them play an essential role in adapting human body to stressors (discussed in detail in Section 1.3.2). Therefore, the auditory system, the central nervous system, and the neuroendocrine system are all connected by these two pathways.  The mechanism of the stress response elicited by sound acting as a stressor can be explained according to the biological basis of these “direct” and “indirect” pathways. The pathway involving the reticular activating system is regarded as the direct pathway because the sound stimuli “directly” evoke the stress response by neural activation of the reticular activating system, which in turn triggers the response of the SAM system and the HPA system. The other pathway, which connects to the cerebral cortex, is referred to as the indirect pathway because whether a sound stimulus is perceived as a stressor or not depends in part on the cognitive processing of the auditory cortex (Ising and Kruppa, 2004; Babisch, 1998; Westman and Walters, 1981).  With respect to the direct pathway, we must consider that the auditory system evolved to arouse or to provide “early warning”; thus in response to sound it activates the reticular activating system automatically and in turn the autonomic-neuroendocrine responses. This “defense” mechanism works unconsciously. In the indirect pathway, it is a person’s cognition of the perceived sound/noise that determines the degree of stress generated by the sound stimuli and this is influenced by both situational parameters (e.g. communication tasks, requirement of concentration) and individual parameters (e.g. coping potential, noise sensibility, attitude towards the sound/noise) (Ising and Kruppa, 2004).  The activation of two pathways is not mutually exclusive. They work simultaneously when there is a sound stimulus, with one pathway predominating over the other, depending on the characteristics of the sound. The direct pathway is thought to be selected more often when the sound stimulus is of high intensity and is novel and unpredictable, while the indirect pathway is selected more often when the sound level of the stimulus is moderate but the signal contains meaningful information, or when other situational or individual parameters are involved (Babisch, 1998; Westman and Walters, 1981). The noise-stress hypothesis response is shown 3  graphically in Figure 1. 1.  Figure 1. 1  1.3.2  Noise and cardiovascular risk-reaction model (from Babisch, 1998)  Stress Response  The general adaptation syndrome, or the stress syndrome, is a defense mechanism of the organism to cope with stress and maintain homeostasis. Many organs and systems of the body are activated to deal with stress under the regulation of the sympathetic nervous system and neuroendocrine system. In Selye’s theory, the fully developed general adaptation syndrome has three stages: an alarm reaction, a stage of resistance and a final stage of exhaustion. Although the stress response exists as a protective mechanism of the organism, excessive activation of this 4  response will lead to adverse disease outcomes. Some severe stress rapidly results in exhaustion, or even death (Selye, 1973).  The biological basis for the stress response was explained in a model proposed by Henry (1993). In his model, two major stress reactions were identified: the defense reaction, which responds actively to cope with stress to the organism; and the defeat reaction, a passive response when the body feels loss of control. The disease outcomes associated with a stress response are thought to relate to the dysregulation of these two reactions.  Henry’s defense reaction, which is equivalent to the fight or flight response, is mainly based in the SAM system. The reaction is activated when the organism’s control is threatened. The sympathetic nervous system plays an essential role in this response and has a major regulating effect on the activity of smooth muscles or cardiac muscles. Through the secretion of norephinephrine, the function of the organs (e.g. heart, vessels, skin, etc), which consist of smooth muscles or cardiac muscles innervated by the sympathetic nervous system, is increased. Furthermore, the activation of the sympathetic nervous system conveys the signal to the adrenal medulla through sympathetic preganglionic fibers from spinal cord, which then causes the adrenal medulla to release norepinephrine and epinephrine. This process further strengthens the sympathetic response in target tissues or organs (Lovallo, 1997). The release of norepinephrine from locus ceruleus in the brain stem to various locations of the brain heightens the organism’s sense of attention and vigilance (Dinan, 1996).  Henry’s defeat reaction is activated when the organism feels loss of control and it is regulated through the hypothalamic-pituitary-adrenocortical (HPA) axis. Among the hormones (ACTH, corticosteroids, catecholamine, etc.) involved in the defeat reaction, cortisol is most important and has numerous central and peripheral effects, such as increasing plasma concentrations of amino acids and glucose, enhancing the actions of the sympathetic nervous system, and increasing the plasma volume, which are necessary for the organism to cope with stress (Vander et al., 1994).  5  1.3.3  The Noise-CVD Biological Model  Based on the hypotheses of the noise-stress and stress-physiology responses, an association between noise exposure and cardiovascular disease has been discussed (Ising and Kruppa, 2004; Maschke et al., 2000; Babisch, 1998).  In the defense reaction, the response of the SAM system is normally transient. Repeated or chronic activation of the system however might increase blood pressure and change the structure of the heart and vessels, which are regarded as important pathophysiological changes in the development of CVD (al’ Absi and Arnett, 2000). In the defeat reaction, cortisol plays an important part and is believed to be associated with cardiovascular disease through its cardiovascular activation effect (al’Absi and Arnett, 2000).  The results from many experimental studies in both humans and animals support the proposed biological model. It was found that exposure to noise caused a significant increase in systolic, diastolic and mean arterial pressure (Andren et al., 1983). In the Sudo et al. (1996) study, levels of epinephrine, norephinephrine, and cortisol were increased in subjects exposed to noise. In animal studies, Peterson et al. (1981) showed increased blood pressure in rhesus monkeys exposed to noise representing an occupational environment. The blood pressure did not go back to the normal level until one month after the cessation of exposure. Similar results have been found in other animals, such as rats and dogs (Fisher and Tucker, 1991; Engeland et al., 1990; Friedman et al., 1967).  It was proposed that, between the two pathways of noise perception, the indirect pathway is selected more frequently when the noise intensity is low and contains content and meaningful information, or when other parameters are involved. On the other hand, if the noise exposure is of high intensity or novel to the receiver, the direct pathway may be the major path. It is unknown which of the two stress reactions is activated more frequently by which pathway. It is possible that either pathways may be related to either reaction and there are complex interactions between the pathways and stress reactions. 6  1.3.4  A More General Biological Model  As discussed, two major stress reaction patterns are activated in stress response and may result in cardiovascular effects. However, the impact of the stress response is not restricted to the circulatory system. Many other organs or systems are involved, working together to adjust the organism to either external or internal stress, such as the nervous system (increased arousal, cognition, and vigilance), respiratory system (more air intake for increased metabolism), and digestive system (inhibited function). Therefore, excessive activation of the stress response and subsequent dysregulation of normal systems may impact various parts of the body, not just the cardiovascular system, and lead to other disease outcomes, for example, diabetes or rheumatoid arthritis, discussed below.  1.3.4.1  Biological Basis for a Noise and Diabetes Association  Diabetes and CVD share risk factors such as obesity, physical inactivity, and hypertension. This indicates that it is possible that these diseases may share similar pathophysiologic pathways. The noise-CVD biological model discussed involves, both the SAM system and the HPA axis in the stress response caused by noise exposure, representing the defense reaction and the defeat reaction respectively (Henry, 1993). Over activation of the HPA axis can result in chronic hypercortisolism, which in turn leads to visceral fat accumulation, causing central obesity, an essential risk factor for type 2 diabetes (Chrousos, 2000). Moreover, high levels of cortisol are associated with insulin resistance and/or hyperinsulinemia, which include heightened gluconeogenesis, visceral fat cell growth, carbohydrate intolerance, increased total cholesterol, LDL, triglycerides, decreased HDL, and impaired insulin secretion. All of these are regarded as important changes in the development of diabetes (Chrousos, 2009; Chrousos, 2000; Maschke et al., 2000). Given the association between noise and hypertension and CVD, it is also possible that there exists an indirect relationship between noise exposure and diabetes using hypertension and/or CVD as an intermediate factor.  7  1.3.4.2  Biological Basis for a Noise and Rheumatoid Arthritis Association  During the stress response, the immune system is involved and its function is influenced by the stress reactions in a very complex way; whether a proinflammatory or an anti-inflammatory effect is expected largely depends upon activity of the HPA axis (Chrousos, 2009; Chrousos, 1995; Sternberg et al., 1992). The increased activity of the HPA axis, which results in increased levels of glucocorticoids (mainly cortisol) and catecholamines, influences the function of leukocytes and accessory immune cells and suppresses the release of proinflammatory cytokines (TNF, IL-1, IL-6, IL-8, IL-12). Therefore, a suppressed function of the immune system is expected to protect the organism from an “overshoot” of the inflammatory response.  Conversely, decreased activity of the HPA axis leads to low levels of plasma cortisol, a situation in which the function of the immune system and the inflammatory effect are heightened due to lack of suppression from the HPA axis. The activation/suppression of the HPA axis depends on many factors such as the duration of stress, the intensity of stress, the perception of stress, and the internal environment of the stressed individual (Straub and Kalden, 2009). Therefore, although there may be a biologically plausible association between noise as an external stressor and rheumatoid arthritis as an autoimmune disease, the association could be quite complex.  1.4  Diabetes  Diabetes is a chronic disease which results when either the pancreas is not able to produce enough insulin or when the body cannot use insulin effectively. Generally speaking, there are two major types of diabetes. Type-1 diabetes occurs due to lack of insulin production by the pancreas and it is more frequently diagnosed in children, teenagers, and young adults. The exact cause of the type-1 diabetes is unknown and therefore it is considered unpreventable.  Type-2 diabetes, which results from ineffective use of insulin, constitutes most diabetes cases and it is most often diagnosed in adults. The risk factors for type-2 diabetes include advanced age, overweight or obesity, physical inactivity, high blood pressure and/or high cholesterol, heart 8  disease, and family history of diabetes (WHO, 2012; PHAC, 2012). Due to the unknown cause of type-1 diabetes, we only focus on type-2 diabetes in the current study.  Globally, 285 million people were diagnosed with diabetes in 2010. This number is expected to increase to 438 million by 2030 (Canadian Diabetes Association, 2013). In 2004, it ranked 11th leading causes of death in the world (WHO, 2004). Within Canada, there were approximately 1.4 million people with diabetes in 2000. Ohinmaa et al. (2004) predicted that this number would reach 2.4 million in 2016 and the total healthcare costs for diabetes were projected to increase from $4.66 billion in 2000 to $8.14 billion in 2016. Currently Canada is facing a challenge in that type-2 diabetes is increasing dramatically for various reasons including an aging population, increasing obesity prevalence, unhealthy lifestyle, and increasing number of new immigrants from high risk populations (Canadian Diabetes Association, 2013). In British Columbia, 6.46% of the population (6.92% of men and 6.01% of women) had diabetes in 2007/2008. The prevalence of diabetes in BC has a slight increasing trend, based on longitudinal data on diabetes from 2001/2002 to 2007/2008 (Provincial Health Services Authority, 2010). In 2007, 996 deaths were caused by diabetes, 3.2% of total deaths in BC (Statistics Canada, 2007).  1.4.1  Epidemiological Studies of Noise and Diabetes  Very few previous population-based studies have investigated this topic. Only one population-based study was found that assessed the direct association between noise exposure and diabetes. Sorensen et al. (2012) conducted a cohort study to investigate whether long-term exposure to residential road traffic noise was associated with increased risk of diabetes. They found incidence rate ratios of 1.11 (95% CI: 1.03 – 1.19) and 1.14 (95% CI: 1.06 – 1.22) per 10 dB increase in road traffic noise at diagnosis and during the 5 years preceding diagnosis of diabetes, respectively. Besides the stress response mechanism discussed in previous sections, they proposed another possible biological mechanism associated with the sleep disturbance caused by road traffic noise leads to diabetes based on a positive association found between type 2 diabetes and deficient sleep quantity and/or poor sleep quality in previous epidemiological studies (Cappuccio et al., 2010). In the environmental scenario, the stress mechanism and sleep 9  disturbance mechanism work together and lead to the diabetes outcome. However, the sleep disturbance mechanism is less likely to play a role in the occupational scenario.  No studies were found focusing on the direct association between diabetes and occupational exposure to noise. Other studies have been conducted to investigate the relationship between work stress and diabetes, however. In Li et al.’s study (2012), work stress was associated with both diabetes (OR: 1.27; 95% CI: 1.02 – 1.58) and prediabetes1 (OR: 1.26; 95% CI: 1.01 – 1.58) in men, but not women. However, Djindjic et al. (2012) found the opposite results; women with higher occupational stress had higher risk of diabetes compared to men. Given the inconsistency from previous studies, Cosgrove et al. (2012) conducted a meta-analysis of nine studies to investigate the association between work-related psychosocial stress and type 2 diabetes. The work-related psychosocial stress was evaluated in the form of high demand, poor decision latitude, poor social support, job strain, or long working hours. No evidence was found to support the hypothesized association. The author suggested future studies aim at using a more rigorous assessment of work-related stress and also including other non-work social and relationship factors as indicators of psychological stress.  To summarize, although the noise-diabetes association seems biologically plausible, there have been limited population-based studies and evidence to support this association. The results from relevant studies were inconsistent. Even though a work stress-diabetes relationship might parallel the noise-diabetes biological model to some extent, the stress caused by noise exposure can be significantly different from general work-related psychological stress due to the complex process of noise perception. Furthermore, other mechanisms may exist to explain the association between noise exposure and diabetes as proposed by Maschke et al. (2000) for the noise-CVD association. It is possible that noise exposure leads to behavioral changes (e.g. decreased physical activity or change of sleep habits), that then result in diabetes. Given the limited evidence and inconsistent results from previous studies, no conclusion can be made at the current stage and more in-depth studies are warranted.  1  Prediabetes is the state in which some but not all of the diagnostic criteria for diabetes are met. People with prediabetes have  glucose levels that are higher than normal but not high enough yet to indicate diabetes.  10  1.5  Rheumatoid Arthritis  Rheumatoid arthritis is an autoimmune disease in which the immune system attacks the body’s own tissue. It influences various parts of the body, especially the joints, causing pain, stiffness, and swelling that limit the normal function of the joint. The cause of the disease is still unknown but genetic factors are thought to be involved. Women are at higher risk of RA than men while for both women and men the onset of RA is highest among people in their sixties. Besides these factors, smoking, reproductive and breastfeeding history are also associated with increased risk of RA onset (Centers for Disease Control and Prevention, 2012; Statistics Canada, 2006). Although RA seldom leads to death, severe cases seriously impact the life quality of the patient by impairing the ability to do basic daily tasks (e.g. dressing, washing) (Statistics Canada, 2006). The prevalence of RA in developed countries varies between 0.3% and 1.0%. In developing countries, there is large variation in reported RA prevalence with some lower prevalence rates similar to those reported in industrialized countries (Symmons et al., 2006). In 2010, 272,000 people had RA in Canada and the prevalence was 0.89%. The prevalence of RA was about two times higher in women than in men. Although the prevalence trend became relatively stable after 2006, the number of people diagnosed with RA is expected to increase to over 549,000 by 2040 (1.34% of the total Canadian population) (Arthritis Alliance of Canada, 2011). In British Columbia, 1.0% of the population (0.6% of men and 1.4% of women) had RA during 2007/2008. According to the longitudinal data from 2001 to 2008, there was an increasing trend in the prevalence of RA from 2001 to 2006 (Provincial Health Services Authority, 2010).  1.5.1  Epidemiological Studies of Noise and Rheumatoid Arthritis  A literature review of population-based epidemiological studies on the direct association between noise exposure and RA did not find any published peer-reviewed studies. Further, only one study investigating the association between stress and RA was found (using the death of a child as an indicator of psychological stress, Li et al., 2005). No statistically significant results were found to support the association between this severe psychological stress and RA. 11  Yet other pathophysiological pathways may exist beside the proposed biological model based on stress response. It is possible that noise exposure modifies risk factors (e.g. smoking status) of RA and therefore, an indirect relationship is expected (Maschke et al., 2000). To better understand a potential noise-RA association, more studies are required.  1.6  Noise Exposure Assessment in Epidemiological Studies  The assessment of occupational noise exposure in epidemiological studies remains a challenge, especially for retrospective studies because the available historical exposure data is limited and no standardized method has been established. In previous studies focusing on the noise – CVD association (McNamee et al., 2006; Davies et al., 2005; Sbihi et al., 2003), based on which we derived the general noise – disease biological model, the noise intensity, exposure duration or the combination of the two were often used to assess occupational noise exposure.  Data on noise intensity may come from either historical or current individual noise measurements, from aggregated information from databases such as job exposure matrix (JEM), or qualitatively from self-reports of exposure. The issues with the individual level data include the limited number of measurements and the fluctuation of exposed noise level; this is more of a problem if the noise in the workplace is highly unstable and small number of measurements might not be able to capture the variation in noise level. Extrapolating current noise measurements “backward” for retrospective studies is also problematic if a temporal trend existed in exposed noise level. In addition, there may be jobs that existed historically, but no longer exist and have never been measured. By aggregating information in the JEM, these problems can be overcome to some extent. Sjostrom et al. (2013) estimated the validity of a JEM for occupational noise and found an 80% agreement in classifying occupational noise exposure between two groups of hygienists without systematic difference. But systematic difference was found when classifying peak level exposure.  12  The exposure duration is often derived from workers’ work history. This information may come from either the workers’ employment records or the self-reported work history. Employment records could provide a relatively objective work history but sometimes they were difficult to access. As for self-reported work history, it might not be as accurate as the employment records and is prone to recall bias and reporting bias (Brigham et al., 2008; Adegoke et al., 2004; Schuz et al., 2003). Using only the noise intensity to assess occupational exposure was seldom found in later studies on the noise – CVD association because the development of CVD is more of a chronic process and the exposure duration appeared to take this into consideration. A large number of recent studies used the exposure duration or cumulative exposure to assess occupational noise (McNamee et al., 2006; Davies et al., 2005; Sbihi et al., 2003).  Another metric used by many later studies was subjective measurement or the self-reported occupational noise, which was often derived from a questionnaire by asking ‘Have you ever worked in a noisy industry or noisy farm environment?’ (Gopinath et al., 2011) or ‘Thinking of all the jobs you have ever had, have you ever been exposed to loud noise at work for at least three months’ (Gan et al., 2011). The exposure duration was quantified from the question ‘how long a period did you work in this industry?’ (Gopinath et al., 2011). Though this metric seemed less accurate in quantifying exposure compared to the two metrics mentioned above, it did have an asset in incorporating some factors involved in the perception of noise, such as personal susceptibility. This was more likely to be the case when exposed noise intensity was moderate, in which case the indirect pathway was assumed to predominate and many situational and individual factors (e.g. communication tasks, requirement of concentration, noise sensibility, attitude towards the sound/noise) were involved. Schlaefer et al. (2009) studied the validity of self-reported occupational noise exposure by comparing it to the exposure estimated from a JEM. Their findings showed a high agreement between self-reported noise exposure and that derived from the JEM, especially at the noise level of 80 dBA.  Another important issue related to occupational noise exposure is the effect of hearing protectors. The influence of hearing protectors on exposure estimates is more likely to be a problem in recent years, as hearing conservation programs were introduced to protect worker’s hearing in 13  the workplace starting about 1980. Babisch (1998) discussed the impact of hearing protectors on the results of studies focusing on occupational noise. Hearing protectors are designed to provide a reduction of 15 to 30 dBA in the perceived noise exposure (Berger, 1993). If the subject wears hearing protector at work, the measured noise level by sound level meter or dosimeter is higher than the actual noise level perceived by the subject. Therefore, it is likely that a systematic misclassification resulting in over-estimates of exposure occurs with the presence of hearing protectors. Talbott et al. (1996) suggested a role of hearing protectors as an effect modifier on study results and Sbihi et al. (2010) studied the influence of the use of hearing protector devices (HPD) in the BC lumber mill workers. Sbihi et al. compared the HPD-adjusted and unadjusted noise metrics on their ability to predict NIHL. They found a 4-fold increase in the noise exposure and hearing loss slope, after adjusting for HPD use. However, adjusting for the effect of hearing protectors in studies remains difficult because the status of hearing protectors use for workers is rarely known. Moreover, as the perceived noise exposure is reduced by wearing hearing protectors, the stress response becomes much more complex. When the perceived noise level decreases due to hearing protectors, the indirect pathway may dominate the potential health outcomes rather than the direct pathway. In this case, it may be the annoyance resulting from the remaining noise that triggers the stress response. This pathway is much more complex compared to the direct pathway as many other environmental or personal parameters are involved. Using a subjective exposure metric might adjust for this effect. But further studies are still needed to provide improved methodology to deal with the impact from hearing protectors.  1.7  Research Objectives  As one of the most ubiquitous hazards in the workplace, noise has the potential to cause adverse health effects in humans. Diabetes and rheumatoid arthritis are two diseases that may lead to serious outcomes and their prevalence is increasing year by year, causing increasing costs in diagnosis and treatment. Previous studies of the association between noise and CVD provide a plausible biological model that can be generalized to the association between noise exposure and these other disease outcomes. Our overarching aim therefore was to examine the association between occupational noise exposure and the health outcomes of diabetes and RA. In addition, 14  the noise – CVD association was examined as a sensitivity analysis to test the face validity of our findings. Specifically, the three main objectives for this thesis research were:  (1) To build a job exposure matrix for noise using measurements from WorkSafeBC’s noise exposure dataset; (2) To develop noise exposure estimates for subjects in the National Population Health Survey based on the job exposure matrix; (3) To examine the association among the subjects in the National Population Health Survey between occupational noise exposure and risk of: a. Diabetes b. Rheumatoid arthritis c. Cardiovascular disease  15  Chapter 2: Methods  Three case-control studies were conducted nested within three sub-cohorts, representing the three health outcomes of interest. The three sub-cohorts were formed based on the original longitudinal cohort of the National Population Health Survey (NPHS). Diabetes, rheumatoid arthritis, and cardiovascular disease cases were determined from self-reports in the NPHS; each self-reported case of disease was matched to five controls who did not have the health outcome at the time the case was diagnosed. Exposure assessment was conducted by building a noise job exposure matrix, assigning the exposure level according to job titles of the subjects and cumulating exposure over self-reported time in job(s). Conditional logistic regression was used to estimate the odds ratios of developing the health outcomes of interest. Although diabetes and RA were our major concern in the study, the noise and CVD association was also assessed to test the face validity of our findings.  2.1  Background Information about the National Population Health Survey  The National Population Health Survey is a Canadian nationwide longitudinal survey established by Statistics Canada in 1994. It is conducted biennially to collect information on the health status of the Canadian population and relevant socio-demographic information, aiming to provide health-related data for both researchers and policy makers. When it was established, the NPHS included two components, a cross-sectional component and a longitudinal component. In the fourth cycle (2000/2001), the cross-sectional component was taken over by the Canadian Community Health Survey (CCHS) and the NPHS became strictly longitudinal. The longitudinal panel was composed of two components, the household component and the institution component. The institution component was terminated after the fifth cycle (2002/2003) because of large number of deaths in the original population. Therefore, the NPHS household component became the only part of NPHS that is still conducted every two years (until the final survey cycle scheduled for completion in 2014/2015; Statistics Canada, 2012). In this study, only the longitudinal household component was used. 16  2.1.1  Target Population  The target population of the NPHS household component was household residents of the ten Canadian provinces in 1994/1995, excluding those who lived on Indian reserves and Crown lands, residents of health institutions, full-time members of the Canadian Forces living on Canadian Forces bases, and residents of some remote areas in Ontario and Quebec. (Statistics Canada, 2012)  2.1.2  Sampling  In all provinces except Quebec, the NPHS used a stratified multi-stage sample design based on the Labour Force Survey. In Quebec, SantéQuébec for the 1992–1993 Enquête sociale et de santéwas used. The detailed sampling method is described in “NPHS, House Component, Cycles 1 to 9, Longitudinal Documentation” by Statistics Canada (2012). In the first cycle, 17,276 people respondents comprise the longitudinal panel. This sample size remains stable in the following cycles. Although the number of people decreased slightly from cycle to cycle due to non-response, the longitudinal sample is still representative of the 1994/1995 panel (Statistics Canada, 2012).  2.1.3  Data Collection  The data collection for the NPHS was conducted using a computer-assisted interview (CAI) system, which allowed the check of the logical flow associated with the questions being asked. This system was designed to control the validity and consistency of the data collected not only within cycle but also from previous cycles. Therefore, the quality of the information collected could be monitored and errors could be located and modified during the interview. In the first cycle, 75% of the subjects were interviewed in person and the rest by telephone. In the following cycles, telephone became the major interview method, covering 95% of the subjects. The interviewers were hired and trained by Statistics Canada to carry out surveys, especially focusing 17  on the content of NPHS. Many strategies were used to minimize non-response, such as assigning reasonable sample sizes to each interviewer to avoid overburden. People who refused to participate were followed up by a senior interviewer, project supervisor or another interviewer to encourage them to return to the survey. To maximize the response rate, subjects who refused to respond in some cycles were still followed up in the subsequent cycles. In each cycle, based on the outcome of the interview, the status of the subjects can be: (1) completed: subjects provided complete response to the interview; (2) deceased: subjects died during the NPHS follow-up; (3) institutionalized: subjects moved into health care institutions; (4) partial response: subjects only responded partly to the NPHS questionnaire; and (5) non-response: none of the above status was assigned. In Cycle 1, 20,095 subjects were sampled to form the longitudinal panel, and 86.0% of them participated. In the following cycles, the response rates (calculated based on the 17,276 participants in the first cycle) were: 92.8% (cycle 2: 1996/1997), 88.3% (cycle 3: 1998/1999), 84.9% (cycle 4: 2000/2001), 80.8% (cycle 5: 2002/2003), 77.6% (cycle 6: 2004/2005), 77.0% (cycle 7: 2006/2007) and 70.7% (cycle 8: 2008/2009). The subjects who refused to respond were the largest source of non-response in the NPHS while failure to trace a subject is another major source of non-response (Statistics Canada, 2012).  2.2  Background Information about the Noise Exposure Data from WorkSafeBC  Since the 1970’s, WorkSafeBC, the provincial workers’ compensation agency, has collected noise measurements from various industries and occupations with the last samples dating from 2004. The occupations and industries represented those with more chance of being exposed to high level of occupational noise. These samples were stored as a dataset that is available to researchers through Population Data BC (a data repository for research purposes). The variables and information contained in the dataset are listed below:    Sampling Time: the year, month, and date when the sample was collected;    Classification Unit (CU) Codes: internal codes used by WorkSafeBC to indicate the industry where the sample was collected;    Job Description: the task/situation/environment where the sample was taken; 18    Sampling Device Type: indicator for whether the measurement was a personal or environmental sample;    2.3  Noise Level: noise exposure level (dBA).  Study Design  Three case-control studies were conducted, nested within three sub-cohorts from the longitudinal cohort in the National Population Health Survey defined by diabetes, rheumatoid arthritis, and cardiovascular disease. The three sub-cohorts were defined according to the inclusion criteria below:  1) Subjects must be in the age range from 18 to 55 in the first cycle (1994/1995). This criterion was used to guarantee that for each subject there were sufficient years of workplace-exposure information; 2) In the sub-cohort for a health outcome of interest, subjects must not have reported this health outcome in any of the first four cycles. For each question related to a health outcome, the answer could be yes (had the outcome) / no (did not have the outcome) or missing value (no valid information provided; see detail in Section 2.4). To minimize potential misclassification by health outcome, only those who responded “no” to the health outcome of interest for each of the first four cycles were retained in the given sub-cohort.  Follow-up started immediately after the fourth cycle (December 31 2001) and ended at the eighth cycle (December 31 2009). The starting point of follow-up was selected to balance time for estimation of the exposure duration (4 cycles prior to follow-up) with time for diseases of interest to develop (the follow-up period). The follow-up for a subject ended at the earliest of the following:  1) The subject died during the follow-up period. The NPHS provided the death year, month, and day for subjects who died; 2) In the sub-cohort for a certain health outcome of interest, a subject became a case; he or she 19  reported being diagnosed with the health outcome. The NPHS provided only the diagnosis year and month for diabetes and rheumatoid arthritis; therefore the diagnosis day was assumed to be the 15th of the diagnosis month. The NPHS did not provide any of the diagnosis year, month, and day for CVD, and so the diagnosis date was set to the midpoint of the two year period covered by the cycle in which the subjects reported cardiovascular disease (e.g. a subject first reported cardiovascular disease in the 6th cycle. The diagnosis time was set to December 31 2004); 3) The subject dropped out from the NPHS/was lost to follow-up.  2.4  Case and Control Definition  Cases were determined based on the self-reported disease status in the NPHS. The self-reported status of diabetes (Dinca-Panaitescu et al., 2012; Neutel et al., 2010; Ross et al., 2010; Millar and Young, 2003), rheumatoid arthritis (Adam et al., 2005), and cardiovascular disease (Chiu et al., 2010; Lee et al., 2009; Patten et al., 2008; Wong and Wong, 2002) in the NPHS was used in previous studies for various research objectives.  For diabetes and CVD, a case was defined as a subject with a positive response to the following questions in the NPHS: 1) “do you have diabetes?” or 2) “do you have heart disease?”  For rheumatoid arthritis, the disease status was determined based on positive response to the question: “do you have arthritis or rheumatism excluding fibromyalgia?” and identification as rheumatoid arthritis from the question “what kind of arthritis do you have (rheumatoid arthritis / osteoarthritis / other)?”  In the cohort for a certain health outcome of interest, the cases were defined at the first time they reported this health outcome during the follow-up period. For each case, five controls were randomly selected from those subjects who did not have the health outcome of interest at that time point. The cases and controls were matched on age (±5 years) and sex.  20  There were 17,276 subjects in the first cycle (1994/1995) of the NPHS. According to the inclusion criteria, subjects who were less than 18 years old or older than 55 years old in the first cycle (n = 7,411) were excluded. Among those subjects aged 18 to 55 years old, 341 subjects reported being diagnosed with diabetes in the first four cycles (1994 to 2001) and 2,308 subjects were not able to provide reliable information on diabetes status for a total of 2,649 subjects excluded from the cohort. Of the eligible cohort (7,216 subjects) for diabetes, 202 new cases were reported during the follow-up period and 1,010 controls were selected (see Figure 2. 1). For rheumatoid arthritis, in order to avoid interference from other types of arthritis, subjects who reported any type of arthritis (rheumatoid arthritis / osteoarthritis / others) in the first four cycles (n = 1,879) and those who were not able to provide reliable information on arthritis status (n = 2,224) were excluded. After applying the age and disease status exclusion criteria, 5,762 subjects from the NPHS cohort were eligible for the case-control analysis of RA, among whom 91 reported rheumatoid arthritis during the follow-up period. To avoid influence from other type of arthritis, the 451 controls were selected only from those who did not report any type of arthritis in the cohort (see Figure 2. 2). For cardiovascular disease, among subjects aged 18 to 55 years old, 369 subjects reported CVD in the first four cycles while 2,303 people did not have reliable CVD status information, which left 7,193 subjects in the study cohort for CVD. Of these, 221 new cases of CVD were reported during the follow-up and 1,105 controls were selected (see Figure 2. 3).  For each sub-cohort there was a sizeable fraction of subjects (diabetes: 2,308, RA: 2,224, CVD: 2,303) whose disease status was unknown; for them the code in the disease status variable was “not stated”, which is a code used by Statistics Canada to show inconsistent or not provided data.  21  Original NPHS Sample (n = 17,276)  Subjects < 18 or > 55 years (n = 7,411) Excluded  Subjects between 18 to 55 years (n = 9,865)  Had diabetes (n = 341) Excluded  Diabetes status unknown (n = 2,308) Excluded  Diabetes cases (n = 202)  Figure 2. 1  Study cohort (n = 7,216)  Controls (n = 1,010)  Flowchart of Selection Cases and Controls for Diabetes  22  Original NPHS Sample (n = 17,276)  Subjects < 18 or > 55 years (n = 7,411) Excluded  Subjects between 18 to 55 years (n = 9,865)  Had arthritis (including RA) (n = 1,879) Excluded  Arthritis status unknown (n = 2,224) Excluded  Study cohort (n = 5,762)  RA cases (n = 91)  Controls* (n = 455)  * The 455 controls only came from those who did not have any type of arthritis in the study cohort Figure 2. 2  Flowchart of Selection Cases and Controls for Rheumatoid Arthritis  23  Original NPHS Sample (n = 17,276)  Subjects < 18 or > 55 years (n = 7,411) Excluded  Subjects between 18 to 55 years (n = 9,865)  Had CVD (n = 369) Excluded  CVD status unknown (n = 2,303) Excluded  Study cohort (n = 7,193)  CVD cases (n = 221) Figure 2. 3  2.5  Controls (n = 1,105)  Flowchart of Selection Cases and Controls for Cardiovascular Disease  Exposure Assessment  Occupational noise exposure for cases and controls was evaluated using cumulative exposure, the product of exposure level and employment duration for each job reported. The exposure level came from the job exposure matrix aggregated from the WorkSafeBC’s noise exposure dataset and the employment duration came from the Labour Force section of the NPHS. The exposure level and duration were linked to the matrix based on the North American Industry Classification System (NAICS 2007) and National Occupational Classification for Statistics (NOC-S 2006) codes.  24  2.5.1  Building the Job Exposure Matrix  Job- and industry-specific noise exposure levels were obtained from the WorkSafeBC dataset. In order to link these exposure measurements to workers and their work history information in the NPHS, the CU codes (an internal industry code used by WorkSafeBC) in the WorkSafeBC data were transcribed into the standardized 4-digit NAICS (2007) codes. Using the job description and industry information provided for each exposure measurement, a corresponding 4-digit NOC-S (2006) code was assigned to each exposure measurement. To reduce error in the code assigned, an indicator variable representing the degree of confidence of assigning the right code was generated by the researcher. This variable had three values, 2 “very confident in the assigned code”, 1 “some uncertainty about whether the appropriate code was assigned”, 0 “do not know which code to assign and no code was assigned”. Only observations having a confidence degree of 2 were kept in the dataset (about 90% of the original dataset).  After the first round of assigning the 4-digit NOC-S code, 500 observations (1.1% of all observations) were randomly selected from the dataset. The NOC-S assigning process was independently repeated for these 500 observations by a second researcher who was familiar with the NOC-S coding system to improve the accuracy of the assigned codes. When there was a dispute with the codes assigned by the two researchers, if a more appropriate code could be decided after discussion, the more appropriate code were selected and corrections were made in the whole dataset. If it was difficult to decide which code was more appropriate, no actions were made.  For each observation in the noise database, a 2-digit NAICS code and a 2-digit NOC-S code were also generated based on the first two digits of the 4-digit NAICS code and the 4-digit NOC-S code respectively. The mean noise levels of each 2-digit NAICS code, 2-digit NOC-S code, 4-digit NAICS code, 4-digit NOC-S code were calculated. If the number of observations for a combination of 4-digit NAICS code and 4-digit NOC-S code was 10 or over, the mean noise exposure level for this combination was also calculated. These calculated means were aggregated into a job-exposure matrix. The JEM representing exposure levels by NOC-S and 25  NAICS classifications was used to assign noise exposure levels to workers based their self-reported work histories in the NPHS.  Besides these calculated mean levels, there were occupations that had very little probability of being exposed to occupational noise. These occupations were identified in the NOC-S coding system and assigned a ‘0’ level2, and these were also added into the JEM. In the final JEM, there were three variables, the NOC-S codes, the NAICS codes and the noise levels. A sample of the JEM is shown in Table 2. 1 and the summary information about the JEM is shown in Table 2. 2. There were 1079 lines in the JEM and approximately 33% of the noise levels were at the most detailed level of the combination of both 4-digit NOC-S and 4-digit NAICS codes.  2.5.2  Estimating Exposure Duration  The NPHS surveys were conducted biennially so that each cycle represented a two year time period. Although the interviews for each cycle could have been conducted at any time over the two year period, it was assumed that all the interviews were conducted at the midpoint of each two year period (e.g. the interviews for the first cycle were conducted at December 31st 1994, etc.). The information necessary for estimating the exposure duration was provided in the Labour Force section of the NPHS. The Labour Force section asked the subjects about information related to their occupations one year prior to the interview. Therefore no data was provided for job status in the second year of cycle and this had to be inferred. To estimate the exposure duration, three variables were derived from the NPHS Labour Force section:  HOUR: the total number of hours the subject worked per week in the past year, estimated from the following question “About how many hours a week do you usually work at your job? If you  2  In reality the exposed sound levels for these “unexposed” occupations might range from 40 to 60 dBA and could not possibly  be 0. However, when calculating the cumulative noise exposure, because of the log scale the difference between 0 and 40 to 60 dBA will not have significant influence on the categorization of noise exposure. So the ‘0’ level was used for convenience, and as a code for “assumed no exposure”.  