UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Exploring spatio-temporal patterns and environmental determinants of pediatric Inflammatory Bowel Disease… Michaux, Mielle 2020

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

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata

Download

Media
24-ubc_2020_november_michaux_mielle.pdf [ 14.69MB ]
Metadata
JSON: 24-1.0394067.json
JSON-LD: 24-1.0394067-ld.json
RDF/XML (Pretty): 24-1.0394067-rdf.xml
RDF/JSON: 24-1.0394067-rdf.json
Turtle: 24-1.0394067-turtle.txt
N-Triples: 24-1.0394067-rdf-ntriples.txt
Original Record: 24-1.0394067-source.json
Full Text
24-1.0394067-fulltext.txt
Citation
24-1.0394067.ris

Full Text

EXPLORING SPATIO-TEMPORAL PATTERNS AND ENVIRONMENTAL DETERMINANTS OF PEDIATRIC INFLAMMATORY BOWEL DISEASE IN BRITISH COLUMBIA by  Mielle Michaux  B.A., The University of British Columbia, 2018  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  MASTER OF SCIENCE in THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES  (Geography)  THE UNIVERSITY OF BRITISH COLUMBIA  (Vancouver)  August 2020  © Mielle Michaux, 2020 ii  The following individuals certify that they have read, and recommend to the Faculty of Graduate and Postdoctoral Studies for acceptance, a thesis/dissertation entitled:  Exploring spatio-temporal patterns and environmental determinants of pediatric Inflammatory Bowel Disease in British Columbia  submitted by Mielle Michaux in partial fulfillment of the requirements for the degree of Master of Science in Geography  Examining Committee: Dr. Brian Klinkenberg, Geography Co-supervisor Dr. Luke Bergmann, Geography Co-supervisor  Dr. Kevan Jacobson, Pediatrics Supervisory Committee Member Dr. Nina Hewitt, Geography Additional Examiner  iii  Abstract Canada has some of the highest rates of pediatric Inflammatory Bowel Diseases (IBD), including Crohn’s disease (CD) and ulcerative colitis (UC), in the world. Environmental factors are known to be important for disease development but are not well understood. This study used two forms of analysis to examine the epidemiology and potential causes of IBD diagnosed before age 17 in the Canadian province of British Columbia from 2001 to 2016. A spatial cluster detection methodology was used to locate disease clusters of high and low incidence rates, the presence of which would highlight potential environmental risk and protective factors. Logistic regression models of case-control data were used to measure the relationship between IBD diagnosis and NO2 air pollution, density of residential and neighborhood vegetation greenness (green spaces), vitamin D adjusted ultraviolet solar radiation, area South Asian and Jewish ethnicity, area self-identification as Aboriginal, and area social and material deprivation. The spatial distributions of IBD, CD, and UC were significantly clustered, with consistent IBD hot spots identified near the main urban centre of the province and cold spots identified in rural areas of south-eastern British Columbia. CD and UC had similar and different hot and cold spots, suggesting both shared and distinct environmental determinants. Most measured associations between variables of interest and IBD were moderate or small; as IBD is a multifactorial disease, these variables may still have a population-level effect on disease risk or interact with other risk factors and should be studied further. NO2 air pollution was a significant risk factor for UC. Area South Asian ethnicity was only a significant risk factor in the univariate analysis, though a small and similar effect was observed in the multivariate analysis which included social and material deprivation. Ultraviolet vitamin D exposure was a protective factor for UC and IBD, especially in winter months. Area Aboriginal identity and area material deprivation (areas with lower iv  socioeconomic status) were significant protective factors for CD, though Aboriginal identity was not significant in a multivariate analysis that included social and material deprivation. No reliable relationship was observed for greenness or area Jewish ethnicity. v  Lay Summary Children in Canada develop Inflammatory Bowel Diseases (IBD) more often than children in other parts of the world. IBD can limit a person’s quality of life, and there is no cure. Characteristics of an individual’s environment, neighborhood, and family could increase or decrease their risk of developing IBD. I examined where children who are diagnosed with IBD tend to live and the characteristics of their home location that might cause or prevent IBD. There were certain locations, often near large cities, that tended to have more IBD patients than they should. Children living in areas with more air pollution and a larger percent of South Asian residents had a higher chance of developing IBD. Conversely, people living in areas with more vitamin D from sunlight, a larger percent of residents who identified as Aboriginal, and lower income and education levels had a lower chance of developing IBD. vi  Preface Clinical data for Inflammatory Bowel Disease patients used in this study was provided by Dr. Kevan Jacobson and Justin Chan from a registry maintained by the Division of Gastroenterology at British Columbia Children’s Hospital. Research activities were planned, conducted, and analyzed by Mielle Michaux with the advice and guidance of her graduate supervisors Dr. Brian Klinkenberg and Dr. Luke Bergmann, committee member Dr. Kevan Jacobson, and Justin Chan.  Ethics approval for this research was obtained from the University of British Columbia Children and Women’s Research Ethics Board (certificate number H19-00739).   Postal code data were provided by DMTI Spatial Inc, via the Canadian Urban Environmental Health Research Consortium (CANUE) under the current SMART Agreement in place with Canadian Universities. Nitrogen dioxide data were indexed to DMTI Spatial Inc. postal codes and were provided by CANUE. Long-term monthly UV data were accessed via the CANUE data portal (https://canuedata.ca). NDVI metrics, indexed to DMTI Spatial Inc. postal codes, were provided by CANUE. Landsat 5 and Landsat 8 TOA data and greenest pixel data were provided by the U.S. Geological Survey and Google respectively, both via Google Earth Engine. Material and Social Deprivation Indices (MSDI), indexed to DMTI Spatial Inc. postal codes, were provided by CANUE. MSDI used by CANUE were provided by: Institut National de Santé Publique du Québec (INSPQ). Indices were compiled for 1991, 1996, 2001 and 2011 Census data by the Bureau d'information et d'études en santé des populations (BIESP).  vii  Table of Contents  Abstract ......................................................................................................................................... iii Lay Summary .................................................................................................................................v Preface ........................................................................................................................................... vi Table of Contents ........................................................................................................................ vii List of Tables ............................................................................................................................... xii List of Figures ............................................................................................................................. xiii List of Abbreviations ................................................................................................................. xiv Acknowledgements .................................................................................................................... xvi Chapter 1: Introduction ................................................................................................................1 1.1 Inflammatory Bowel Disease .......................................................................................... 1 1.2 Literature review ............................................................................................................. 2 1.2.1 Global distribution of IBD incidence ...................................................................... 2 1.2.2 Causes of IBD ......................................................................................................... 3 1.2.3 Spatial cluster analysis and IBD ............................................................................. 5 1.2.3.1 Spatial cluster analysis and spatial autocorrelation ............................................ 5 1.2.3.2 Selected forms of spatial cluster analysis ........................................................... 5 1.2.3.3 Local Indicators of Spatial Association .............................................................. 7 1.2.3.4 Spatial analysis of IBD ....................................................................................... 8 1.2.3.5 Spatial cluster analysis of IBD ............................................................................ 9 1.2.4 Spatial methods for measuring associations between IBD and risk factors ........... 9 1.2.5 Physical environment and IBD ............................................................................. 11 viii  1.2.5.1 Latitude, sunlight, and vitamin D ..................................................................... 11 1.2.5.2 Air pollution and nitrogen dioxide .................................................................... 13 1.2.6 Socioeconomic status, ethnicity, immigration, urban residence, and IBD ........... 14 1.2.6.1 Socioeconomic status ........................................................................................ 14 1.2.6.2 Ethnicity ............................................................................................................ 14 1.2.6.3 Trends in immigrants to Canada ....................................................................... 15 1.2.6.4 Urban residence ................................................................................................ 16 1.2.7 Greenness and health status .................................................................................. 16 1.2.7.1 Green spaces, greenness, and health ................................................................. 16 1.2.7.2 Measuring greenness ......................................................................................... 17 1.2.8 Key ideas resulting from review of literature ....................................................... 19 1.3 Research objectives ....................................................................................................... 20 1.4 Thesis structure ............................................................................................................. 21 Chapter 2: Data ............................................................................................................................23 2.1 Patient data .................................................................................................................... 23 2.2 Canadian Urban Environmental Health Research Consortium data ............................. 23 2.2.1 Airborne NO2 concentration ................................................................................. 24 2.2.2 Vitamin D weighted UV ....................................................................................... 25 2.2.3 NVDI greenness data ............................................................................................ 26 2.2.4 Material and social deprivation ............................................................................. 28 2.3 Census data: ethnicity, Aboriginal identity, age, and population ................................. 28 Chapter 3: Spatial analysis .........................................................................................................31 3.1 Introduction ................................................................................................................... 31 ix  3.2 Data and methods .......................................................................................................... 32 3.2.1 Data ....................................................................................................................... 32 3.2.1.1 Clinical data ...................................................................................................... 32 3.2.1.2 Population and geographic data ........................................................................ 33 3.2.2 Methods................................................................................................................. 36 3.2.2.1 Spatial methods ................................................................................................. 36 3.3 Results ........................................................................................................................... 38 3.3.1 Incidence ............................................................................................................... 38 3.3.1.1 Incidence Rates by diagnosis ............................................................................ 38 3.3.1.2 Incidence Rates by age group ........................................................................... 39 3.3.1.3 Incidence Rates by Health Authority ................................................................ 41 3.3.2 Spatial cluster analysis .......................................................................................... 42 3.3.2.1 Provincial spatial patterning ............................................................................. 42 3.3.2.2 Local spatial patterning ..................................................................................... 44 3.4 Discussion ..................................................................................................................... 51 Chapter 4: Modelling of potential risk and protective factors ................................................55 4.1 Introduction ................................................................................................................... 55 4.2 Data and methods .......................................................................................................... 55 4.2.1 Data ....................................................................................................................... 55 4.2.1.1 Patient data ........................................................................................................ 56 4.2.1.2 Physical environment, ethnicity, and socioeconomic status data ..................... 57 4.2.1.3 Associating patients with physical environment and socioeconomic data ....... 58 4.2.2 Methods................................................................................................................. 58 x  4.2.3 Models................................................................................................................... 60 4.3 Results ........................................................................................................................... 61 4.3.1 Assessing models with Tjur’s R2 ......................................................................... 61 4.3.2 Assessing spatial autocorrelation of model residuals with Moran’s “I” ............... 62 4.3.3 Interpreting model outputs: Odds ratios and statistical significance .................... 63 4.3.4 Physical environment ............................................................................................ 64 4.3.4.1 NO2 pollution ................................................................................................... 64 4.3.4.2 Vitamin D UV ................................................................................................... 65 4.3.4.3 Greenness .......................................................................................................... 65 4.3.5 Ethnicity ................................................................................................................ 66 4.3.5.1 Area South Asian ethnicity ............................................................................... 66 4.3.5.2 Area Jewish ethnicity ........................................................................................ 66 4.3.5.3 Area Aboriginal identity ................................................................................... 66 4.3.6 Socioeconomic status ............................................................................................ 67 4.3.6.1 Material and social deprivation ......................................................................... 67 4.4 Discussion ..................................................................................................................... 68 4.4.1 Physical environment ............................................................................................ 68 4.4.1.1 NO2 pollution ................................................................................................... 68 4.4.1.2 Vitamin D UV ................................................................................................... 68 4.4.1.3 Greenness .......................................................................................................... 70 4.4.2 Ethnicity ................................................................................................................ 70 4.4.2.1 Area South Asian ethnicity ............................................................................... 70 4.4.2.2 Area Jewish ethnicity ........................................................................................ 71 xi  4.4.2.3 Area Aboriginal identity ................................................................................... 71 4.4.3 Socioeconomic status ............................................................................................ 73 4.4.3.1 Material and social deprivation ......................................................................... 73 4.5 Conclusion .................................................................................................................... 74 Chapter 5: Conclusion .................................................................................................................77 References .....................................................................................................................................80 Appendices ....................................................................................................................................99 Appendix A ............................................................................................................................... 99 A.1 Formulas ................................................................................................................... 99 A.2 IBD incidence rate maps ......................................................................................... 100 Appendix B ............................................................................................................................. 107 B.1 Formulas ................................................................................................................. 107 B.2 Data maps ................................................................................................................ 108 B.3 Regression results ................................................................................................... 116  xii  List of Tables  Table 3.1 Incidence rate per 100,000 for IBD and by disease type .............................................. 39 Table 3.2 Incidence rate per 100,000 by age group ...................................................................... 39 Table 3.3 Incidence per 100,000 by Health Authority .................................................................. 41 Table 3.4 BC Moran’s I spatial cluster analysis results ................................................................ 42 Table 4.1 Predictor variables used in regression modelling ......................................................... 56 Table 4.2 Multivariate regression models ..................................................................................... 60 Table 4.3 BC regression analyses ................................................................................................. 60 Table 4.4 Fraser Health and urban (Fraser Health and Vancouver Coastal Health) analyses ...... 60 Table 4.5 Summary of selected modelling results (statistically significant results in bold) ......... 64  xiii  List of Figures  Figure 1.1 Study area: province of British Columbia (Teucher, Hazlitt, Albers, & Province of British Columbia, 2017)................................................................................................................ 20 Figure 3.1 Health Authorities of BC ............................................................................................. 35 Figure 3.2 Local Health Areas of BC ........................................................................................... 35 Figure 3.3 Incidence rates per 100,000 by disease type ............................................................... 38 Figure 3.4 Incidence rate per 100,000 by Health Authority ......................................................... 41 Figure 3.5 Selected LHAs of interest, Lower Mainland ............................................................... 44 Figure 3.6 Selected LHAs of interest, Vancouver Island ............................................................. 45 Figure 3.7 Smoothed IBD incidence per 100,000 for 2001 – 2005, 2006 – 2010, and 2011 - 2016....................................................................................................................................................... 46 Figure 3.8 Hot and cold spots for 2001 – 2005, 2006 – 2010, and 2011 – 2016 .......................... 47 Figure 3.9 Smoothed incidence per 100,000 for IBD, CD, and UC, 2001 – 2016 ....................... 48 Figure 3.10 Hot and cold spots for IBD, CD, and UC SIR, 2001 – 2016 ..................................... 49 Figure 4.1 Percent of the population in each LHA who identified as Aboriginal in the 2016 census (Provincial Health Services Authority, n.d.) ..................................................................... 72 Figure 4.2 Population density per square kilometre as of the 2016 census (Provincial Health Services Authority, n.d.) ............................................................................................................... 