FROM EDUCATION TO TUMOUR CHARACTERISTICS IN COLORECTAL CANCER: AN ANALYSIS OF THE PATHWAYS

FROM EDUCATION TO TUMOUR CHARACTERISTICS IN COLORECTAL CANCER: AN ANALYSIS OF THE PATHWAYS By Parisa Airia A thesis submitted in conformity with the requirements for the degree of PhD Dalla Lana School of Public Health University of Toronto Copyright by Parisa Airia 2013

From education to tumour characteristics in colorectal cancer: an analysis of the pathways Abstract Parisa Airia Doctor of Philosophy Dalla Lana School of Public Health University of Toronto 2013 Background: Genetic and environmental factors have been associated with colorectal cancer (CRC) risk. However, their association with prognosis has been less studied. Methods: Path analysis was employed to examine causal pathways from education to environmental (diet, alcohol, smoking, physical activity) and personal factors (screening), and then to obesity and ultimately to tumour characteristics (stage, grade, microsatellite instability (MSI), and site) that are associated with CRC prognosis. Data came from the Ontario Familial Colon Cancer Registry. Pathways were evaluated for effect modification by sex and two indicators of CRC genetic susceptibility (Bethesda criteria and newly identified familial cancer clusters). Results: Four food patterns (healthy foods, high-fat foods, sweet and processed foods, and oriental foods) and four nutrient patterns (total macronutrient, fat vs. carbohydrate, and micronutrients from supplements and from foods) were identified. Education was associated positively with healthy lifestyle factors (e.g. healthy foods factor) and negatively with unhealthy factors (e.g. smoking). As expected, high body mass index (BMI) was associated with lower physical activity and higher fat vs. carbohydrate factor. Unexpectedly, BMI was positively associated with the healthy foods factor among Bethesda positive patients and men. An association between education and BMI was mediated by the healthy foods factor and by physical activity. Important poor prognostic factors, higher grade and stage, were associated with smoking and not being screened. However, unexpected associations included a positive ii

association of physical activity with tumour grade among Bethesda positive patients and a positive association of healthy foods with stage among Bethesda negative patients. Patients with right-sided tumours were more likely to receive micronutrients from supplements, and screening and less likely to smoke, and for men, to have a high BMI, high fat diet and healthy food diet. Conclusion: Some unhealthy lifestyle factors, such as smoking and a high fat food dietary pattern, are associated with adverse CRC tumour characteristics and so may affect the prognosis. Family history may modify some associations though the findings require independent confirmation. iii

Acknowledgments I would like to thank my thesis supervisor, Dr. Gail Eyssen, for her support, patience, advice and guidance throughout my studies and for the countless hours of her time that she contributed to this thesis. I would like to extend my heartfelt thanks to all my committee members, Drs. Pierre Cote, Loraine Marrett, Andrew Seary, and Steven Gallinger for all their support and contributions without whose help I could not complete this thesis. I will also remain indebted to Drs. Elizabeth Badley, Anthony Miller, John McLaughlin, and Polly Newcomb for reviewing my thesis and providing me with their invaluable input. Many thanks to Darshana Daftary and Dr. Michelle Cotterchio and Rose Sarvaria for patiently responding to my questions and helping me with obtaining and understanding my data. This thesis is dedicated to my parents, my daughter, and my sister. Words fail to express my gratitude to my parents Azam Malekzadeh and Daryoush Airia for their unending love and support, beyond imagination, who made it possible for me to continue my studies under all life circumstances; and to my amazing daughter Roxana for her patience and for sacrificing her fun and her time with me to make it possible for me to complete this work. Thanks to my sister, Atoosa, for her continuous emotional support throughout my life. iv