26  usually work extra hours, paid or unpaid, please include these hours”;  Table 2. 1  Sample of the Noise Job Exposure Matrix, showing (i) entries of 4-digit NAICS and 4-digit  NOC-S (only where n>10); (ii) 4-digit NOC-S only; (iii) 4-digit NAICS only; (iv) 4-digit NOC-S – NO EXPOSURE; (v) 2-digit NOC-S only; and (vi) 2-digit NAICS only NOC-S  NAICS  Noise Level (dBA)  H523  3211  87.2  J141  3211  95.4  H523  3212  87.6  J193  3212  95.0  …  …  …  H523  N/A  88.0  J141  N/A  95.3  J193  N/A  94.9  …  …  …  N/A  3211  92.7  N/A  3212  90.0  …  …  …  A011  N/A  0.0  B011  N/A  0.0  …  …  …  H5  N/A  88.2  J1  N/A  93.4  …  …  …  N/A  32  91.7  N/A  33  91.9  …  …  …  Table 2. 2  Summary of the Composition of the Job Exposure Matrix. Number of Noise Levels in Each Category Based  on Different Detail Level of NOC-S and/or NAICS Codes (the most detailed has both 4-digit NOC-S and 4-digit NAICS)  Frequency  Percent (%)  Both 4-digit NOC-S and NAICS  362  33.55  4-digit NOC-S, no NAICS  264  24.47  4-digit NAICS, no NOC-S  132  12.23  “0” exposure 4-digit NOC-S  255  23.63  2-digit NOC-S, no NAICS  43  3.99  2-digit NAICS, no NOC-S  23  2.13  1,079  100  Total  WEEK: the total number of weeks the subject worked in the past year, estimated from the question “During the past 52 weeks, how many weeks did you do any work at a job or a business? 27  (Include paid vacation leave, paid maternity leave, and paid sick leave”; TWOYEAR: a variable indicating whether, within a two year period covered by each cycle, the work status one year prior to the (assumed mid-point) interview was the same as that one year after the interview. The values of the TWOYEAR variable could be:    “0” (derived variable indicating did not work in the past 12 months and did not work at the time of the interview) the subject did not have significant occupational noise exposure in the two year period;    “1” (derived variable indicating that subject worked in the past 12 months but did not work at the time of the interview or the opposite) the subject had one year of occupational noise exposure;    “2” (derived variable indicating that subject worked both in the past 12 months and at the time of the interview) the subject had occupational noise exposure status over the two year period;    “missing” (indicating that the subjects did not provide reliable information on this variable, details on dealing with missing value stated in Section 2.8.2).  The exposure duration within each two year period was calculated using the following formula:  EXPOSURE DURATION (years) = (HOUR * WEEK * TWOYEAR) / 2000  (1)  where 2000 represents the working hours for a worker with a normal shift in a year (50 weeks/year, 40 hours/week). The unit of the EXPOSURE DURATION is year.  2.5.3  Assigning Noise Exposure Level  In each cycle, subjects provided a description of their job and industry in the one year period prior to the survey and based on these descriptions, a 4-digit NAICS code and a 4-digit NOC-S code were assigned by Statistics Canada. The noise exposure level in the JEM from WorkSafeBC was linked to each subject in each cycle based on the NAICS and NOC-S codes. Each subject 28  was assigned a noise exposure level (variable name: MEAN, unit: dBA) in each cycle, using the following assignment procedure (also Figure 2. 4):  1) If the combination of the 4-digit NOC-S code and 4-digit NAICS code in the NPHS could be found in the job exposure matrix, the corresponding noise exposure level was assigned. Exposure periods which were assigned a noise level in this step got an “assignment confidence rank’ of one to indicate these had the highest validity; 2) If the combination could not be found in the job exposure matrix, the noise exposure level was assigned only according to the 4-digit NOC-S code. An assignment confidence rank of two was given to exposure periods assigned a noise level in this step; 3) If the job exposure matrix did not have the exposure level for the 4-digit NOC-S code in NPHS, then the noise exposure level was assigned based on the 4-digit NAICS code. Exposure periods which got a noise level in this step were given an assignment confidence rank of three; 4) The noise levels assigned in the first three steps all came from the available information from the JEM. Priority was given to them over the zero exposure occupations determined by expert judgment. After the first three steps, there were still some subjects who did not have a corresponding noise exposure level. According to their NOC-S codes, a portion of these subjects were assigned a zero exposure based on expert judgment because their job in that cycle had very little probability of being exposed to occupational noise. Subjects who were assigned a noise level in this step got a rank of four; 5) Among the subjects left from the previous four steps, the noise exposure level was assigned at the 2-digit NOC-S level. A rank of five was given to subjects who got a noise level in this step; 6) If the job exposure matrix did not have the exposure level for the 2-digit NOC-S code in subjects in the step 5, the noise exposure level was assigned according to the 2-digit NAICS code. Subjects who were assigned a noise level in this step got a rank of six.  29  Figure 2. 4  2.5.4  Noise Assignment Procedures – note (4) refers to 4-digit NOC-S that are unexposed  Calculating the Cumulative Noise Exposure  The entire exposure period for each subject starts from the first cycle (1994/1995) to the end of follow-up. Within each two year period represented by each cycle, both the noise exposure level (MEAN) and the exposure duration (from HOUR, WEEK, TWOYEAR) were known. The total cumulative noise exposure (CNE) was estimated by adding up the cumulative noise exposure from all the cycles within the entire exposure period. The detailed formula is shown below:  (2)  where i denotes the ordinal number of the cycle. The unit of CNE is dBA-years.  30  2.5.5  Categorizing Exposure Groups  Subjects were divided into three exposure groups as low, medium, high exposure using cut points of 85 dBA-year and 95 dBA-year. The cut points were selected based on the distribution of the calculated cumulative noise exposure and to ensure that there were enough subjects distributed in each exposure category.  2.6  Covariates  Factors that could be potential confounders or effect modifiers of the relationship between noise exposure and each disease outcome were identified according to the existing literature and then included in the analysis, as outlined below. The subjects’ status for all these variables (except high blood pressure) came from the last cycle during the follow-up period to use the most up-to-date data, and were treated as constant in the analysis.  2.6.1  Income and Education  Lower social-economic status (e.g. education, income, occupation) is associated with an increased risk of diabetes (Maty et al., 2005; Mokdad et al., 2000), RA (Kwon et al., 2012), and CVD (Johnson-Lawrence et al., 2013; Beebe-Dimmer et al., 2004). Income and education are two important components of socio-economic status. Previous studies indicated that they were correlated with occupational noise exposure (Gan et al., 2011). Therefore, they were treated as confounders in previous studies focusing on noise and disease association and their effect was adjusted in the regression model (Sorensen et al., 2012, Gan et al., 2011, Selander et al., 2009, Wang et al., 2000).  The income information was obtained from the household income variable in the NPHS. In the original dataset the total household income was collapsed into deciles. In this study, the total household income was re-categorized into three categories at the cut points of $40,000 and 31  $80,000. The cut points were selected to ensure there were enough subjects in each category to build the regression model.  The NPHS provided several different ways to categorize the subject’s education status but in the current study, the four-level categorization was used: less than secondary school graduation, secondary school graduation, some post-secondary, and post-secondary graduation.  2.6.2  Obesity  Obesity is an important risk factor for both diabetes (Maty et al., 2005; Mokdad et al., 2000) and CVD (WHO, 2013) and its effect was adjusted in the regression model of previous studies focusing on noise – diabetes and noise – CVD association (Sorensen et al., 2012; Gan et al., 2011; Gopinath et al., 2011; Selander et al., 2009; McNamee et al., 2006). For RA, the effect of obesity as a confounder was explored in our study as we speculated that being obese might increase the burden and damage of joints. Also having obesity status might prevent the subjects from participating in certain occupations, which would be associated with their noise exposure. Obesity status was derived from the Body Mass Index (BMI) of each subject using the international standard recommended by Health Canada and the World Health Organization (Health Canada, 2005). The BMI was calculated according to the height and weight of the subject using the following formula:  BMI = Weight (kg) / (Height (m))2  (3)  The international standard BMI classification system for adults aged 18 or over had six categories: “Underweight” BMI <= 18.5; “Normal Weight” BMI 18.5 – 24.9; “Overweight” BMI 25.0 – 29.9; “Obese Class I” BMI 30.0 – 34.9; “Obese Class II” BMI 35.0 – 39.9; “Obese Class III” BMI ≥ 40.0.  32  To avoid small number of observations in some categories (Statistics Canada has a policy on releasing results with small number in certain categories), the “Underweight and Normal Weight” were merged into one category for all three health outcomes of interest. For the health outcomes of cardiovascular disease and rheumatoid arthritis, the three different classes of obese were merged into one category due to small numbers.  2.6.3  Physical Activity  Physical inactivity is another risk factor for both diabetes (Maty et al., 2005; Mokdad et al., 2000) and CVD (WHO, 2013) so its effect was adjusted in the regression models of previous studies (Sorensen et al., 2012; Gan et al., 2011; Gopinath et al., 2011; Selander et al., 2009). Though there was no previous study on the confounder effect of physical activity on the noise – RA relationship, we studied it in the current study as being highly physically active might damage the joints and physical activity may be to some extent associated with the subjects’ occupation. The subjects’ physical activity status was determined based on the physical activity index, which was derived from Energy Expenditure (EE). EE was calculated according to the frequency and time per session of physical activity as well as its energy cost, the metabolic energy cost expressed as a multiple of the resting metabolic rate. The EE value for all activities in a day was calculated as follows:  EE (kcal/kg/day) = Sum of ((Ni × Di × MET value) / 365)  (4)  Where Ni was the number of times a respondent engaged in an activityi over a 12 month period, Di was the average duration in hours of the activityi, and MET was the energy cost of the activity expressed as kilocalories expended per kilogram of body weight per hour of activity (kcal/kg per hour) / 365 (to convert yearly data into daily data). Based on the calculated EE, the subjects’ physical activity was divided into three categories: “Active” EE ≥ 3.0; “Moderate” 1.5 ≤ EE < 3.0; “Inactive” EE < 1.5 (Statistics Canada, 2012).  33  2.6.4  Smoking  Smoking is regarded as the risk factor for both RA (Oliver and Silman, 2006) and CVD (WHO, 2013) and was treated as a confounder in previous studies (Li et al., 2012; Gan et al., 2011; Gopinath et al., 2011; Selander et al., 2009; McNamee et al., 2006). Its confounding effect was also indicated in studies focusing on the direct association between noise and diabetes (Sorensen et al., 2012). In NPHS, six types of smokers were defined: daily smoker, occasional smoker but former daily smoker, always an occasional smoker, former daily smoker, former occasional smoker, never smoked. The six types were re-categorized for the analysis. The “daily smoker”, “occasional smoker but former daily smoker”, and “always occasional smoker” were merged into one category “current smoker”. The “former daily smoker” and “former occasional smoker” were merged into one “former smoker” while the “never smoked” remained the same. Therefore in this study, three types of smoker were defined: “Current Smoker”, “Former Smoker”, and “Never Smoked”.  2.6.5  Drinking  It was suggested in some previous studies that drinking might confound the noise – diabetes association and noise – CVD association (Sorensen et al., 2012; Gan et al., 2011). In our study, four categories of drinker were used based on the types of drinkers defined in the NPHS: “Regular Drinker3”, “Occasional Drinker4”, “Former Drinker”, and “Never Drank”. In this study, the “Occasional Drinker” and “Never Drank” were merged as one due to limited number of subjects in these categories and similar risk.  3  The frequency of drinking alcohol beverage for a regular drinker ranged from once a month to 4 to 6 times a month.  4  Occasional drinker were people who drank less than once a month.  34  2.6.6  High Blood Pressure  High blood pressure is an important risk factor for both diabetes and CVD (PHAC, 2012; PHAC, 2009) and its effect was considered in the models in the current study. Two categories were defined as yes or no for: “do you have high blood pressure?”.Considering the fact that hypertension might be an intermediate factor between noise exposure and CVD, the subjects’ hypertension status was obtained from the only first cycle, i.e. we assumed that high blood pressure after the first cycle might be due to noise exposure. This did not necessarily avoid this problem, however, as noise exposure prior to the first cycle was unknown.  2.6.7  Covariates for the Three Health Outcomes of Interest  The inclusion of the covariates for income, education, physical activity, obesity, smoking status, drinking status and high blood pressure varied depending upon the health outcome under investigation, and the available evidence of an association with the health outcomes, and is summarized in Table 2. 3.  2.7  Statistical Analysis  The data analyses were conducted using STATA, version 10.0.  2.7.1  Descriptive Statistics  The descriptive statistics on noise exposure and covariate status for cases and controls were calculated using cross-tabulations. Since the variables for exposure groups and covariates were all categorical variables, the Pearson chi-square test and the Fisher’s exact test were used to compare whether the distributions of variables differed between cases and controls. Differences were considered significant when p-value < 0.05.  35  2.7.2  Bivariate Analysis  Bivariate analysis was conducted: (1) to provide a general view of the association between exposure and covariates (using Pearson chi-square test and the Fisher’s exact test; (2) to see whether exposure variable or covariates were associated with the health outcome (using simple conditional logistic regression); and (3) to check the collinearity between two covariates (using Spearman’s correlation coefficient).  2.7.3  Multiple Logistic Regression Analysis  To obtain adjusted odds ratios for different exposure groups, three multiple conditional logistic regression models were built, one each for the three health outcomes of interest. The multiple regression model was built with the exposure variable as the primary predictor and other covariates to control for potential confounding. The likelihood ratio test was used to test the hypothesis that all the variable coefficients in the model were equal to zero while the Wald test was used to test the significance of each single predictor in the model.  2.7.4  Selection of Predictors  The selection of predictors for the multiple conditional logistic regression model was conducted using a manual forward stepwise method. Specifically, the process started with the noise exposure group variable as the only predictor. The socioeconomic variables, education and family income, were forced to enter the model because they were the a priori confounders in epidemiological studies. Other covariates were then added into the model one at a time to see their influence on the ORs of different exposure groups. If adding a predictor resulted in significant change in ORs (10% or more), the predictor was regarded as a confounder and remained in the model. If adding a predictor had a minor effect on the ORs but this predictor was an a priori confounder or risk factor of the health outcome of interest in previous studies, it was also retained in the model. After this selection process, adjusted ORs were reported according to 36  this final model.  Table 2. 3 Disease Outcome  List of Covariates for Different Health Outcomes  Variable Obesity  Diabetes  Variable Values Underweight or Normal weight / Overweight / Obese I / Obese II / Obese III  Physical Activity Index  Inactive/ Moderate / Active  Type of Smoker  Current Smoker / Former Smoker / Never Smoked  Type of Drinker High Blood Pressure  Regular drinker / Former drinker / Occasional drinker or Never drank Yes / No Less than Secondary School Graduation / Secondary  Education  School Graduation / Some Post-secondary / Post-secondary Graduation  Rheumatoid Arthritis  Total Household Income  < $40,000 / $40,000 – $79,999 / >= $80,000  Obesity  Underweight or Normal weight / Overweight / Obese  Physical Activity Index  Inactive/ Moderate / Active  Type of Smoker  Current Smoker / Former Smoker / Never Smoked  Education  Cardiovascular Disease  Secondary School Graduation or Less / Some Post-secondary / Post-secondary Graduation  Total Household Income  < $40,000 / $40,000 – $79,999 / >= $80,000  Obesity  Underweight or Normal weight / Overweight / Obese  Physical Activity Index  Inactive/ Moderate / Active  Type of Smoker  Current Smoker / Former Smoker / Never Smoked  Type of Drinker High Blood Pressure  Regular drinker / Former drinker / Occasional drinker or Never drank Yes / No Less than Secondary School Graduation / Secondary  Education  School Graduation / Some Post-secondary / Post-secondary Graduation  Total Household Income  2.8  < $40,000 / $40,000 – $79,999 / >= $80,000  Dealing with Missing Values  Missing values were generated from the NPHS data and were categorized as follow:  Not Applicable: Some questions are only asked of people who meet certain requirements or people must respond ‘yes’ to certain questions to be asked subsequent ‘nested’ questions. For example, people who reported having a job were able to go to the questions regarding the job 37  description; people must respond yes to the question asking whether they have arthritis to be then asked further questions like what kind of arthritis they had. Otherwise the variable is coded as “not applicable”;  Don’t Know: When the subjects don’t know the answer to the question, the relevant variable is coded as “don’t know”;  Refusal: When the subjects refuse to answer the question, the relevant variable is coded as “refusal”;  Not Stated: In order to improve the quality of the collected data, consistency checks are conducted during and after the data collection process by Statistics Canada. If there is inconsistency in responses across question, corrective action is implemented by the coders of the survey. If a correction can not resolve the discrepancy, the questions involved in the inconsistency are coded as “not stated”.  All the missing values in the NPHS fall into the four categories mentioned above. A subject’s response falling into any of the four categories above was treated as a missing value. The detailed strategies to fill in these missing values are described in the following sections.  2.8.1  Missing Health Outcome Measurement  To avoid health outcome misclassification, subjects with missing values in the variables determining the health outcome were dropped.  2.8.2  Missing Exposure Assessment  Missing exposure data were filled through the procedures shown in Table 2. 4. Different procedures were applied, depending on the availability of information on the four variables necessary for estimating the cumulative noise exposure (MEAN, HOUR, WEEK, and 38  TWOYEAR). For example, for the seven situations belonging to category I, gaps were filled following the four steps outlined for that category in the “procedure” column. For situations in category II, CNE was calculated depending on the value of the TWOYEAR variable and for situations in category III, the population mean was assigned.  2.8.3  Missing Covariates  In the NPHS, the questions providing information on the selected covariates were asked in each cycle. Since people might not know the answer, refuse to answer, or even miss the entire cycle, in some cycles a variable might have missing values. Though the status of the covariates could change over time, it was treated as static in the study and the information on a certain covariate was obtained from the relevant variable in the cycle when the follow-up was stopped to guarantee that the information proximal to the diagnosis was used. If the variable for a certain covariate in the latest cycle had a missing value, the relevant variable in the previous cycle was checked to see if it had non-missing information on the covariate to fill in the blank. If not, then the variable in the previous cycle was referenced until the missing value was filled. If the information from the variables in all eight cycles still could not fill in the missing value, since all covariates were categorical, the category with the largest number of observations was assigned to the observations with missing value.  39  Table 2. 4  Procedures of Dealing with Missing Values in Exposure Assessment  TWOYEARic  WEEKi  HOURi  MEANi  Category  Variable  Procedures  1. a  A  A  M  b  A  If NOC-Si-1 = NOC-Si = NOC-Si+1, then:  (1) If TWOYEARi = 0  CNEi = 0 (2) If TWOYEARi = 1  CNEi = CNEi-1 / 2 (3) If TWOYEARi = 2  CNEi = CNEi-1 / 2 + CNEi+1 / 2  A  M  A  A  (4) If TWOYEARi = M  CNEi = CNEi-1 / 2 + CNEi+1 / 2 2.  A  M  M  A  If NOC-Si-1 = NOC-Si & NOC-Si ≠ NOC-Si+1, then:  (1) If TWOYEARi = 0  CNEi = 0 (2) If TWOYEARi = 1  CNEi = CNEi-1 / 2 (3) If TWOYEARi = 2  CNEi = CNEi-1  A  I  A  A  M  (4) If TWOYEARi = M  CNEi = CNEi-1 3.  If NOC-Si-1 ≠ NOC-Si & NOC-Si = NOC-Si+1, then:  (1) If TWOYEARi = 0  CNEi = 0 A  A  M  M  (2) If TWOYEARi = 1  CNEi = CNEi+1 / 2 (3) If TWOYEARi = 2  CNEi = CNEi+1 (4) If TWOYEARi = M  CNEi = CNEi+1  A  M  A  M  4.  