73  xiv  List of Abbreviations  BC  British Columbia BCCH  British Columbia Children’s Hospital CANUE  Canadian Urban Environmental Health Research Consortium CD  Crohn’s Disease CI   Confidence Interval GI  Gastrointestinal HA  Health Authority IBD  Inflammatory Bowel Disease IBD-U  Inflammatory Bowel Disease-unclassified LHA  Local Health Area LISA  Local Indicator of Spatial Association NDVI  Normalized Difference Vegetation Index NIR  Near-Infrared Radiation NO2  Nitrogen Dioxide OR  Odds Ratio PM  Particulate Matter RED  Visible Red Light SIR  Standardized Incidence Ratio UC  Ulcerative Colitis UK  United Kingdom UV  Ultraviolet xv  UVR  Ultraviolet Radiation xvi  Acknowledgements   I want to thank my supervisors, Dr. Brian Klinkenberg and Dr. Luke Bergmann, for their continued support of my research and learning process; thank you for sharing your wisdom with me. This project would also not have been possible without valuable input from Dr. Kevan Jacobson and continued guidance from Justin Chan.  This research was supported by a Graduate Support Initiative Award and International Tuition Award from the Faculty of Arts, and by funding from Dr. Brian Klinkenberg and Dr. Kevan Jacobson. I would like to acknowledge the entire UBC Geography community of students, staff, and faculty (particularly Professor Sally Hermansen and Dr. Brian Klinkenberg) who have shaped me as a student for the past seven years. Suzanne Lawrence deserves credit/blame for initially suggesting this to me.  A very special thank you to my lab mates in the Lab for Advanced Spatial Analysis for their warm friendship and advice, including Emily Acheson, Peter Whitman, Micheal Jerowsky, Kevin Hu, and Mollie Holmberg. You have all made this experience (and my research) unquantifiably better. Thank you to my communities in Vancouver and Seattle for their friendship and especially for keeping me tethered to something outside my work. Kristian Castañeda deserves particular recognition for selflessly encouraging me every single day.   Finally, I would like to thank my family (particularly my parents) for their unwavering support, encouragement, patience, and inspiration. Truly, this could not have happened without you two. 1  Chapter 1: Introduction  1.1 Inflammatory Bowel Disease Chronic inflammatory disorders are a serious emerging health challenge in high-income countries, particularly in urban areas (Rook, 2013). Canada is no exception, with Canadian children having some of the highest global rates of Inflammatory Bowel Disease (IBD) (Carroll et al., 2016). IBD is a heterogenous disorder characterized by chronic inflammation of the gastrointestinal (GI) tract. Most patients with IBD have either Crohn’s disease (CD) or ulcerative colitis (UC), though a small percentage of patients do not meet the criteria for CD or UC (Xia, Crusius, Meuwissen, & Peña, 1998). Up to one quarter of all IBD cases are established before adulthood (Benchimol et al., 2011; Diefenbach & Breuer, 2006), and there is evidence that rates of pediatric-onset IBD are increasing faster than in other age groups (Abraham & Cho, 2009; Diefenbach & Breuer, 2006). In addition, pediatric cases present during an important time of growth and development, and the disease tends to be extensive (Kolho et al., 2015). There are likely multiple immune pathways that lead to the development of IBD, where some cases are characterized by lowered immune function and others by an abnormal proinflammatory immune response (Peloquin, Goel, Villablanca, & Xavier, 2016). IBD is complex, currently understood to originate from a combination of a baseline genetic susceptibility, immune dysregulation, and environmental initiators (Kapoor, Bhatia, & Sibal, 2016; Peloquin et al., 2016; Ramos & Papadakis, 2019). Environment is often defined quite broadly in the literature, and here I define environmental risk factors as anything other than an individual’s genome or immune system that influences the likelihood of developing IBD (e.g. diet, air quality, or socioeconomic status). I 2  will use the term physical environment to refer to characteristics of the environment such as air quality, water, sunlight, and green spaces.   1.2 Literature review 1.2.1 Global distribution of IBD incidence   Higher incidence of pediatric IBD has been documented in Europe (especially Scandinavia) and Canada (Benchimol et al., 2011), and rates of IBD are increasing in developed countries and countries undergoing intensified urbanization and development (Benchimol, Kaplan, et al., 2017; Ramos & Papadakis, 2019). Reported IBD incidence for most countries and regions has been under 4.4 per 100,000, while for some areas of Canada and Scandinavia measured incidence has been almost three times as high (Benchimol et al., 2011). In Canada, incidence rates appear to be rising in children under five but remain stable in other pediatric age groups, and will be discussed more in Chapter 3 (Benchimol, Bernstein, et al., 2017). Incidence in Canadian adults is likewise relatively high (Bernstein et al., 2006; Kaplan et al., 2019) but has stabilized (Kaplan et al., 2019). Pediatric IBD rates are not available for the majority of countries, particularly those in Africa, South America, and Asia (Benchimol et al., 2011). As such, it is difficult to make claims about international patterns of the disease. However, a systematic review of available publications found that globally, the incidence of pediatric IBD is increasing (Benchimol et al., 2011). This is mostly due to rising incidence of CD, as the majority of studies examining pediatric UC found no statistically significant increase (Benchimol et al., 2011).   3  1.2.2 Causes of IBD It has been postulated that there is an altered interaction between the intestinal mucosal immune system and intestinal microbiome resulting in intestinal dysbiosis and intestinal inflammation in people who develop IBD (Abraham & Cho, 2009). The intestinal epithelium and mucous layer which provide a protective barrier between intestinal luminal content and the mucosa is also altered in IBD patients, enabling microbes to cross this barrier and further perpetuate inflammation (Abraham & Cho, 2009; Ramos & Papadakis, 2019). Essentially, patients with IBD have both inappropriate immune responses to GI microbes and incompletely protected GI tracts.  A family history of IBD is a stronger risk factor than any single known environmental determinant, but can only explain a relatively small percentage of IBD risk (Jostins et al., 2012). Both twin studies and genetic studies have been used to assess how genes may influence IBD risk; twin studies may overestimate the importance of genetics because twins can also share similar environments, while some genetic studies may result in underestimates (Gordon, Trier Moller, Andersen, & Harbord, 2015). Twin studies have reported a wide range of concordance rates (where both twins diagnosed with IBD), which are higher for identical twins and for CD. Concordance rates for CD varied from 20 – 55% for identical twins and 0 – 3.6% for fraternal twins, while for UC rates varied from 6.3 – 17% for identical twins and 0 – 6.3% for fraternal twins (Gordon et al., 2015). A large genetic study found that 13.6% of CD and 7.5% of UC variance could be explained by known genotypes (Jostins et al., 2012). These results suggest that genetic variation is perhaps the most important known single contributor to IBD risk, but that the majority of variation in IBD risk cannot be explained by genetic background alone (Gordon et al., 2015; Jostins et al., 2012).  4  Environmental factors that are not well understood also appear to play an important role in the development of IBD (Aamodt, Johnsen, Bengtson, Moum, & Vatn, 2008; Abraham & Cho, 2009; Benchimol et al., 2011; Chu, 2017; Diefenbach & Breuer, 2006; Pinsk et al., 2007; Ramos & Papadakis, 2019). Environmental influences can include features of the physical environment (e.g. air quality) or socioeconomic characteristics (e.g. income and education). They can also include individual lifestyle factors such as smoking, diet, and breastfeeding– all of which may influence risk of IBD (Diefenbach & Breuer, 2006). Moreover, medications including oral contraceptives and non-steroidal anti-inflammatory drugs have also been implicated for increasing risk (AnanthAnaakrishnan, 2013). Based on the diversity and relatively small effects of known risk factors, it seems likely that there are multiple additional unknown risk factors and that multiple risk factors must be present or interact to trigger disease. No single environmental risk factor appears to have a dramatic effect on disease risk. The timing of exposure to different risk factors may also be important, and environmental exposures in childhood appear to be a critical factor in the development of IBD– making childhood an ideal period to study environmental contributors (Benchimol, Kaplan, et al., 2017). The microbiome, or miniature ecosystem of bacteria, viruses, fungi, and other microbes that co-exists with the human body, is widely accepted as the link between the outside environment and the GI tract (Prescott, Wegienka, Logan, & Katz, 2018; Ramos & Papadakis, 2019; Rook, 2013). Environmental factors are thought to potentially influence IBD risk through changes to the microbiome. In particular, the gut microbiome of young children which is still evolving during early life appears to be particularly sensitive to environment and lifestyle factors (Benchimol, Kaplan, et al., 2017; Ramos & Papadakis, 2019).   5  1.2.3 Spatial cluster analysis and IBD 1.2.3.1 Spatial cluster analysis and spatial autocorrelation Spatial analyses that empirically identify areas of high and low IBD incidence can be used to identify novel risk and protective factors because spatial and temporal clusters can be indicative of underlying genetic, social, and/or physical environment conditions that influence IBD risk. Once clusters have been identified, the analyst’s background knowledge of characteristics of study area may be used to guide follow-up investigations into specific variables. Spatial cluster detection is a group of spatial analysis techniques used to identify areas that are close in space and have similar attributes. Almost any attribute can be used in a cluster analysis, but in a health context it is common to use rates of disease in order to highlight areas with a higher than expected burden of disease. Spatial analyses are common in infectious disease research but are also used to find patterns of chronic illness (Caprarelli & Fletcher, 2020). Asthma and cancer are some examples of chronic diseases that have been analyzed with a cluster detection methodology (Al-Ahmadi & Al-Zahrani, 2013; Ouédraogo et al., 2018).  1.2.3.2 Selected forms of spatial cluster analysis Depending on the type of data available and the desired output, there are many possible forms of spatial cluster analysis (Caprarelli & Fletcher, 2020; Kirby, Delmelle, & Eberth, 2017); I will discuss several in this section in order to provide context for my selection of analytical techniques. One way to categorize cluster detection methods is by how they represent the disease or outcome of interest. Point-based methods represent the locations of an illness as points in space and analyze the spatial density of these locations. A common point-based method is the use of density maps of disease created by mathematically estimating a continuous field of disease 6  location density; these can be used to visualize the spatial arrangement of disease in a study area and visually check for high-density disease clusters (Caprarelli & Fletcher, 2020). Another example of a point-based method is the spatial scan statistic (Aamodt et al., 2008; Green, Elliott, Beaudoin, & Bernstein, 2006), where ellipses encompassing different proportions of cases are created in order to signal the location and probability of potential clusters (Aamodt et al., 2008; Kulldorff, 1997); quantifying the probability of suspected clusters can be an advantage of the spatial scan statistic over a visual analysis of density maps. In contrast to the point-based methods outline above, which analyze the density of disease locations, area-based methods measure the intensity of disease rates as they vary through space. Area-based methods typically use area disease rates standardized by population, and are what I chose to use for this study (Kirby et al., 2017). A disadvantage of this category of methods is that they are extremely dependent on the chosen geographic unit of analysis and how these units aggregate individual cases; for example, it is likely that I would have found at least somewhat different results if I had chosen areas of a different size or shape for the cluster analysis (Nelson & Brewer, 2015). In geography, this is termed the Modifiable Areal Unit Problem, where the geographic scale, size, shape, and distribution of the geographic regions in a given analysis can strongly influence the results (Nelson & Brewer, 2015). Despite the Modifiable Areal Unit Problem, there are several features of area-based analyses that made them a good fit for this study. By detecting clusters of area disease rates adjusted by population (rather than individual disease case locations) I was able to use multi-year average rates that reduced the instability in disease rates caused by small case counts; this will be discussed in more detail in Chapter 3. Another advantage to choosing an area-based cluster method was that it balanced the privacy of study participants with analytical needs by obscuring exact patient location while 7  preserving some degree of local spatial patterns. By choosing an area-based method, I was also able to increase the transparency of my analysis by publishing the disease rates that I used as well as the cluster output (see Appendix); with a point-based cluster method, I could not publish the raw case locations used in the analysis. Finally, unlike many point-based methods that focus on areas of high density of disease, the area-based method that I selected can also identify areas of low incidence. In summary, the ability to use multi-year average incidence, better protect patient privacy, and identify multiple types of spatial patterning were all factors in selecting a method of spatial analysis. The cluster detection method that I used (and an alternative) will be discussed in the following section.   1.2.3.3 Local Indicators of Spatial Association Spatial autocorrelation is a term used to describe the spatial distribution of phenomena.  When a disease is spatially autocorrelated, it means the distribution in space is not random. The Moran’s “I” statistic is a commonly used measure of spatial autocorrelation for a study area (Anselin, 1995) which is similar to a correlation coefficient (r) that is used in traditional statistics. The global Moran’s “I” describes the overall spatial pattern of an entire data set but does not indicate local variation in that pattern. In effect, the global Moran’s “I” will indicate whether a study area is clustered or dispersed but will not identify the location of potential clusters.   Local Indicators of Spatial Association (LISA) are a class of spatial statistics used to identify statistically significant spatial clusters (Anselin, 1995). A LISA must meet several criteria. One is that for every observation or data point, the LISA denotes if neighbouring data points have similar attribute values, indicating a spatial cluster (Anselin, 1995). Another is that if 8  the LISA statistics for every data point are summed, this figure is the same as the global indicator of spatial association (Anselin, 1995). LISAs can be useful for both describing local spatial patterns in the data and identifying how small-scale patterns contribute to the global autocorrelation of the study area (Anselin, 1995). Using the Moran’s “I” statistic as a LISA is an established method, and has been used to identify high and low asthma clusters in Canada (Ouédraogo et al., 2018), high-low clusters of vestibular schwannomas (rare benign tumors that can affect hearing) in Scotland (Caulley et al., 2017), and high-high clusters of cancer incidence in Saudi Arabia (Al-Ahmadi & Al-Zahrani, 2013)— among others. The Getis-Ord Gi* statistic is another local statistic that can be used to identify disease clusters and assess the statistical significance of those clusters (Anselin, 1995; Caprarelli & Fletcher, 2020; Khan et al., 2017). However, I selected the Moran’s “I” LISA as it can also be used to highlight spatial outliers—areas where incidence is quite different from neighboring incidence (e.g. low incidence surrounded by high incidence) and because it is directly related to the global Moran’s “I” statistic that I used to measure the provincial level of spatial autocorrelation (Anselin, 1995).  1.2.3.4 Spatial analysis of IBD The most common form of geographic analysis of IBD is based on latitude. North-south gradients of IBD incidence have been well documented, and mostly studied in the context of solar vitamin D exposure (Armitage et al., 2004; Holmes, Xiang, & Lucas, 2015; Khalili et al., 2012; Stein et al., 2016). These studies generally focus only on large-scale patterns in the distribution of IBD from north to south rather than more fine-scale spatial patterns. For example, pediatric IBD incidence in Scotland was found to be higher in northern postal areas (Armitage et al., 2004). Not all spatial studies focus on latitude, and several studies have investigated overall 9  spatial variation in IBD without searching for clusters or hot spots. In one, incidence rates of Forward Sortation Areas (three digit postal codes) were quite varied, suggesting spatial heterogeneity in the distribution of IBD in Manitoba (Blanchard, Bernstein, Wajda, & Rawsthorne, 2001). Similarly, a United States study found that IBD hospitalizations varied between states, with higher rates in northern and rural (no definition given) states (Sonnenberg, Mccarty, & Jacobsen, 1991).   1.2.3.5 Spatial cluster analysis of IBD Few studies have attempted to identify specific spatial clusters or hot spots of IBD. A French study visually analyzed maps of Relative Risk to identify clusters, and used a global Moran’s “I” statistic to confirm that the data exhibited a clustered pattern (Christophe et al., 2010). CD and UC appeared to have different spatial patterns (Christophe et al., 2010). Two studies in Manitoba, Canada and Norway found CD clusters using the spatial scan statistic, while only the Manitoba study found UC clusters (Aamodt et al., 2008; Green et al., 2006). The results of spatial IBD studies on latitude, spatial heterogeneity, and spatial clusters all suggest that IBD has a patterned spatial distribution. It is unclear if this is true for British Columbia (BC) or for the pediatric IBD population. A variety of explanations for spatial variation in IBD have been proposed, including spatial patterns of urbanization and urbanicity, education, income, ethnicity, and intestinal infections (Aamodt et al., 2008; Green et al., 2006).  1.2.4 Spatial methods for measuring associations between IBD and risk factors  There are a variety of spatial methods for measuring associations between a disease outcome and potential risk or protective factors; I will briefly discuss three (spatial lag models, 10  spatial error models, and geographically weighted regression) and why I chose an alternative aspatial method for this study. An important strength of these spatial methods is that they account for spatial autocorrelation (as described in section 1.2.3.3) and so are appropriate when data is spatially autocorrelated. Spatially autocorrelated phenomena may violate the assumption of many aspatial methods that the observations are independent (Kirby et al., 2017). Spatial lag models include a spatially lagged version of the dependent variable as a regression term and are used in situations where the outcome of interest displays spatial autocorrelation (Anselin, 2001, 2002; Kirby et al., 2017). Spatial error models are used when the model residuals are spatially autocorrelated (Anselin, 2001; Kirby et al., 2017) to minimize the effects of spatial autocorrelation on the model outcome (Anselin, 2001). Geographically weighted regression is a local regression approach where an individual regression model is calculated for every location of analysis to see how the relationship between the outcome of interest and potential explanatory variables varies through space (Kirby et al., 2017).  While these can be extremely useful analytical techniques, I ultimately decided to choose an aspatial model (logistic regression, discussed in greater detail in Chapter 4) for several reasons. As I was already choosing one spatial method (LISA cluster detection) that is likely unfamiliar to many clinicians and researchers in the field of IBD, I wanted to complement the spatial analysis with a method that would be more commonplace outside fields that routinely use spatial methods. In addition, the output of a geographically weighted regression can be somewhat difficult to interpret as it produces results that vary across space. In contrast, coefficients calculated by logistic regression models that summarize the relationship between predictor and outcome variables had a much simpler interpretation than the spatial methods described above. Spatial lag and spatial error models do produce a summary output coefficient 11  for the full dataset, and a spatial error model in particular could be a productive future analytical technique to reduce potential bias from spatial autocorrelation. However, I ultimately decided that the use of logistic regression would make this research more directly comparable to other research on environmental determinants of IBD, as this type of model has been widely used to assess environmental determinants of the disease (Aamodt et al., 2008; Bernstein et al., 2016; Bernstein, Burchill, Targownik, Singh, & Roos, 2019; Bernstein et al., 2001; Kaplan et al., 2010; Opstelten et al., 2016; Stein et al., 2016). This could increase the future potential for this research to add to the state of knowledge of IBD through comparison and meta-analysis with the results of other studies.    1.2.5 Physical environment and IBD When spatial clusters of high or low IBD incidence have been identified, the characteristics of those areas might be indicative of potential risk or protective factors for CD, UC, or overall IBD. In the following section, I will review proposed risk and protective factors for pediatric IBD that are relevant to this study.  1.2.5.1 Latitude, sunlight, and vitamin D A variety of studies on latitude and IBD on all age groups found that living at a lower latitude (closer to the equator) was correlated with lower rates of IBD (Jaime et al., 2017; Khalili et al., 2012). In Europe and North America, higher IBD rates have often been observed in countries or regions at higher latitudes (Armitage et al., 2004; Khalili et al., 2012; Shivananda et al., 1996; Vernia et al., 2018)– though some areas of southern Europe have been associated with high incidence as well (Shivananda et al., 1996) and methodological differences between studies 12  make it difficult to definitively summarize the relationship numerically (Holmes et al., 2015). Sunlight, specifically Ultraviolet (UV) B exposure, has been proposed as one of the explanatory factors underlying this pattern (Holmes et al., 2015; Jaime et al., 2017; Khalili et al., 2012). A study of IBD and sun exposure in Italy found that patients with IBD had significantly less sun exposure than controls (Vernia et al., 2018). There is a high degree of variation in how studies measure and conceptualize sunlight and vitamin D exposure; some assess seasonal differences (Stein et al., 2016), while others examine longer-term (from annual to multi-decade) patterns in location and average exposure (Jaime et al., 2017; Jantchou et al., 2014; Khalili et al., 2012). From my literature review, it has not been established if there are sensitive periods for exposure.  