Table of Contents Acknowledgments... iv Table of Contents... v List of Tables... xi List of Figures... xiv List of Appendices... xv Chapter 1. Introduction... 1 1.1 Background... 1 1.2 Research Problem... 1 1.3 Study Objectives... 7 1.4 Importance of this Study... 9 Chapter 2. Literature Review... 11 2.1 Introduction... 11 2.2 Tumour characteristics related to CRC survival... 11 2.2.1 Clinicopathologic tumour characteristics... 11 2.2.1.1 Stage... 11 2.2.1.2 Grade... 13 2.2.1.3 Site... 13 2.2.2 Genetic tumour characteristics... 15 2.2.2.1 Chromosomal Instability pathway (CIN) or microsatellite stable (MSS)... 15 2.2.2.2 Microsatellite Instability (MSI)... 15 2.3 Familial patterns related to the occurrence of CRC... 20 2.3.1 HNPCC (Hereditary Non-Polyposis CRC)... 21 2.3.2 Newly identified familial clusters of CRC and other cancers... 24 v

2.4 Personal factors associated with CRC incidence and survival... 25 2.4.1 Diet in relation to CRC incidence and survival... 25 2.4.1.1 Diet and CRC incidence... 25 2.4.1.2 Diet and CRC survival... 44 2.4.1.3 Diet and tumour characteristics... 47 2.4.2 Physical activity in relation to CRC incidence and survival... 48 2.4.2.1 Physical activity and CRC incidence... 48 2.4.2.2 Physical activity and CRC Survival... 49 2.4.3 Obesity in relation to CRC incidence and survival... 50 2.4.3.1 Obesity and CRC incidence... 50 2.4.3.2 Obesity and CRC Survival... 51 2.4.4 Socioeconomic Status (SES) in relation to CRC incidence and survival... 53 2.4.4.1 SES and CRC incidence... 53 2.4.4.2 SES and CRC survival... 54 2.4.5 Screening in relation to CRC incidence and survival... 56 2.4.5.1 Screening and CRC incidence... 56 2.4.5.2 Screening and CRC survival... 57 2.4.6 Smoking in relation to CRC incidence and survival... 57 2.4.6.1 Smoking and CRC incidence... 57 2.4.6.2 Smoking and CRC survival... 58 2.4.7 Age and CRC survival... 58 2.4.8 Sex and CRC survival... 59 2.5 Measurement issues... 59 2.5.1 Issues with the measurement of tumour characteristics... 59 2.5.1.1 Tumour grade... 60 vi

2.5.1.2 MSI and its associated genetic mutations... 60 2.5.1.3 Tumour site... 62 2.5.1.4 Tumour stage... 63 2.5.2 Issues with the measurement of obesity... 63 2.5.3 Issues with the measurement of physical activity... 64 2.5.4 Issues with the measurement of diet... 64 2.5.5 Issues with the measurement of SES... 65 2.6 Pathways to tumour characteristics... 66 2.6.1 Possible mechanisms for the association of Lifestyle factors and CRC... 66 Chapter 3. Methods... 70 3.1 Introduction... 70 3.2 Study Design... 72 3.3 Study Sample... 72 3.3.1 Structure and methods of OFCCR... 73 3.3.1.1 Cases... 74 3.3.1.2 Family members... 75 3.3.1.3 Controls... 75 3.4 Data collection and Variable definitions... 75 3.4.1 Personal history questionnaire (PHQ)... 76 3.4.1.1 Physical activity... 76 3.4.1.2 Smoking... 76 3.4.1.3 BMI... 77 3.4.1.4 Screening... 77 3.4.1.5 Demographic characteristics (SES, age, sex)... 77 3.4.2 Family history questionnaire... 78 vii

3.4.2.1 Family history... 78 3.4.3 The food frequency questionnaire... 80 3.4.3.1 Alcohol consumption... 81 3.4.3.2 Dietary patterns... 81 3.4.4 Review of medical records... 81 3.4.4.1 Grade... 82 3.4.4.2 Stage... 82 3.4.4.3 Site... 82 3.4.4.4 MSI... 83 3.5 Analysis... 83 3.5.1 Data cleaning... 83 3.5.2 Descriptive analysis... 84 3.5.2.1 Principal component analysis... 84 3.5.2.2 Distribution of data... 84 3.5.3 Analysis of associations... 84 3.5.3.1 Standard analysis... 85 3.5.3.2 Path Analysis... 88 Chapter 4. Results... 90 4.1 Introduction... 90 4.2 Descriptive Analysis... 91 4.2.1 Distribution of the variables... 91 4.2.2 Principal Component Analysis... 95 4.2.2.1 Food Patterns... 96 4.2.2.2 Nutrient Patterns... 101 4.2.3 Summary of Section... 107 viii