If NOC-Si-1 ≠ NOC-Si & NOC-Si ≠ NOC-Si+1, then:  (1) If TWOYEARi = 0  CNEi = 0 (2) If TWOYEARi = 1  CNEi = mean CNE of population A  M  M  M  (3) If TWOYEARi = 2  CNEi = mean CNE of population (4) If TWOYEARi = M  CNEi = mean CNE of population  II  III  M  A  A  A  M  M  A  A  M  A  M  A  M  M  M  A  M  A  M  M  M  M  A  M  M  A  A  M  M  M  M  M  (1) If TWOYEARi = 0  CNEi = 0 (2) If TWOYEARi = 1  CNEi = mean CNE of population (3) If TWOYEARi = 2  CNEi = mean CNE of population  CNEi = mean CNE of population  a.  ‘A’ means the information for that variable is available;  b.  ‘M’ means the information for that variable is missing;  c.  i denotes the cycle number.  40  Chapter 3: Results  3.1  Job Exposure Matrix  Before data cleaning, there were 49,776 observations in the WorkSafeBC noise exposure dataset collected between 1970 and 2004. The recorded minimum and maximum noise levels were 50 dBA and 150 dBA respectively. The mean noise exposure level of the whole dataset was 91 dBA and the standard deviation was 9 dBA. There were 276 unique CU code categories and 27,867 unique job descriptions.  Based on the degree of confidence on the assigned NOC-S code, only observations that had a confidence degree of 2 were retained in the dataset (45,046 observations, 90% of the original dataset). The minimum (50 dBA) and maximum noise level (150 dBA) remained the same. The mean value and standard deviation of the noise level were 91.4 dBA and 8.8 dBA respectively. Because of the logarithmic characteristic of the decibel unit, the collected noise levels were expected to follow a normal distribution, and this was supported by the histogram and Q-Q plot of the exposure levels (Figure 3. 1 and Figure 3. 2). The transformation procedure of the CU codes into 4-digit NAICS codes gave 132 different industrial categories. In terms of occupation,  Leq 100  .03 .02  50  .01 0  Density  .04  .05  150  264 4-digit NOC-S codes were assigned according to the job descriptions in the database.  50  Figure 3. 1  100 Leq  150  Histogram of Noise level (N=45,046)  60  Figure 3. 2  80  100 Inverse Normal  120  140  Q-Q Plot of Noise Level (N=45,046)  41  3.1.1  Noise Exposure by Occupations  The noise exposure levels by occupation at the one-digit NOC-S level5 are shown in Table 3. 1. Approximately 97% of all samples were collected in the following occupational categories: occupations unique to processing, manufacturing and utilities; trades, transport and equipment operators and related occupations; and occupations unique to primary industries. The highest mean noise exposure levels were also found in these three occupation groups: 92.6 dBA, 89.4 dBA, and 91.2 dBA respectively. The standard deviations of the noise levels were relatively large, especially in natural and applied sciences and related occupations (11.5 dBA).  Table 3. 1  Noise Exposure Levels by Occupations (1-digit NOC-S Level)  Occupation (NOC-S 1 digit)  Percent  Mean  SD  (%)  (dBA)  (dBA)  30,519  67.75  92.6  8.6  10,883  24.16  89.4  7.8  2,341  5.20  91.2  8.2  Sales and Service Occupations  544  1.21  81.9  8.7  Business, Finance and Administrative Occupations  218  0.48  75.8  10.7  Natural and Applied Sciences and Related Occupations  202  0.45  85.3  11.5  Management Occupations  148  0.33  74.6  8.5  Health Occupations  108  0.24  75.2  10.4  Occupations in Social Science, Education, Government  70  0.16  85.2  5.3  13  0.03  84.2  5.9  Occupations Unique to Processing, Manufacturing and  Frequency  Utilities Trades, Transport and Equipment Operators and Related Occupations Occupations Unique to Primary Industry  Service and Religion Occupations in Art, Culture, Recreation and Sport  5  The noise levels are shown at one digit level for ease of presentation of the information. Otherwise the table will be too huge to  show.  42  3.1.2  Noise Exposure by Industries  The noise exposure levels by industry are shown in Table 3. 2 at 2-digit NAICS level. Approximately 94% of all samples were collected in the following three industries: manufacturing industry; agriculture, forestry, fishing and hunting industry; and construction industry. The mean noise exposure levels found in most industries were in the range between 80 dBA and 90 dBA, however, the highest mean noise exposure level was found in education services (92.2 dBA), followed by manufacturing (91.7 dBA), agriculture, forestry, fishing and hunting (91.1 dBA), construction (91.1 dBA), mining, quarrying, oil and gas extraction (91.1 dBA). The standard deviations of the noise levels were relatively large, especially in administrative & support, waste management, etc. (10.1 dBA).  Table 3. 2  Noise Exposure Levels by Industries (2-digit NAICS Level)  Industry (NAICS 2 digits)  Frequency  Percent  Mean  SD  (%)  (dBA)  (dBA)  Manufacturing  37,608  83.49  91.7  8.8  Agriculture, Forestry, Fishing and Hunting  3,296  7.32  91.1  8.0  Construction  1,268  2.81  91.1  7.4  Public Administration  856  1.90  89.0  9.6  Transportation and Warehousing  569  1.26  87.8  9.7  Education Services  379  0.84  92.2  9.8  Other Services (except Public Administration)  294  0.65  87.7  6.9  Retail Trade  195  0.43  83.1  7.0  Information and Cultural Industries  135  0.30  88.3  8.5  Accommodation and Food Services  120  0.27  83.8  6.2  Health Care and Social Assistance  119  0.26  81.3  7.1  Mining, Quarrying, Oil and Gas Extraction  89  0.20  91.1  8.5  Wholesale Trade  37  0.08  83.0  9.1  Profession, Scientific, and Technical Services  20  0.04  85.9  5.2  Finance and Insurance  19  0.04  82.2  7.1  Administrative & Support, waste management, etc  17  0.04  89.8  10.1  Utilities  16  0.04  87.9  8.4  Arts, Entertainment and Recreation  7  0.02  83.1  3.5  Real Estate and Rental and Leasing  2  0.00  84.6  2.5  43  3.1.3  Grab Sample versus Personal Sample  Around 62 % of all samples were “grab samples” collected by sound level meter while the rest were personal samples collected by dosimeter. The mean noise levels of grab samples and personal samples were 91.4 dBA and 91.2 dBA respectively. There was only a 0.2 dBA difference between the two mean levels though the t-test showed that the difference was statistically significant (p = 0.018) because of the large sample size. (Table 3. 3)  Table 3. 3  Comparison between Grab Sample and Personal Sample  Frequency  Percent (%)  Mean (dBA)  SD (dBA)  Grab Sample  28,009  62.18  91.4  9.6  Personal Sample  17,037  37.82  91.2  7.2  t-test p = 0.018  When comparing the noise levels collected by sound level meter and dosimeter in different occupations (1-digit NOC-S level), the differences were quite small in three occupation groups: 1) occupations unique to processing, manufacturing and utilities (0.3 dBA), 2) trades, transport and equipment operators and related occupations (0.5 dBA), 3) occupations unique to primary industry (0.2 dBA) regardless of the results of the t-test. For other occupation groups, the sound level meter always yielded a relatively lower noise level and the difference ranged between 2.9 dBA (Occupations in Social Science, Education, Government Service and Religion) to 12.0 dBA (Management Occupations). (Table 3. 4)  The situation for different industries (at 2-digit NAICS) was complex. The difference of the noise level between sound level meter and dosimeter methods was small in the following industries: manufacturing (0.1 dBA), agriculture, forestry, fishing and hunting (0.5 dBA), transportation and warehousing (0.4 dBA), information and cultural industries (0.6 dBA), and health care and social assistance (0.6 dBA). The difference was large and the dosimeter found a higher noise levels in the three industries: construction (1.1 dBA), retail trade (0.9 dBA), mining, quarrying, oil and gas extraction (1.9 dBA). For the following industries the sound level meter yielded a higher noise level: public administration (3.4 dBA), education services (8.9 dBA), other services except public administration (4.4 dBA), accommodation and food services (7.9 44  dBA), and wholesale trade (2.4 dBA); (Table 3. 5).  In the JEM, these differences between grab sample and personal sample were ignored because dropping the measurements collected by sound level meter meant losing the noise exposure information for a large number of occupations and industries. Both grab sample and personal sample were used to build the JEM.  Table 3. 4  Comparison between Grab Sample and Personal Sample by Occupations  Occupation (NOC-S 1 digit)  Sampling Method  N  a  Mean  SD  (dBA)  (dBA)  Occupations Unique to Processing, Manufacturing  G  19,381  92.7  9.4  and Utilities  P  11,138  92.4  7.1  Trades, Transport and Equipment Operators and  G  6,279  89.2  8.8  Related Occupations  P  4,604  89.7  6.2  G  1,720  91.2  8.1  P  621  91.0  8.5  G  198  79.4  11.6  P  346  83.4  6.1  Business, Finance and Administrative  G  150  73.9  10.6  Occupations  P  68  80.0  9.7  Natural and Applied Sciences and Related  G  116  82.6  12.8  Occupations  P  86  89.0  8.1  G  97  70.5  6.8  P  51  82.5  5.1  G  61  72.7  11.9  P  47  78.4  6.9  Occupations in Social Science, Education,  G  6  82.5  10.0  Government Service and Religion  P  64  85.4  4.7  Occupations in Art, Culture, Recreation and Sport  G  1  82.0  /  b  P  12  84.4  6.1  Occupations Unique to Primary Industry Sales and Service Occupations  Management Occupations Health Occupations  a.  “G” means grab sample and “P” means personal sample;  b.  T-test could not be done due to small sample size.  p-value  0.0032 0.0014 0.5266 <0.0001 0.0001 0.0001 <0.0001 0.004 0.192 /  45  Table 3. 5  Comparison between Grab Sample and Personal Sample by Industries  Industry (NAICS 2 digits)  Sampling  N  Method a Manufacturing Agriculture, Forestry, Fishing and Hunting Construction Public Administration Transportation and Warehousing Education Services Other Services (except Public Administration) Retail Trade Information and Cultural Industries Accommodation and Food Services Health Care and Social Assistance Mining, Quarrying, Oil and Gas Extraction Wholesale Trade Profession, Scientific, and Technical Services b Finance and Insurance  b  Administrative & Support, waste management, etc  b  b  Mean  SD  (dBA)  (dBA)  p-value  G  23,410  91.6  9.7  P  14,198  91.7  7.0  G  2,670  91.0  7.7  P  626  91.5  8.8  G  306  90.2  10.4  P  962  91.3  6.1  G  599  90.0  9.7  P  257  86.6  8.7  G  392  87.9  10.8  P  177  87.5  6.8  G  312  93.8  9.8  P  67  84.9  5.0  G  47  91.4  9.1  P  247  87.0  6.2  G  62  82.5  9.1  P  133  83.4  5.8  G  60  88.7  10.7  P  75  88.1  6.3  G  11  91.0  6.3  P  109  83.1  5.8  G  39  80.9  9.1  P  80  81.5  5.9  G  59  90.5  7.7  P  30  92.4  10.0  G  19  84.2  11.6  P  18  81.8  5.6  /  /  /  /  /  /  /  /  /  /  /  /  /  /  /  0.6377 0.1216 0.0201 <0.0001 0.6443 <0.0001 0.0001 0.3668 0.6972 <0.0001 0.6693 0.3154 0.4399  /  /  /  /  /  Arts, Entertainment and Recreation  b  /  /  /  /  /  Real Estate and Rental and Leasing  b  /  /  /  /  /  Utilities  a.  “G” means grab sample and “P” means personal sample;  b.  The t-test could not be conducted due to small sample size or no sample in “G” or “P” group.  46  3.1.4  Temporal Trend  The noise exposure level showed a decreasing trend over time but that the decrease was very slight – see scatter plot (Figure 3. 3). Based on the results of simple linear regression, there was a 0.15 dBA decrease every year. However, only 0.55% of the total variance in noise exposure level could be explained by year (R2 = 0.0055).  100 50  Noise Level (dBA)  150  Temporal Trend of Noise Level  1970  1980 Noise Level  Figure 3. 3  1990 Year  2000  2010  Fitted values  Scatter Plot of Noise Exposure Level by Year  By visualizing the scatter plots shown in Figure 3. 4, there is no obvious temporal trend, neither increasing nor decreasing, found in the four groups of occupations: a. occupations unique to processing, manufacturing and utilities, b. trades, transport and equipment and operators and related occupations, c. occupations unique to primary industry, d. sales and service occupations. In all other occupation groups, the temporal trend was undetectable due to limited observations or clustered observations in certain time period (Appendix A: Figure A. 1).  47  Figure 3. 4  Scatter Plots of Noise Level in Certain Occupations  In terms of temporal trend by industries, no obvious temporal trend was found in the following three industries: manufacturing; construction; and agriculture, forestry, fishing and hunting (Figure 3.5). However, an apparent temporal trend can be seen in public administration industry. As for other industries, the temporal trend was difficult because of limited observations and clustered observation during certain time period (Appendix A: Figure A. 2)  48  Figure 3. 5  3.1.5  Scatter Plots of Noise Level by Industries  Quality of the Assigned NOC-S Codes  Among the 500 observations selected for the quality assurance assessment, 82% were given the same NOC-S code by both coders. There existed different opinions on the assigned codes in 6% of observations. The divergences were discussed and a more suitable code was selected, and then reassigned to all observations. The rest (12%) were regarded as ambiguous and difficult to decide which assigned code was more appropriate. No change was made to these assigned codes based on the second reviewer’s codes.  49  3.2  National Population Health Survey  3.2.1  3.2.1.1  Diabetes  Characteristics of Cases and Controls  There were 202 diabetes cases reported during follow-up and 1,010 selected controls matched by sex and age. The characteristics of cases and controls are shown in Table 3. 6. The distribution of noise exposure groups among cases and controls was quite similar (p = 0.915). Compared to controls, cases were more likely to be physically inactive and overweight or obese. There was a higher proportion of cases than controls among regular drinkers. The distribution of categories of smoking status among cases and controls was similar. There was a higher proportion of cases who had hypertension in the first cycle compared to controls. In terms of the socioeconomic status, controls were more likely to have higher education and higher family income compared to cases.  3.2.1.2  Collinearity among Covariates  According to the results from the spearman’s correlation, the strongest correlation was found between family income and education (r = 0.31). There was no collinearity between the covariates used for model building and all covariates were offered to the multivariable model. (Table 3. 7)  3.2.1.3  Association between Noise Exposure and Covariates  Based on the results of the chi-square tests between the noise exposure variable and other covariates, strong associations were found between noise exposure and sex (most men were in the high exposure group while most women were in low or medium exposure group, p < 0.001), age (more people aged 50 years and younger were in the high exposure group while more people 50  aged over 50 years were in the medium exposure group, p < 0.001), education (people who had some post-secondary education were more likely to be in the medium exposure group while the people belonging to other education categories were more likely to be in the high exposure group, p = 0.04), and family income (people who had lower income tended to have low exposure while those who had higher income had high exposure, p < 0.001). The detailed bivariate analysis results between noise exposure and covariates are shown in Table 3. 8.  3.2.1.4  Associations between Noise Exposure, Other Covariates and Diabetes  The unadjusted ORs calculated from bivariate analysis between diabetes and all predictors using a simple logistic regression model are listed in Table 3. 9. Obese status, lower physical activity, and hypertension were significantly associated with diabetes. Noise exposure, smoking, drinking, income and education were not.  After the predictor selecting process, the final model contained all offered predictors and was used to estimate adjusted ORs. For detailed information on how the final multiple regression model was selected, please see Appendix B:, Table B. 1. According to the adjusted ORs, no statistically significant association was found between noise exposure and diabetes. In terms of covariates, only obesity was associated with diabetes. The association showed an exposure-response relationship. For physical activity, smoking and hypertension, though no statistically significant results were found, there existed associations between diabetes and these covariates: being physically inactive, a current smoker, and having hypertension were all associated with an increased risk of diabetes. The association between drinking and diabetes was not clear since neither a statistical significance nor a dose-response relationship was found. In terms of socio-economic status, higher SES (better education, higher income) was associated with a lower risk of diabetes (Table 3. 9)  51  Table 3. 6  Characteristics of Diabetes Cases and Controls Cases  Percent  Controls  Percent  Chi-square  N = 202  (%)  N = 1010  (%)  Test  Low-exposure (< 85 dBA-year)  60  29.7  286  28.3  Medium-exposure (85 – 95 dBA-year)  64  31.7  331  32.8  High-exposure (>95 dBA-year)  78  38.6  393  38.9  Male  117  57.9  585  57.9  Female  85  42.1  425  42.1  < 30  19  9.4  95  9.4  31 – 40  39  19.3  195  19.3  41 – 50  94  46.5  470  46.5  > 50  50  24.8  250  24.8  Underweight / Normal Weight  25  12.4  314  31.1  Overweight  76  37.6  453  44.9  Obese I  60  29.7  179  17.7  Obese II  25  12.4  42  4.1  Obese III  16  7.9  22  2.2  Active  43  21.3  280  27.7  Moderate  45  22.3  281  27.8  Inactive  114  56.4  449  44.5  Current Smoker  37  18.3  205  20.3  Former Smoker  111  55.0  515  51.0  Never Smoked  54  26.7  290  28.7  Regular Drinker  115  56.9  709  70.2  Former Drinker  36  17.8  121  12.0  Occasional Drinker / Never Drank  51  25.3  180  17.8  Yes  30  14.9  66  6.5  No  172  85.1  944  94.5  Less than Secondary School Graduation  35  17.3  164  16.2  Secondary School Graduation  34  16.8  120  11.9  Some Post-secondary  63  31.2  273  27.0  Post-secondary Graduation  70  34.7  453  44.9  < $ 40,000  61  30.2  265  26.2  $ 40,000 - $ 79,999  86  42.6  353  35.0  > $ 80,000  55  27.2  392  38.8  Exposure Groups p = 0.915  Sex p=1  Age  p=1  Obese  p < 0.001  Physical Activity p = 0.008  Smoke p = 0.584  Drinking Status p = 0.001  Hypertension p < 0.001  Education  p = 0.037  Family Income p = 0.007  52  Table 3. 7 Age  Sex  Obese  Correlation Matrix for Covariates  Physical Activity  Smoking Status  Drinking Status  Hypertension  Education  Age  1  Sex  -0.0096  1  Obese  -0.0341  -0.0872  1  Physical Activity  -0.0216  0.0422  0.0888  1  Smoking Status  0.0515  0.1072  0.0619  -0.0664  1  Drinking Status  0.1018  0.1302  0.1117  0.1332  0.0435  1  Hypertension  -0.1119  0.0272  -0.1491  -0.1007  -0.0499  -0.1047  1  Education  -0.1488  0.0524  -0.097  -0.1353  0.127  -0.1367  0.0662  1  Family Income  -0.2623  -0.1497  -0.0608  -0.1787  0.0713  -0.3044  0.1073  0.3081  Family Income  1  53  Table 3. 8  Associations between Exposure Groups and Covariates Exposure Group  Chi-square  Low  %  Medium  %  High  %  Male  128  37.0  194  49.1  380  80.7  Female  218  63.0  201  50.9  91  19.3  < 30  29  8.4  50  12.7  35  7.4  31 – 40  50  14.5  75  19.0  109  23.1  41 – 50  159  46.0  175  44.3  230  48.8  > 50  108  31.1  95  24.0  97  20.7  Underweight / Normal Weight  109  31.5  114  28.9  116  24.6  Overweight  133  38.4  174  44.1  222  47.1  Obese  104  30.1  107  27.0  133  28.3  Active  100  28.9  100  25.3  124  26.3  Moderate  91  26.3  113  28.6  131  27.8  Inactive  155  44.8  182  46.1  216  45.9  Current Smoker  57  16.5  86  21.8  99  21.0  Former Smoker  176  50.9  199  50.4  251  53.3  Never Smoked  113  32.6  110  27.8  121  25.7  Regular Drinker  216  62.4  272  68.9  336  71.3  Former Drinker  54  15.6  45  11.4  58  12.3  Occasional Drinker / Never Drank  76  22.0  78  19.7  77  16.4  Yes  35  10.1  24  6.1  37  7.9  No  311  89.9  371  93.9  434  92.1  52  15.0  55  13.9  92  19.5  Secondary School Graduation  42  12.1  49  12.4  63  13.4  Some Post-secondary  104  30.1  126  31.9  106  22.5  Post-secondary Graduation  148  42.8  165  41.8  210  44.6  < $ 40,000  117  33.8  121  30.6  88  18.7  $ 40,000 - $ 79,999  118  34.1  138  34.9  183  38.9  > $ 80,000  111  32.1  136  34.5  200  42.4  Test  Sex p < 0.001  Age  p < 0.001  Obese p = 0.108  Physical Activity p = 0.849  Smoke p = 0.141  Drinking Status p = 0.071  Hypertension p = 0.127  Education Less than Secondary School Graduation p = 0.040  Family Income p < 0.001  54  Table 3. 9  Summary of Results for Diabetes from Single Logistic Regression and the Final Multivariate Logistic Regression Unadjusted OR (95% CI)  Adjusted OR (95% CI)  1  1  Medium-exposure (85 – 95 dBA-year)  0.92 (0.62 – 1.36)  0.93 (0.61 - 1.41)  High-exposure (>95 dBA-year)  0.94 (0.63 – 1.40)  1.04 (0.67 - 1.59)  1  1  Overweight  2.06 (1.27 – 3.34)  2.14 (1.31 - 3.51)  Obese I  4.22 (2.53 – 7.01)  4.03 (2.39 - 6.80)  Obese II  7.72 (4.02 – 14.8)  6.88 (3.47 - 13.7)  Obese III  9.46 (4.39 – 20.4)  7.40 (3.26 - 16.8)  1  1  Moderate  1.01 (0.65 – 1.58)  0.93 (0.58 - 1.49)  Inactive  1.73 (1.17 – 2.54)  1.37 (0.91 - 2.06)  1  1  Former Smoker  1.16 (0.81 – 1.66)  1.09 (0.74 - 1.59)  Current Smoker  0.97 (0.62 – 1.52)  1.22 (0.75 - 1.99)  1  1  Former Drinker  1.01 (0.62 - 1.65)  1.01 (0.60 - 1.71)  Regular Drinker  0.55 (0.38 - 0.81)  0.74 (0.48 - 1.14)  No  1  1  Yes  2.67 (1.64 – 4.34)  1.62 (0.94 - 2.78)  1  1  Secondary School Graduation  1.32 (0.78 – 2.26)  1.60 (0.90 - 2.84)  Some Post-secondary  1.09 (0.68 – 1.72)  1.37 (0.84 - 2.25)  Post-secondary Graduation  0.71 (0.45 – 1.12)  1.05 (0.64 - 1.72)  1  1  $ 40,000 - $ 79,999  1.02 (0.70 – 1.48)  1.21 (0.80 - 1.84)  > $ 80,000  0.57 (0.37 – 0.87)  0.86 (0.53 - 1.39)  Exposure Low-exposure (< 85 dBA-year)  Obese Status Underweight / Normal Weight  Physical Activity Active  Smoke Never Smoked  Drinking Status Occasional Drinker / Never Drank  Hypertension at Cycle 1  Education Less than Secondary School Graduation  Family Income < $40,000  55  3.2.2  3.2.2.1  Rheumatoid Arthritis  Characteristics of Cases and Controls  There were 91 reported rheumatoid arthritis cases during follow-up and 455 controls matched by sex and age. The characteristics of cases and controls are shown in Table 3. 10. There was little difference in noise exposure between cases and controls (p = 0.974). Controls were more likely to have higher education and family income compared to cases. In terms of the obese status, physical activity, and smoking status, the distribution of cases and controls among different categories was relatively similar.  3.2.2.2  Collinearity among Covariates  The correlation matrix for covariates is shown in Table 3. 