Potential confounding factors that may explain correlations between higher IBD rates and higher latitudes include diet and migration patterns between areas with differential risk (Holmes et al., 2015). However, in addition to observed correlations, there are several biological mechanisms that could explain links between sunlight and IBD. UV exposure results in skin synthesis of vitamin D, which is virtually non-existent during high latitude winters (Holmes et al., 2015). Exposure to sunlight and vitamin D can also produce immunoregulatory responses by suppressing inflammation (Holmes et al., 2015; Khalili et al., 2012). In animal models of colitis, animals with vitamin D deficiency or inhibited vitamin D receptors had higher levels of inflammation (Khalili et al., 2012; Ryz et al., 2015). Additionally, vitamin D may play a role in regulating the GI microbiome by facilitating a more successful targeting of pathogenic microbial agents (Holmes et al., 2015). Pediatric and adult patients with IBD have been observed to be vitamin D deficient, particularly those suffering from CD (Diefenbach & Breuer, 2006; Holmes et al., 2015; Vernia et al., 2018). However, it is currently uncertain if vitamin D deficiency is a potential cause, contributing factor, or a consequence of IBD (Holmes et al., 2015).  13   1.2.5.2 Air pollution and nitrogen dioxide  On a global scale, air pollution appears to be distributed quite differently from IBD as the vast majority of premature deaths attributed to outdoor pollution occur in low and middle income countries (World Health Organization, 2018). Nevertheless, various types of air pollution have been implicated as risk factors for IBD which suggests that there may be additional factors that interact with pollutants in order to trigger disease. Pediatric IBD patients may be the most appropriate patient group for examining possible links with air pollution as studies have established that children are more sensitive to pollutant exposure and develop worse pollution-related health outcomes than adults (Kaplan et al., 2010). Children also tend to engage in more outdoor time and absorb more pollutants in the GI tract (Kaplan et al., 2010).  Particulate matter (PM) is a class of air pollutants that has been linked with gastrointestinal disease. PM has been shown to induce oxidant dependent GI epithelial cell death, disruption of tight junction proteins, intestinal inflammation, and increased gut permeability (Mutlu et al., 2011) as well as alteration in the composition of the gut microbiome (Mutlu et al., 2018, 2011). The evidence for correlations between exposure to another pollutant, nitrogen dioxide (NO2), and IBD is mixed. A positive correlation between IBD and regional average NO2  for 2001 was reported in a United Kingdom (UK) study in patients under 25, but not other age groups (Kaplan et al., 2010). A more fine-scale study of annual residential average concentrations of a variety of pollutants in four European countries found no significant relationship between NO2 and IBD, though there were non-significant positive correlations (Opstelten et al., 2016). A recent study of NO2 and pediatric IBD in Canada using similar exposure data to my study observed no correlation between IBD and outdoor residential prenatal 14  exposure or childhood exposure prior to diagnosis (Elten et al., 2020). I chose to include NO2 in this study as it is a traffic-related pollutant (Hystad et al., 2011) present in higher concentrations in urban areas (the link between urban areas and IBD will be discussed in a following section), and was a quality dataset with a high spatial resolution available for the study period.   1.2.6 Socioeconomic status, ethnicity, immigration, urban residence, and IBD 1.2.6.1 Socioeconomic status High socioeconomic status at both the individual and neighborhood levels is considered a risk factor for IBD (Armitage et al., 2004; Bernstein et al., 2019; Christophe et al., 2010; Green et al., 2006; Kaplan et al., 2010). For example, areas with the highest incidence of IBD in Norway also had populations with high education levels, but income did not appear to be significantly correlated (Aamodt et al., 2008). Conversely, a Manitoba study using individual-level socioeconomic data from surveys and Statistics Canada did not find that participants with IBD had higher educational attainment at time of diagnosis or time of survey, or that they had a higher income (as measured through census income data from one census year), though this included only adult patients (Bernstein et al., 2001). It is unclear why populations with high socioeconomic status appear to have an increased risk for developing IBD. One of the main explanations for this phenomenon is the hygiene hypothesis, where more sanitary environments during childhood lead to chronic disease in adulthood (Bernstein et al., 2019; Green et al., 2006).   1.2.6.2 Ethnicity There have been repeated observations that Jewish people are more likely to develop IBD, especially those of middle European Ashkenazi Jewish descent (Xia et al., 1998). In 15  Canada, areas of Manitoba with the largest percentage of Jewish ethnic background have reported a higher likelihood of receiving an IBD diagnosis (Green et al., 2006). Notably, children of immigrants from South Asia to Canada and the UK have more recently been reported to be a population of interest because of the dramatic increase in their incidence of IBD compared with that of their parents and parental countries of origin (Arebi, Misra, Faiz, Munkholm, & Burisch, 2018; Benchimol et al., 2011; Carroll et al., 2016; Pinsk et al., 2007). In contrast, a Manitoba study of all age groups found that self-identification as a person of Aboriginal descent may be a protective factor for IBD, as people living in areas that were not located on First Nations reserves and those with the smallest percentages of people of Aboriginal descent were more likely to be diagnosed with IBD (Green et al., 2006).   1.2.6.3 Trends in immigrants to Canada In an Ontario population-based cohort study (high quality longitudinal study design where a population is followed and all cases that develop are compared to non-cases), most children born to immigrants to Canada had a lower likelihood of being diagnosed with IBD than children of Canadian-born parents (Benchimol et al., 2015). However, the difference between children with immigrant and non-immigrant parents was much smaller than the difference in incidence between immigrants and their Canadian-born counterparts (Benchimol et al., 2015). For immigrants, the older a person was at initial arrival to Canada, the less likely they were to receive an IBD diagnosis (Benchimol et al., 2015). IBD was most likely in children of immigrants from North America and Western Europe, similar in non-immigrant children and those with parents from the Middle East and North Africa, and South Asia, and lower in children from Latin America and the Caribbean, Sub-Saharan Africa, and Eastern Europe and Central 16  Asia (Benchimol et al., 2015). Notably, children with parents from East Asia and the Pacific had extremely low rates of IBD, roughly a tenth of those from North America and Western Europe (Benchimol et al., 2015).   1.2.6.4 Urban residence In Canada, living in a rural area has been associated with a lower risk of developing pediatric IBD (Benchimol, Kaplan, et al., 2017; Green et al., 2006). Residence in rural areas before the age of five had a particularly robust protective association (Benchimol, Kaplan, et al., 2017). Findings were consistent across multiple definitions of rurality (various combinations of population size, density, and connection with metropolitan areas) (Benchimol, Kaplan, et al., 2017), and were supported by a Norwegian study that found higher rates of IBD in urbanized areas (Aamodt et al., 2008). However, it is still unclear how environmental, genetic, lifestyle, or dietary differences between urban and rural groups may influence IBD risk (Aamodt et al., 2008; Benchimol, Kaplan, et al., 2017).   1.2.7 Greenness and health status 1.2.7.1 Green spaces, greenness, and health One difference between urban and rural areas is that rural areas generally have more green spaces (e.g. trees, parks, lawns, or gardens), but green space has not been included in current published studies of IBD. Asthma, another inflammatory disease, exhibits a similar rural-urban divide in incidence (Ouédraogo et al., 2018); reduced rates of asthma have been observed near green space and higher asthma hospitalizations have been associated with lower vegetation cover (Ayres-Sampaio et al., 2014; Gascon et al., 2016). However, a systematic review of the 17  relationship between residential greenness (green vegetation cover) and pediatric asthma found conflicting results between studies and a lack of consistent study designs precluded any firm conclusions (Lambert et al., 2017). In Vancouver, BC, living near green space has been linked to positive birth outcomes such as higher birth weight and lower likelihood of preterm birth, even when adjusting for noise, air pollution, nearest distance to parks, and walkability—though maternal education and income at the census Dissemination Area level weakened the association (Hystad et al., 2014). Multiple hypotheses have been proposed for how green spaces and greenness can support health, including lowered stress, increased physical activity, increased psychological wellbeing, increased social interaction, lower exposure to air and noise pollution, and lower air temperature (Casey, James, Rudolph, Wu, & Schwartz, 2016). An alternative hypothesis is that greenness can support human health through exposure to a wide variety of microbes (Prescott et al., 2018; Rook, 2013). Microbes play a key role in regulating the immune system by conditioning it to target harmful inputs (such as pathogenic microbes) by producing inflammation. Conversely, a properly conditioned immune system is also critical for avoiding inappropriate inflammatory responses that can damage the body (Lorimer, 2017; Prescott et al., 2018; Rook, 2013; Rook, Raison, & Lowry, 2014).    1.2.7.2 Measuring greenness Multiple methods for assessing greenness and green spaces have been used in health research (Abelt & McLafferty, 2017; Rugel, Henderson, Carpiano, & Brauer, 2017). One approach is to use mapped land use data that demarcates green spaces (such as parks and street trees) and to then measure the amount, quality, and accessibility of these green spaces (Rugel et 18  al., 2017). These types of exposure assessment can be useful for examining potentially key details about the types and amounts of available green space, and for excluding or categorizing green spaces that may not be directly accessible (Rugel et al., 2017). However, I avoided this approach as it relies on the availability of consistent, high quality data for the full study period and some sources of green vegetation (such as yards) would not be included in similar datasets for the province of BC.  Another method of quantifying greenness is the Normalized Difference Vegetation Index (NDVI) derived from satellite measurement of reflectance from the Earth’s surface (Abelt & McLafferty, 2017; Casey et al., 2016; Cilluffo et al., 2018; Dzhambov, Dimitrova, & Dimitrakova, 2014; Markevych, Thiering, et al., 2014; Persson et al., 2018; Reid, Kubzansky, Li, Shmool, & Clougherty, 2018). A ratio of visible red (RED) to near infrared (NIR) measured reflectance is used to calculate NDVI (see Appendix B for formula) (Persson et al., 2018). NDVI is expressed as values from -1 to 1, with higher values representing denser green vegetation and values near zero representing bare surfaces such as rock or snow (Cilluffo et al., 2018; Gascon et al., 2016; Hystad et al., 2014; Markevych, Fuertes, et al., 2014). NDVI is not the only remote sensing-derived indicator of greenness. For example, the Enhanced Vegetation Index is a comparatively newer index that overcomes a potential drawback of using NDVI—that it can become oversaturated when measuring very densely vegetated areas (Rugel et al., 2017). Despite the lack of detail on form and quality of green space and the possibility of oversaturation in densely green areas, I selected NDVI because it has been frequent used in health research  (facilitating future comparisons and meta-analysis) (Abelt & McLafferty, 2017) and was consistently availability for the study period at the correct geography for analysis.  19  A variety of health conditions including birth weight, asthma symptoms, and body mass index have been assessed in relation to NDVI, and methodologies are generally similar across studies. Typically, buffers in a range of distances (often some combination of 100, 250, and 500 metres) around each patient’s home location are created (Abelt & McLafferty, 2017; Dzhambov et al., 2014; Persson et al., 2018). NDVI values that fall inside the buffer are then averaged to obtain a single NDVI value for each buffer distance. In BC, six-digit postal codes have been used in lieu of the patient’s home location. In this case, researchers buffered the centroid of each postal code polygon to calculate average NDVI (Hystad et al., 2014). Depending on the health condition of interest, researchers may use yearly or seasonal averages of NDVI, or NDVI values for the months preceding a particular health event (Abelt & McLafferty, 2017; Casey et al., 2016; Cilluffo et al., 2018; Dzhambov et al., 2014; Markevych, Fuertes, et al., 2014; Persson et al., 2018; Reid et al., 2018).   1.2.8 Key ideas resulting from review of literature In summary, spatial cluster analysis could be an important tool for understanding the distribution of IBD and for identifying novel risk and protective factors, and considerably more spatial research on IBD (particularly pediatric IBD) is needed. Environmental determinants are thought to be crucial for triggering or preventing disease, though no single factor appears to strongly determine IBD risk. NO2 air pollution, urban residence, being of South Asian or Jewish ethnic origin, and high socioeconomic status have been observed to increase IBD risk. Conversely, protective associations have been measured for solar UVR, vitamin D, rural residence, low socioeconomic status, and self-identification as Aboriginal. However, more research is also needed to explore these links and establish if they are relevant for the BC IBD 20  population and particularly the BC pediatric IBD population. In addition, residential greenness should be analyzed to determine if it could be correlated with IBD.   1.3 Research objectives  Figure 1.1 Study area: province of British Columbia (Teucher, Hazlitt, Albers, & Province of British Columbia, 2017)  A more robust understanding of the physical environment and its relationship to IBD development could be key for explaining variations in IBD incidence (Ng et al., 2013). Based on my literature review, spatial analysis is an exploratory tool that is underrepresented in IBD 21  research but has the potential to provide insights into the distribution and potential causes of the disease. At the same time, a more robust understanding of the non-genetic causes of IBD can lead to better treatment and prevention in the future. In this thesis, I was given the opportunity to analyze 1,183 patients who were diagnosed with IBD at British Columbia Children’s Hospital (BCCH) between 2001 and 2016. To my knowledge, no study has analyzed spatial clustering of pediatric IBD, and I have observed consistent calls in the literature for a deeper understanding of environmental triggers of IBD. Therefore, the objectives of this research were to explore spatiotemporal patterns and associations with the physical environment and ethnicity in pediatric IBD in BC (study area in Figure 1.1). Specifically: 1) Identify and describe spatiotemporal patterns in pediatric IBD in BC.  2) Model associations between pediatric IBD and NO2 air pollution, residential greenness, vitamin D-adjusted solar ultraviolet radiation, and area ethnicity. Quantifying patterns of pediatric IBD in space and time will increase understandings of the development of IBD in BC’s youth population and could lead to the identification of additional relevant factors that influence the disease. The outcomes of this research will also enhance the state of knowledge of IBD by building on previous research examining IBD risk in relation to the physical environment and ethnicity. Objective 1 was addressed using spatial cluster detection of IBD incidence. Objective 2 was addressed using a case-control methodology.   1.4 Thesis structure In this chapter, I reviewed relevant literature on the distribution and causes of IBD, previous spatial studies of IBD and relevant spatial statistical methods, and relevant known and novel risk and protective factors. Chapter 2 includes details of data used in this study. 22  Methodology and relevant results will be described in Chapter 3 for the spatial cluster analysis, and Chapter 4 for the case-control study of the physical environment and ethnicity. I will finish in Chapter 5 by discussing the implications of this research and suggesting future avenues of study.    23  Chapter 2: Data 2.1 Patient data Patient data was taken from clinical records of patients at BCCH in Vancouver, BC. All patients who received an upper endoscopy or colonoscopy at BCCH between 2001 and 2016 and were diagnosed with CD, UC, or IBD-unclassified (IBD-U, previously referred to as indeterminant colitis) were included as cases. Control patients were identified from a pool of patients that had received a diagnostic colonoscopy or upper endoscopy and did not receive a diagnosis of CD, UC, or IBD-U. All cases, whether diagnosed with CD, UC, or IBD-U, were considered to have IBD. Patients were under the age of 17 at diagnosis (cases) or when the upper endoscopy or colonoscopy was performed (controls). Case and control data will be discussed in more detail in Chapter 3 (cases) and Chapter 4 (controls). Patient six-digit postal code of residence was used to associate each patient with relevant data related to the physical environment, area ethnicity, and area socioeconomic status.   2.2 Canadian Urban Environmental Health Research Consortium data Physical environment and socioeconomic status data were obtained from the Canadian Urban Environmental Health Research Consortium (CANUE). All data from CANUE was georeferenced to the appropriate six-digit postal code represented as a point location expressed in latitude-longitude coordinates (DMTI Spatial Inc., 2015; DMTI Spatial Inc, 2016). According to CANUE, “There is no single representative areal measure of the spatial footprint represented by each postal code. Postal codes are point locations and do not have defined boundaries – especially in an historical context”; in cases where a single postal code was associated with multiple point locations (more common in rural areas), CANUE used the point location that best 24  represented the area population (as determined by DMTI) (Canadian Urban Environmental Health Research Consortium, 2018d). Likewise, quality census population estimates are not tabulated for six-digit postal codes so it is unknown how many people may be represented by a postal code. Six-digit postal codes may represent areas that range from a single apartment building up to (in some rural areas) a small town (Canadian Urban Environmental Health Research Consortium, 2018c, 2018d). Datasets from CANUE have been extensively documented elsewhere, and a summary of each dataset with citations for previous documentation is available below.   2.2.1 Airborne NO2 concentration I selected annual average nitrogen dioxide (NO2) concentration in parts per billion (ppb) from CANUE as the measure of air pollution for my study (Canadian Urban Environmental Health Research Consortium, 2018b; Hystad et al., 2011; Weichenthal, Pinault, & Burnett, 2017). Land use regression models of NO2 concentration were created by Dr. Perry Hystad of Oregon State University using concentration estimates from remotely sensed satellite data combined with geographic data on highways and major roads (Hystad et al., 2011). Annual NO2 concentration data was available from 1996 to 2012 (Canadian Urban Environmental Health Research Consortium, 2018b). As the NO2 values from CANUE were smoothed annually from a limited number of actual air pollution measurements, they may not always capture the highest pollution levels that could be present near high traffic areas or the peaks and valleys of hourly, daily, and weekly concentration levels. These fine-scale temporal and spatial variations in pollution concentration could be important for understanding potential links between air quality and IBD but were not reflected in this dataset. The values also do not reflect indoor air pollution, 25  which may not correlate with outdoor measurements. For patients included in this study, residential outdoor NO2 concentration values ranged from 0.1 to 44.1 parts per billion (ppb), with a mean value of 13.16 ppb. Provincial air quality objectives created to protect health and the environment for 2017 – 2024 set a target of annual average NO2 concentration at 17 ppb, which will be lowered to 12 ppb in 2025 (B.C. Ministry of Environment, 2017, 2020). Some study participants were exposed to NO2 concentrations well above provincial guidelines, and it is possible that negative health effects can occur below provincial objectives.   2.2.2 Vitamin D weighted UV To measure potential vitamin D from UV sources, I used monthly mean daily dose of UV weighted by vitamin D in joules per metre (m) squared (Canadian Urban Environmental Health Research Consortium, 2020; Fioletov et al., 2004; Fioletov, McArthur, Mathews, & Marrett, 2010). UV data was originally created by Environment Canada based on work by Environment Canada and Cancer Care Ontario; it was associated with six-digit postal codes and provided to me by CANUE (Canadian Urban Environmental Health Research Consortium, 2020). Methodology used to create this dataset was modified from a procedure developed for assessing UV radiation in the context of sunburns, which was based on shortwave solar radiation data collected by pyranometers distributed throughout the U.S. and Canada (Fioletov et al., 2010). The complete methodology has been described previously (Fioletov et al., 2010). Full spectrum solar radiation (between roughly 300 and 3,000 nanometres) was combined by CANUE with the ozone data (derived from satellite and ground based total ozone values) to calculate UV irradiance. Global solar radiation data was integrated with solar zenith angle and dew point temperature to produce calculations of ambient UV for each pyranometer location used in the 26  study. 