4.3 Network Analysis for Identifying Familial Clustering... 107 4.4 Standard Analysis of Associations... 108 4.4.1 Associations among lifestyle factors... 108 4.4.2 Association between education and lifestyle factors... 110 4.4.3 Associations among education, lifestyle and personal factors and BMI and tumour characteristics... 112 4.4.3.1 Associations with BMI... 113 4.4.3.2 Summary of associations with tumour characteristics... 124 4.4.3.3 Detailed associations with tumour characteristics... 130 4.5 Path analysis... 130 4.5.1 Comparison of the results of Path analysis and Standard analysis... 131 4.5.2 Summary of the section... 132 Chapter 5. Discussion... 141 5.1 Introduction... 141 5.2 Identified dietary patterns... 142 5.2.1 Limitations of the identified patterns... 145 5.3 Associations among lifestyle factors... 148 5.4 Relationship of Education with diet, other lifestyle factors, and screening... 149 5.4.1 Education with diet... 149 5.4.2 Education with other lifestyle factors... 150 5.4.3 Education with screening... 151 5.5 Predictors of BMI... 152 5.5.1 Diet... 152 5.5.2 Lifestyle factors... 154 5.5.3 Education... 155 5.5.4 Mediated pathways to BMI... 156 ix

5.6 Predictors of tumour characteristics... 158 5.6.1 Predictors of tumour Grade... 159 5.6.2 Predictors of tumour MSI... 165 5.6.3 Predictors of tumour site... 167 5.6.4 Predictors of tumour stage... 171 5.6.5 Summary... 173 5.7 Comparison of results of path analysis with standard analysis... 175 5.8 Study limitations... 177 5.8.1 Data limitations... 177 5.8.2 Design limitations... 182 5.8.3 Analysis limitations... 184 5.9 Implications of the study... 185 5.10 Recommendations for Future Research... 186 References... 188 x

List of Tables Table 2.1 Dukes system and its association with TNM system [71]... 12 Table 2.2 Distribution of colon and rectal cancer [75]... 14 Table 2.3 Clinicopathologic differences between MSI-H and MSS tumours [43]... 17 Table 2.4 Clinicopathologic differences between MSI-H tumours in sporadic and HNPCC cases [43]... 19 Table 2.5 Frequency of MSI and MSS tumours [43]... 19 Table 2.6 Amsterdam criteria I and II for HNPCC [43]... 22 Table 2.7 Bethesda Guidelines for MSI Testing: Tumours From Any of the Following Should be Tested for MSI (by immunohistochemistry) and Then Positive Patients Should Continue for MMR Testing [128]... 23 Table 2.8 Modified Bethesda Guidelines for MSI Testing: Tumours From Any of the Following Should be Tested for MSI (by immunohistochemistry) and Then Positive Patients Should Continue for MMR Testing [128]... 23 Table 2.9 Newly identified familial clusters of CRC... 24 Table 2.10 Nnutritional factors associated with CRC risk... 27 Table 2.11 Strength of associations between foods/nutrients and CRC risk: results of metaanalyses or range of estimates from individual studies when meta-analysis has not been performed [19].... 38 Table 2.12 Dietary Patterns associated with CRC/Adenoma risk.... 42 Table 3.1 Regression methods used for the analysis of associations.... 86 Table 4.1: Mean and frequency distribution of variables in the grade sample.... 94 xi