11. The strongest correlation was found between education and family income (r = 0.24). There was no significant collinearity found between any of the two covariates that would influence the validity of the regression model and all covariates were offered to the multivariable model.  3.2.2.3  Association between Noise Exposure and Covariates  Similar to the diabetes sub-cohort, strong associations were found between noise exposure and sex (most men were in the high exposure group while most women were in low or medium exposure group, p < 0.001), age (people aged between 35 and 44 years were more likely to have high exposure while those not belonging to this age range tended to have medium exposure, p = 0.016), obesity (people who were overweight or obese were more likely to have high exposure while those who had normal weight or were underweight were more likely to have the medium exposure, p = 0.043), and education (people who had less education tended to have high exposure while people who had more education tended to be in the low or medium exposure group, p = 0.021). The detailed bivariate analyses between noise exposure and covariates are 56  shown in Table 3. 12.  3.2.2.4  Association between Noise Exposure and Rheumatoid Arthritis  Unadjusted ORs estimated from bivariate analyses between rheumatoid arthritis variable and all predictors using a simple logistic regression model are listed in Table 3. 13. Only family income had a significant association with rheumatoid arthritis.  After the predictor selection process, the final model contained all offered predictors and was used to estimate adjusted ORs. For detailed information on how the final multiple regression model was selected, please find in Appendix B:, Table B. 2. In the multiple logistic regression model, there was no statistically significant association between noise exposure and rheumatoid arthritis. In terms of covariates, people who were current smokers or had some post-secondary education were at higher risk of rheumatoid arthritis, while moderate physical activity or higher family income were associated with a lower risk of rheumatoid arthritis. The association between obesity and RA was not statistically significant. (Table 3. 13)  57  Table 3. 10  Characteristics of Rheumatoid Arthritis Cases and Controls Cases  Percent  Controls  Percent  Chi-square  N = 91  (%)  N = 455  (%)  Test  Low-exposure (< 85 dBA-year)  25  27.5  120  26.4  Medium-exposure (85 – 95 dBA-year)  34  37.4  171  37.6  High-exposure (>95 dBA-year)  32  35.1  164  36.0  Male  43  47.3  215  47.3  Female  48  52.7  240  52.7  < 35  29  31.9  145  31.9  35 – 44  33  36.3  165  36.3  >= 45  29  31.8  145  31.8  Underweight or Normal Weight  32  35.2  168  36.9  Overweight  37  40.6  180  39.6  Obese  22  24.2  107  23.5  Active  26  28.6  107  23.5  Moderate  19  20.9  139  30.6  Inactive  46  50.5  209  45.9  Current Smoker  23  25.3  84  18.5  Former Smoker  44  48.3  228  50.1  Never Smoked  24  26.4  143  31.4  Secondary School Graduation or Less  25  27.5  125  27.5  Some Post-secondary  34  37.4  115  25.3  Post-secondary Graduation  32  35.1  215  47.2  < $ 40,000  24  26.4  90  19.8  $ 40,000 - $ 79,999  39  42.8  158  34.7  > $ 80,000  28  30.8  207  45.5  Exposure Groups p = 0.974  Sex p=1  Age p=1  Obese p = 0.951  Physical Activity p = 0.166  Smoke p = 0.290  Education p = 0.039  Family Income p = 0.034  58  Table 3. 11 Age  Sex  Correlation Matrix for Covariates  Obese  Physical Activity  Smoking Status  Education  Age  1  Sex  -0.1379  1  Obese  0.1274  -0.2093  1  Physical Activity  -0.0049  -0.0213  0.1637  1  Smoking Status  0.131  0.0845  0.0096  -0.041  1  Education  -0.1405  0.0686  -0.0435  -0.0236  0.1229  1  Family Income  -0.1773  -0.0985  -0.0737  -0.1222  0.0857  0.2403  Family Income  1  59  Table 3. 12  Associations between Exposure Groups and Covariates Exposure Group  Chi-square  Low  %  Medium  %  High  %  Male  44  30.3  70  34.1  144  73.5  Female  101  69.7  135  65.9  52  26.5  < 35  41  28.3  79  38.5  54  27.6  35 – 44  49  33.8  63  30.7  86  43.9  >= 45  55  37.9  63  30.8  56  28.5  Underweight or Normal Weight  58  40.0  82  40.0  60  30.6  Overweight  46  31.7  84  41.0  87  44.4  Obese  41  28.3  39  19.0  49  25.0  Active  36  24.8  46  22.4  51  26.0  Moderate  39  26.9  60  29.3  59  30.1  Inactive  70  48.3  99  48.3  86  43.9  Current Smoker  24  16.6  41  20.0  42  21.4  Former Smoker  77  53.1  97  47.3  98  50.0  Never Smoked  44  30.3  67  32.7  56  28.6  43  29.7  47  22.9  60  30.6  Some Post-secondary  50  34.5  57  27.8  42  21.4  Post-secondary Graduation  52  35.8  101  49.3  94  48.0  < $ 40,000  36  24.8  50  24.4  28  14.3  $ 40,000 - $ 79,999  53  36.6  72  35.1  72  36.7  > $ 80,000  56  38.6  83  40.5  96  49.0  Test  Sex p < 0.001  Age p = 0.016  Obese p = 0.043  Physical Activity p = 0.848  Smoke p = 0.705  Education Secondary School Graduation or Less  p = 0.021  Family Income p = 0.060  60  Table 3. 13  Summary of Results for Rheumatoid Arthritis from Single Logistic Regression and the Final Multivariate Logistic Regression Unadjusted OR (95% CI)  Adjusted OR (95% CI)  1  1  0.95 (0.54 – 1.68)  1.01 (0.56 - 1.82)  0.93 (0.51 – 1.70)  1.07 (0.57 - 2.01)  1  1  Overweight  1.09 (0.63 – 1.86)  1.09 (0.63 - 1.90)  Obese  1.08 (0.60 - 1.96)  1.11 (0.59 - 2.08)  1  1  Moderate  0.57 (0.30 - 1.08)  0.48 (0.25 - 0.93)  Inactive  0.91 (0.54 - 1.52)  0.76 (0.44 - 1.32)  1  1  Former Smoker  1.67 (0.87 - 3.21)  1.00 (0.56 - 1.78)  Current Smoker  1.17 (0.67 - 2.01)  1.48 (0.75 - 2.94)  1  1  Some Post-secondary  1.53 (0.84 - 2.77)  1.86 (1.01 - 3.44)  Post-secondary Graduation  0.74 (0.41 - 1.31)  0.91 (0.50 - 1.66)  1  1  $ 40,000 - $ 79,999  0.90 (0.49 - 1.62)  0.98 (0.53 - 1.80)  > $ 80,000  0.48 (0.26 - 0.90)  0.50 (0.26 - 0.96)  Exposure Low-exposure (< 85 dBA-year) Medium-exposure High-exposure  (85 – 95 dBA-year)  (>95 dBA-year)  Obese Status Underweight / Normal Weight  Physical Activity Active  Smoke Never Smoked  Education Secondary School Graduation or Less  Family Income < $40,000  61  3.2.3  3.2.3.1  Cardiovascular Disease  Characteristics of Cases and Controls  There were 221 reported cardiovascular disease cases during follow-up and 1,105 controls matched by sex and age. The characteristics of cases and controls are shown in Table 3. 14. Although the proportion of controls belonging to the high noise exposure group was higher compared to that of cases, the difference was very slight. Compared to controls, cases were more likely to be physically inactive and overweight or obese. The proportion of cases who were current or former smoker was higher compared to that of controls. A higher proportion of cases than controls belonged to the “occasional drinker / never drank” category. Cases were more likely to have hypertension in the first cycle compared to controls. Controls were more likely to have higher education and family income compared to cases.  3.2.3.2  Collinearity among Covariates  The results of correlations between covariates are listed in Table 3. 15. The strongest correlation was found between education and family income (r = 0.36). There was no significant collinearity found between any of the two covariates that would influence the validity of the regression model and all covariates were offered to the multivariable model.  3.2.3.3  Association between Noise Exposure and Covariates  Based on the results of the chi-square tests between the noise exposure group variable and other covariates, strong associations were found between noise exposure and sex (most men were in the high exposure group while most women were in low or medium exposure group, p < 0.001), age (more people aged younger than 50 were in the high exposure group while more people aged elder than 50 were in the low exposure group, p < 0.001), smoking status (more people who were current/former smoker were in the high exposure group while more people who never smoked 62  were in the low exposure group, p = 0.010), drinking status (people who drank were more likely to be in the high exposure group while those who seldom drank were more likely to be in the low exposure group, p < 0.001), and family income (people who had lower income tended to have low exposure while those who had higher income had high exposure, p < 0.001). The detailed bivariate analysis between noise exposure and covariates are shown in Table 3. 16.  3.2.3.4  Association between Noise Exposure and Cardiovascular Disease  The results of the bivariate analysis between the CVD and model predictors are shown as unadjusted ORs in Table 3. 17. Smoking status drinking status, hypertension, and family income were found to be significantly associated with CVD.  After the predictor selection process, the final model contained all offered predictors and was used to estimate adjusted ORs. For detailed information on how the final multiple regression model was selected, please find in Appendix B:, Table B. 3. The calculated adjusted ORs from the multiple logistic regression model showed no statistically significant association between noise exposure and CVD (Table 3. 17). In terms of covariates, people who used to smoke or had hypertension were at higher risk of CVD. Being a regular drinker or having relatively higher family income was associated with reduced risk of CVD. Although no statistically significant association was found between CVD and obesity, a mild dose-response relationship was observed.  63  Table 3. 14  Characteristics of CVD Cases and Controls Cases  Percent  Controls  Percent  Chi-square  N = 221  (%)  N = 1105  (%)  Test  Low-exposure (< 85 dBA-year)  76  34.4  339  30.7  Medium-exposure (85 – 95 dBA-year)  69  31.2  347  31.4  High-exposure (>95 dBA-year)  76  34.4  419  37.9  Male  115  52.0  575  52.0  Female  106  48.0  530  48.0  < 30  20  9.1  100  9.1  31 – 40  52  23.5  260  23.5  41 – 50  88  39.8  440  39.8  > 50  61  27.6  305  27.6  Underweight / Normal Weight  58  26.3  329  29.8  Overweight  94  42.5  495  44.8  Obese  69  31.2  281  25.4  Active  52  23.5  243  22.0  Moderate  54  24.4  351  31.8  Inactive  115  52.1  511  46.2  Current Smoker  41  18.6  215  19.5  Former Smoker  137  62.0  574  52.1  Never Smoked  43  19.4  314  28.4  Regular Drinker  126  57.0  744  67.3  Former Drinker  39  17.7  141  12.8  Occasional Drinker / Never Drank  56  25.3  220  19.9  Yes  30  13.6  60  5.4  No  191  86.4  1045  94.6  40  18.1  160  14.5  Secondary School Graduation  29  13.1  145  13.1  Some Post-secondary  68  30.8  302  27.3  Post-secondary Graduation  84  38.0  498  45.1  < $ 40,000  72  32.6  277  25.1  $ 40,000 - $ 79,999  89  40.3  372  33.7  > $ 800,000  60  27.1  456  41.2  Exposure Groups p = 0.490  Sex p=1  Age  p=1  Obese p = 0.188  Physical Activity p = 0.093  Smoke p = 0.011  Drinking Status p = 0.012  Hypertension p < 0.001  Education Less than Secondary School Graduation p = 0.215  Family Income p < 0.001  64  Table 3. 15 Age  Sex  Obese  Correlation Matrix for Covariates  Physical Activity  Smoking Status  Drinking Status  Hypertension  Education  Age  1  Sex  -0.0349  1  Obese  -0.0105  -0.1316  1  Physical Activity  -0.0547  0.0032  0.13  1  Smoking Status  0.1444  0.1036  0.0355  -0.0755  1  Drinking Status  0.0923  0.1692  0.0309  0.0914  0.0992  1  Hypertension  -0.0987  -0.017  -0.0857  -0.0112  -0.0137  -0.0106  1  Education  -0.1757  0.0067  -0.0351  -0.1223  0.1133  -0.1269  0.0155  1  Family Income  -0.2954  -0.1645  -0.0172  -0.1054  0.0472  -0.2374  0.0445  0.3561  Family Income  1  65  Table 3. 16  Associations between Exposure Groups and Covariates Exposure Groups  Chi-square  Low  %  Medium  %  High  %  Male  132  31.8  177  42.5  381  77.0  Female  283  68.2  239  57.5  114  23.0  < 30  34  8.2  44  10.6  42  8.5  31 – 40  92  22.2  95  22.8  125  25.3  41 – 50  136  32.8  164  39.4  228  46.1  > 50  153  36.8  113  27.2  100  20.1  Underweight / Normal Weight  132  31.8  125  30.0  130  26.3  Overweight  183  44.1  170  40.9  236  47.7  Obese  100  24.1  121  29.1  129  26.0  Test  Sex p < 0.001  Age  p < 0.001  Obese p = 0.141  Physical Activity Active  89  21.4  90  21.6  116  23.4  Moderate  130  31.3  127  30.5  148  29.9  Inactive  196  47.3  199  47.9  231  46.7  Current Smoker  61  14.7  82  19.7  113  22.8  Former Smoker  225  54.2  219  52.6  267  53.9  Never Smoked  129  31.1  115  27.7  115  23.3  Regular Drinker  237  57.1  276  66.3  357  72.1  Former Drinker  64  15.4  49  11.8  67  13.5  Occasional Drinker / Never  114  27.5  91  21.9  71  14.3  Yes  34  8.2  29  7.0  27  5.5  No  381  91.8  387  93.0  468  94.5  63  15.2  55  13.2  82  16.6  Secondary School Graduation  63  15.2  48  11.5  63  12.7  Some Post-secondary  119  28.6  119  28.6  132  26.7  Post-secondary Graduation  170  41.0  194  46.7  218  44.0  < $ 40,000  138  33.3  123  29.6  88  17.8  $ 40,000 - $ 79,999  138  33.3  144  34.6  179  36.2  > $ 800,000  139  33.6  149  35.8  228  46.0  p = 0.949  Smoke p = 0.010  Drinking Status  p < 0.001  Drank Hypertension p = 0.258  Education Less than Secondary School Graduation p = 0.442  Family Income p < 0.001  66  Table 3. 17  Summary of Results for Cardiovascular Disease from Single Logistic Regression and the Final Multivariate Logistic Regression Unadjusted OR (95% CI)  Adjusted OR (95% CI)  1  1  Medium-exposure (85 – 95 dBA-year)  0.87 (0.61 – 1.26)  0.87 (0.59 - 1.28)  High-exposure (>95 dBA-year)  0.77 (0.53 – 1.14)  0.85 (0.57 - 1.27)  1  1  Overweight  1.08 (0.75 - 1.55)  1.12 (0.77 - 1.63)  Obese  1.39 (0.95 - 2.04)  1.18 (0.79 - 1.78)  1  1  Moderate  0.71 (0.47 – 1.08)  0.65 (0.42 - 1.01)  Inactive  1.05 (0.73 – 1.50)  0.91 (0.62 - 1.34)  1  1  Former Smoker  1.76 (1.21 – 2.56)  1.81 (1.23 - 2.67)  Current Smoker  1.40 (0.87 – 2.24)  1.30 (0.80 - 2.13)  1  1  Former Drinker  1.08 (0.69 - 1.70)  1.03 (0.64 - 1.65)  Regular Drinker  0.64 (0.45 - 0.92)  0.68 (0.46 - 1.00)  No  1  1  Yes  2.73 (1.71 – 4.36)  2.86 (1.74 - 4.69)  1  1  Secondary School Graduation  0.80 (0.47 – 1.34)  1.00 (0.59 - 1.71)  Some Post-secondary  0.89 (0.57 – 1.40)  1.12 (0.70 - 1.80)  Post-secondary Graduation  0.67 (0.44 – 1.02)  0.96 (0.60 - 1.52)  1  1  $ 40,000 - $ 79,999  0.87 (0.61 – 1.24)  0.95 (0.65 - 1.39)  > $ 80,000  0.46 (0.31 – 0.68)  0.52 (0.34 - 0.81)  Exposure Low-exposure (< 85 dBA-year)  Obesity Status Underweight / Normal Weight  Physical Activity Active  Smoke Never Smoked  Drinking Status Occasional / Never Drank  Hypertension at Cycle 1  Education Less than Secondary School Graduation  Family Income < $40,000  67  Chapter 4: Discussion  4.1  4.1.1  Summary of Key Findings  Diabetes  No association was observed between occupational noise exposure and diabetes in the present study. The unadjusted ORs for diabetes associated with medium and high noise exposure groups were 0.92 (95% CI: 0.62 – 1.36) and 0.94 (95% CI: 0.63 – 1.40) respectively. After adjusting for other covariates, the direction of the OR for high exposure group changed (1.04, 95% CI: 0.67 – 1.59), but the magnitude of change was quite small and the OR was not statistically significant.  A review of the literature indicated that there has been only one study of the direct association between noise exposure and diabetes, and that was a broader environmental study. Sorensen et al. (2012) investigated long-term exposure to residential road traffic noise and incidence of diabetes in a large cohort study. They found incidence risk ratios of 1.11 (95% CI: 1.03 – 1.19) and 1.14 (95% CI: 1.06 – 1.22) per 10-dB increase in road traffic noise at diagnosis and during the 5 years preceding diagnosis, respectively. The associations were slightly stronger when using a longer period of exposure (5 years) compared to shorter period of exposure (1 year). Although this study provided some support to a direct association between noise and diabetes, the results may not be generalized to occupational noise since environmental noise may work by a different mechanism in which the stress response may not dominate. Sorensen et al. proposed that the sleep disturbances (e.g. short sleep duration, reduced sleep quality, changes in sleep stages) resulting from night-time exposure to traffic noise might be a possible intermediate stage which led to the diabetes outcome.  There are however studies demonstrating a positive association between stress and diabetes (Djindjic et al., 2012; Li et al., 2012; Novak et al., 2012). Djindjic et al. and Li et al. studied work-related stress and Novak et al. studied the stress from various aspects of life (both at work 68  and at home). However, the measured stress in these studies was not a surrogate of noise exposure; their results provided support for the second part of the proposed biological model, the stress-disease relationship. The perception of noise (i.e. the noise-stress response) is a complex process in which two major pathways (direct and indirect) interact in a very complicated way and many other parameters are involved (e.g. noise level, personal susceptibility, attitude towards noise) which were not addressed in these studies. Although they did not examine the direct association between noise and diabetes, they did provide some ideas to explain the negative results of our study. Since the stress – diabetes association was supported by these studies and a noise – stress response has been demonstrated in previous studies, it would be expected that there is a positive association between noise exposure and diabetes, yet we did not find this in our study. Assuming then that there is a true relationship, why might our study have not seen it? It is possible that the noise in our study was not stressful enough (high enough or of sufficient duration or both) to trigger the stress response (i.e. noise exposure was not cumulated sufficiently due to limited time period to estimate exposure duration.). It is also possible that some biases (e.g. exposure misclassification) in the study may dilute the association. Given the relatively well-studied association between occupational noise exposure and CVD, we used CVD in this study as a sensitivity test to examine that the validity of the methods of this study. Since no association was found between occupational noise exposure and CVD in the subjects of NPHS, this supports the idea that biases in the study diluted the estimated association. The biases that may influence the validity of the methods and their sources are discussed in following sections.  Subjects’ obesity status was strongly associated with diabetes and a clear trend was found where the more severe the obesity, the higher the risk of diabetes (overweight: 2.14, obese I: 4.03, obese II: 6.88, obese III: 7.40). Being physically inactive increased the risk of diabetes (1.37, 95% CI: 0.91 – 2.06) though the positive association was not statistically significant. Hypertension was also positively associated with diabetes (1.62, 95% CI: 0.94 – 2.78). These associations were consistent with the risk factors described elsewhere for diabetes (Canadian Diabetes Association, 2013), which means that the measurement of the diabetes outcome and these covariates were relatively reliable.  69  4.1.2  Rheumatoid Arthritis  Our results indicate that there was no observed association between occupational noise exposure and RA in the current study. The unadjusted ORs for RA associated with the medium and high noise exposure groups showed decreased but non-statistically significant association between occupational noise exposure and RA (medium exposure group: 0.95, 95% CI: 0.54 – 1.68; high exposure group: 0.93, 95% CI: 0.51 – 1.70). After adjusting for other covariates, the direction of the association changed but the association was weak and remained non-statistically significant (medium exposure group: 1.01, 95% CI: 0.56 – 1.82; high exposure group: 1.07, 95% CI: 0.57 – 2.01). The change of direction occurred when adding the education and family income covariates, indicating that they might be confounders.  A literature review found no previous studies that focused on the association between noise and RA for comparison, though findings of an unpublished recent environmental noise study that studied RA was also negative (personal communication with Hugh Davies). Our negative findings may reflect a true lack of association between noise and RA, or might result from biases in the study; these are discussed in later sections.  Looking at other risk factors, compared to the high activity group, having moderate physical activity was associated with a decreased risk of RA (0.48, 95% CI: 0.25 – 0.93); this is consistent with our speculation that some benefits of moderate physically activity but more extreme levels causing more damage to a joint. Current smokers were at higher risk of RA though the OR was not statistically significant (1.48, 95% CI: 0.75 – 2.94). This is consistent with previous relevant information on RA (Centers for Disease Control and Prevention, 2012; Statistics Canada, 2006). Having higher family income was associated with a decreased risk (0.50, 95% CI: 0.26 – 0.96). This is also consistent with a previous study that used the first cycle of the NPHS in a cross-sectional analysis (Wang et al., 2000).  70  4.1.3  Cardiovascular Disease as “Sensitivity Analysis”  The ORs for CVD in medium and high noise exposure groups were 0.87 (95% CI: 0.61 – 1.26) and 0.77 (95% CI: 0.53 – 1.14) respectively. After adjusting for other covariates, the direction of the ORs remained the same and the magnitude of the ORs only changed slightly (medium exposure group: 0.87, 95% CI: 0.59 – 1.28; high exposure group: 0.85, 95% CI: 0.57 – 1.27). The results indicated a decreased risk of heart disease associated with occupational noise exposure although the association was not statistically significant, that is opposite to the direction hypothesized.  The association between noise exposure and CVD has been studied for decades and, as relevant research methodologies became more and more sophisticated, a positive association has gradually been accepted even though negative studies have still occurred in recent years (Suadicani et al., 2012). Most studies focusing on the noise-CVD association were conducted within two major domains: occupational noise and environmental noise. Although many previous studies focusing on environmental noise have found positive association between noise and CVD (Eriksson et al., 2010; Babisch and van Kamp, 2009; Kluizenaar et al., 2007; Babisch et al., 2005), they are not discussed further here because environmental noise likely works by a different mechanism.  