324 nanometre UV-A irradiance was used, as little radiation is absorbed by atmospheric ozone at this frequency. As latitude and snow can both affect UV irradiance, data on both were included by CANUE in the next stage to produce the most accurate UV estimation. The resulting values were then interpolated from the 97 pyranometer locations to a 1 degree grid (Fioletov et al., 2010). UV dose was adjusted to the altitude of each grid cell (Canadian Urban Environmental Health Research Consortium, 2020). One set of long-term monthly UV values was used, so UV data did not change over time (Canadian Urban Environmental Health Research Consortium, 2020). For participants in this study, vitamin D weighted UV values for the highest UV month (July) ranged from 4,617 to 7,959 joules per m squared, with a mean of 6,542. Values for the lowest UV month (December) were considerably lower and varied from 22 to 250 joules per m squared, with a mean of 163.4.   2.2.3 NVDI greenness data I used NDVI data for each postal code point obtained from CANUE to measure residential and neighbourhood greenness. Annual average NDVI was selected over growing season (May 1st through August 31st) average or maximum values because there is not yet evidence to suggest that NDVI seasonality or extreme values are of particular significance for IBD. Average NDVI values were calculated by CANUE researchers from 30 m resolution Landsat satellite imagery retrieved from Google Earth Engine for a variety of distance buffers around each postal code point (Canadian Urban Environmental Health Research Consortium, 2018a; DMTI Spatial Inc., 2015; “Landsat 5 TM Annual Greenest-Pixel TOA Reflectance Composite, 1984 to 2012,” n.d.; “Landsat 8 Annual Greenest-Pixel TOA Reflectance Composite, 2013 to 2015,” n.d.; “USGS Landsat 5 TM TOA Reflectance (Orthorectified), 1984 to 2011,” 27  n.d.; “USGS Landsat 8 TOA Reflectance (Orthorectified), 2013 to 2017,” n.d.). The Landsat satellites are a series of satellites that measure reflected solar radiation from Earth’s surface for a variety of wavelengths, imaging the same location once every 16 days. I selected Landsat from a variety of satellite options because it had the finest spatial resolution of available CANUE satellite data, and has been used frequently in comparable studies (Abelt & McLafferty, 2017; Gascon et al., 2016; Hystad et al., 2014). Areas of water were excluded before NDVI calculation (Canadian Urban Environmental Health Research Consortium, 2018a). Available values included annual average NDVI at the postal code point location (30m Landsat grid cell), and 100, 250, 500, and 1000 m buffer distances. To examine potential links between greenness and IBD, I used 100 m buffer green space values as a measure of residential greenness. I selected the 1000 m buffer as a measure of average annual neighbourhood greenness. NDVI data was available every year from 1996 to 2015, with the exception of 2012 Landsat NDVI data that did not exist due to a gap in coverage of between the Landsat 5 and Landsat 8 satellites. For patients who would be matched with NDVI data from 2012, data from 2013 was used instead. NDVI values for study participants at postal code point ranged from -0.17 to 0.76 with a mean of 0.34. At the 100 m buffer distance, values ranged from 0.04 to 0.73 with a mean of 0.38, while values at the 1000m buffer were slightly higher and ranged from 0.096 to 0.68 with a mean of 0.41. Study participants associated with the lowest NDVI values lived in areas with almost no measured green vegetation, while those with the highest values lived near highly dense green vegetation.   28  2.2.4 Material and social deprivation To measure socioeconomic status, I included Canadian Material Deprivation Index and Canadian Social Deprivation Index values for each postal code (Canadian Urban Environmental Health Research Consortium, 2017; Gamache, Hamel, & Pampalon, 2017; Pampalon et al., 2012). The indices were created from Canadian census data on employment, income, education and family structure for use in public health research (Gamache et al., 2017). The material deprivation index was composed of three census measurements: people without a high school diploma, ratio of employment to population, and average personal income; the social deprivation index included measurements of people living alone, people separated, divorced, or widowed, and single parent families (Pampalon et al., 2012). Index data was initially calculated at the census Enumeration Area (1996) or Dissemination Area (2001 – 2016) level but was linked with six-digit postal codes by CANUE researchers (Canadian Urban Environmental Health Research Consortium, 2017). Dissemination Areas are the smallest geographic census areas for which all census data is released; they have an average population ranging from 400 – 700 and are comprised of at least one connected Dissemination Block (Statistics Canada, 2016, 2018). Material deprivation index values ranged from -0.15 (least deprived, highest socioeconomic status) to 0.17 (most deprived, lowest socioeconomic status) with a mean of -0.003; social deprivation index values ranged from -0.12 to 0.14 with a mean of -0.01.   2.3 Census data: ethnicity, Aboriginal identity, age, and population Statistics Canada census tabulations of total population, self-identified South Asian (e.g. Punjabi, Bangladeshi, East Indian) and Jewish ethnicities, and self-identified Aboriginal identity were used to measure the composition of ethnicity at each study participant’s location. For the 29  purpose of the census, Aboriginal identity was defined as persons who “identified with the Aboriginal peoples of Canada. This includes those who are First Nations (North American Indian), Métis or Inuk (Inuit) and/or those who are Registered or Treaty Indians (that is, registered under the Indian Act of Canada), and/or those who have membership in a First Nation or Indian band. Aboriginal peoples of Canada are defined in the Constitution Act, 1982, Section 35 (2) as including the Indian, Inuit and Métis peoples of Canada” (Statistics Canada, 2017a). This definition was similar to previous census years. In this thesis, I use the term “Aboriginal” (instead of Indigenous) in order to be precise when referencing a specific census variable. For participants in this study, the percent of South Asian residents in their local Dissemination/Enumeration Area ranged from 0 to 96.5%, with a mean of 6.7% The percent of residents of Jewish ethnic origin ranged from 0 to 17.9% with a mean of 0.7%. For both South Asian and Jewish ethnicities, the median percentage of residents of that ethnicity was 0 as most census areas had no South Asian or Jewish-identified census respondents. For Aboriginal identity, most study participants did live in an area with some Aboriginal-identified census respondents; percentages ranged from 0 to 99.3% with a mean of 3.2%. In addition to data on ethnicity and Aboriginal identity, I also downloaded total population counts and age-stratified population counts for each Dissemination/Enumeration Area and each Health Authority and Local Health Area (described in more detail in Chapter 3). Dissemination/Enumeration Areas can be quite variable in size. For example, in the 2016 census, the largest Dissemination Area in BC was 114,400.5 km2, while the smallest was 2,012.7 m2 (mean size 120.5 km2). In terms of population, the most populous Dissemination Area had 8,778 people while some Dissemination Areas had no population or fewer than five persons (Statistics Canada, n.d.-g). I retrieved tabular census data at the Dissemination/Enumeration Area level 30  from the Computing in the Humanities and Social Sciences Data Centre of the University of Toronto. Geospatial Dissemination Area boundaries were also obtained from Statistics Canada via the Abacus Dataverse Network. Each ethnicity count was converted to a percentage of the appropriate census universe population (total number of respondents to relevant ethnicity and Aboriginal identity census questions). Next, I performed a spatial join to associate postal codes with the appropriate Dissemination/Enumeration Area location; to do this, I overlaid the Dissemination/Enumeration Area boundaries for each census year over all BC postal code point locations that were active in the five years surrounding that census. For example, data from the 2006 census was linked with postal codes that existed in any year during the period of 2004 – 2008. As census geography is not necessarily consistent over time, data from different Dissemination/Enumeration areas might be needed depending on the year of interest; the data linkage procedure that I used was designed to associate the correct census Dissemination/Enumeration Area data from the proper year with each postal code. Finally, I linked patients with census ethnicity data for the appropriate census year. I used data from census years 1996 (Statistics Canada, n.d.-a), 2001 (Statistics Canada, n.d.-b), 2006 (Statistics Canada, n.d.-c, n.d.-d), and 2016 (Statistics Canada, n.d.-h, n.d.-i), and the 2011 National Household Survey (Statistics Canada, n.d.-e, n.d.-f). For use in the spatial analysis, I retrieved census population data aggregated to the province, Health Authorities, and Local Health Areas from BC Stats (BC Stats, 2019). Corresponding geographic boundaries were retrieved from DataBC (Ministry of Health, 2019).    31  Chapter 3: Spatial analysis 3.1 Introduction Current incidence and spatial patterning of IBD in BC are not publicly available. A previous study of five Canadian provinces for 1999 to 2010 found that incidence was high but fairly stable for children under 16 (9.68 per 100,000), with increasing incidence in children diagnosed at age five or younger (Benchimol, Bernstein, et al., 2017). Incidence rates for patients under 20 from 1998 – 2000 was found to be considerably lower in the province of BC than five other Canadian provinces at 5.4 per 100,000 for CD (other provinces ranged from 6.9 to 12) and 3.2 per 100,000 for UC (other provinces from 4.1 to 5.7) (Bernstein et al., 2006), but more recent incidence data is needed to see if BC continues to have a low incidence. Moreover, quantifying how disease incidence has changed through time and space can lead to a better understanding of this clinical population and where services are needed. Understanding the spatial patterning of IBD may also lead to the identification of novel environmental determinants or protective factors. While many studies have examined the existence of a north-south gradient of IBD (Armitage et al., 2004; Balzola, Cullen, Ho, & Russell, 2013; Holmes et al., 2015; Jaime et al., 2017; Khalili et al., 2012), few studies have attempted to identify specific spatial clusters of IBD, and no study has focused on detecting disease clusters in the pediatric IBD population. IBD incidence was found to be spatially clustered in Northern France, and CD and UC appeared to have different spatial patterns (Christophe et al., 2010). Two studies in Manitoba and Norway found CD clusters, while only the Manitoba study found UC clusters (Aamodt et al., 2008; Green et al., 2006). Considered together, the results of these studies suggest that IBD is not randomly distributed in space (i.e. there is some underlying spatial pattern to disease cases) and that more research on specific spatial patterns of IBD is necessary for understanding trends in IBD 32  distribution. If no spatial patterning (as measured with a Moran’s “I” statistic close to 0) or significant clustering is observable in this analysis, it could indicate that IBD risk in BC was distributed evenly throughout the youth population and that the size of the youth population in a given area would be enough information to predict the number of cases in that area. However, this seems unlikely given that many known and hypothesized risk factors and at-risk populations are not randomly distributed in space. A lack of observed spatial patterning could also indicate that the geographic scale of analysis or the way that spatial relationships were defined was not appropriate to describe the “true” spatial patterning of IBD, and that using different geographical units of analysis or exploring different spatial processes would be necessary to describe the spatial distribution of IBD. Conversely, observed spatial patterning in the results would reinforce the importance of environmental factors in determining IBD risk. This study used patient data obtained from clinical records from BCCH for 2001 to 2016. The aims of this study were to provide an exploration of how pediatric incidence in BC changed over the study period and updated incidence rates of pediatric IBD in BC. Spatial methods were implemented to discern small area patterns in standardized incidence ratios (SIR) of IBD.  3.2 Data and methods 3.2.1 Data 3.2.1.1 Clinical data Patient data from BCCH in Vancouver, Canada was used to determine incidence. As BCCH is the only tertiary care pediatric institution with academic pediatric gastroenterologists in the province, the majority of children in BC are diagnosed at BCCH. A small number of cases are diagnosed in the community, use health services from a different province or, in the case of 33  older patients, are diagnosed by adult GI physicians. Cases for this study were selected from a clinical register of patients who were diagnosed with or received care for IBD at BCCH. The register is maintained by the Division of Gastroenterology at BCCH. IBD patients from the register were excluded from the study if they were diagnosed outside the study period (2001 to 2016) or over age 16.9, did not have a valid address or postal code on file, or had a postal code associated with the location of BCCH as this was likely not their permanent residential address. Patient six-digit postal code point location was joined with latitude and longitude coordinate data from DMTI Spatial Inc. obtained from the CANUE (Canadian Urban Environmental Health Research Consortium, 2018c; DMTI Spatial Inc., 2015; DMTI Spatial Inc, 2016).  3.2.1.2 Population and geographic data Age-stratified population data was taken from BC Stats, which aggregated Canadian census data and intercensal estimates for each study year (BC Stats, 2019). To reduce the temporal instability of incidence caused by small case counts, multi-year averages of incidence for 2001 – 2005, 2006 – 2011, and 2011 – 2016 were used to examine change over time, as well as averages for the full study period (2001 – 2016). Age groups 0 – 4.9, 5 – 9.9, 10 – 14.9, 15 – 19.9, and 0 – 16.9 were used to look at patterns in age of diagnosis. I calculated incidence for the province and for each of BC’s five contiguous and mutually exclusive Health Authorities (HA). For the spatial cluster analysis, BC incidence was used as a reference to calculate SIRs for BC’s 89 Local Health Areas (LHA). A SIR is the ratio (expressed as a percent) of the actual case count to the case count that would be expected if the area had the same incidence as the reference population-- in this case, the provincial incidence (see formula in Appendix A). A SIR over 100% indicates more cases than expected when provincial incidence is used as a reference, and a 34  SIR under 100% indicates fewer cases than expected. HAs and LHAs were chosen because their geographic boundaries were stable over the study period (facilitating cross-year comparisons) and quality population estimates were available for the study years. Geographic data was downloaded from DataBC (Ministry of Health, 2019) in the BC Albers projection (EPSG:3005) and is presented in Figures 3.1 (HA) and 3.2 (LHA) below. 35   Figure 3.1 Health Authorities of BC Figure 3.2 Local Health Areas of BC 36  3.2.2 Methods 3.2.2.1 Spatial methods Spatial autocorrelation is a term used to describe the spatial distribution of phenomena.  When a disease is spatially autocorrelated, it means the distribution in space has some detectable patterning. The Moran’s “I” statistic is a commonly used measure of spatial autocorrelation for a study area (Anselin, 1995). Values obtained from calculating the statistic generally fall between 1 and -1; values closer to 1 signal the presence of positive spatial autocorrelation, or clustering, while values closer to -1 indicate negative autocorrelation, or dispersion (Kirby et al., 2017). When results are close to 0, this indicates that spatial distribution is not probabilistically indistinguishable from random and there is not a spatial pattern underlying the distribution of the data that is distinguishable with the methods and scale of analysis.  The global Moran’s “I” statistic was used to determine if IBD SIRs observed in this study exhibited a statistically significant clustered pattern at the provincial level during the study period (see Appendix A for formula). The local Moran statistic was used to quantify small area patterns in SIRs (see Appendix A for formula). Both the global and local Moran statistics require a mathematical definition of proximity for the purpose of determining which LHAs would be close enough in space to be considered a potential cluster. For this study, neighboring LHAs were defined as any adjacent LHAs that shared part of a border (either line segment or point). Provincial land and sea borders were treated the same, which resulted in LHAs near the provincial border having fewer neighbors. This is an analytical simplification of real life adjacency, as in reality areas near land borders do have additional neighbors (in other provinces or the United States) that are not included in the analysis because rates for those areas were not part of this study. To calculate the local Moran statistic, the SIR at each LHA was expressed as 37  deviations from mean incidence (Anselin, 1995). The incidence for each location was then compared with other incidence values for neighbouring LHAs. A pseudo-p value was calculated for each LHA using 999 Monte Carlo randomizations. With this Monte Carlo method, the SIR at each LHA remained constant while the SIRs for other LHAs were randomly switched. The Moran statistic was re-calculated for each of these 999 randomized permutations, allowing for the approximate assessment of how extreme each value was (the pseudo-p value) (Anselin, 1995). A Holm correction to adjust the pseudo p-values was applied to account for multiple comparisons. In this adjustment, p values were ordered from least to greatest and sequentially compared with the alpha level (0.05) divided by the number of remaining observations, implemented with the p.adjust R function (Aickin & Gensler, 1996; Holm, 1979). This was desirable as the large number of tests performed during the LISA cluster analysis increased the likelihood of producing a statistically significant result by chance alone. The local Moran statistic quantified how similar the SIR in each LHA was to the SIR of surrounding LHAs. I used the statistic to identify clusters of high SIRs (“hot spots”) or low SIRs (“cold spots”), as well as spatial outliers with incidence that differed from neighbouring LHAs. The pseudo p-values allowed for an approximation of the likelihood that each cluster was the result of chance rather than a reflection of true underlying spatial patterns (Anselin, 1995).  To minimize the instability of SIRs caused by low case counts and populations, multi-year average SIRs were smoothed prior to cluster detection using spatial linear empirical Bayes estimation. This method was based on the process described by Marshall (Marshall, 1991). In a comparison of multiple smoothing methods for mapping SIR, this was found to correlate closely with the most accurate known rates while minimizing root mean square error and computing efficiently (Yasaitis, Arcaya, & Subramanian, 2015). Youth population counts varied 38  considerably between LHAs (from 115 to 93,098) and as areas with larger populations were more likely to reflect “true” incidence, only LHAs with a population of less than 10,000 were smoothed. Data cleaning and analysis were completed in the R (version 3.6.3) environment in R Studio (version 1.1.456) using the packages spdep for spatial analysis (Bivand & Wong, 2018), tmap for mapping (Tennekes, 2018), and ggplot2 for visualization (Wickham, 2016).   3.3 Results 3.3.1 Incidence 3.3.1.1 Incidence Rates by diagnosis   Figure 3.3 Incidence rates per 100,000 by disease type 39  Diagnosis 2001 – 2005 2006 – 2010 2011 – 2016 IBD  7.32 9.97 10.23 CD 4.94 6.7 6.53 UC 1.37 2.49 2.79 IBD-U 1.0 0.78 0.9 Table 3.1 Incidence rate per 100,000 for IBD and by disease type  All incidence values were calculated per 100,000 population of the appropriate age group. IBD incidence increased during the study period, with the majority of growth between 2001 – 2005 and 2006 – 2010. CD represented the majority of incident pediatric IBD cases and followed a similar trend, but with a slight decrease between 2006 – 2010 and 2011 – 2016. CD incidence increased from the first to the third study period. UC incidence rates increased sharply from 2001–2005 to 2011– 2016. The majority of growth in UC incidence occurred between 2001–2005 to 2006 – 2010. Growth appeared to slow in the period of 2011 – 2016. IBD-U incidence declined slightly over the study period, perhaps as a result of more precise diagnostic practices.   3.3.1.2 Incidence Rates by age group Age group 2001 – 2005 2006 – 2010 2011 – 2016 2001 – 2016 0 – 4.9 2.02 1.88 1.92 1.94 5 – 9.9 5.23 6.36 7.06 6.27 10 – 14.9 13.1 17.41 17.84 16.22 15 – 16.9 8.03 15.88 17.91 14.19 Table 3.2 Incidence rate per 100,000 by age group 40  Incidence rates during the study period were lowest for the 0 – 4.9 age group and remained fairly stable but with a slight decline in 2006 – 2010. The 5 – 9.9 year old age groups saw a small but steady increase in incidence rates for each of the three time periods. Age 10 – 14.9 years had a much steeper increase between 2001 – 2005 and 2006 – 2010, with only a slight increase in 2011 – 2016. Notably, incidence rates in the 15 – 16.9 year old age group increased sharply from 2001 – 2005 to 2006 – 2010 and continued to increase in 2011 – 2016, resulting in the highest incidence rate for any age group observed during the study period (17.9 per 100,000). The two older age groups had considerably higher incidence rates than the two younger groups at all points in the study.  41  3.3.1.3 Incidence Rates by Health Authority   Figure 3.4 Incidence rate per 100,000 by Health Authority Health Authority 2001 – 2005 2006 – 2010 2011 – 2016 2001 – 2016 Fraser 9.15 11.35 12.06 10.93 Vancouver Coastal 5.68 11.85 10.21 9.31 Vancouver Island 7.01 9.05 9.02 8.4 Interior 6.61 6.7 8.02 7.17 Northern 5.0 6.27 7.26 6.24 Table 3.3 Incidence per 100,000 by Health Authority 42  I observed a large amount of variation in incidence rates between the five BC Health Authorities. The most variable study period was 2006 – 2010 (coefficient of variation 28.4%) while 2001 – 2005 and 2011 – 2016 and the full study period had similar coefficients of variation (23.7%, 20.3%, and 21.8%, respectively). Notably, Fraser Health had the highest incidence in 2001 – 2005 and 2011 – 2016, but the rate was similar to Vancouver Coastal Health in 2006 – 2010. Vancouver Coastal Health experienced the largest increase in incidence of any Health Authority from 5.68 in 2001 – 2005 to 11.85 in 2006 – 2010 with a slight decline in 2011 – 2016. Fraser Health and Vancouver Island Health had a similar increase between the first two study periods, while only Fraser Health continued to have a slight increase into 2011 – 2016. In contrast, Interior Health and Northern Health had comparatively low IBD incidence rates in all three study periods, and Interior Health had the smallest increase from the first to the third study period. However, Northern Health still had a substantial increase in incidence during the study period, the majority of which occurred from 2001 – 2005 to 2006 – 2010.   3.3.2 Spatial cluster analysis 3.3.2.1 Provincial spatial patterning Diagnosis Year Global Moran’s I Pseudo–p IBD 2001–2005 0.47 0.001 IBD 2006 – 2010 0.52 0.001 IBD 2011 – 2016 0.36 0.001 IBD 2001 – 2016 0.65 0.001 CD 2001 – 2016 0.58 0.001 UC 2001 – 2016 0.3 0.001 Table 3.4 BC Moran’s I spatial cluster analysis results 43  The global Moran’s I statistic calculated from SIRs for the full study period was 0.65, for 2001 – 2005 it was 0.47, for 2006 – 2010 it was 0.52, and for 2011 – 2016 it was 0.36. For CD, the global Moran’s I was 0.58, and for UC it was 0.3. All cluster results were statistically significant at an alpha level of 0.05. These results suggest the presence of positive spatial autocorrelation, or spatial clusters, for pediatric IBD in BC. The full study period had the highest level of clustering, while the three sub-periods were slightly lower. Study periods 2001 – 2005 and 2006 – 2010 had quite similar Moran’s “I” values for the whole IBD cohort, while 2011 – 2016 seemed to have a less clustered distribution. CD for 2001 – 2016 was more clustered than UC for the same period. Positive spatial autocorrelation can be indicative of clusters of high and/or low incidence.         44  3.3.2.2 Local spatial patterning  Figure 3.5 Selected LHAs of interest, Lower Mainland 45    Figure 3.6 Selected LHAs of interest, Vancouver Island     46     Figure 3.7 Smoothed IBD incidence per 100,000 for 2001 – 2005, 2006 – 2010, and 2011 - 2016    47      Figure 3.8 Hot and cold spots for 2001 – 2005, 2006 – 2010, and 2011 – 2016    48      Figure 3.9 Smoothed incidence per 100,000 for IBD, CD, and UC, 2001 – 2016     49      Figure 3.10 Hot and cold spots for IBD, CD, and UC SIR, 2001 – 2016  Maps of incidence per 100,000 and statistically significant hot and cold spots for the same periods are presented above. I identified at least one statistically significant hot and cold 50  spot for each study period, and no statistically significant outliers (areas of high incidence surrounded by low incidence, or vice versa). Surrey and Delta, LHAs from the Fraser Health Authority, were the only significant hot spots across every study period. For 2001– 2016, these LHAs had some of the highest provincial average SIRs of 150% and 173% (age adjusted incidence per 100,000 of 13.85 and 16), respectively. These were the only hot spots in the Lower Mainland (BC’s major population centre) for 2001 – 2005 and the only provincial hot spot for 2011 – 2016. For 2006 – 2010, the Lower Mainland hot spot also included Richmond and New Westminster. The full study period hot spot was the same as 2006 – 2010, but with the addition of Maple Ridge/ Pitt Meadows. A hot spot at Cowichan Valley South was identified on Vancouver Island for 2001 – 2005. For 2001 – 2016 and 2001 – 2005, parts of the Okanagan region in south-central BC were distinguished as a hot spot. The hot spot was considerably larger for 2001 – 2005. In the analysis stratified by disease type, the UC hot spot included Delta, Surrey, and New Westminster. These LHAs were also identified in a CD hot spot, which was almost the same hot spot as in the IBD 2001 – 2016 analysis but with Langley instead of New Westminster included. Comox Valley, a LHA on Vancouver Island, was also identified as a hot spot for CD and for IBD (with the addition of the Oceanside LHA) in 2006 – 2010. The East Kootenay area of BC was consistently identified as a “cold spot” with clustered low incidence for the full study period and each of the three time periods. Based on the clinical experience of Dr. Kevan Jacobson, division head of Pediatric Gastroenterology, Hepatology, and Nutrition at the University of British Columbia and Dr. Matthew Caroll, pediatric gastroenterologist at Stollery Children’s Hospital in Edmonton, some patients choose to visit the neighbouring province of Alberta for healthcare if they live far from Vancouver. In light of this, one possible contributor to the cold spot along the BC-Alberta border is that it is in part related to 51  the local population seeking healthcare elsewhere rather than true low incidence. A unique cold spot consisting of the Central Coast LHA in northern Vancouver Coastal Health was observed for 2006 – 2010. One cold spot in Telegraph Creek was highlighted in Northern BC for the full study period and for CD. In addition, a second CD cold spot was identified in northeastern BC. UC had a similar cold spot cluster in southeastern BC as the rest of the groups, in addition to a cold spot on the northwestern coast of BC consisting of Prince Rupert and Kitimat.    3.4 Discussion I observed a higher incidence of CD than UC for all study periods. IBD incidence in BC documented in this study was similar to previously reported Canadian incidence from a study of pediatric IBD in five other Canadian provinces between 1999 and 2010 (Benchimol, Bernstein, et al., 2017). While BC incidence for IBD, CD, and UC in 2001 – 2005 was lower than the all-province average, by 2006 – 2010 incidence for each was slightly higher than the multi-province average observed in the Canada study. As the previous study defined pediatric IBD as diagnosed before age 16, high and increasing IBD diagnosis in the oldest age group of my study likely inflated BC incidence in comparison with the under-16 study. Overall IBD incidence increased during the study period, particularly from 2001 – 2005 to 2006 – 2010. CD and UC had somewhat different temporal patterns, with UC experiencing a steeper increase than CD over the study period. Interestingly, this is in contrast to other studies of pediatric IBD that have reported rising global CD incidence over UC incidence (Benchimol et al., 2011). These findings suggest that IBD and CD incidence in BC may have reached a plateau by 2011 – 2016, while the increase in UC incidence slowed considerably during the same period. It is possible that some of 52  this increase was due to the presence of more pediatric gastroenterologists recruited to BCCH with increased access to care or changing diagnostic practices.  I observed considerable variation in incidence between age groups. By the end of the study period, ages 15 to 16.9 had both the highest observed incidence and largest increase during the study period. Ages 10 – 14.9 had the highest incidence for 2001 – 2005 and 2006 – 2010, with only slightly lower incidence than the oldest group in 2011 – 2016. The very early onset IBD group (0 – 4.9) had the lowest incidence, which remained fairly consistent. This is in contrast to the results of the five-province study, which found a steep increase in onset before age five in the time period of 1999 – 2010 (Benchimol, Bernstein, et al., 2017). The findings of my study confirm the results of a previous study suggesting that pediatric IBD incidence in BC was comparatively low in the early 2000s, though my reported incidence is different (Bernstein et al., 2006). However, incidence in BC has since increased considerably.  Incidence increased in every Health Authority over the study period. I observed consistently higher incidence for Fraser Health than for most other Health Authorities, while Vancouver Coastal Health had the largest increase. As with provincial incidence, the majority of that increase for most Health Authorities occurred between 2001 – 2005 and 2006 – 2011, supporting the idea that incidence stabilized to some extent by 2010 – 2016. Only Interior Health had a larger increase in incidence between 2006 – 2010 and 2011 – 2016. Interior and Northern Health are considered rural Health Authorities (Elliott & Copes, 2011; Simkin, Woods, & Elliott, 2017), and in all but one case had the lowest observed incidence. Fraser Health had the highest average population density for the study period, followed by Vancouver Coastal Health and Vancouver Island Health. While Health Authority population density should not be substituted for population density at an individual’s residence or other measures of rurality, my observation 53  of lower incidence in more rural, lower-density Health Authorities is consistent with a previously documented relationship between rural residence and low incidence of IBD in other Canadian provinces (Benchimol, Kaplan, et al., 2017). People living in rural areas may have difficulty accessing health care due to lack of local services, prohibitive travel time and costs to visit a care provider, and access to transportation (Select Standing Committee on Health, 2017) which could also lead to the under-diagnosis of IBD in rural areas.  The spatial distribution of IBD in BC was significantly clustered during the study period of 2001 to 2016. Some consistent hot spots of high SIRs and cold spots of low SIRs were identified throughout the study sub-periods and for IBD, CD, and UC. Other clusters were unique to certain time periods or disease types. These findings suggest the presence of underlying environmental or socioeconomic conditions that may influence IBD risk. Differences between CD and UC clustering may be indicative of some disparate place-based risk factors for the two diseases, though the presence of some shared hot and cold spots also suggests common risk factors and at-risk populations. The consistent identification of a hot spot in the main urban centre of the province and cold spots in more rural areas is further evidence of previous research suggesting rural residence as a protective factor in the development of IBD (Benchimol, Kaplan, et al., 2017). As urban areas were not uniformly identified as hot spots, there may be additional relevant risk factors beyond urbanization. Surrey and Delta are home to much of BC’s South Asian population— 45.1% and 5.5%, respectively (versus 11.2% and 2.2% of total population) (Statistics Canada, 2017c). Previous research has suggested that people of South Asian origin are at a higher risk of developing IBD (Carroll et al., 2016; Pinsk et al., 2007). The Lower Fraser Valley airshed, where the Lower Mainland IBD hot spots were located, is also an area of concern for air quality due to its topography and proximity to multiple sources of air pollution (Vingarzan 54  & Li, 2006). Evidence is conflicting for the link between pediatric IBD and air pollution (Elten et al., 2020; Kaplan et al., 2010; Opstelten et al., 2016), and more exploration of multiple types of air pollution as risk factors for IBD is necessary.  In conclusion, I found the spatial distribution of pediatric IBD incidence in BC to be clustered and to have increased during the period of 2001 to 2016. Most of the increase occurred during the beginning of the study period, suggesting that pediatric incidence may have reached a plateau. UC had a much larger increase than CD or overall IBD. Incidence rates increased in all age groups except for very early onset IBD (age 0 – 4.9) and increased the most in adolescents (age 15 – 16.9).  The distribution of IBD was spatially clustered, with consistent hot spots in the most populous area of BC and cold spots in more rural areas of BC, particularly along the border with Alberta. Socioeconomic characteristics and neighbourhood features of these hot and cold spots should be investigated as potential risk and protective factors for IBD, and future research should compare spatial patterns of pediatric and adult onset IBD. Now that I have described my estimation of the spatial patterns of BC’s pediatric IBD population, I will look in more detail at potential environmental triggers of IBD related to a person’s physical environment, and neighbourhood ethnicity and socioeconomic status.     55  Chapter 4: Modelling of potential risk and protective factors   4.1 Introduction While the results of the LISA cluster analysis were useful for identifying areas of interest for pediatric IBD, that methodology was not designed to evaluate potential risk and protective factors that could be responsible for pockets of high and low incidence. To examine possible triggers of IBD, I compared cases (IBD patients) and controls (patients without IBD) on their exposure to several potential risk and protective factors. I selected features of the physical environment and neighbourhood ethnicity as variables of interest for this analysis, including nitrogen dioxide air pollution (NO2), greenness (measured with NDVI), vitamin D weighted UV radiation (referred to in this section as vitamin D UV), and percent of the area population of South Asian, Jewish, and Aboriginal ethnicities. Area material and social deprivation were also included to potentially reduce confounding from socioeconomic status.   4.2 Data and methods 4.2.1 Data Category Variable Measurement Physical environment NO2  air pollution Average annual outdoor ground NO2 concentration (ppb) at postal code. Vitamin D UV Vitamin D adjusted UV average monthly UV dose in joules per m squared. Greenness Mean yearly NDVI at postal code, and for 100 and 1000 m buffers. 56  Ethnicity South Asian ethnicity Percent of census Dissemination/Enumeration Area self-identified as being of South Asian ethnicity for nearest census year. Jewish ethnicity Percent of census Dissemination/Enumeration Area self-identified as being of Jewish ethnicity for nearest census year. Aboriginal identity Percent of census Dissemination/Enumeration Area self-identified as Aboriginal for nearest census year. Information is collected individually for each household member; parents may respond for their children. Socioeconomic status Material deprivation Index composed of people without a high school diploma, ratio of employment to population, and average personal income in each Dissemination/Enumeration Area. Social deprivation Index composed of people living alone, people separated, divorced, or widowed, and single parent families in each Dissemination/Enumeration Area. Table 4.1 Predictor variables used in regression modelling  4.2.1.1 Patient data Cases used in the regression analysis were the same as those used for the LISA cluster analysis detailed in Chapter 3 (minus 46 patients that were missing all environment data). Controls were compared to cases on the basis of their exposure to the variables of interest to see if there were differences between them and IBD patients, and were selected from BCCH’s clinical records of patients who had received a diagnostic upper endoscopy or colonoscopy and were not given a diagnosis of IBD. Control patients had some form of gastrointestinal complaint or disease that prompted the upper endoscopy or colonoscopy (e.g. abdominal pain, potential irritable bowel syndrome, Eosinophilic Esophagitis, and celiac disease). Some control patients had normal scope results and were given no diagnosis. This makes them an appropriate control population for this study as potential confounding from illness-related behavior was reduced 57  (Vernia et al., 2018). Though control patients did not have IBD at the time of endoscopy or colonoscopy, they were not followed into adulthood to ascertain if IBD developed later; if some controls eventually developed IBD, the results of this study could be diluted. If potential cases were missing from this study because of prohibitive travel distance to BCCH, that spatial bias was likely replicated in the control population. A propensity score matching procedure based on year of birth and year of diagnosis was used to match controls to IBD cases at a 3:1 ratio. Propensity score matching can be briefly summarized as a procedure where the measured effect of the matching variables (year of birth and year of diagnosis) on the probability of the outcome (case or control) were used to balance the distribution of the matching variables between the case and control groups (Månsson, Joffe, Sun, & Hennessy, 2007). Propensity matching was used as it provided the most equivalent control groups without eliminating cases. Controls for the CD and UC analyses were selected at a 4:1 ratio from the initially selected 3:1 control group.   4.2.1.2 Physical environment, ethnicity, and socioeconomic status data  As discussed in the Chapter 2, NO2, NDVI, UV, and material and social deprivation data were obtained from CANUE. Ethnicity data was downloaded at the Dissemination/Enumeration Area level and spatially joined with patient postal codes. Mean annual NDVI values were used in this analysis as I did not find convincing theoretical or biological justification for using maximum NDVI values. To assess residential greenness, I used average NDVI within 100m of each residential postal code point locations. 100m buffers have been used in a previous study of Vancouver as they most closely approximate a city block (Hystad et al., 2014), and over 97% of both cases and controls from this study lived at postal codes representing a city block. 1000m buffers of NDVI were used to approximate neighbourhood greenness. 500m was also considered 58  as a buffer distance but was rejected as it was highly correlated (ρ = .91) with the 1000m values. Annual average NDVI at postal code point location was used as a control variable in the NO2 analysis because NO2 values were also calculated for annual average at postal code.   4.2.1.3 Associating patients with physical environment and socioeconomic data Case and control residential addresses were from the time of the endoscopic procedures. Associating patients with data from their year of birth using their address at time of the procedures would have introduced a high level of uncertainty into the analysis by assuming patients had lived at the same address for their entire life (particularly for adolescent patients). Five years prior to diagnosis or endoscopic procedures was selected as the period of interest in order to balance this uncertainty with the need to examine potential social and environmental determinants well before the likely onset of symptoms. If a patient was under the age of five at diagnosis, they were joined with data from their year of birth. If attribute data was unavailable for the appropriate year, data from the closest available year was used. According to the 2016 census, the majority of children in the 10 – 14 and 15 – 19 year age groups (58% and 59%) did not move in the previous five years (Statistics Canada, 2017b). Just under half of children age 5 – 9 years  (48%) did not move in the same time period, though 77% remained in the same Census Metropolitan Area (Statistics Canada, 2017b).   4.2.2 Methods I used a generalized linear model with a binomial probability distribution and logit link function, often referred to as a logistic regression model, to measure associations between risk or protective factors and diagnosis of IBD (see Appendix B for full model equation). To rule out 59  associations between variables of interest and IBD that were actually a result of other factors, I ran several multiple regression models that included control variables. The conceptual model of my regression equations can be summarized as:  log odds of diagnosis with IBD = effect of variable of interest + effect of control variables Through the multiple regression analyses, I was able examine measured associations in more detail by assessing potential confounding from additional variables (multivariate models in Table 4.2 below). To evaluate the potential influence of multicollinearity, I assessed the Spearman rank correlation of each variable pair that would be combined in a multiple regression. The highest correlation was -0.5 (NO2 and 1000m NDVI) and the next highest was just under 0.3 (percent Aboriginal identity and social deprivation), so I decided to proceed with my initial models. As my research objective was to quantify the relationship between my variables of interest and IBD, rather than to build a comprehensive picture of IBD risk, I chose not to include models with multiple variables of interest. I did not have access to key risk factors such as family disease history or diet that would have strengthened my ability to predict IBD diagnosis. Related, I also chose not to include interaction effects in my modelling as I judged that the most productive examination of interaction between variables would have been between the environmental variables of this study and individual or family risk factors which were not present in my dataset. All regression models were fitted in the R  (version 3.6.3) environment using the glm function (R Core Team, 2020).    NO2 model NDVI 100m model NDVI 1000m model Percent South Asian model Percent Jewish model Percent Aboriginal model 60  Variable of interest NO2  NDVI 100m NDVI 1000m Percent South Asian Percent Jewish Percent Aboriginal Control variables Material deprivation Material deprivation Material deprivation Material deprivation Material deprivation Material deprivation Social deprivation Social deprivation Social deprivation Social deprivation Social deprivation Social deprivation NDVI at PC NO2  NO2     Table 4.2 Multivariate regression models  4.2.3 Models  Age of diagnosis:  0 - 16.9 Age of diagnosis:  0 - 9.9 Age of diagnosis:  10 – 16.9 cases Controls cases controls cases controls IBD 1137 3389 286 851 851 2538 CD 751 3023 157 669 594 2354 UC 274 1092 89* 353 185 739  * sample size deemed too small Table 4.3 BC regression analyses   Fraser Health 0 – 16.9 Urban (Fraser Health and  Vancouver Coastal Health) 0 – 16.9 cases controls cases controls IBD 542 1651 790  2395 CD 348 1469 511 2140  UC 144 532 203 776 Table 4.4 Fraser Health and urban (Fraser Health and Vancouver Coastal Health) analyses 61  Each of the univariate and multivariate regression models were fitted separately for IBD (combined CD, UC, and IBD-U), CD, and UC. Each disease type model was then stratified into groups based on diagnosis before or after age 10 (Table 4.3, above). To increase sample size and statistical power, age groups were broader than in the incidence analysis from Chapter 3. In order to evaluate urban areas only, I also produced models using patients from Fraser Health and Vancouver Coastal Health (Table 4.4, above). Not all parts of Fraser Health and Vancouver Coastal Health are urban areas, but these two HAs had the highest population density and contained the majority of the province’s population. As Fraser Health had higher IBD incidence than the other Health Authorities and contained the main provincial IBD hot spot, I also fitted models with just Fraser Health cases and controls.   