In Virkkunen et al. (2005), exposure to occupational noise was associated with a moderate but statistically significant increase in coronary heart disease risk among workers that persisted even after they retired. McNamee et al. (2006) found increases in the risk of ischemic heart disease in one of two studied industrial plants; the negative result in the second plant was attributed to poor exposure assessment and subsequent misclassification. Gan et al. (2011) studied the noise and CVD association in the general population of the USA and found an increased risk of angina pectoris (OR = 2.91, 95% CI: 1.35 – 6.26), coronary heart disease (OR = 2.04, 95% CI: 1.16 – 3.58), and isolated diastolic hypertension (OR = 2.23, 95% CI: 1.21 – 4.12). Lee et al. (2009) studied 530 workers at a metal manufacturing factory in Korea and observed an increase of systolic blood pressure in all three noise exposed groups. In British Columbia, a retrospective 71  cohort study of lumber mill workers found that long-term exposure to occupational noise was associated with an increased risk of acute myocardial infarction death (Davies et al., 2005). In another examination of the same BC sawmill workers, a positive association between noise exposure above 85 dBA and hypertension was found (Sbihi et al., 2008). In a meta-analysis based on 15 papers from 1950 to 2008, a statistically significant increase in systolic and diastolic blood pressure was found in workers in the high exposure group. In this exposure group, the prevalence of hypertension and ECG abnormalities were also higher (Tomei et al., 2010).  In our study, being overweight (OR: 1.12, 95% CI: 0.59 – 1.28) and obese (OR: 1.18, 95% CI: 0.79 – 1.78) were associated with an increased risk of CVD, though the associations were not statistically significant. People who smoked were at higher risk and the positive association was more obvious in former smokers (OR: 1.81, 95% CI: 1.23 – 2.67) than current smokers (OR: 1.30, 95% CI: 0.80 – 2.13). Having hypertension was significantly associated with increased risk of CVD (OR: 2.86, 95% CI 1.74 – 4.69). All these factors (obesity, smoking, and hypertension) are known risk factors for CVD (WHO, 2012). The consistency between our results and those of prior studies indicates that the significant bias in the measurement of CVD and the covariates was unlikely; thus if there is a true relation between noise and CVD as evidenced by earlier studies, the most likely source of bias that might obscure the relation in our study is from the exposure assessment.  4.2  Potential Impact of Bias  In the current study, there were potential biases that could have influenced the exposure assessment as well as the measurement of health outcome and covariates, and which may have had further impact on the validity of the results. The major biases and their effect on the study results are discussed in this section.  72  4.2.1  Misclassification in Health Outcome and Covariates  The measurement of health outcome was based on self-reported diagnosis by a health professional. Therefore, the accuracy of the reported information determined the degree of misclassification in the health outcomes. It is true that information provided by some subjects on the same question varied among different cycles (e.g. one reported CVD in the first cycle, then became non-CVD in the second and third cycles, finally had CVD in the forth cycle), which indicated some quality issues about the reported information. In subjects who reported disease status, it is thought that the severe cases may remember relevant information about their disease (e.g. diagnosis time) and provide more reliable information compared to those who only had mild symptoms (Coughlin, 1990). On the other hand, not reporting a health outcome did not ensure that the subject was disease free; for example, people might be at the early stage of the disease or their disease symptoms were not obvious enough to draw their attention to seek a health professional’s advice. Baker et al. (2011) used the Ontario Health Insurance Plan (OHIP) as a relatively reliable ‘gold standard’ and compared the medical record in its system to the health status reported by the NPHS subjects in the province of Ontario. They found correlation coefficients of 0.71 and 0.17 between OHIP records and self-reported health status in NPHS for diabetes and arthritis respectively. After quantifying the measurement error, the ratios of the error variance to the total variance were 0.35 and 0.88 for diabetes and arthritis respectively. When decomposing the error variance, for diabetes, the large proportion of variance came from false negatives (72%) while more error variance was due to false positives for arthritis (93%). In the thirteen health conditions (heart disease was not studied) Baker studied, diabetes had the smallest ratio of the error variance to the total variance while that for arthritis was the second largest. This study revealed some problems with the validity of self-reported health status in the NPHS. The direction and magnitude of the misclassification in health outcome measurement was largely determined by the ratio of false negatives to false positives and the distribution of these false negatives and false positives among different exposure groups. Therefore, without knowing how the false negatives and false positives distributed among different exposure groups, the effect of misclassification in measurement of health outcomes on the regression estimates was difficult to assess. For diabetes, if we assumed the misclassification was non-differential among exposure 73  groups, the ORs of the current study were likely overestimated. But the estimates in the study were very close to 1. So it was more likely that the biases in exposure assessment were more of a problem responsible for these negative results. In terms of RA, because the arthritis studied in Baker et al.’s study included all types of arthritis (osteoarthritis, rheumatoid, and other) and RA constituted a small proportion of the total, the misclassification in measurement of RA becomes more complex and hard to evaluate. As for CVD, although it was not studied in Baker et al.’s study, we might assume a major source of error from false negatives and a similar effect of misclassification on the estimates.  Another source of misclassification in the measurement of diabetes might come from the failure to exclude other diabetes types, especially type-1 diabetes. Ng et al. (2008) developed an algorithm to differentiate diabetes types in the Canadian Community Health Survey and it was adapted and modified in Dinca-Panaitescu et al.’s study (2012) to the NPHS longitudinal data to distinguish type-2 diabetes from other types of diabetes. In their study, the type-1 diabetes cases were excluded if they were younger than 30 at interview and started using insulin within 6 month of diagnosis. In the current study, although some type-1 diabetes cases might be excluded because of the inclusion criterion of aged between 18 and 55, others were inevitably retained in the sub-cohort and misclassified as type-2 diabetes. However, given the small proportion (2%) of type-1 diabetes (n = 27) in total diabetes cases (n = 1340) indicated in Dinca-Panaitescu et al.’s study, the impact of this misclassification was likely mild.  In terms of covariates, all the covariates were treated as time-stable even though some of them could be time-varying. In order to assure the most up-to-date data was obtained, the information on these covariates was from the latest cycle before the end of follow-up period. Some missing values were filled by obtaining information from earlier cycles. For these reasons, as well as others, some misclassification in the measurement of covariates was inevitable. However in the multivariable models, the associations between the health outcomes and covariates were consistent with current knowledge and therefore, the impact of misclassification on the internal associations between outcomes and covariates was assumed to be mild.  74  4.2.2  Biases in Exposure Assessment  If the influence from misclassification in health outcome and covariates on the regression estimates was mild, then those biases associated with exposure assessment could be the major reason resulting in the negative results. The exposure assessment part is always a challenging part in occupational epidemiological studies. Generally, the exposure assessment method of the current study can be broken into five stages: sampling, coding, building the JEM, linking two datasets, and estimating CNE. Biases and errors that could occur each of these stages are discussed in following section.  4.2.2.1  Sampling Stage  A potentially substantial source of bias in this stage was related to the sampling devices used to collected noise measurements. The noise samples in WorkSafeBC’s noise exposure dataset were collected either by sound level meter or dosimeter. To estimate the cumulative noise exposure for each subject in the NPHS, noise exposure level at individual level based on the subject’s NOC-S and/or NAICS codes was required. Therefore, personal sample was preferable to area or “grab” sample. The dosimeter is expected to yield a more accurate estimation of personal exposure because the dosimeter is attached to the subject being sampled and it takes into account the different tasks performed by the subject and the variability of noise exposure (CCOHS, 2004). Accurate noise levels for occupation and/or industry are required to estimate the cumulative noise exposure for each individual in NPHS and reduce exposure misclassification. Therefore, personal exposure is preferable in this study but the majority of samples (62 %) were collected using a sound level meter. Overall, there was only a 0.2 dBA difference between the mean levels of the two sampling type (statistically significant, but the n was very large).  However, when comparing the difference between the two sampling methods within occupations, the sound level meter tended to underestimate the noise level in some occupational groups at the 1-digit NOC-S level and the situation was more varied when more specific occupations (3-digit or 4-digit NOC-S level) were considered and the situation was similar when comparing the two 75  sampling methods in different industries. Therefore, it is difficult to conclude how the different sampling methods influenced the results but it appears likely that the use of sampling type was randomly distributed and thus resulting misclassification is non-differential.  4.2.2.2  Coding Stage  Each collected samples was assigned a CU code and then aggregated into WorkSafeBC’s original noise exposure dataset. These CU codes were then transcribed into 4-digit NAICS codes based on a CU – NAICS codes “cross-walk” that was created by other UBC researchers and has been evolved and refined through use in several other studies of WorkSafeBC data (personal communication with Lillian Tamburic).  In terms of occupation codes, the assignment of NOC-S codes was conducted based mainly on the job description variable, from which a corresponding 4-digit NOC-S code was assigned. Two strategies were applied to reduce error and assure the quality of the assigned NOC-S codes: (1) only measurements that had the highest level of confidence (on a 3-part scale) for assigned NOC-S code were retained in the dataset for JEM building (90% of original dataset); and (2) after the code assigning process, a fraction of the measurements were duplicate coded by a second hygienist independently and these codes were compared with those assigned in the first round. The comparison showed a relatively high consistency (82% of the 500 selected measurements). Moreover, when divergence was found, the more appropriate code was selected and replaced all the incorrect ones in the dataset, which also reduce the error in the assigned NOC-S codes.  4.2.2.3  JEM Building Stage  The building of the JEM was conducted by calculating the mean noise levels for the NOC-S and NAICS codes at four digit and two digit level. Considering that the noise exposure for the same occupation (i.e. same NOC-S) in different industries (i.e. different NAICS) could be different, the mean noise levels were also calculated for each combination of 4-digit NOC-S and 4-digit 76  NAICS, but only if the number of measurements for that NOC-S NAICS pair was larger than ten. The JEM provided noise levels for subjects in the NPHS and therefore its quality had significant impact on the final calculated cumulative noise exposure. Two issues with the quality of the JEM are explored in more detail below:  4.2.2.3.1  Time Axis in the JEM  The noise measurements in the dataset were collected between 1970 and 2004. About 98% of samples were collected before 1994. However, the period covered by NPHS was from 1994 to 2004. Therefore, if there existed an obvious temporal trend in the noise levels, the mean noise levels before and after 1994 could be different and it would be necessary to add a time (year) axis to the JEM. Restriction of measurements to those after 1994 was not possible as most of the samples were collected before that.  Considering the whole dataset, the noise levels decreased at a rate of 0.15 dBA per year. So the total decrease in the noise level during the 34 years covered by the dataset was 5.1 dBA. The association between noise exposure level and sampling year was negligible (R2 = 0.0055). The temporal trend was also investigated in the following four occupation groups (1-digit NOC-S level, accounts for 98% of all measurements) in which there were enough observations across time to allow for the trend analysis: (1) occupations unique to processing, manufacturing and utilities, (2) trades, transport and equipment and operators and related occupations, (3) occupations unique to primary industry, and (4) sales and service occupations. No obvious increasing or decreasing trend was found in the four occupation groups. So in these groups, the influence from time (year) was weak, indicating that adding the time axis in JEM was not necessary. In other occupation groups where there were small number of observations and cluster of observations in certain time period, the temporal trend based on linear regression could not be assessed. In terms of temporal trend by different industries, a decreasing temporal trend was found in public administration only.  77  Given the findings above, it was decided that it was not necessary to add a time axis in the JEM for these occupation groups. However, since there was an obvious decreasing temporal trend in public administration industry, biased noise exposure might be estimated in individuals who belonged to this industry category and overestimated exposure could be expected. But in each cycle, only a small proportion of subjects (less than 2%) were assigned a noise level based on their industry code of public administration. The impact on the overall population could be very slight.  4.2.2.3.2  Noise Exposure Levels  The minimum and maximum noise exposure levels collected in the dataset were 50 dBA and 150 dBA, which means this dataset covered occupations/industries with low exposure as well as some exposed to extremely high levels of noise. Considering the mean noise level of 91.4 dBA and distribution of the collected noise levels, the majority of occupations/industries contained in this dataset had relatively high exposure to noise. This is to some extent consistent with the compliance-orientation of the WorkSafeBC’s noise exposure dataset. Those occupations and/or industries with more chance of being exposed to high level of noise were more likely to be sampled.  High mean noise exposure levels were found in those occupations or groups of occupations expected to have higher exposure, for example, occupations unique to processing, manufacturing and utilities (92.6 dBA), and occupations unique to primary industry (91.2 dBA). However, for management occupations, the mean noise exposure level (74.6 dBA) seems a little high; typical office exposure would be expected to be around 60 dBA or less. Overestimation of noise levels would result in underestimated ORs, while underestimated noise levels would lead to overestimated ORs.  78  In terms of noise levels by industries, there was relatively small variance among the calculated mean noise levels by industries (2-digit NAICS level). However, an unexpectedly high mean noise exposure level was found in the education services industry (92.2 dBA). One explanation for this could be that the samples might be collected in trade schools or departments, where the noise levels could be high. Since the WorkSafeBC’s noise dataset was compliance oriented, those industries with high levels of noise exposure were more likely to be sampled. And within an industry, occupations having more chance of being exposed to high levels of noise were more likely to be sampled. This indicates that the mean noise levels by occupations may provide a more reasonable and reliable estimation for an individual’s noise exposure than by industries. Using the mean noise level of the combination of both occupation and industry should yield the best estimation. Therefore, when linking the noise level to individuals in NPHS, priority was given to noise levels that could be assigned by a combination of both occupation and industry of employment, followed by occupations only, and followed by industries alone.  4.2.2.4  Linking Health and Exposure Stage  The noise levels in the JEM were assigned to each subject in NPHS based on NOC-S and NAICS codes. The confidence of the assignment was ranked one to six based on the level of NAICS or NOC-S codes available; ranks 1 to 4, indicated that the linkage was based on 4-digit NOC-S and/or NAICS, and were considered to provide better noise estimation compared to ranks 5 and 6 (linkage based on 2-digit NOC-S or NAICS). If a subject’s noise level in each of eight cycles was assigned at least a rank of 4, the sum of the ranks from all cycles was 32, and we used this as a threshold for a sensitivity test. The main analyses were re-run within subjects whose sum of ranks was 32 or less. For the CVD, the direction of calculated ORs for medium and high exposure groups remained the same while the association did become weaker. For diabetes and RA, the direction of ORs did not change but the association became stronger (though still not statistically significant). This result indicates that the different assigned ranks had some influence on the calculated ORs but the effects were negligible to the overall conclusion.  79  4.2.2.5  Estimating Cumulative Noise Exposure Stage  The cumulative noise exposure for each subject was estimated based on exposure level (from the JEM) and exposure duration (from the NPHS). Two issues that influenced the validity of the results were selection of exposure metric and missing values.  4.2.2.5.1  Selection of Exposure Metric  In previous studies on the noise – CVD association (Suadicani et al., 2012; Gan et al., 2011; Gopinath et al., 2011; Selander et al., 2009; Sbihi et al., 2008; McNamee et al., 2006; Babisch et al., 2005; Davies et al., 2005), different exposure metrics were used to estimate people’s noise exposure. In general, there were two major methods, subjective and objective. In the former method, the subjects report whether they think they were exposed to noise and how many years they were exposed. Therefore, the reported information could be regarded as qualitative and less accurate. The objective method measured the exposed noise level of subjects and is able to provide a more accurate estimation. Despite the likely more accurate estimation of noise exposure intensity using objective methods, some argue that if the noise-disease association is mainly based on the stress response and the response is susceptible to personal parameters (e.g. personal susceptibility to noise, attitude to the noise), the subjective method is able to adjust for these parameters and better reflects the stress response (Babisch, 1998). Neitzel et al. (2009) compared the perceived and quantitative measures of occupational noise exposure and found that using the subjective assessment might improve the contrast and precision of epidemiological exposure estimates for workers with highly variable noise exposure. They also suggested combining the subjective and objective estimates to reduce bias and improve precision.  In this study, the cumulative noise exposure combined measured exposure levels from a JEM and exposure duration estimated from self-reported job history. Because there was no question in the NPHS regarding the subjects’ subjective perception of workplace noise, subjective noise exposure was not assessed in our study. 80  The cut points for different exposure groups were selected mainly to ensure enough cases in each exposure group (or the results can not be released). However, in previous studies, increased secretion of stress hormones (e.g. adrenaline, nordrenaline, cortisol) was found in subjects exposed to occupational noise and threshold for the increased secretion was around 60 to 80 dBA (Ising and Braun, 2000; Melamed et al., 1996; Sudo et al., 1996; Cavatorta et al., 1987). This indicated that the cut points in the current study might not be able to distinguish the unexposed group and low exposure group. An analysis was done re-categorizing subjects into a definite unexposed group (<60 dBA, around 20% of all subjects) and a possible or definite exposed group (e.g. “ever/never”). This method was also suggested by Dosemeci and Stewart (1996) to reduce exposure misclassification by placing the subjects whose exposure status was in doubt into the category with the largest number of subjects. For CVD, the results of the multiple regression using this metric showed an OR of 1.2 (95% CI: 0.8 – 1.9) for the possible or definite exposed group though it was not statistically significant. For diabetes, the ORs for the exposed group were 0.9 (95% CI: 0.7 – 2.74) (The noise and RA association became stronger. However, the results can not be released due to limited sample size as the definite unexposed group was only a fraction of the original low exposure group). This crude estimation based on NOC-S codes may have reduced misclassification and revealed a result that was more consistent with previous studies of CVD.  4.2.2.5.2  Missing Values  In order to examine the influence from missing values in the exposure assessment, sensitivity tests were conducted. The proportion of subjects who did not have any missing values in the four variables (MEAN, HOUR, WEEK, TWOYEAR, detail shown in Section 2.8.2) needed for estimating the CNE across the eight cycles over the total subjects for diabetes, RA, and CVD were all around 50%. When estimated through person-year, the proportion of missing exposure period over total exposure period for diabetes, RA, and CVD were 28%, 21%, and 28% respectively. When re-running the multi-variable conditional logistic regression restricted to subjects who did not have any missing values in the four variables across the eight cycles, the direction of the ORs for different exposure groups remained the same, but there was a relatively 81  large change in the magnitude of the ORs (Results can not be released due to small sample size). Given the relatively large number of subjects with missing values affecting exposure assessment and the stronger association with improved exposure assessment, it might be inferred that the missing values may have resulted in a misclassification bias that attenuated observed results.  4.2.3  Hearing Protectors  The issues related to hearing protection devices (HPD) are discussed in the introduction section. In the current study, lack of information on hearing protectors among subject in the NPHS meant that we could not adjust for their influence. Wearing hearing protectors is more likely to influence subjects who had relatively high noise exposures at work. Therefore, without information on hearing protectors, some subjects who worked in noisy environments, and wore HPD and whose real exposure was therefore moderate (and who should have been in the medium exposure group) may have been categorized into the high exposure group. An analysis was conducted merging the medium and high exposure groups into one. For the three health outcomes of interest, the calculated ORs for this new exposure group were very close to one and not statistically significant, indicating the impact of wearing hearing protectors could be mild.  4.2.4  Healthy Worker Effect  The healthy worker effect is another source of bias. Workers with certain disease outcomes may die of the disease or leave their positions due to the impact of the disease on their ability to work. Moreover, people who are susceptible to noise and those who already had some disease symptoms may not be willing to work in a noisy environment (Babisch, 1998). This is more likely to be the case for the outcome of CVD and RA compared to diabetes because CVD may result in more severe outcomes which restrain workers’ daily activities. The healthy worker effect would dilute the magnitude of the effect estimate and bias the estimates towards the null.  82  4.3  4.3.1  Strengths and Limitations  Strengths  There were very few previous studies focusing on the association between noise exposure and the health outcomes of diabetes and RA. Though the biological plausibility was studied and proposed based on some experimental studies, information on the past population-based studies was missing to support the biological mechanism. This study is the first to investigate the direct association between occupational noise and diabetes and RA. Although no significant association was found, it laid the basis and provided some implications for future studies.  Most previous studies focusing on occupational noise selected workers in a certain industry as the study population to form a relatively homogeneous occupational group. This is to some extent supported by Babisch’s (1998) argument on white and blue collar workers. He argued that the working environment could be quite different between white collar workers and blue collar workers. Many factors (e.g. heat, cold, organic solvents) in the workplace were associated with noise exposure and also possibly related to the health outcome. Moreover, white and blue collar workers differ in lifestyle, socio-economic status, etc. All these factors may play a role as potential confounders and therefore, a relatively homogeneous occupational group was preferred to reduce the impact of confounding on the relationship between noise and disease. However, a narrow study population may compromise the generalizability of the study results. Previous studies using the general population to estimate the noise-CVD association found positive results (Gan et al., 2011; Gopinath et al., 2011). Although the generalizablity of the current study was restricted because survey weights were not applied to the findings, various occupations were included in the NPHS covering both white collar workers and blue collar workers. Furthermore, with sufficient information provided on subjects’ lifestyle and socio-economic status, the effects of the potential confounders were adjusted in the regression model.  83  WorkSafeBC’s noise exposure dataset provided noise levels in a wide variety of occupations and industries, which enabled quantitative exposure assessment for subjects in the NPHS. Given the large number of observations in the noise exposure dataset, it provided a relatively reliable estimation of the noise level in certain occupations and industries. Although the distribution of noise samples was skewed towards certain occupations and industries (e.g. manufacturing industries and occupations unique to them), even in those occupations/industries where the number of samples was small (n around 10), the number of samples was still larger than some studies using only very few or even single noise measurement to quantify exposure (Chang et al., 2009a; Chang et al., 2009b; Lee et al., 2009; Selander et al., 2009).  4.3.2  Limitations  Many of the most important limitations have been discussed in the prior section on bias.  Although the NPHS subjects were a representative sample of Canadian population in 1994/1995, the study samples used were subsamples of the original population. Inclusion/exclusion criteria were applied for the purposes of these analyses and cases and controls were selected from the study cohort. Moreover, a large number of NPHS subjects were dropped due to unknown disease status (diabetes: N = 2,308, RA: N = 2,224, CVD: N = 2,303). Comparison between them and the study subjects of the current study (both cases and controls) showed that they were younger and they were more likely to have lower education and income. Therefore, selection bias was possible and may have some impact on the external validity.  Some previous studies found that many other work-related factors may have joint effects with noise on the risk of health outcomes. Co-exposure of noise with air pollutants (e.g. particulate matter, nitrogen dioxide, nitric oxide), organic solvents, and job strain may increase the risk of CVD and we did not have data on these co-exposures (Gan et al., 2012; Beelen et al., 2009; Selander et al., 2009; Chang et al., 2009a; Chang et al., 2009b; Selander et al., 2007). For the health outcome of diabetes and RA, although no previous studies have been conducted to suggest the existence of joint effects, given their similar pathophysiological mechanism and the general 84  noise-disease biological model, it is quite possible that the joint effect of other factors with noise on diabetes and/or RA may exist. However, due to lack of information on relevant factors, joint effects were not considered in this study.  Another limitation in the current study relates to Statistics Canada’s policy to protect participants’ privacy by limiting the reporting of findings based on small cell sizes. Results based on fewer than 15 respondents do not meet the requirement and are not eligible for disclosure. Therefore, some categories were collapsed for the analysis, including for example, the three different classes of obesity were merged into the “obese” category for CVD and RA due to small sample sizes, and the “occasional drinker” and “never drank” categories were collapsed into one for diabetes and CVD. Due to the collapse of categories, the dose-response relationship between health outcome and covariates might be diluted or distorted and this may also influence the effects in the multiple regression model. However, the influence is expected to be mild.  4.4  Future Perspectives  No statistically significant associations were found between occupational noise exposure and the health outcomes of diabetes, RA, and CVD as we hypothesized. Because the quality of the noise exposure assessment in this study is uncertain, and sensitivity tests suggested that improvements to the exposure assessment influenced the strength of association, further studies are needed to assess these associations, especially for diabetes and RA, given the sparse evidence from previous studies. There are some issues that future researchers should consider.  (1) Selection of exposure metric: Given the stress response as the biological basis for the noise-disease association, using a subjective metric does have some advantages in incorporating the personal parameters that can influence the stress response. But its estimates are crude compared to the objective metric, which is able to provide a relatively accurate quantitative assessment of noise exposure. There is no conclusion about which one is better and it deserves further study to investigate how to integrate the two methods.  85  (2) Other sources of noise exposure: Noise exists almost everywhere in people’s daily life. The exposure from workplace only contributes a proportion of the total exposure. Although noise from other sources (e.g. traffic noise, MP3 players) may not be as stable or as high as occupational noise, its contribution should not be ignored especially if it is more frequent.  (3) Joint effects or “side pathways”: There are other factors that may have joint effects with noise on the disease outcome. It is a point worth exploring and its results may contribute to understanding mechanisms of the proposed biological model. Also side pathways may exist beside the stress response model. For example, the noise exposure in the workplace may modify workers’ lifestyles and further influence the health outcome.  (4) Using the general population as study population: Although this study did not find positive associations between occupational noise exposure and the three health outcomes of interest, our attitude about using the general population to study the noise-disease association in future studies is positive, given the strengths of using the general population (better generalizability, sufficient information on potential confounders) and previous positive studies based on general population samples.  Although no positive results were found in this study, considering the biological plausibility of the associations between noise exposure and a variety of well-demonstrated and potential health outcomes (e.g. noise-induced hearing loss, sleep disturbance, CVD, diabetes, RA) and the ubiquitous nature of noise exposure, the relationship between noise and disease outcomes warrants further attention by policy makers, researchers, and other relevant health professionals. Studies should be conducted to assess the effectiveness of hearing conservation programs to see if any improvement can be made to allow them to cost-effectively reach their aims. Relevant policies or bylaws should be reinforced to lower the noise exposure not only in occupational but also in environmental scenarios (traffic noise, community noise, etc). There are also responsibilities of engineers to improve current engineering control methods or design more effective methods to control noise, preferably at the source. 86  4.5  Conclusions  No statistically significant associations were found between occupational noise exposure and the two major health outcomes of interest, diabetes and RA. However, in the main sensitivity analysis, the study on the noise - CVD association, no positive associations were found. There are multiple possible explanations for the negative findings: 1) the negative results could result from study biases such as misclassification in exposure and/or the health outcome; or 2) there was no real association between noise and the health outcomes. It seems more likely that biases diluted the measured association, given the large amount of evidence from previous studies, especially for CVD. As for diabetes and rheumatoid arthritis, given the possible biases in this study and limited evidence from previous studies, future studies are warranted to improve the exposure assessment methods to investigate the hypothesized relationship between noise and the two disease outcomes.  87  Bibliography  Adam V, St-Pierre Y, Fautrel B, Clarke AE, Duffy CM, Penrod JR. (2005). What is the impact of adolescent arthritis and rheumatism? Evidence from a national sample of Canadians. Journal of Rheumatology, 32(2): 354 – 361.  Adegoke OJ, Blair A, Ou SX, Sanderson M, Addy CL, Dosemeci M, et al. (2004). Agreement of job-exposure matrix (JEM) assessed exposure and self-reported exposure among adult leukemia patients and controls in Shanghai. American Journal of Industrial Medicine, 45: 281 – 288.  al’Absi M, Arnett DK. (2000). Adrenocortical responses to psychological stress and risk for hypertension. Biomedicine & Pharmacotherapy, 54(4): 234 – 244.  Andren L, Hansson L, Eggertsen T, Hedner T, Karlberg BE. (1983). Circulatory effects of noise. Acta Medica Scandinavica, 213: 31 – 35.  Arthritis Alliance of Canada. The impact of arthritis in Canada: today and over the next 30 years. (2011). Accessed on Jan 20th 2013, < http://www.arthritisnetwork.ca/downloads/20111022_Impact_of_arthritis.pdf>  Babisch W. (1998). Epidemiological studies of the cardiovascular effects of occupational noise- a critical appraisal. Noise & Health, 1: 24-39.  Babisch W, Beule B, Schust M, Kersten N, Ising H. (2005). Traffic noise and risk of myocardial infarction. Epidemiology, 16: 33 – 40.  Babisch W, van Kamp I. (2009). Exposure-response relationships of the association between aircraft noise and the risk of hypertension. Noise & Health, 11: 161 – 168.  88  Baker M, Stabile M, Deri C. (2001) What do self-reported, objective, measures of health measure? Working Paper 8419. Accessed on Feb 12th 2013, <http://www.nber.org/papers/w8419.pdf?new_window=1>  Beebe-Dimmer J, Lynch J, Turrell G, Lustgarten S, Raghunathan T, Kaplan G. (2004). Childhood and adult socioeconomic conditions and 31-year mortality risk in women. American Journal of Epidemiology, 159(5): 481 – 490.  Berger EH. (1993) The naked truth about NRR's. Cited in Community noise. Document prepared for the World Health Organisation. Berglund B. & Lindvall T. eds. (1995) Archives of the Center for Sensory Research 2, Stockholm, appendix 1, pp5  Brigham J, Lessov-Schlaggar CN, Javitz HS, McElroy M, Krasnow R, Swan GE. (2008). Reliability of adult retrospective recall of lifetime tobacco use. Nicotine & Tobacco Research, 10: 287 – 299.  Canadian Diabetes Association. The prevalence and costs of diabetes. (2013). Accessed on Jan 20th 2013, <http://www.diabetes.ca/diabetes-and-you/what/prevalence/>  Cappuccio FP, D’Elia L, Strazzullo P, Miller MA. (2010). Quantity and quality of sleep and incidence of type 2 diabetes: a systematic review and meta-analysis. Diabetes Care, 33: 414 – 420.  Cavatorta A., Falzoi M., Romanelli A., Cigala F., Ricco M., Bruschi G., Franchini I. and Borghetti A. (1987) Adrenal response in the pathogenesis of arterial hypertension in workers exposed to high noise levels. Journal of Hypertension 5 (Suppl. 5): 463-466.  89  Centers for Diseases Control and Prevention. Rheumatoid Arthritis. (2012). Accessed on Jan 20th 2013, < http://www.cdc.gov/arthritis/basics/rheumatoid.htm#4>  Chang TY, Wang VS, Lin SY, Yen HY, Lai JS, Liu CS. (2009). Co-exposure to noise, N,N-Dimethylformanide, and toluene on 24-hour ambulatory blood pressure in synthetic leather workers. Journal of Occupational and Environmental Hygiene, 7: 14 – 22.  Chang TY, Lai YA, Hsieh HH, Lai JS, L CS. (2009). Effects of environmental noise exposure on ambulatory blood pressure in young adults. Environmental Research, 109: 900 – 905.  Chiu M, Austin PC, Manuel DG, Tu JV. (2010). Comparison of cardiovascular risk profiles among ethnic groups using population health surveys between 1996 and 2007. CMAJ, 182(8): E301 – E310.  Chrousos GP. (2009). Stress and disorders of the stress system. Nature Reviews Endocrinology, 5: 374 – 381.  Chrousos GP. (2000). The role of stress and the hypothalamic-pituitary-adrenal axis in the pathogenesis of the metabolic syndrome: neuro-endocrine and target tissue-related causes. International Journal of Obesity, 24: S50 – S55.  Chrousos GP. (1995). The hypothalamic-pituitary-adrenal axis and immune-mediated inflammation. The New England Journal of Medicine, 332(20): 1351 – 1362.  Cohen, A. (1977). Extraauditory effects of acoustic stimulation. In: Handbook of Physiology-Reactions to Envrionmental Agents. American Physiological Society, Bethesda, Md.  90  Cosgrove MP, Sargeant LA, Caleyachetty R, Griffin SJ. (2012). Work-related stress and type 2 diabetes: systematic review and meta-analysis. Occupational Medicine, 62: 167 – 173.  Coughlin SS. (1990). Recall bias in epidemiological studies. Journal of Clinical Epidemiology, 43: 87 – 91.  Davies HW, Teschke K, Kennedy S, et al. (2005). Occupational exposure to noise and mortality from acute myocardial infarction. Epidemiology, 16: 25-32.  Dinan TG. (1996). Serotonin and the regulation of hypothalamic-pituitary-adrenal axis function. Life Science, 58: 567 – 572.  Dinca-Panaitescu M, Dinca-Panaitescu S, Raphael D, Bryant T, Pilkington B, Daiski I. (2012). The dynamics of the relationship between diabetes incidence and low income: longitudinal results from Canada’s National Population Health Survey. Maturitas, 72: 229 – 235.  Djindjic N, Jovanovic J, Djindjic B, Jovanovic M, Jovanovic JJ. (2012). Associations between the occupational stress index and hypertension, type 2 diabetes mellitus, and lipid disorders in middle-aged men and women. The Annals of Occupational Hygiene, 56(9): 1051 – 1062.  Dosemeci M, Stewart PA. (1996). Recommendations for reducing the effects of exposure misclassification on relative risk estimates. Occupational Hygiene, 3: 169 – 176.  EASHW. (2000). Monitoring the state of occupational safety and health in the European Union—pilot study. Luxembourg.  Engeland WC, Miller P, Gann DS. (1990). Pituitary-adrenal and adrenomedullary responses to noise in awake dogs. American Journal of Physiology, 27: 672 – 677.  91  Eriksson C, Bluhm G, Hilding A, Ostenson C, Pershagen G. (2010). Aircarft noise and incidence of hypertension-gender specific effects. Environmental Research, 110: 764 – 772.  Fisher LD, Tucker DC. (1991). Air jet noise exposure rapidly increases blood pressure in young borderlin hypertensive rats. Journal of Hypertension, 9(3): 275 – 282.  Friedman M, Byers SO, Brown AE. (1967). Plasma lipid response of rats and rabbits to an auditory stimulus. American Journal of Physiology, 212(5): 1174 – 1178.  Gan WQ, Davies HW, Demers PA. (2011). Exposure to occupational noise and cardiovascular disease in the United States: the National Health and Nutrition Examination Survey 1999 – 2004. Occupational and Environmental Medicine, 68: 183 – 190.  Gopinath B, Thiagalingam A, Teber E, Mitchell P. (2011). Exposure to workplace noise and the risk of cardiovascular disease events and mortality among older adults. Preventive Medicine, 53: 390 – 394.  Health Canada. Canadian guidelines for body weight classification in adults. (2005). Accessed on March 10th 2013, <http://www.hc-sc.gc.ca/fn-an/nutrition/weights-poids/guide-ld-adult/qa-qr-pub-eng.php>  Henry JP. (1993). Biological basis of stress response. News in Physiological Science, 8: 69 – 73.  Ising H, Braun C. (2000). Acute and chronic endocrine effects of noise: review of the research conducted at the institute for water, soil and air hygien. Noise & Health, 7: 7 – 24.  Ising H, Kruppa B. (2004). Health effects caused by noise: evidence in the literature from the past 25 years. Noise and Health, 22: 5-13.  92  Johnson-Lawrence V, Kaplan G, Galea S. (2013). Socioeconomic mobility in adulthood and cardiovascular disease mortality. Annals of Epidemiology, 23: 167 – 171.  Kluizenaar Y, Gansevoort RT, Miedema HME. de Jong PE. (2007). Hypertension and road traffic noise exposure. Journal of Occupational and Environmental Medicine, 49(5): 484 – 492.  Kwon JM, Rhee J, Ku H, Lee EK. (2012). Socioeconomic and employment status of patients with rheumatoid arthritis in Korea. Epidemiology and Health, 34: e2012003. doi: 10.4178/epih/e2012003. Accessed on March 26th 2013, <http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3350820/pdf/epih-34-e2012003.pdf>  Lee DS, Chiu M, Manuel DG, et al. (2009). Trends in risk factors for cardiovascular disease in Canada: temporal, socio-demographic and geographic factors. CMAJ, 181(3-4): E55 – 66.  Lee JH, Kang W, Yaang SR, Choy N, Lee CR. (2009). Cohort study for the effect of chronic noise exposure on blood pressure among male workers in Busan, Korea. American Journal of Industrial Medicine, 52: 509 – 517.  Li J, Jarczok MN, Loerbroks A, Schollgen I, Siegrist J, Bosch jA, Wilson MG, Mauss D, Fischer JE. (2012). Work stress is associated with diabetes and prediabetes: cross-sectional results from the MIPH industrial cohort studies. International Society of Behavioral Medicine, online. Accessed on Jan 20th 2013, < http://link.springer.com/content/pdf/10.1007%2Fs12529-012-9255-0>  Li J, Schiottz-Christensen B, Olsen J. (2005). Psychological stress and rheumatoid arthritis in parents after death of a child: a national follow-up study. Scandinavian Journal of Rheumatology, 34: 448 – 450.  93  Lovallo WR. (1997). Stress & health: biological and psychological interactions. Thousand Oaks.  Maschke C, Rupp T, Hecht K. (2000). The influence of streesors on biochemical reactions – a review of present scientific findings with noise. International Journal of Hygiene and Environmental Health, 203: 45 – 53.  Melamed S. and Bruhis S. (1996) The effects of chronic industrial noise exposure on urinary cortisol, fatigue, and irritability. Journal of Occupational and Environmental Medicine, 38: 252-256.  McNamee R, Burgess G, Dippnall WM, Cherry N. (2006). Occupational noise exposure and ischaemic heart disease mortality. Occupational and Environmental Medicine, 63: 813-819.  Millar WJ, Young TK. (2003). Tracking diabetes: prevalence, incidence and risk factors. Health Reports, 14(3): 35 – 47.  Mokdad AH. Ford ES, Bowman BA, Nelson DE, Engelgau MM, Vinicor F, Marks JS. (2000). Diabetes trends in the U.S.: 1990 – 1998. Diabetes Care, 23: 1278 – 1283.  Morata TC, Fiorini AC, Fischer FM, et al. (1997). Toluene-induced hearing loss among rotogravure printing workers. Scandinavian Journal of Work, Environment & Health, 23(4): 289 – 298.  Mukherjee AK, Nag DP, Prakash MN, et al. (1995). Noise level monitoring in a watch factory ion Bangalore. Indian Journal of Industrial Medicine, 41(2): 42 – 44.  