4.3 Results 4.3.1 Assessing models with Tjur’s 𝑹𝟐 There are multiple alternatives for binary logistic regression to the 𝑅" values used to assess explanatory power in “typical” logistic regression models with continuous outcome variables (e.g. Ordinary Least Squares) (Tjur, 2009). I used Tjur’s 𝑅" statistic, which he terms the coefficient of discrimination, as it is the most analogous to the “traditional” 𝑅" values mentioned above (Tjur, 2009). This statistic is essentially a measure of how well my models discriminated between cases and controls (as coded in the data as 1s and 0s). The 𝑅" values I obtained for my models were generally quite low (among models discussed below, they ranged from 0.001 to 0.023), but I believe this is acceptable for two reasons. First, my research objective in building these models was to measure the association between the variables of interest and the likelihood of IBD diagnosis, not to develop a predictive model for IBD; I am missing many key 62  variables that are known to influence IBD risk and would not have expected high 𝑅" values. Second, the effect sizes I observed were relatively small, so a small 𝑅" value is not surprising. In my results, models with larger effect sizes often had larger 𝑅" values. 𝑅" values for all models are available in Appendix B.  4.3.2 Assessing spatial autocorrelation of model residuals with Moran’s “I” I calculated the global Moran’s “I” statistic (described in Chapter 3 and Appendix A) for the residuals of each multivariate regression model to assess if the residuals displayed spatial autocorrelation. If the residuals were autocorrelated, it could indicate that the assumption that observations are independent was violated, or that an important autocorrelated variable was not included in the model. Generally, I observed very little statistically significant (p value below 0.05) spatial autocorrelation of the model residuals. I was able to measure low levels of spatial autocorrelation in the residuals for the provincial CD multivariate models of NO2 and NDVI at 100 and 1000 m, and the Fraser Health IBD multivariate model of NO2. Though statistically significant, the Moran’s “I” statistics for each of the models were quite small (around 0.09 for the three CD models, and 0.02 for the Fraser Heath IBD model), suggesting a low but detectable level of spatial autocorrelation in the residuals. It is possible that I would have detected more spatial autocorrelation at a different geographic scale or by measuring it in a different way; however, the very low levels of spatial autocorrelation that I measured suggest that if autocorrelation patterns were present in the data, they may have been captured in the spatial distribution of the included explanatory variables and were consequently not observed in the residuals.  63  4.3.3 Interpreting model outputs: Odds ratios and statistical significance Not all model results are presented here, and the model for UC diagnosed under 10 was excluded due to small sample size. P values below 0.05 were considered statistically significant, though both the scientific value and proper method for using p values to interpreting statistical results are subject to debate in many scientific communities (Amrhein, Greenland, & McShane, 2019). One potential issue in this analysis is that I ran a large number of models, which increased the likelihood of identifying statistically significant results by chance (type 1 error). Applying some form of multiple testing correction (as described in Chapter 3) would have been a method for reducing potential type 1 error and should perhaps be used for similar analyses in the future. However, the use of a strict cutoff for identifying statistical significance has also been criticized (Amrhein et al., 2019). In an attempt to balance these concerns and provide a more holistic description of the results of my exploratory analysis, I will present Confidence Intervals (CI) and discuss some models with p values above 0.05 while still using an alpha of 0.05 to assess statistical significance.  Selected estimated Odds Ratios (OR) are reported in the text below. For full model results, see Appendix B. An OR is the measurement of the estimated effect each variable has on the odds of developing IBD. If the OR is greater than one, the variable of interest is a potential risk factor that appears to increase the odds of developing IBD; if the OR is less than one, it is a potential protective factor that decreases the odds of developing IBD. For example, an OR of 1.05 would indicate that I observed a 5% increase in the odds of developing IBD for each one-unit increase in exposure to the variable of interest. A one-unit increase is an increase of one unit of the variable of interest, e.g. a change of one ppb for NO2 concentration or 1% for census ethnicity percentages.  64   Variable of interest Model type Effect Model Notes NO2 univariate risk UC, IBD: Fraser Health  multivariate risk univariate risk UC, UC age 10 and over: BC UC, IBD: urban multivariate risk Vitamin D UV univariate protective UC, IBD: urban, Fraser Health Larger effect for winter months Percent South Asian univariate risk UC age 10 and over: BC  multivariate risk Percent Aboriginal univariate protective CD, CD age 10 and over: BC  multivariate protective NDVI 1000 m univariate protective UC: Fraser Health  multivariate risk Table 4.5 Summary of selected modelling results (statistically significant results in bold)  4.3.4 Physical environment 4.3.4.1 NO2 pollution I observed a nonsignificant positive association between NO2 pollution and UC from the results of my modelling. The observed relationship was not particularly strong (OR 1.02, CI 1 – 1.05, p 0.057), but was consistent in the multivariate model (OR 1.02, CI 0.99 – 1.04, p 0.187). A similar association was also measured in the UC over 10 year age group (univariate OR 1.03, 65  CI 1 – 1.06, p 0.077, multivariate OR 1.02, CI 0.99 – 1.05, p 0.256). I observed the same pattern in the urban model (urban model OR 1.03, CI 1.00 – 1.06, p 0.06) which was slightly stronger and statistically significant in the Fraser Health model (OR 1.07, CI 1.03 – 1.11). For IBD, NO2 pollution was a significant risk factor in the Fraser Health model (OR 1.03, CI 1.01 – 1.04) that remained consistent in the multivariate regression. In the urban model, NO2 was not significant (univariate OR 1.01, CI 1.00 – 1.02, p 0.188, multivariate OR 1.01, CI 1.00 – 1.03, p 0.082). This means that at most, a unit increase in NO2 exposure was associated with a 7% increase in the odds of UC diagnosis.  4.3.4.2 Vitamin D UV In the urban UC model, all months of vitamin D UV were statistically significant with small ORs (range 0.97 – 1). A similar pattern was observed in the UC Fraser Health model, with slightly lower ORs (0.94 – 1) that were lower for winter months. At most, a one unit increase in December vitamin D UV exposure was associated with a 6% decrease in the odds of receiving a UC diagnosis (OR .94). In the IBD Fraser Health model, vitamin D UV was a significant explanatory variable for all months, but with small ORs ranging from 0.97 – 1. In the urban model for IBD (February through September) and provincial UC model (June), vitamin D adjusted UV was significant but the effect was so slight that the OR was 1.  4.3.4.3 Greenness Neighbourhood (1000m) annual average NDVI was a nonsignificant protective factor in the univariate Fraser Health UC model (OR 0.11, CI 0.01 – 1.01, p 0.053) but became a nonsignificant risk factor with a large confidence interval in the multivariate model. This 66  suggests that the association measured in the univariate model was potentially confounded by the negative correlation between NO2 and NDVI, as NO2 was a significant risk factor for Fraser Health UC and was somewhat negatively correlated with NDVI.   4.3.5 Ethnicity 4.3.5.1 Area South Asian ethnicity In UC in just the over 10 year age group, percent of patients’ Dissemination/Enumeration Area who identified as being of South Asian origin had a small but significant positive association with UC diagnosis (OR 1.01, CI 1 – 1.02), meaning a one unit increase in the percent of South Asian-identified residents was associated with a 1% increase in the odds of being diagnosed with UC. In the multivariate model with material and social deprivation, the OR remained the same but the association was no longer significant (p 0.256).  4.3.5.2 Area Jewish ethnicity I observed no statistically significant association between the percent of patient Dissemination Area who identified as being of Jewish ethnicity and IBD.  4.3.5.3 Area Aboriginal identity In the 0 – 16.9 CD age group, I observed a small but significant protective effect for the percent of patients’ Dissemination/Enumeration Area that identified as Aboriginal (OR 0.98, CI 0.96 – 1). This means that each unit increase in the percent of area residents who identified as Aboriginal was associated with a 2% decrease in the odds of developing CD. A similar but nonsignificant pattern appeared in the multivariate model (p 0.075). Percent Aboriginal was also 67  significantly protective in the under 10 year age group (OR 0.94, CI 0.88 – 0.98) but was not significant in the multivariate model (OR 0.95, CI 0.89 – 1, p 0.088).   4.3.6 Socioeconomic status 4.3.6.1 Material and social deprivation For CD only, material deprivation had a strong, significant protective effect (OR 0.05, CI 0 – 0.55) that remained constant in the multivariate model, meaning that each unit increase in material deprivation was associated with a 95% decrease in the odds of receiving a CD diagnosis. For the CD over 10 year age group, material deprivation was not significantly protective in the univariate model (OR 0.09, CI 0.01 – 1.26, p 0.075) with a similar pattern noted in the multivariate models. Material deprivation was also not significantly protective in the under 10 year age group (OR 0.01, CI 0 – 1.3, p 0.064). This is the only CD group for which a consistent significant protective effect for social deprivation was also observed (OR 0, CI 0 – 0.37, similar in multivariate). For the urban and Fraser Health models for CD, material and social deprivation had small, nonsignificant protective effects. Both material and social deprivation showed nonsignificant protective associations with IBD in single and multivariate models. When stratified by age, the over 10 year age group had even less of an association. In contrast, in the under 10 year age group, social deprivation was significant (OR 0.01, CI 0 – 0.34).   68  4.4 Discussion 4.4.1 Physical environment 4.4.1.1  NO2 pollution Previous research on the link between NO2 and IBD has had mixed results (Elten et al., 2020; Kaplan et al., 2010; Opstelten et al., 2016). In this study, NO2 was a significant risk factor in Fraser Health for UC and IBD, but not CD. The highest observed OR was for UC (1.07) in Fraser Health, while the IBD OR was smaller (1.03). Several non-significant positive associations between NO2 and IBD were observed with similar ORs (ranging from 1.01 to 1.03). If exposure to NO2 pollution is a risk factor for the development of IBD, its effect appeared to be moderate and more important for UC. As NO2 pollution was much higher in the Vancouver Coastal and Fraser Health Authorities during the study period, it may be that NO2 concentrations in other areas were not high enough to influence IBD risk. Interesting, while Vancouver Coastal Health had the highest average NO2 concentration, controls in that area had higher average exposure than UC cases (16.1 ppb vs 14.4 ppb, respectively). Fraser Health had the second highest average NO2 exposure, with 15.6 ppb for UC patients and 14.2 ppb for controls. NO2 exposure inside a child’s residence or at other locations where they might spend substantial amounts of time (e.g. school or recreational activities) was not measured in this study but could also be important for determining risk.   4.4.1.2 Vitamin D UV Many previous studies on solar radiation and IBD have found protective associations, mostly for CD or overall IBD (Holmes et al., 2015; Jantchou et al., 2014; Limketkai, Bayless, Brant, & Hutfless, 2014; Stein et al., 2016; Vernia et al., 2018), though an ecological study of 69  IBD-related hospital admissions found a protective association with UC (Jaime et al., 2017). I observed a small statistically significant protective association between vitamin D UV and UC in the urban and Fraser Health models, particularly for winter months, and a weaker effect in the same areas for IBD. However, variation in monthly UV values within Fraser Health and Vancouver Coastal Health was low, and the difference in mean UV between cases and controls for those areas was small.  Sunlight is the most important source of vitamin D for most people (Jaime et al., 2017), though vitamin D from supplements and dietary sources was a potential unmeasured confounding factor in this analysis. A 2013 report on blood levels of vitamin D in Canadians found that children tended to have adequate vitamin D levels (defined as 50 nanomoles per litre of blood) (Janz & Pearson, 2013). Of those studied, 89% of participants ages 3 – 5, 76% of those ages 6 – 11, and 71% of those ages 12 – 19 were found to have sufficient vitamin D levels (Janz & Pearson, 2013). It is possible that UV exposure is not a critical determinant of vitamin D levels for BC children, and therefore vitamin D UV did not have much (if any) of an impact on IBD risk during the study period. If there was an effect, it appeared to be strongest for UC patients in Fraser Health and Vancouver Coastal Health.  Though ambient environmental UVR will be the same for everyone at a given location, individual behavior and biology also influence actual vitamin D levels (Fioletov et al., 2010). Skin tone is one of the biological determinants of how much vitamin D the body produces from UV exposure, and people with darker skin require more UV exposure to produce adequate vitamin D (Fioletov et al., 2010). Ambient UV in BC is considerably lower in the winter. Consequently, though the doses of vitamin D weighted UV used in this study were accurate for each location, they did not reflect differences in vitamin D production related to individual sun-70  protection behavior or skin tone. These differences could be important for determining IBD risk related to solar vitamin D. BC residents of South Asian descent have been documented to have an elevated risk of developing IBD (Pinsk et al., 2007), and often have darker skin tones that could make it difficult to produce sufficient vitamin D from UV exposure during much of the year. It is possible that their risk could be compounded by vitamin D deficiency, and future research using individual skin tone and serum vitamin D levels would be necessary for addressing this area of research.  4.4.1.3 Greenness I observed no results that appeared to be indicative of a true association between greenness and IBD.   4.4.2 Ethnicity 4.4.2.1 Area South Asian ethnicity Living in a Dissemination/Enumeration Area with a higher percentage of South Asian residents was a small, statistically significant risk factor (OR 1.01) for the UC over 10 year age group. The OR remained the same but lost statistical significance once material and social deprivation were added in the multivariate analysis. This observation is somewhat in line with a previous study where the pediatric South Asian population of BC was found to have a high incidence of IBD and comparatively high proportion of Crohn’s disease (Pinsk et al., 2007). However, as my analysis looked at area ethnicity rather than individual ethnicity, my results should be interpreted cautiously to avoid the ecological fallacy. The percentage of an ethnic group in a Dissemination/Enumeration Area was not necessarily indicative of a patient’s ethnic 71  background, though it was more likely in some areas– especially those where the vast majority of the population identified as being of South Asian ethnic origin.  4.4.2.2 Area Jewish ethnicity Though it has been established that people of Jewish, especially Ashkenazi Jewish, origin are at a higher risk for IBD (Xia et al., 1998), I did not observe any correlation between percent of the population of Jewish ethnic origin and IBD. There are several possible reasons for this. The Canadian census question of ethnic origin is different than that of religious affiliation, and both the number and percentage of respondents who reported being of Jewish ethnicity has declined in every census since 1996, particularly in 2016 (Smith & McLeish, 2019). Because of this, census ethnicity may not be a very accurate measure of Jewish ethnic origins. In addition, as almost all the values for percent Jewish ethnicity were quite low and the majority of participants lived in a Dissemination Area where no census respondents identified as being of Jewish ethnicity, it is unlikely that the area ethnicity had a good chance of capturing individual ethnicity.  4.4.2.3 Area Aboriginal identity  The percent of residents in each patient’s Dissemination/Enumeration Area that identified as Aboriginal had a significant protective association in the all ages and under 10 CD groups (OR 0.98 and 0.94, respectively). This association was similar but not statistically significant in the multivariate analysis. My results resemble those of a Manitoba ecological study which found that areas with a higher population self-identifying as Aboriginal had lower incidence of IBD (Green et al., 2006). Whether this association is due to genetic, socioeconomic, or other factors, more research is needed on the possible relative rarity of IBD in Aboriginal communities. As 72  areas with a higher percent of the population identifying as Aboriginal are often in more rural areas of the province (see Figures 4.1 and 4.2 below), it is possible that rurality was fully or partially responsible for the observed protective association with CD. Related to the hygiene hypothesis, First Nations communities in Canada are also less likely to have access to safe drinking water and are consequently at a higher risk for waterborne infections (Richmond & Cook, 2016). Despite the possible lower incidence of IBD, poorer health outcomes and inequitable access to clean water and health care in Indigenous communities are serious public health issues that must be addressed.   Figure 4.1 Percent of the population in each LHA who identified as Aboriginal in the 2016 census (Provincial Health Services Authority, n.d.)  73   Figure 4.2 Population density per square kilometre as of the 2016 census (Provincial Health Services Authority, n.d.)  4.4.3 Socioeconomic status 4.4.3.1 Material and social deprivation Material and social deprivation had protective associations with CD and IBD (meaning higher socioeconomic status was a risk factor), in line with previous studies that found a higher risk of developing IBD in groups with higher socioeconomic status (Bernstein et al., 2019; Kaplan et al., 2010). Associations were often not statistically significant. The strongest statistically significant protective relationship was observed for CD and material deprivation. Social deprivation appeared to be more important than material deprivation for the IBD group, particularly for patients diagnosed before age 10. These results were not unexpected, as children of families with higher socioeconomic status have been linked to an elevated risk of IBD 74  (Bernstein et al., 2019), and lower socioeconomic status has been observed to be more protective for CD than UC (Kaplan et al., 2010).  4.5 Conclusion In summary, measured associations were often different for IBD, CD, and UC (especially CD and UC), suggesting that some environmental triggers of the diseases may not be shared. UC was associated with the most variables of interest: a positive association with NO2  and South Asian ethnicity, and a negative association with vitamin D UV. IBD had similar but weaker associations with NO2 and vitamin D in the same geographic areas, an effect which could be mostly due to the presence of UC patients. CD and UC had different associations with ethnicity: a positive association between UC and percentage of South Asian residents, and a negative association with CD and percentage of self-identified Aboriginal residents. This is potentially suggestive of genetic and/or environmental determinants of IBD in these two populations, though without knowing patient ethnicity it is difficult to make a clear connection. Material deprivation was a consistently strong protective factor for CD, but not UC. An important consideration in interpreting these results is the various scales of the variables of interest (range and average values were discussed in Chapter 2). All reported ORs are for a one-unit increase in the exposure of interest; for some variables (material and social deprivation, and NDVI), a one-unit increase would encompass a large proportion of all possible data values. For others (vitamin D UV, census ethnicity variables, and NO2), a one-unit increase would cover only a small amount of possible data values. For example, a one-unit increase in the percent of Dissemination/Enumeration area respondents who self-identified as Aboriginal would only be a difference of  1%, while a one-unit increase in NDVI would mean going from an NDVI value of 75  0 (no vegetation) to 1 (total dense green vegetation coverage). In the future, building regression models using scaled variables would be useful for comparing the effect size of different variables; my models are useful for describing the effect size of individual variables as they are easier to interpret in the original units of that variable. There was some uncertainty in linking attribute data to participants because of the possibility that they did not live at the same residential address five years prior to diagnosis or that postal code location was not representative of actual home location. Uncertainty was likely higher in the under 10 group. However, census data on childhood mobility and high positional accuracy for the vast majority of participant postal codes adds to the validity of this study. Geographic region and age of diagnosis appeared to have a role in determining associations between IBD and the variables of interest. The Fraser Health models often returned the most significant associations, which could be related to the comparatively high incidence of IBD observed in that region. As results differed by age for some variables, there may be sensitive periods for certain environmental exposures. It is possible that this study lacked the statistical power to observe certain associations, particularly in the under 10, urban, and Fraser Health groups. Another possibility is that the control population was too similar to the case population, and the use of healthy controls might have returned different results. Methodological choices including the cluster detection method, scale and geographic regions of analysis, and use of logistic regression modelling over previously discussed spatial methods also influenced my results, and a different study design could have led to different findings. Besides material deprivation as a risk factor for CD, most of the statistically significant associations I observed were relatively weak. In terms of clinical significance, there may not be much applicability of these results. Nevertheless, it is accepted that IBD develops from a complicated interaction of 76  environmental and biological factors (Kapoor et al., 2016; Peloquin et al., 2016; Ramos & Papadakis, 2019); interactions between multiple risk factors could meaningfully alter disease risk and each risk and protective factor should be conceptualized as pieces of an individual’s complex risk profile. Further study of the relationship between IBD, ethnicity, and the physical environment, and particularly interactions between risk factors, could be valuable for gaining a better understanding of the etiology of CD and UC.     77  Chapter 5: Conclusion The objectives of this study were to 1) identify and describe spatiotemporal patterns in pediatric IBD in BC and 2) model associations between pediatric IBD and NO2 air pollution, residential greenness, vitamin D-adjusted solar UV, and area ethnicity. Objective 1 was addressed in a spatial hot spot analysis using Moran’s “I” as a Local Indicator of Spatial Association. A series of logistic regression analyses comparing cases and controls on their exposure to the variables of interest were used to examine Objective 2. In some ways, the results of the cluster analysis and regression models seem somewhat conflicting. Strong spatial patterning of IBD would suggest the importance of environmental determinants, but none of the modelled explanatory factors had a particularly large effect on individual-level IBD risk. However, neither set of findings is at odds with previous research on IBD— environmental determinants are known to be influential but observed individual factors often have small effects. It seems likely that various forms of interaction between biological systems and environmental factors are necessary to trigger IBD, and that future modelling should attempt to reproduce interactive relationships between variables. Exactly how interactions should be modelled (both statistically and conceptually) is less clear, but current understandings of the most likely biological mechanisms for interaction could inform which interactions should be initially studied and how this relationship should be represented numerically.  Surrey and Delta in the Fraser Health Authority were consistently identified as hot spots for overall IBD, CD, and UC. NO2 air pollution may be a risk factor for UC and overall IBD, a relationship that was most strongly observed in the Fraser Health Authority. This area is also home to a large South Asian population, a group which is known to have a high incidence of pediatric IBD. Potential disproportionate exposure to air pollution for the South Asian 78  community would be an environmental justice issue and should be investigated further. In addition, air pollution should be explored from an epigenetic perspective as a potential trigger for IBD-implicated genes that are more common in people of South Asian descent. Prevention and treatment resources should be concentrated in areas with a disproportionate burden of disease. The rural Health Authorities of Interior and Northern Health had lower IBD incidence and most of the identified cold spots, suggesting that BC is similar to other provinces in its rural-urban distribution of IBD. Lower observed incidence of IBD in more rural areas could be indicative of barriers to accessing healthcare that would decrease the odds of an accurate diagnosis. While IBD care resources should be concentrated in high incidence areas, it is also important to investigate if rural areas have access to adequate health services. The slight lower incidence of CD in areas with a higher percentage of the population that identified as Aboriginal was consistent with a previous Canadian study, and possible genetic, lifestyle, environmental, and healthcare access factors that could explain this observation should be investigated. Incidence of IBD may be lower in areas with higher Aboriginal populations, but disproportionate morbidity and mortality from other causes between BC’s Indigenous and non-Indigenous populations must be addressed (Office of the Provincial Health Officer & First Nations Health Authority, 2018).  Though IBD is a rare disease, its strongly negative impact on quality of life makes BC’s increasing pediatric incidence cause for concern. A better understanding of environmental determinants of IBD has implications for future prevention efforts and may lead to a better understanding of how IBD is triggered biologically. Though the effect sizes I observed in this study were fairly small to be clinically meaningful, they could have a stronger measurable effect at the population level or in concert with other risk factors. In particular, NO2 air pollution and 79  ethnicity (specifically people who self-identify as South Asian or Aboriginal) are worth including in future studies, especially high-quality cohort studies that incorporate data on individual ethnicity and control for diet, serum vitamin D, and family disease history. Increasing the sample size to boost statistical power would perhaps allow for the observation of additional effects that were too subtle to be statistically significant in this analysis. A larger sample would be better accomplished by expanding to additional provinces than by including adult patients from BC, as the use of adult patients might increase uncertainty in assigning environmental exposures and mask age-dependent environmental effects. If adult populations were included in future studies, it could be productive to compare pediatric and adult populations in terms of their respective IBD spatio-temporal distributions and environmental exposures to determine similarities and differences between pediatric and adult onset IBD.  Though I completed the spatial analysis and risk factor modelling separately, several statistical techniques could be used in the future to integrate these two approaches. Spatial lag models (described in Chapter 1) could be a method for including spatial relationships in models of potential environmental determinants of IBD. In addition, Poisson regression modelling to measure the association between environmental variables and area IBD incidence would be a way to more directly explore possible environmental explanations for the spatial variations observed in the cluster analysis, particularly if the same geographic units and study period were used. This approach has been used previously in the context of adult-onset IBD, and similar methods employed for the pediatric IBD population could complement the existing literature (Aamodt et al., 2008; Green et al., 2006).    80  References  Aamodt, G., Johnsen, J., Bengtson, M. B., Moum, B., & Vatn, M. H. (2008). Geographic distribution and ecological studies of inflammatory bowel disease in southeastern Norway in 1990-1993. Inflammatory Bowel Diseases, 14(7), 984–991. https://doi.org/10.1002/ibd.20417 Abelt, K., & McLafferty, S. (2017). Green streets: Urban green and birth outcomes. International Journal of Environmental Research and Public Health, 14(7). https://doi.org/10.3390/ijerph14070771 Abraham, C., & Cho, J. H. (2009). Inflammatory Bowel Disease. New England Journal of Medicine, 361(21), 2066–2078. https://doi.org/10.1056/NEJMra0804647 Aickin, M., & Gensler, H. (1996). Adjusting for multiple testing when reporting research results: The Bonferroni vs Holm methods. American Journal of Public Health, 86(5), 726–728. https://doi.org/10.2105/AJPH.86.5.726 Al-Ahmadi, K., & Al-Zahrani, A. (2013). Spatial autocorrelation of cancer incidence in Saudi Arabia. International Journal of Environmental Research and Public Health, 10(12), 7207–7228. https://doi.org/10.3390/ijerph10127207 Amrhein, V., Greenland, S., & McShane, B. (2019, March 21). Scientists rise up against statistical significance. Nature. Nature Publishing Group. https://doi.org/10.1038/d41586-019-00857-9 AnanthAnaakrishnan, A. N. (2013). Environmental Risk Factors for Inflammatory Bowel Disease. Gastroenterology & Hepatology (Vol. 9). Retrieved from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3736793/pdf/GH-09-367.pdf 81  Anselin, L. (1995). Local Indicators of Spatial Association-LISA. Geographical Analysis, 27(2), 93–115. Anselin, L. (2001). Spatial Econometrics. In B. H. Baltagi (Ed.), A Companion to Theoretical Econometrics (pp. 310–330). Blackwell Publishing Ltd. Retrieved from http://web.pdx.edu/~crkl/WISE/SEAUG/papers/anselin01_CTE14.pdf Anselin, L. (2002). Under the hood issues in the specification and interpretation of spatial regression models. Agricultural Economics, 27, 247–267. https://doi.org/10.1016/S0169-5150(02)00077-4 Arebi, N., Misra, R., Faiz, O., Munkholm, P., & Burisch, J. (2018). Epidemiology of inflammatory bowel disease in racial and ethnic migrant groups. World Journal of Gastroenterology, 24(3), 424–437. https://doi.org/10.3748/wjg.v24.i3.424 Armitage, E. L., Aldhous, M. C., Anderson, N., Drummond, H. E., Riemersma, R. A., Ghosh, S., & Satsangi, J. (2004). Incidence of juvenile-onset Crohn’s disease in Scotland: Association with northern latitude and affluence. Gastroenterology, 127(4), 1051–1057. https://doi.org/10.1053/J.GASTRO.2004.06.024 Ayres-Sampaio, D., Teodoro, A. C., Sillero, N., Santos, C., Fonseca, J., & Freitas, A. (2014). An investigation of the environmental determinants of asthma hospitalizations: An applied spatial approach. Applied Geography, 47, 10–19. https://doi.org/10.1016/j.apgeog.2013.11.011 B.C. Ministry of Environment. (2017). Guidance on Application of Provincial Air Quality Objectives for NO2 . Retrieved from https://www2.gov.bc.ca/assets/gov/environment/air-land-water/air/reports-pub/no2_aqo-implementation_guide.pdf B.C. Ministry of Environment. (2020). British Columbia Ambient Air Quality Objectives. 82  Retrieved from http://airquality-qualitedelair.ccme.ca/en/. Balzola, F., Cullen, G., Ho, G. T., & Russell, R. K. (2013, May 3). The role of the environment in the development of pediatric inflammatory bowel disease. Inflammatory Bowel Disease Monitor. Remedica Medical Education and Publishing. https://doi.org/10.1007/s11894-013-0326-4 BC Stats. (2019). British Columbia - Population Estimates. Retrieved from https://bcstats.shinyapps.io/popApp/ Benchimol, E. I., Bernstein, C. N., Bitton, A., Carroll, M. W., Singh, H., Otley, A. R., … Kaplan, G. G. (2017). Trends in Epidemiology of Pediatric Inflammatory Bowel Disease in Canada: Distributed Network Analysis of Multiple Population-Based Provincial Health Administrative Databases. The American Journal of Gastroenterology, 112(7), 1120–1134. https://doi.org/10.1038/ajg.2017.97 Benchimol, E. I., Fortinsky, K., Gozdyra, P., Van Den Heuvel, M., Van Limbergen, J., & Griffiths, A. M. (2011). Epidemiology of pediatric inflammatory bowel disease: A systematic review of international trends. Inflammatory Bowel Diseases, 17(1), 423–439. https://doi.org/10.1002/ibd.21349 Benchimol, E. I., Kaplan, G. G., Otley, A. R., Nguyen, G. C., Underwood, F. E., Guttmann, A., … Bernstein, C. N. (2017). Rural and Urban Residence during Early Life is Associated with a Lower Risk of Inflammatory Bowel Disease: A Population-Based Inception and Birth Cohort Study. American Journal of Gastroenterology, 112(9), 1412–1422. https://doi.org/10.1038/ajg.2017.208 Benchimol, E. I., Mack, D. R., Guttmann, A., Nguyen, G. C., To, T., Mojaverian, N., … Manuel, D. G. (2015). Inflammatory bowel disease in immigrants to Canada and their children: A 83  population-based cohort study. American Journal of Gastroenterology, 110(4), 553–563. https://doi.org/10.1038/ajg.2015.52 Bernstein, C. N., Banerjee, A., Targownik, L. E., Singh, H., Ghia, J. E., Burchill, C., … Roos, L. L. (2016). Cesarean Section Delivery Is Not a Risk Factor for Development of Inflammatory Bowel Disease: A Population-based Analysis. Clinical Gastroenterology and Hepatology, 14(1), 50–57. https://doi.org/10.1016/j.cgh.2015.08.005 Bernstein, C. N., Burchill, C., Targownik, L. E., Singh, H., & Roos, L. L. (2019). Events Within the First Year of Life, but Not the Neonatal Period, Affect Risk for Later Development of Inflammatory Bowel Diseases. Gastroenterology, 156(8), 2190–2197. https://doi.org/10.1053/j.gastro.2019.02.004 Bernstein, C. N., Kraut, A., Blanchard, J. F., Rawsthorne, P., Yu, N., & Walld, R. (2001). The relationship between inflammatory bowel disease and socioeconomic variables. American Journal of Gastroenterology, 96(7), 2117–2125. https://doi.org/10.1016/S0002-9270(01)02509-6 Bernstein, C. N., Wajda, A., Svenson, L. W., Mackenzie, A., Koehoorn, M., Jackson, M., … Blanchard, J. F. (2006). The Epidemiology of Inflammatory Bowel Disease in Canada: A Population-Based Study. American Journal of Gastroenterology, 101(8), 1559–1568. https://doi.org/10.1111/j.1572-0241.2006.00603.x Bivand, R. S., & Wong, D. W. S. (2018). Comparing implementations of global and local indicators of spatial association. Test, 27, 716–748. https://doi.org/10.1007/s11749-018-0599-x Blanchard, J. F., Bernstein, C. N., Wajda, A., & Rawsthorne, P. (2001). Small-area Variations and Sociodemographic Correlates for the Incidence of Crohn’s Disease and Ulcerative 84  Colitis (Vol. 154). Retrieved from https://academic.oup.com/aje/article-abstract/154/4/328/61953 Canadian Urban Environmental Health Research Consortium. (2017). CANUE Metadata Material and Social Depriviation Index (MSDI) V1.0. Retrieved from www.canue.ca Canadian Urban Environmental Health Research Consortium. (2018a). CANUE Metadata NDVI Landsat. Retrieved from http://canue.ca/wp-content/uploads/2018/02/CANUE-Metadata-NDVI-Landsat.pdf Canadian Urban Environmental Health Research Consortium. (2018b). CANUE Metadata NO2 LUR. Retrieved from https://canue.ca/wp-content/uploads/2019/05/CANUE-Metadata-NO2-LUR-Annual.pdf Canadian Urban Environmental Health Research Consortium. (2018c). CANUE Metadata Postal Codes. Retrieved from https://canue.ca/wp-content/uploads/2018/11/CANUE-Metadata-PostalCodes-1.pdf Canadian Urban Environmental Health Research Consortium. (2018d). Postal Code User Guide. Retrieved from http://canue.ca/wp-content/uploads/2018/03/CANUE-Postal-Code-User-Guide.pdf Canadian Urban Environmental Health Research Consortium. (2020). CANUE Metadata Ultra Violet Dose. https://doi.org/10.1016/jphotbiol.2010.05.002 Caprarelli, G., & Fletcher, S. (2020). A brief review of spatial analysis concepts and tools used for mapping, containment and risk modelling of infectious diseases and other illnesses. Parasitology, 141(5), 581–601. https://doi.org/10.1017/S0031182013001972 Carroll, M. W., Hamilton, Z., Gill, H., Simkin, J., Smyth, M., Espinosa, V., … Jacobson, K. (2016). Pediatric Inflammatory Bowel Disease among South Asians Living in British 85  Columbia, Canada: A Distinct Clinical Phenotype. Inflammatory Bowel Diseases, 22(2), 387–396. https://doi.org/10.1097/MIB.0000000000000651 Casey, J. A., James, P., Rudolph, K. E., Wu, C. Da, & Schwartz, B. S. (2016). Greenness and birth outcomes in a range of Pennsylvania communities. International Journal of Environmental Research and Public Health, 13(3), 311. https://doi.org/10.3390/ijerph13030311 Caulley, L., Sawada, M., Hinther, K., Ko, Y. T. I., Crowther, J. A., & Kontorinis, G. (2017). Geographic distribution of vestibular schwannomas in West Scotland between 2000-2015. PLoS ONE, 12(5), e0175489. https://doi.org/10.1371/journal.pone.0175489 Christophe, D., Gower-Rousseau, C., Vernier-Massouille, G., Salleron, J., Baldé, M., Poirier, G., … Colombel, J.-F. (2010). Mapping of inflammatory bowel disease in northern France: Spatial variations and relation to affluence. Inflammatory Bowel Diseases, 16(5), 807–812. https://doi.org/https://doi.org/10.1002/ibd.21111 Chu, H. (2017). Host gene-microbiome interactions: Molecular mechanisms in inflammatory bowel disease. Genome Medicine, 9(69). https://doi.org/10.1186/s13073-017-0459-4 Cilluffo, G., Ferrante, G., Fasola, S., Montalbano, L., Malizia, V., Piscini, A., … La Grutta, S. (2018). Associations of greenness, greyness and air pollution exposure with children’s health: a cross-sectional study in Southern Italy. Environmental Health, 17(86), 1–12. https://doi.org/10.1186/s12940-018-0430-x Diefenbach, K. A., & Breuer, C. K. (2006). Pediatric inflammatory bowel disease. World Journal of Gastroenterology, 12(20), 3204–3212. https://doi.org/10.3748/wjg.v12.i20.3204 DMTI Spatial Inc. (2015). CanMap Postal Code Suite v2015.3. Markham: DMTI Spatial Inc. DMTI Spatial Inc. (2016). CanMap Postal Code Suite v2016.3. Markham: DMTI Spatial Inc. 86  Dzhambov, A. M., Dimitrova, D. D., & Dimitrakova, E. D. (2014). Association between residential greenness and birth weight: Systematic review and meta-analysis. Urban Forestry & Urban Greening, 13, 621–629. https://doi.org/10.1016/j.ufug.2014.09.004 Elliott, C. T., & Copes, R. (2011). Burden of Mortality Due to Ambient Fine Particulate Air Pollution (PM 2.5 ) in Interior. Journal of Public Health / Revue Canadienne de Santé Publique, 102(5), 390–393. Retrieved from www.jstor.org/stable/41995637 Elten, M., Benchimol, E. I., Fell, D. B., Kuenzig, M. E., Smith, G., Chen, H., … Lavigne, E. (2020). Ambient air pollution and the risk of pediatric-onset inflammatory bowel disease: A population-based cohort study. Environment International, 138, 105676. https://doi.org/10.1016/j.envint.2020.105676 Fioletov, V. E., Kimlin, M. G., Krotkov, N., McArthur, L. J. B., Kerr, J. B., Wardle, D. I., … Kaurola, J. (2004). UV index climatology over the United States and Canada from ground-based and satellite estimates. Journal of Geophysical Research D: Atmospheres, 109(22), 1–13. https://doi.org/10.1029/2004JD004820 Fioletov, V. E., McArthur, L. J. B., Mathews, T. W., & Marrett, L. (2010). Estimated ultraviolet exposure levels for a sufficient vitamin D status in North America. Journal of Photochemistry and Photobiology B: Biology, 100(2), 57–66. https://doi.org/10.1016/j.jphotobiol.2010.05.002 Gamache, P., Hamel, D., & Pampalon, R. (2017). The Material and Social Deprivation Index: A Summary. Quebec: Institut National de Santé Publique du Québec. Retrieved from http://canue.ca/wp-content/uploads/2017/11/MSDI_UserGuide_EN.pdf Gascon, M., Cirach, M., Martínez, D., Dadvand, P., Valentín, A., Plasència, A., & Nieuwenhuijsen, M. J. (2016). Normalized difference vegetation index (NDVI) as a marker 87  of surrounding greenness in epidemiological studies: The case of Barcelona city. Urban Forestry and Urban Greening, 19, 88–94. https://doi.org/10.1016/j.ufug.2016.07.001 Gordon, H., Trier Moller, F., Andersen, V., & Harbord, M. (2015, April 3). Heritability in inflammatory bowel disease: From the first twin study to genome-wide association studies. Inflammatory Bowel Diseases. Lippincott Williams and Wilkins. https://doi.org/10.1097/MIB.0000000000000393 Green, C., Elliott, L., Beaudoin, C., & Bernstein, C. N. (2006). A Population-based Ecologic Study of Inflammatory Bowel Disease: Searching for Etiologic Clues. American Journal of Epidemiology, 164(7), 615–623. https://doi.org/10.1093/aje/kwj260 Holm, S. (1979). A Simple Sequentially Rejective Multiple Test Procedure. Scandinavian Journal of Statistics (Vol. 6). Retrieved from http://www.jstor.com/stable/4615733 Holmes, E. A., Xiang, F., & Lucas, R. M. (2015). Variation in Incidence of Pediatric Crohn’s Disease in Relation to Latitude and Ambient Ultraviolet Radiation. Inflammatory Bowel Diseases, 21(4), 809–817. https://doi.org/10.1097/MIB.0000000000000320 Hystad, P., Davies, H. W., Frank, L., Van Loon, J., Gehring, U., Tamburic, L., & Brauer, M. (2014). Residential Greenness and Birth Outcomes: Evaluating the Influence of Spatially Correlated Built-Environment Factors. Environmental Health Perspectives, 122(10), 1095–1102. https://doi.org/10.1289/ehp.1308049 Hystad, P., Setton, E., Cervantes, A., Poplawski, K., Deschenes, S., Brauer, M., … Demers, P. (2011). Creating National Air Pollution Models for Population Exposure Assessment in Canada. Environmental Health Perspectives, 119(8), 1123–1129. https://doi.org/10.14288/1.0220728 Jaime, F., Riutort, M. C., Alvarez-Lobos, M., Hoyos-Bachiloglu, R., Camargo, C. A., & 88  Borzutzky, A. (2017). Solar radiation is inversely associated with inflammatory bowel disease admissions. Scandinavian Journal of Gastroenterology, 52(6–7), 730–737. https://doi.org/10.1080/00365521.2017.1307444 Jantchou, P., Clavel-Chapelon, F., Racine, A., Kvaskoff, M., Carbonnel, F., & Boutron-Ruault, M. C. (2014). High residential sun exposure is associated with a low risk of incident Crohn’s disease in the prospective E3N cohort. Inflammatory Bowel Diseases, 20(1), 75–81. https://doi.org/10.1097/01.MIB.0000436275.12131.4f Janz, T., & Pearson, C. (2013). Vitamin D blood levels of Canadians. Retrieved from https://www150.statcan.gc.ca/n1/pub/82-624-x/2013001/article/11727-eng.pdf Jostins, L., Ripke, S., Weersma, R. K., Duerr, R. H., McGovern, D. P., Hui, K. Y., … Whittaker, P. (2012). Host-microbe interactions have shaped the genetic architecture of inflammatory bowel disease. Nature, 491, 119–124. https://doi.org/10.1038/nature11582 Kaplan, G. G., Bernstein, C. N., Coward Msc, S., Bitton, A., Murthy, S. K., Nguyen, G. C., … Benchimol, E. I. (2019). The Impact of Inflammatory Bowel Disease in Canada 2018: Epidemiology. Journal of the Canadian Association of Gastroenterology, 2(S1), S6–S16. https://doi.org/10.1093/jcag/gwy054 Kaplan, G. G., Hubbard, J., Korzenik, J., Sands, B. E., Panaccione, R., Ghosh, S., … Villeneuve, P. J. (2010). The Inflammatory Bowel Diseases and Ambient Air Pollution: A Novel Association. American Journal of Gastroenterology, 105(11), 2412–2419. https://doi.org/10.1038/ajg.2010.252 Kapoor, A., Bhatia, V., & Sibal, A. (2016). Pediatric Inflammatory Bowel Disease. Indian Pediatrics, 53(11), 993–1002. https://doi.org/10.1007/s13312-016-0975-0 Khalili, H., Huang, E. S., Ananthakrishnan, A. N., Higuchi, L., Richter, J. M., Fuchs, C. S., & 89  Chan, A. T. (2012). Geographical variation and incidence of inflammatory bowel disease among US women. Gut, 61(12), 1686–1692. https://doi.org/10.1136/GUTJNL-2011-301574 Khan, D., Rossen, L. M., Hamilton, B. E., He, Y., Wei, R., & Dienes, E. (2017). Hot spots, cluster detection and spatial outlier analysis of teen birth rates in the U.S., 2003–2012. Spatial and Spatio-Temporal Epidemiology, 21, 67–75. https://doi.org/10.1016/j.sste.2017.03.002 Kirby, R. S., Delmelle, E., & Eberth, J. M. (2017). Advances in spatial epidemiology and geographic information systems. Annals of Epidemiology, 27(1), 1–9. https://doi.org/10.1016/j.annepidem.2016.12.001 Kolho, K. L., Korpela, K., Jaakkola, T., Pichai, M. V. A., Zoetendal, E. G., Salonen, A., & De Vos, W. M. (2015). Fecal microbiota in pediatric inflammatory bowel disease and its relation to inflammation. American Journal of Gastroenterology, 110(6), 921–930. https://doi.org/10.1038/ajg.2015.149 Kulldorff, M. (1997). A spatial scan statistic. Communications in Statistics - Theory and Methods, 26(6), 1481–1496. https://doi.org/10.1080/03610929708831995 Lambert, K. A., Bowatte, G., Tham, R., Lodge, C., Prendergast, L., Heinrich, J., … Erbas, B. (2017). Residential greenness and allergic respiratory diseases in children and adolescents – A systematic review and meta-analysis. Environmental Research, 159, 212–221. https://doi.org/10.1016/j.envres.2017.08.002 Landsat 5 TM Annual Greenest-Pixel TOA Reflectance Composite, 1984 to 2012. (n.d.). Retrieved July 20, 2017, from https://explorer.earthengine.google.com/#detail/LANDSAT%2FLT5_L1T_ANNUAL_GREENEST_TOA 90  Landsat 8 Annual Greenest-Pixel TOA Reflectance Composite, 2013 to 2015. (n.d.). Retrieved July 20, 2017, from https://explorer.earthengine.google.com/#detail/LANDSAT%2FLC8_L1T_ANNUAL_GREENEST_TOA Limketkai, B. N., Bayless, T. M., Brant, S. R., & Hutfless, S. M. (2014). Lower regional and temporal ultraviolet exposure is associated with increased rates and severity of inflammatory bowel disease hospitalisation. Alimentary Pharmacology & Therapeutics, 40(5), 508–517. https://doi.org/10.1111/apt.12845 Lorimer, J. (2017). Parasites, ghosts and mutualists: a relational geography of microbes for global health. Transactions of the Institute of British Geographers, 42(4), 544–558. https://doi.org/10.1111/tran.12189 Månsson, R., Joffe, M. M., Sun, W., & Hennessy, S. (2007). On the estimation and use of propensity scores in case-control and case-cohort studies. American Journal of Epidemiology, 166(3), 332–339. https://doi.org/10.1093/aje/kwm069 Markevych, I., Fuertes, E., Tiesler, C. M. T., Birk, M., Bauer, C. P., Koletzko, S., … Heinrich, J. (2014). Surrounding greenness and birth weight: Results from the GINIplus and LISAplus birth cohorts in Munich. Health and Place, 26, 39–46. https://doi.org/10.1016/j.healthplace.2013.12.001 Markevych, I., Thiering, E., Fuertes, E., Sugiri, D., Berdel, D., Koletzko, S., … Heinrich, J. (2014). A cross-sectional analysis of the effects of residential greenness on blood pressure in 10-year old children: Results from the GINIplus and LISAplus studies. BMC Public Health, 14. https://doi.org/10.1186/1471-2458-14-477 Marshall, R. J. (1991). Mapping disease and mortality rates using empirical Bayes estimators. 