Neitzel R, Daniell W, Sheppard L, Davies H, Seixas N. (2009). Comparison of perceived and quantitative measures of occupational noise exposure. Annals of Occupational Hygiene, 53(1): 41 – 54.  94  Nelson DI, Nelson RY, Choncha-Barrientos M, et al. (2005). The global burden of occupational noise-induced hearing loss. American Journal of Industrial Medicine, 48: 446-458.  Neutel CI, Campbell NR, Morrison HI. (2010). Trends in diabetes treatment in Canadians, 1994 – 2004. Chronic Disease in Canada, 30(3): 107 – 111.  Ng E, Dasgupta K, Johnson J. (2008). An algorithm to differentiate diabetic respondents in the Canadian Community Health Survey. Health Reports, 19(1): 71 – 79.  Novak M, Bjorck L, Giang KW, Heden-Stahl C, Wilhelmsen L, Rsengren A. (2012). Perceived stress and incidence of type 2 diabetes: a 35-year follow-up study of middle-aged Swedish men. Diabetic Medicine, 30: e8 – e16. Accessed on Jan 20th 2013, < http://onlinelibrary.wiley.com/doi/10.1111/dme.12037/pdf>  Ohinmaa A, Jacobs P, Simpson S, Johnson JA. (2004). The projection of prevalence and cost of diabetes in Canada: 2000 to 2016. Canadian Journal of Diabetes, 28(2): online. Accessed on Jan 20th 2013, <http://www.diabetes.ca/Files/JohnsonCJDJune2004.pdf>  Oleru UG, Ijaduola GTA, Sowho EE. (1990). Hearing thresholds in an auto assembly plant: Prospects for hearing conservation in an Nigerian factory. International Archives of Occupational and Environmental Health, 62: 199 – 202.  Oliver JE, Silman AJ. (2006). Risk factors for the development of rheumatoid arthritis. Scandinavian Journal of Rheumatology, 35: 169 – 174.  Patten SB, Williams JV, Lavorato DH, Modgill G, Jette N, Eliasziw M. (2008). Major depression as a risk factor for chronic disease incidence: longitudinal analyses in a general population cohort. General Hospital Psychiatry, 30(5): 407 – 413.  95  Peterson ETA, Augenstein JS, Tanis DC, Augenstein DG. (1981). Noise raises blood pressure without impairing auditory sensitivity. Science, 211(4489): 1450 – 1452.  Provincial Health Services Authority. Summary report on health for British Columbia from regional, longitudinal and gender perspectives. (2010). Accessed on Dec 12th 2012, < http://www.llbc.leg.bc.ca/public/pubdocs/bcdocs2010/465409/bchealth_indicators_report.pdf >  Public Health Agency of Canada. Diabetes facts, risk factors. (2012). Accessed on Jan 20th 2013, < http://www.phac-aspc.gc.ca/cd-mc/diabetes-diabete/risk_factors-facteurs_de_risque-eng.php>  Public Health Agency of Canada. Minimizing the risk factors of cardiovascular disease. (2009). Accessed on March 15th 2013, < http://www.phac-aspc.gc.ca/cd-mc/cvd-mcv/risk-risques-eng.php>  Ross NA, Gilmour H, Dasqupata K. (2010). 14-year diabetes incidence: the role of socio-economic status. Health Reports, 21(3): 19 – 28.  Sbihi H, Davies HW, Dmers PA. (2008). Hypertension in noise-exposed sawmill workers: a cohort study. Occupational and Environmental Medicine, 65: 643 – 646.  Sbihi H, Teschke K, Macnab YC, Davies HW. (2010). An investigation of the adjustment of retrospective noise exposure for use of hearing protection device. Annals of Occupational Hygiene, 54: 329 – 339.  Schlaefer K, Schlehofer B, Schuz J. (2009). Validity of self-reported occupational noise exposure. European Journal of Epidemiology, 24(8): 469 – 475.  Schuz J, Spector LG, Ross JA. (2003). Bias in studies of parental self-reported occupational exposure and childhood cancer. American Journal of Epidemiology, 158: 710 – 716.  96  Selander J, Bluhm G, Nilsson M, Hallqvist J, Theorell T, Willix P, Pershagen G. (2007). Joint effects of job strain and road-traffic and occupational noise on myocardial infarction. Scandinavian Journal of Work, Environment & Health, 33(6): 470 – 478. Acceessed on Oct 19 2012, < http://www.sjweh.fi/show_abstract.php?abstract_id=3324>  Selander J, Nilsson ME, Bluhm G, Rosenlund M, Lindqvist M, Nise G, Pershagen G. (2009). Long-term exposure to road traffic noise and myocardial infarction. Epidemiology, 20: 272 – 279.  Selye H. (1973). The Evolution of the Stress Concept: The originator of the concept traces its development from the discovery in 1936 of the alarm reaction to modern therapeutic applications of syntoxic and catatoxic hormones. American Scientist, 61(6): 692 – 699.  Shaikh GH. (1996). Noise problem in a polyester fiber plant in Pakistan. Industrial Health, 34: 427 – 431.  Sjostrom M, Lewne M, Alderling M, Willix P, Berg P, Gustavsson P, Svartengren M. (2013). A job-exposure matrix for occupational noise: development and validation. Annals of Occupational Hygiene, online. Accessed on March 18th 2013, <http://annhyg.oxfordjournals.org/content/early/2013/02/01/annhyg.met001.full.pdf+html>  Sorensen M, Andersen ZJ, Nordsborg RB, Becker T, Tjonneland A, Overvad K, Raaschou-Nielsen O. (2012). Long-term exposure to road traffic noise and incident diabetes; a cohort study. Environmental Health Perspectives, online. Accessed on Jan 20th 2013, < http://ehp.niehs.nih.gov/wp-content/uploads/2012/12/ehp.1205503.pdf>  Statistics Canada. National Population Health Survey Household Component Cycles 1 to 9 (1994/1995 to 2010/2011), longitudinal documentation. (2012). Accessed on Jan 15th 2013, <http://abacus.library.ubc.ca/bitstream/10573/42748/11/Cycles_1_to_9_NPHS_Household_Lon gdoc_E.pdf >  97  Statistics Canada. National Population Health Survey – Household Component Documentation for the Derived Variables and the Constant Longitudinal Variable, Cyles1 to 9. (2012). Accessed on March 10th 2013, <http://www.google.ca/url?sa=t&rct=j&q=national+population+health+survey+documentation+f or+derived+variables&source=web&cd=1&cad=rja&ved=0CDAQFjAA&url=http%3A%2F %2Fabacus.library.ubc.ca%2Fjspui%2Fbitstream%2F10573%2F42748%2F7%2FC_1-9_D V_Doc_E.pdf&ei=bao_Ubj6FsiyqQGC34HYAQ&usg=AFQjCNGnxZ9q_eVhkop87B-Miz0 MdN4UnQ&bvm=bv.43287494,d.aWM>  Statistics Canada. Mortality, summary list of causes. (2007). Accessed on Dec 12th 2012, <http://www.statcan.gc.ca/pub/84f0209x/84f0209x2007000-eng.pdf>  Statistics Canada. Rheumatoid Arthritis. (2006). Accessed on Jan 20th 2013, < http://www.statcan.gc.ca/pub/82-619-m/2006003/4053552-eng.htm>  Sternberg EM, Chrousos GP, Wilder RL, Gold PW. (1992). The stress response and the regulation of inflammatory disease. Annals of Internal Medicine, 117(10): 854 – 866.  Straub RH, Kalden JR. (2009). Stress of different types increases the proinflammatory load in rheumatoid arthritis. Arthritis Research & Therapy, 11: 114.  Suadicani P, Hein HO, Gyntelberg F. (2012). Occupational noise exposure, social class, and risk of ischemic heart disease and all-cause mortality – a 16-year follow-up in the Copenhagen Male Study. Scandinavian Journal of Work, Environment & Health, 38(1): 19 – 26.  Sudo A, Eggertsen R, Magnusson M, Andren L. (1996). Persisting cardiovascular effects of noise exposure in essential hypertension. Journal of Hypertension, 5(suppl5): S459 – S461.  98  Sudo A., Luong N.A., Jonai H., Matsuda S., Villanueva M.B.G., Sotoyama M., Cong N.T., Trinh L.V., Hien H.M., Trong N.D. and Sy N. (1996) Effects of earplugs on catecholamine and cortisol excretion in noise-exposed textile workers. Industrial Health 34: 279-286.  Symmons D, Mathers C, Pfleger B. The global burden of rheumatoid arthritis in the year 2000. (2006). Accessed on Jan 20th 2013, < http://www.who.int/healthinfo/statistics/bod_rheumatoidarthritis.pdf>  Tak S, Davies RR, Calvert GM. (2009). Exposure to hazardous workplace noise and use of hearing protection devices among US workers—NHANES, 1999-2004. American Journal of Industrial Medicine, 52: 358-371.  Tomei G, Fioravanti M, Cerratti D, et al. (2010). Occupational exposure to noise and cardiovascular system: a meta-analysis. Science of the Total Environment, 408: 681 – 689.  Vander AJ, Sherman JH, Luciano DS. (1994). Human physiology. New York: McGraw-Hill.  Virkkunen H, Kauppinen T, Tenkanen L. (2005). Long-term effect of occupational noise on the risk of coronary heart disease. Scandianavian Journal of Work, Environment and Health, 31(4): 291-299.  Wang PP, Elsbett-Koeppen R, Geng GY, Badley EM. (2000). Arthritis prevalence and place of birth: findings from the 1994 Canadian National Population Health Survey. American Journal of Epidemiology, 152(5): 442 – 445.  Westman JC, Walters JR. (1981). Noise and stress: a comprehensive approach. Environmental Health Perspectives, 41: 291 – 309.  99  Wong J, Wong S. (2002). Trends in lifestyle cardiovascular risk factors in women: analysis from the Canadian National Population Health Survey. International Journal of Nursing Studies, 39(2): 229 – 242.  WorkSafeBC. Annual Hearing Test Program, 2003. (2011). Accessed on Dec 12th 2012, < http://www2.worksafebc.com/PDFs/hearing/statistics/category_2011.pdf>  World Health Organization. Cardiovascular Disease. (2013). Accessed on March 26th 2013, < http://www.who.int/mediacentre/factsheets/fs317/en/>  World Health Organization. Diabetes. (2012). Accessed on Jan 20th 2013, < http://www.who.int/mediacentre/factsheets/fs312/en/>  World Health Organization. The Global Burden of Disease, 2004 update. (2004). Accessed on Dec 12th 2012, <http://www.who.int/healthinfo/global_burden_disease/GBD_report_2004update_full.pdf>  100  Appendices Appendix A: Temporal Trend of Noise Levels by Different Occupations / Industries  Figure A. 1  Temporal Trend of Noise Level by Occupations  A – Management Occupations; B – Business, Finance and Administrative Occupations; C – Natural and Applied Sciences and Related Occupations; D – Health Occupations; E – Occupations in Social Science, Education, Government Service and Religion; F –Occupations in Art, Culture, Recreation and Sport; G – Sales and Service Occupations; H – Trades, Transport and Equipment Operators and Related Occupations; I – Occupations Unique to Primary Industry; J – Occupations Unique to Processing, Manufacturing and Utilities.  101  Figure A. 2  Temporal Trend of Noise Levels by Industries  11 – Agriculture, Forestry, Fishing and Hunting; 21 – Mining, Quarrying, and Oil and Gas Extraction; 22 – Utilities; 23 – Construction; 31 / 32 / 33 – Manufacturing; 41 – Wholesale Trade; 44 / 45 – Retail Trade; 48 / 49 – Transportation and Warehousing; 51 – Information and Cultural Industries; 52 – Finance and Insurance; 53 – Real Estate and Rental and Leasing; 54 – Professional, Scientific and Technical Services; 56 – Administrative and Support, Waste Management and Remediation Services; 61 – Educational Services; 62 – Health Care and Social Assistance; 71 – Arts, Entertainment and Recreation; 72 – Accommodation and Food Services; 81 – Other Services (except Public Administration); 91 – Public Administration.  102  Appendix B: The Model Selection Process  Table B. 1  Model Selection Process for Diabetes Unadjusted OR  Adjusted OR (95% CI)  (95% CI)  Model 1  Model 2  1  1  1  Medium-exposure (85 – 95 dBA-year)  0.92 (0.62 – 1.36)  0.90 (0.61 - 1.34)  0.91 (0.60 - 1.38)  High-exposure (>95 dBA-year)  0.94 (0.63 – 1.40)  0.94 (0.63 - 1.41)  1.02 (0.67 - 1.56)  Exposure Low-exposure (< 85 dBA-year)  Obese Status 1  1  Overweight  2.06 (1.27 – 3.34)  2.14 (1.31 - 3.49)  Obese I  4.22 (2.53 – 7.01)  4.24 (2.53 - 7.09)  Obese II  7.72 (4.02 – 14.83)  7.60 (3.91 - 14.75)  Obese III  9.46 (4.39 – 20.36)  8.67 (3.94 - 19.07)  Underweight / Normal Weight  Physical Activity Active  1  Moderate  1.01 (0.65 – 1.58)  Inactive  1.73 (1.17 – 2.54)  Smoke Never Smoked  1  Former Smoker  1.16 (0.81 – 1.66)  Current Smoker  0.97 (0.62 – 1.52)  Drinking Status Occasional Drinker / Never Drank  1  Former Drinker  1.01 (0.62 - 1.65)  Regular Drinker  0.55 (0.38 - 0.81)  Hypertension at Cycle 1 No  1  Yes  2.67 (1.64 – 4.34)  Education 1  1  1  Secondary School Graduation  1.32 (0.78 – 2.26)  1.42 (0.82 - 2.43)  1.50 (0.85 - 2.63)  Some Post-secondary  1.09 (0.68 – 1.72)  1.19 (0.74 - 1.92)  1.33 (0.81 - 2.16)  Post-secondary Graduation  0.71 (0.45 – 1.12)  0.85 (0.53 - 1.36)  0.95 (0.59 - 1.53)  1  1  1  $ 40,000 - $ 79,999  1.02 (0.70 – 1.48)  1.00 (0.68 - 1.46)  1.05 (0.70 - 1.56)  > $ 80,000  0.57 (0.37 – 0.87)  0.61 (0.39 - 0.94)  0.68 (0.43 - 1.08)  Less than Secondary School Graduation  Family Income < $40,000  103  Table B. 1  Model Selection Process for Diabetes (continued) Unadjusted OR  Adjusted OR (95% CI)  (95% CI)  Model 3  Model 4  1  1  1  Medium-exposure (85 – 95 dBA-year)  0.92 (0.62 – 1.36)  0.90 (0.59 - 1.37)  0.90 (0.59 - 1.37)  High-exposure (>95 dBA-year)  0.94 (0.63 – 1.40)  1.02 (0.67 - 1.56)  1.02 (0.67 - 1.56)  1  1  1  Overweight  2.06 (1.27 – 3.34)  2.16 (1.32 - 3.52)  2.18 (1.33 - 3.56)  Obese I  4.22 (2.53 – 7.01)  4.21 (2.52 - 7.05)  4.29 (2.55 - 7.23)  Obese II  7.72 (4.02 – 14.83)  7.53 (3.85 - 14.73)  7.67 (3.90 - 15.08)  Obese III  9.46 (4.39 – 20.36)  8.21 (3.69 - 18.25)  8.33 (3.70 - 18.72)  1  1  1  Moderate  1.01 (0.65 – 1.58)  0.92 (0.58 - 1.47)  0.92 (0.58 - 1.48)  Inactive  1.73 (1.17 – 2.54)  1.44 (0.96 - 2.17)  1.43 (0.95 - 2.16)  Exposure Low-exposure (< 85 dBA-year)  Obese Status Underweight / Normal Weight  Physical Activity Active  Smoke 1  1  Former Smoker  1.16 (0.81 – 1.66)  1.07 (0.73 - 1.57)  Current Smoker  0.97 (0.62 – 1.52)  1.15 (0.71 - 1.87)  Never Smoked  Drinking Status Occasional Drinker / Never Drank  1  Former Drinker  1.01 (0.62 - 1.65)  Regular Drinker  0.55 (0.38 - 0.81)  Hypertension at Cycle 1 No  1  Yes  2.67 (1.64 – 4.34)  Education 1  1  1  Secondary School Graduation  1.32 (0.78 – 2.26)  1.50 (0.85 - 2.64)  1.52 (0.86 - 2.68)  Some Post-secondary  1.09 (0.68 – 1.72)  1.33 (0.82 - 2.17)  1.34 (0.82 - 2.19)  Post-secondary Graduation  0.71 (0.45 – 1.12)  0.98 (0.60 - 1.58)  0.99 (0.61 - 1.61)  1  1  1  $ 40,000 - $ 79,999  1.02 (0.70 – 1.48)  1.13 (0.75 - 1.69)  1.13 (0.76 - 1.71)  > $ 80,000  0.57 (0.37 – 0.87)  0.75 (0.47 - 1.19)  0.75 (0.47 - 1.20)  Less than Secondary School Graduation  Family Income < $40,000  104  Table B. 1  Model Selection Process for Diabetes (continued) Unadjusted OR  Adjusted OR (95% CI)  (95% CI)  Model 5  Model 6  1  1  1  Medium-exposure (85 – 95 dBA-year)  0.92 (0.62 – 1.36)  0.91 (0.60 - 1.39)  0.93 (0.61 - 1.41)  High-exposure (>95 dBA-year)  0.94 (0.63 – 1.40)  1.03 (0.67 - 1.58)  1.04 (0.67 - 1.59)  1  1  1  Overweight  2.06 (1.27 – 3.34)  2.16 (1.32 - 3.54)  2.14 (1.31 - 3.51)  Obese I  4.22 (2.53 – 7.01)  4.17 (2.48 - 7.02)  4.03 (2.39 - 6.80)  Obese II  7.72 (4.02 – 14.83)  7.31 (3.71 - 14.43)  6.88 (3.47 - 13.67)  Obese III  9.46 (4.39 – 20.36)  7.68 (3.40 - 17.31)  7.4 (3.26 - 16.80)  1  1  1  Moderate  1.01 (0.65 – 1.58)  0.92 (0.58 - 1.48)  0.93 (0.58 - 1.49)  Inactive  1.73 (1.17 – 2.54)  1.41 (0.94 - 2.12)  1.37 (0.91 - 2.06)  1  1  1  Former Smoker  1.16 (0.81 – 1.66)  1.10 (0.75 - 1.61)  1.09 (0.74 - 1.59)  Current Smoker  0.97 (0.62 – 1.52)  1.19 (0.73 - 1.94)  1.22 (0.75 - 1.99)  1  1  1  Former Drinker  1.01 (0.62 - 1.65)  0.99 (0.59 - 1.66)  1.01 (0.60 - 1.71)  Regular Drinker  0.55 (0.38 - 0.81)  0.71 (0.46 - 1.08)  0.74 (0.48 - 1.14)  Exposure Low-exposure (< 85 dBA-year)  Obese Status Underweight / Normal Weight  Physical Activity Active  Smoke Never Smoked  Drinking Status Occasional Drinker / Never Drank  Hypertension at Cycle 1 No  1  1  Yes  2.67 (1.64 – 4.34)  1.62 (0.94 - 2.78)  Education 1  1  1  Secondary School Graduation  1.32 (0.78 – 2.26)  1.56 (0.88 - 2.77)  1.60 (0.90 - 2.84)  Some Post-secondary  1.09 (0.68 – 1.72)  1.38 (0.84 - 2.25)  1.37 (0.84 - 2.25)  Post-secondary Graduation  0.71 (0.45 – 1.12)  1.03 (0.63 - 1.68)  1.05 (0.64 - 1.72)  1  1  1  $ 40,000 - $ 79,999  1.02 (0.70 – 1.48)  1.22 (0.81 - 1.85)  1.21 (0.80 - 1.84)  > $ 80,000  0.57 (0.37 – 0.87)  0.84 (0.52 - 1.37)  0.86 (0.53 - 1.39)  Less than Secondary School Graduation  Family Income < $40,000  105  Table B. 2  Model Selection Process for Rheumatoid Arthritis Unadjusted OR  Adjusted OR (95% CI)  (95% CI)  Model 1  Model 2  1  1  1  0.95 (0.54 – 1.68)  1.01 (0.57 - 1.82)  1.01 (0.56 - 1.81)  0.93 (0.51 – 1.70)  1.05 (0.56 - 1.96)  1.03 (0.55 - 1.94)  Exposure Low-exposure (< 85 dBA-year) Medium-exposure High-exposure  (85 – 95 dBA-year)  (>95 dBA-year)  Obese Status Underweight / Normal Weight  1  Overweight  1.09 (0.63 – 1.86)  Obese  1.08 (0.60 - 1.96)  Physical Activity Active  1  Moderate  0.57 (0.30 - 1.08)  Inactive  0.91 (0.54 - 1.52)  Smoke 1  1  Former Smoker  1.67 (0.87 - 3.21)  1.04 (0.59 - 1.82)  Current Smoker  1.17 (0.67 - 2.01)  1.44 (0.73 - 2.83)  Never Smoked  Education 1  1  1  Some Post-secondary  1.53 (0.84 - 2.77)  1.71 (0.93 - 3.13)  1.76 (0.96 - 3.24)  Post-secondary Graduation  0.74 (0.41 - 1.31)  0.85 (0.47 - 1.55)  0.90 (0.49 - 1.64)  1  1  1  $ 40,000 - $ 79,999  0.90 (0.49 - 1.62)  0.97 (0.53 - 1.78)  0.99 (0.54 - 1.82)  > $ 80,000  0.48 (0.26 - 0.90)  0.51 (0.27 - 0.97)  0.53 (0.28 - 1.00)  Secondary School Graduation or Less  Family Income < $40,000  106  Table B. 2  Model Selection Process for Rheumatoid Arthritis (continued) Unadjusted OR  Adjusted OR (95% CI)  (95% CI)  Model 3  Model 4  1  1  1  0.95 (0.54 – 1.68)  1.01 (0.56 - 1.82)  1.01 (0.56 - 1.82)  0.93 (0.51 – 1.70)  1.04 (0.55 - 1.94)  1.07 (0.57 - 2.01)  1  1  1  Overweight  1.09 (0.63 – 1.86)  1.11 (0.64 - 1.92)  1.09 (0.63 - 1.90)  Obese  1.08 (0.60 - 1.96)  1.09 (0.59 - 2.00)  1.11 (0.59 - 2.08)  Exposure Low-exposure (< 85 dBA-year) Medium-exposure High-exposure  (85 – 95 dBA-year)  (>95 dBA-year)  Obese Status Underweight / Normal Weight  Physical Activity 1  1  Moderate  0.57 (0.30 - 1.08)  0.48 (0.25 - 0.93)  Inactive  0.91 (0.54 - 1.52)  0.76 (0.44 - 1.32)  Active  Smoke 1  1  1  Former Smoker  1.67 (0.87 - 3.21)  1.03 (0.59 - 1.81)  1.00 (0.56 - 1.78)  Current Smoker  1.17 (0.67 - 2.01)  1.43 (0.73 - 2.83)  1.48 (0.75 - 2.94)  1  1  1  Some Post-secondary  1.53 (0.84 - 2.77)  1.77 (0.96 - 3.25)  1.86 (1.01 - 3.44)  Post-secondary Graduation  0.74 (0.41 - 1.31)  0.89 (0.49 - 1.63)  0.91 (0.50 - 1.66)  1  1  1  $ 40,000 - $ 79,999  0.90 (0.49 - 1.62)  0.99 (0.54 - 1.82)  0.98 (0.53 - 1.80)  > $ 80,000  0.48 (0.26 - 0.90)  0.53 (0.28 - 1.01)  0.50 (0.26 - 0.96)  Never Smoked  Education Secondary School Graduation or Less  Family Income < $40,000  107  Table B. 3  Model Selection Process for Cardiovascular Disease Unadjusted OR  Adjusted OR (95% CI)  (95% CI)  Model 1  Model 2  1  1  1  Medium-exposure (85 – 95 dBA-year)  0.87 (0.61 – 1.26)  0.87 (0.60 - 1.27)  0.86 (0.59 - 1.24)  High-exposure (>95 dBA-year)  0.77 (0.53 – 1.14)  0.80 (0.54 - 1.18)  0.80 (0.54 - 1.18)  Exposure Low-exposure (< 85 dBA-year)  Obesity Status 1  1  Overweight  1.08 (0.75 - 1.55)  1.10 (0.76 - 1.59)  Obese  1.39 (0.95 - 2.04)  1.35 (0.92 - 2.01)  Underweight / Normal Weight  Physical Activity Active  1  Moderate  0.71 (0.47 – 1.08)  Inactive  1.05 (0.73 – 1.50)  Smoke Never Smoked  1  Former Smoker  1.76 (1.21 – 2.56)  Current Smoker  1.40 (0.87 – 2.24)  Drinking Status Occasional / Never Drank  1  Former Drinker  1.08 (0.69 - 1.70)  Regular Drinker  0.64 (0.45 - 0.92)  Hypertension at Cycle 1 No  1  Yes  2.73 (1.71 – 4.36)  Education 1  1  1  Secondary School Graduation  0.80 (0.47 – 1.34)  0.85 (0.50 - 1.44)  0.88 (0.52 - 1.49)  Some Post-secondary  0.89 (0.57 – 1.40)  1.00 (0.63 - 1.58)  1.03 (0.65 - 1.63)  Post-secondary Graduation  0.67 (0.44 – 1.02)  0.84 (0.54 - 1.31)  0.86 (0.55 - 1.34)  1  1  1  $ 40,000 - $ 79,999  0.87 (0.61 – 1.24)  0.89 (0.62 - 1.28)  0.89 (0.62 - 1.28)  > $ 80,000  0.46 (0.31 – 0.68)  0.48 (0.32 - 0.74)  0.49 (0.32 - 0.75)  Less than Secondary School Graduation  Family Income < $40,000  108  Table B. 3  Model Selection Process for Cardiovascular Disease (continued) Unadjusted OR  Adjusted OR (95% CI)  (95% CI)  Model 3  Model 4  1  1  1  Medium-exposure (85 – 95 dBA-year)  0.87 (0.61 – 1.26)  0.86 (0.59 - 1.24)  0.85 (0.59 - 1.24)  High-exposure (>95 dBA-year)  0.77 (0.53 – 1.14)  0.80 (0.54 - 1.18)  0.80 (0.54 - 1.19)  1  1  1  Overweight  1.08 (0.75 - 1.55)  1.13 (0.78 - 1.63)  1.09 (0.75 - 1.58)  Obese  1.39 (0.95 - 2.04)  1.34 (0.90 - 1.99)  1.29 (0.86 - 1.92)  1  1  1  Moderate  0.71 (0.47 – 1.08)  0.69 (0.45 - 1.06)  0.69 (0.45 - 1.05)  Inactive  1.05 (0.73 – 1.50)  0.93 (0.64 - 1.36)  0.95 (0.65 - 1.39)  Exposure Low-exposure (< 85 dBA-year)  Obesity Status Underweight / Normal Weight  Physical Activity Active  Smoke 1  1  Former Smoker  1.76 (1.21 – 2.56)  1.73 (1.19 - 2.54)  Current Smoker  1.40 (0.87 – 2.24)  1.23 (0.76 - 2.00)  Never Smoked  Drinking Status Occasional / Never Drank  1  Former Drinker  1.08 (0.69 - 1.70)  Regular Drinker  0.64 (0.45 - 0.92)  Hypertension at Cycle 1 No  1  Yes  2.73 (1.71 – 4.36)  Education 1  1  1  Secondary School Graduation  0.80 (0.47 – 1.34)  0.89 (0.53 - 1.51)  0.95 (0.56 - 1.61)  Some Post-secondary  0.89 (0.57 – 1.40)  1.05 (0.66 - 1.67)  1.07 (0.67 (1.70)  Post-secondary Graduation  0.67 (0.44 – 1.02)  0.88 (0.56 - 1.38)  0.92 (0.59 - 1.46)  1  1  1  $ 40,000 - $ 79,999  0.87 (0.61 – 1.24)  0.89 (0.62 - 1.28)  0.87 (0.60 - 1.26)  > $ 80,000  0.46 (0.31 – 0.68)  0.49 (0.32 - 0.75)  0.47 (0.31 - 0.72)  Less than Secondary School Graduation  Family Income < $40,000  109  Table B. 3  Model Selection Process for Cardiovascular Disease (continued) Unadjusted OR  Adjusted OR (95% CI)  (95% CI)  Model 5  Model 6  1  1  1  Medium-exposure (85 – 95 dBA-year)  0.87 (0.61 – 1.26)  0.89 (0.61 - 1.30)  0.87 (0.59 - 1.28)  High-exposure (>95 dBA-year)  0.77 (0.53 – 1.14)  0.83 (0.56 - 1.24)  0.85 (0.57 - 1.27)  1  1  1  Overweight  1.08 (0.75 - 1.55)  1.12 (0.77 - 1.63)  1.12 (0.77 - 1.63)  Obese  1.39 (0.95 - 2.04)  1.28 (0.86 - 1.90)  1.18 (0.79 - 1.78)  1  1  1  Moderate  0.71 (0.47 – 1.08)  0.67 (0.44 - 1.03)  0.65 (0.42 - 1.01)  Inactive  1.05 (0.73 – 1.50)  0.91 (0.62 - 1.34)  0.91 (0.62 - 1.34)  1  1  1  Former Smoker  1.76 (1.21 – 2.56)  1.81 (1.23 - 2.66)  1.81 (1.23 - 2.67)  Current Smoker  1.40 (0.87 – 2.24)  1.29 (0.79 - 2.11)  1.30 (0.80 - 2.13)  1  1  1  Former Drinker  1.08 (0.69 - 1.70)  1.04 (0.66 - 1.66)  1.03 (0.64 - 1.65)  Regular Drinker  0.64 (0.45 - 0.92)  0.68 (0.47 - 1.00)  0.68 (0.46 - 1.00)  Exposure Low-exposure (< 85 dBA-year)  Obesity Status Underweight / Normal Weight  Physical Activity Active  Smoke Never Smoked  Drinking Status Occasional / Never Drank  Hypertension at Cycle 1 No  1  1  Yes  2.73 (1.71 – 4.36)  2.86 (1.74 - 4.69)  Education 1  1  1  Secondary School Graduation  0.80 (0.47 – 1.34)  0.99 (0.58 - 1.69)  1.00 (0.59 - 1.71)  Some Post-secondary  0.89 (0.57 – 1.40)  1.11 (0.70 - 1.78)  1.12 (0.70 - 1.80)  Post-secondary Graduation  0.67 (0.44 – 1.02)  0.97 (0.61 - 1.54)  0.96 (0.60 - 1.52)  1  1  1  $ 40,000 - $ 79,999  0.87 (0.61 – 1.24)  0.91 (0.63 - 1.33)  0.95 (0.65 - 1.39)  > $ 80,000  0.46 (0.31 – 0.68)  0.52 (0.33 - 0.80)  0.52 (0.34 - 0.81)  Less than Secondary School Graduation  Family Income < $40,000  110  

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.24.1-0071919/manifest

Comment

Related Items