91  Journal of the Royal Statistical Society. Series C, Applied Statistics, 40(2), 283–294. https://doi.org/10.2307/2347593 Ministry of Health. (2019). Local Health Area Boundaries. Mutlu, E. A., Comba, I. Y., Cho, T., Engen, P. A., Yazıcı, C., Soberanes, S., … Mutlu, G. M. (2018). Inhalational exposure to particulate matter air pollution alters the composition of the gut microbiome. Environmental Pollution, 240, 817–830. https://doi.org/10.1016/j.envpol.2018.04.130 Mutlu, E. A., Engen, P. A., Soberanes, S., Urich, D., Forsyth, C. B., Nigdelioglu, R., … Mutlu, G. M. (2011). Particulate matter air pollution causes oxidant-mediated increase in gut permeability in mice. Particle and Fibre Toxicology, 8, 19. https://doi.org/10.1186/1743-8977-8-19 Nelson, J. K., & Brewer, C. A. (2015). Evaluating data stability in aggregation structures across spatial scales: revisiting the modifiable areal unit problem. Cartography and Geographic Information Science, 44(1), 35–50. https://doi.org/10.1080/15230406.2015.1093431 Ng, S. C., Bernstein, C. N., Vatn, M. H., Lakatos, P. L., Loftus, E. V, Tysk, C., … Colombel, J.-F. (2013). Geographical variability and environmental risk factors in inflammatory bowel disease. Gut, 62, 630–649. https://doi.org/10.1136/GUTJNL-2012-303661 Office of the Provincial Health Officer, & First Nations Health Authority. (2018). Indigenous Health and Well-being. Retrieved from https://www.fnha.ca/Documents/FNHA-PHO-Indigenous-Health-and-Well-Being-Report.pdf Opstelten, J. L., Beelen, R. M. J., Leenders, M., Hoek, G., Brunekreef, B., van Schaik, F. D. M., … Oldenburg, B. (2016). Exposure to Ambient Air Pollution and the Risk of Inflammatory Bowel Disease: A European Nested Case–Control Study. Digestive Diseases and Sciences, 92  61(10), 2963–2971. https://doi.org/10.1007/s10620-016-4249-4 Ouédraogo, A. M., Crighton, E. J., Sawada, M., To, T., Brand, K., & Lavigne, E. (2018). Exploration of the spatial patterns and determinants of asthma prevalence and health services use in Ontario using a Bayesian approach. PLOS ONE, 13(12), e0208205. https://doi.org/10.1371/journal.pone.0208205 Pampalon, R., Hamel, D., Gamache, P., Philibert, M. D., Raymond, G., & Simpson, A. (2012). An Area-based Material and Social Deprivation Index for Public Health in Québec and Canada. Journal of Public Health / Revue Canadienne de Santé Publique, 103(Supplement 2), S17–S22. Retrieved from https://www.jstor.org/stable/41995684 Peloquin, J. M., Goel, G., Villablanca, E. J., & Xavier, R. J. (2016). Mechanisms of Pediatric Inflammatory Bowel Disease. Annual Review of Immunology, 34, 31–64. https://doi.org/10.1146/annurev-immunol-032414-112151 Persson, Pyko, A., Lind, T., Bellander, T., Östenson, C. G., Pershagen, G., … Lõhmus, M. (2018). Urban residential greenness and adiposity: A cohort study in Stockholm County. Environment International, 121(1), 832–841. https://doi.org/10.1016/j.envint.2018.10.009 Pinsk, V., Lemberg, D. A., Grewal, K., Barker, C. C., Schreiber, R. A., & Jacobson, K. (2007). Inflammatory bowel disease in the South Asian pediatric population of British Columbia. American Journal of Gastroenterology, 102(5), 1077–1083. https://doi.org/10.1111/j.1572-0241.2007.01124.x Prescott, S. L., Wegienka, G., Logan, A. C., & Katz, D. L. (2018). Dysbiotic drift and biopsychosocial medicine: How the microbiome links personal, public and planetary health. BioPsychoSocial Medicine, 12(1), 1–12. https://doi.org/10.1186/s13030-018-0126-z Provincial Health Services Authority. (n.d.). BC Community Health Atlas. Retrieved June 17, 93  2020, from http://maps.gov.bc.ca/ess/hm/cha/ R Core Team. (2020). R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. Retrieved from https://www.r-project.org/ Ramos, G. P., & Papadakis, K. A. (2019). Mechanisms of Disease: Inflammatory Bowel Diseases. Mayo Clinic Proceedings, 94(1), 155–165. https://doi.org/10.1016/j.mayocp.2018.09.013 Reid, C. E., Kubzansky, L. D., Li, J., Shmool, J. L., & Clougherty, J. E. (2018). It’s not easy assessing greenness: A comparison of NDVI datasets and neighborhood types and their associations with self-rated health in New York City. Health and Place, 54, 92–101. https://doi.org/10.1016/j.healthplace.2018.09.005 Richmond, C. A. M., & Cook, C. (2016). Creating conditions for Canadian aboriginal health equity: The promise of healthy public policy. Public Health Reviews, 37, 2. https://doi.org/10.1186/s40985-016-0016-5 Rook, G. (2013). Regulation of the immune system by biodiversity from the natural environment: an ecosystem service essential to health. Proceedings of the National Academy of Sciences of the United States of America, 110(46), 18360–18367. https://doi.org/10.1073/pnas.1313731110 Rook, G., Raison, C. L., & Lowry, C. A. (2014). Microbial “old friends”, immunoregulation, and socioeconomic status. Clinical and Experimental Immunology, 177, 38–46. https://doi.org/10.1111/cei.12269 Rugel, E. J., Henderson, S. B., Carpiano, R. M., & Brauer, M. (2017). Beyond the Normalized Difference Vegetation Index (NDVI): Developing a Natural Space Index for population-level health research. Environmental Research, 159, 474–483. 94  https://doi.org/10.1016/j.envres.2017.08.033 Ryz, N. R., Lochner, A., Bhullar, K., Ma, C., Huang, T., Bhinder, G., … Vallance, B. A. (2015). Dietary vitamin D3 deficiency alters intestinal mucosal defense and increases susceptibility to Citrobacter rodentium-induced colitis. American Journal of Physiology-Gastrointestinal and Liver Physiology, 309(9), G730–G742. https://doi.org/10.1152/ajpgi.00006.2015 Select Standing Committee on Health. (2017). Looking Forward: Improving Rural Health Care, Primary Care, and Addiction Recovery Programs . Victoria, BC. Retrieved from https://www.leg.bc.ca/content/CommitteeDocuments/40th-parliament/6th-session/health/Health_20170301_Report_LookingForward.pdf Shivananda, S., Lennard-Jones, J., Logan, R., Fear, N., Price, A., Carpenter, L., & Van Blankenstein, M. (1996). Incidence of inflammatory bowel disease across Europe: Is there a difference between north and south? Results of the European collaborative study on inflammatory bowel disease (EC-IBD). Gut, 39(5), 690–697. https://doi.org/10.1136/gut.39.5.690 Simkin, J., Woods, R., & Elliott, C. (2017). Cancer mortality in Yukon 1999–2013: elevated mortality rates and a unique cancer profile. International Journal of Circumpolar Health, 76(1), 1324231. https://doi.org/10.1080/22423982.2017.1324231 Smith, T., & McLeish, S. (2019). Technical report on changes in response related to the census ethnic origin question: Focus on Jewish origins, 2016 Census integrated with 2011 National Household Survey. Retrieved from https://www150.statcan.gc.ca/n1/pub/89-657-x/89-657-x2019009-eng.htm Sonnenberg, A., Mccarty, D. J., & Jacobsen, S. J. (1991). Geographic Variation of Inflammatory Bowel Disease Within the United States. GASTROENTEROLOGY (Vol. 100). 95  Sperandei, S. (2014). Understanding logistic regression analysis. Biochemia Medica, 24(1), 12–18. https://doi.org/10.11613/BM.2014.003 Statistics Canada. (n.d.-a). 1996 Census / Profile of Enumeration Areas / Ethnic and Aboriginal origins, and visible minorities. Canadian Census Analyser (database). Using CHASS (distributor). Retrieved from http://dc1.chass.utoronto.ca/cgi-bin/census/1996/displayCensusDA.cgi?c=eth Statistics Canada. (n.d.-b). 2001 Census / Profile of Dissemination Areas / Citizenship, immigration, birthplace, generation status, ethnic origin, visible minorities and aboriginal peoples. Canadian Census Analyser (database). Using CHASS (distributor). Retrieved from http://dc1.chass.utoronto.ca/cgi-bin/census/2001/displayCensusDA.cgi?c=cip Statistics Canada. (n.d.-c). 2006 Census / Profile of Dissemination Areas / Aboriginal peoples. Retrieved February 10, 2020, from http://dc1.chass.utoronto.ca/cgi-bin/census/2006/displayCensusDA.cgi?c=abo Statistics Canada. (n.d.-d). 2006 Census / Profile of Dissemination Areas / Ethnic origin and visible minorities. Retrieved February 10, 2020, from http://dc1.chass.utoronto.ca/cgi-bin/census/2006/displayCensusDA.cgi?c=eth Statistics Canada. (n.d.-e). 2011 National Household Survey (NHS) / Profile of Census Disseminations Areas / Aboriginal population. Canadian Census Analyser (database). Using CHASS (distributor). Retrieved from http://dc1.chass.utoronto.ca/cgi-bin/census/2011nhs/displayCensus.cgi?year=2011&geo=da Statistics Canada. (n.d.-f). 2011 National Household Survey (NHS) / Profile of Census Disseminations Areas / Ethnic origin population. Retrieved February 10, 2020, from http://dc1.chass.utoronto.ca/cgi-bin/census/2011nhs/displayCensus.cgi?year=2011&geo=da 96  Statistics Canada. (n.d.-g). 2016 Census Profiles / Profile of Census Dissemination Areas / Population and dwelling counts. Canadian Census Analyser (database). Using CHASS (distributor). Retrieved from http://dc1.chass.utoronto.ca/cgi-bin/census/2016/displayCensus.cgi?year=2016&geo=da Statistics Canada. (n.d.-h). 2016 Census Profiles / Profile of Census Disseminations Areas / Aboriginals and Visible Minorities. Canadian Census Analyser (database). Using CHASS (distributor). Retrieved from http://dc1.chass.utoronto.ca/cgi-bin/census/2016/displayCensus.cgi?year=2016&geo=da Statistics Canada. (n.d.-i). 2016 Census Profiles / Profile of Census Disseminations Areas / Ethnic origin. Canadian Census Analyser (database). Using CHASS (distributor). Retrieved from http://dc1.chass.utoronto.ca/cgi-bin/census/2016/displayCensus.cgi?year=2016&geo=da Statistics Canada. (2016, November 16). Dictionary, Census of Population, 2016 - Dissemination area (DA). Retrieved July 25, 2020, from https://www12.statcan.gc.ca/census-recensement/2016/ref/dict/geo021-eng.cfm Statistics Canada. (2017a). Aboriginal Peoples Reference Guide, Census of Population, 2016. Modified 2019-01-03. Retrieved from https://www12.statcan.gc.ca/census-recensement/2016/ref/guides/009/98-500-x2016009-eng.cfm Statistics Canada. (2017b). Mobility Status 5 Years Ago (9), Mother Tongue (8), Marital Status (9), Age(18) and Sex (3) for the Population Aged 5 Years and Over in Private Households of Canada, Provinces and Territories, Census Metropolitan Areas and Census Agglomerations, 2016 Cens. Catalogue number 98-400-X2016311. Last modified 2019-06-17. 97  Statistics Canada. (2017c). Surrey, CY [Census subdivision], British Columbia, Delta, DM [Census subdivision], British Columbia and British Columbia [Province] (table). Ottawa. Retrieved from https://www12.statcan.gc.ca/census-recensement/2016/dp-pd/prof/index.cfm?Lang=E Statistics Canada. (2018). Dissemination area: Detailed definition. Retrieved July 25, 2020, from https://www150.statcan.gc.ca/n1/pub/92-195-x/2011001/geo/da-ad/def-eng.htm Stein, A. C., Gaetano, J. N., Jacobs, J., Kunnavakkam, R., Bissonnette, M., & Pekow, J. (2016). Northern Latitude but Not Season Is Associated with Increased Rates of Hospitalizations Related to Inflammatory Bowel Disease: Results of a Multi-Year Analysis of a National Cohort. PloS One, 11(8), e0161523. https://doi.org/10.1371/journal.pone.0161523 Tennekes, M. (2018). Tmap: Thematic maps in R. Journal of Statistical Software, 84(6), 1–39. https://doi.org/10.18637/jss.v084.i06 Teucher, A., Hazlitt, S., Albers, S., & Province of British Columbia. (2017). bcmaps. Environmental Reporting BC. Retrieved from https://github.com/bcgov/bcmaps Tjur, T. (2009). Coefficients of Determination in Logistic Regression Models- A New Proposal: The Coefficient of Discrimination. The American Statistician, 63(4), 366–372. https://doi.org/10.1198/tast.2009.08210 USGS Landsat 5 TM TOA Reflectance (Orthorectified), 1984 to 2011. (n.d.). Retrieved July 20, 2007, from https://explorer.earthengine.google.com/#detail/LANDSAT%2FLT5_L1T_TOA USGS Landsat 8 TOA Reflectance (Orthorectified), 2013 to 2017. (n.d.). Retrieved July 20, 2007, from https://explorer.earthengine.google.com/#detail/LANDSAT%2FLC8_L1T_TOA Vernia, P., Burrelli Scotti, G., Dei Giudici, A., Chiappini, A., Cannizzaro, S., Afferri, M. T., & 98  de Carolis, A. (2018). Inadequate sunlight exposure in patients with inflammatory bowel disease. Journal of Digestive Diseases, 19(1), 8–14. https://doi.org/10.1111/1751-2980.12567 Vingarzan, R., & Li, S. M. (2006). The Pacific 2001 Air Quality Study-synthesis of findings and policy implications. Atmospheric Environment, 40(15), 2637–2649. https://doi.org/10.1016/j.atmosenv.2005.09.083 Weichenthal, S., Pinault, L. L., & Burnett, R. T. (2017). Impact of Oxidant Gases on the Relationship between Outdoor Fine Particulate Air Pollution and Nonaccidental, Cardiovascular, and Respiratory Mortality. Scientific Reports, 7(1), 16401. https://doi.org/10.1038/s41598-017-16770-y Wickham, H. (2016). ggplot2: Elegant Graphics for Data Analysis. New York: Springer-Verlag. Retrieved from https://ggplot2.tidyverse.org World Health Organization. (2018). Ambient (outdoor) air pollution. Retrieved from https://www.who.int/news-room/fact-sheets/detail/ambient-(outdoor)-air-quality-and-health Xia, B., Crusius, J. B. A., Meuwissen, S. G. M., & Peña, A. S. (1998). Inflammatory bowel disease: Definition, epidemiology, etiologic aspects, and immunogenetic studies. World Journal of Gastroenterology, 4(5), 446–458. https://doi.org/10.3748/wjg.v4.i5.446 Yasaitis, L. C., Arcaya, M. C., & Subramanian, S. V. (2015). Comparison of estimation methods for creating small area rates of acute myocardial infarction among Medicare beneficiaries in California. Health and Place, 35, 95–104. https://doi.org/10.1016/j.healthplace.2015.08.003  99  Appendices  Appendix A    Materials in Appendix A relate to the spatial cluster analysis described in Chapter 3.   A.1 Formulas The Moran’s “I” statistic can be used to measure global spatial autocorrelation for a study area, and is calculated as: 𝐼 = ' 𝑛𝑆#*	,.$ ,.% 𝑤$%𝑧$𝑧% 	/	,.$ 𝑧$" Where 𝑛 = number of spatial units (Local Health Areas, in this study). 𝑆# = aggregate of all spatial weights. 𝑤$% = spatial weight between location 𝑖 and location 𝑗. 𝑧$ = deviation from the attribute (Standardized Incidence Ratio, in this study) at location 𝑖 from its mean. The calculation of the Moran’s “I” statistic relies on a spatial weight matrix to mathematically summarize the spatial relationship between features. In this study, directly adjacent Local Health Areas were assigned a weight of 1, while non-adjacent Local Health Areas were assigned a weight of 0.  The local Moran’s “I” statistic is a Local Indicator of Spatial Association that can be used to identify local patterns in spatial autocorrelation and assess the contribution of a particular local area to the global Moran’s “I” statistic. If all the local Moran’s “I” statistics for a given study 100  area are summed, they equal the global Moran’s “I” for that study area. The local Moran’s “I” for a given study region is calculated as: 𝐼𝑖 = 𝑧$ 	,𝑤$%𝑧%%   The Standardized Incidence Ratio (SIR) was used to measure how much the number of cases in each Local Health Area (LHA) differed from what would be expected if the incidence in that area was the same as the province of BC. The SIR is calculated as: SIR = (number	of	cases	in	area	𝑖)	(reference	rate	 × 	population	in	area	𝑖) 		× 	100 For each Local Health Area in this study, the SIR was calculated as: SIR = (LHA	case	count)(BC	incidence	 × 	LHA	population) 	× 	100	 For example: if a hypothetical Local Health Area had two IBD cases and a population of 10,000, and the BC incidence was 10 per 100,000 (0.0001), the SIR would be 200% and would signify that the example Local Health Area had double the expected IBD cases.  A.2 IBD incidence rate maps The following maps depict incidence rates at the Local Health Area level that were used to calculate Standardized Incidence Ratios for cluster detection (see Chapter 3). To protect patient privacy and reduce uncertainty caused by low case counts and small populations, incidence rates have been smoothed according to the procedure described in Chapter 3.  101   102   103  104   105   106           107  Appendix B   Material in Appendix B relate to the modelling of potential environmental and social determinants of IBD described in Chapter 4.   B.1 Formulas The Normalized Difference Vegetation Index (NDVI) was used to measure greenness in this study. It is calculated from a ratio of remotely-sensed visible red (RED) reflectance to near infrared (NIR) measured reflectance is used to calculate NDVI with the following formula (Persson et al., 2018):  NDVI = (NIR − RED)	(NIR + RED)   Generalized linear models with binomial probability distribution and logit link (also known as logistic regression models) were used to model the association between a binary outcome variable (in this study, diagnosis with IBD vs. no diagnosis with IBD) and various explanatory variables (in this study, NO2, UV vitamin D, NDVI, ethnicity, Aboriginal identity, and socioeconomic status) (Sperandei, 2014). They can be calculated as: log ' 𝜌1 − 	𝜌* = 	𝛽& +	𝛽'𝑥' +	𝛽"𝑥"…	𝛽(𝑥( Where 𝜌 is the probability of the event (diagnosis with IBD) mathematically represented as 1.  1 − 	𝜌 is probability of the alternative (no diagnosis with IBD) mathematically represented as 0.  Z )'*	)[ is the odds of diagnosis with IBD.  𝛽& is the y-intercept, where the regression curve crosses the Y axis. 𝛽' is the regression coefficient associated with explanatory variable 𝑥'. 108   B.2 Data maps The following maps depict data from CANUE and the Canadian Census used in regression modelling of the relationship between IBD and variables of interest. Data from CANUE was provided at postal code points; to protect patient privacy, I used a random sample of BC postal codes (rather than case and control locations) from 2008 to visualize the distribution of each CANUE variable in the province. Census variables are depicted for every available Dissemination Area from 2006.       109    110    111  112   113  114    115        116  B.3 Regression results  All univariate and multivariate logistic regression model results are presented below. Variable names are explained in the following table.   Name Variable Notes NO2LUR_A_02 NO2 air pollution  GRLAN_01 NDVI greenness at postal code location  GRLAN_02 NDVI greenness at 100 m buffer  GRLAN_03 NDVI greenness at 250 m buffer Results not included in findings due to correlation with 100 m NDVI. GRLAN_05 NDVI greenness at 1000 m buffer  MDVD_Jan Vitamin D UV for January Month name abbreviated after MDVD_ MSD_08 Material deprivation  MSD_09 Social deprivation  perSA Percent of Dissemination/ Enumeration area of South Asian ethnic origin  perAb Percent of Dissemination/ Enumeration area who self-identified as Aboriginal  perJewish Percent of Dissemination/ Enumeration area of Jewish ethnic origin       117  Provincial models of IBD diagnosed between ages 0 to 16.9    118   119   120  Provincial models of IBD diagnosed between ages 0 – 9.9    121   122   123  Provincial models of IBD diagnosed between ages 10 – 16.9    124   125   126  Provincial models of CD diagnosed between ages 0 – 16 127  128   129  Provincial models of CD diagnosed between ages 0 – 9.9 130  131   132  Provincial models of CD diagnosed between ages 10 – 16.9    133   134   135  Provincial models of UC diagnosed between ages 0 – 16.9 136  137   138  Provincial models of UC diagnosed between ages 0 – 9.9 139  140   141  Provincial models of UC diagnosed between ages 10 – 16.9142  143   144  Urban (Fraser Health and Vancouver Coastal Health) models of IBD diagnosed between ages 0 – 16.9 145  146   147  Urban (Fraser Health and Vancouver Coastal Health) models of CD diagnosed between ages 0 – 16.9    148   149   150  Urban (Fraser Health and Vancouver Coastal Health) models of UC diagnosed between ages 0 – 16.9 151  152   153  Fraser Health models of IBD diagnosed between ages 0 – 16.9 154  155   156  Fraser Health models of CD diagnosed between ages 0 – 16.9  157   158   159  Fraser Health models of UC diagnosed between ages 0 – 16.9    160   161   

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}]}"
                            data-media="{[{embed.selectedMedia}]}"
                            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:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0394067/manifest

Comment

Related Items