Table 4.2: Component matrix showing the component score coefficients for each food item loading on the first food pattern (healthy foods) in grade sample.... 98 Table 4.3: Component matrix showing the component score coefficients for each food item loading on the second food pattern (high-fat foods) in grade sample.... 99 Table 4.4: Component matrix showing the component score coefficients for each food item loading on the third food pattern (Oriental foods) in grade sample.... 100 Table 4.5: Component matrix showing the component score coefficients for each food item loading on the fourth food pattern (Sweet and processed foods) in grade sample.... 101 Table 4.6: Component matrix showing the component score coefficients for each macronutrient loading on the first macronutrient pattern (total macronutrients) in grade sample.... 103 Table 4.7Component matrix showing the component score coefficients for each macronutrient loading on the second macronutrient pattern (fat vs. carbohydrate) in grade sample.... 104 Table 4.8: Component matrix showing the component score coefficients for each micronutrient loading on the first micronutrient pattern (total micronutrients from food)... 105 Table 4.9: Component matrix showing the component score coefficients for each micronutrient loading on the first micronutrient pattern (total micronutrients from food) in grade sample.... 106 Table 4.10 Newly identified familial clusters of CRC.... 108 Table 4.11: Association of education and continuously measured life-style factors (beta from linear regression models) in grade sample, N=926.... 111 Table 4.12: Association of education and binary life-style factors (odds ratio or exponential of beta from logistic regression models) in grade sample, N=926.... 111 Table 4.13: Education, lifestyle factors, food patterns in association with BMI, considering effect modification by Bethesda family history, sex, and familial clustering of cancer in the grade tumour sample.... 120 xii

Table 4.14: Education, lifestyle factors, nutrient patterns in association with BMI, considering effect modification by Bethesda family history, sex, and familial clustering of cancer in the grade tumour sample.... 122 Table 4.15: Comparison of significant or borderline associations among all variables between standard and path analysis in Grade sample (n=926), including food patterns, and considering Bethesda interaction.... 133 Table 4.16: Comparison of significant or borderline associations among all variables between standard and path analysis in Grade sample (n=926), including food patterns, and considering sex interaction.... 134 Table 4.17: Comparison of significant or borderline associations among all variables between standard and path analysis in Grade sample (n=926), including food patterns, and considering familial clustering interaction.... 135 Table 4.18: Comparison of significant or borderline associations among all variables between standard and path analysis in Grade sample (n=926), including nutrient patterns, and considering Bethesda interaction.... 137 Table 4.19: Comparison of significant or borderline associations among all variables between standard and path analysis in Grade sample (n=926), including nutrient patterns, and considering sex interaction.... 138 Table 4.20: Comparison of significant or borderline associations among all variables between standard and path analysis in Grade sample (n=926), including nutrient patterns, and considering familial clustering interaction.... 139 xiii

List of Figures Figure 1-1: Assumed pathways of associations between the modifiable lifestyle and personal factors and survival from CRC. 2 Figure 1-2 Conceptual Model 9 Figure 4-1: Association among education, personal and life style factors including food patterns and BMI in Grade sample. 115 Figure 4-2: Association among education, personal and life style factors including nutrient patterns and BMI in Grade sample. 116 Figure 4-3: Association of education, lifestyle and personal factors including screening and BMI with tumour grade. 126 Figure 4-4 Association of education, lifestyle and personal factors including screening and BMI with tumour site. 129 Figure 4-5 Association of education, lifestyle and personal factors including screening and BMI with tumour stage. 130 Figure 5-1: Assumed pathways of associations between the modifiable lifestyle and personal factors and survival from CRC. 183 xiv

List of Appendices Appendix 1: Case recruitment in OFCCR... 250 Appendix 2: Family Member Recruitment in OFCCR... 254 Appendix 3: Control Recruitment in OFCCR... 255 Appendix 4: Data cleaning procedures for food/nutrient variables... 256 Appendix 5: Mean and frequency distribution of variables.... 257 Appendix 6: Frequency distribution of familial clusters within each tumour characteristic sample.... 263 Appendix 7: Average daily intake of important dietary items within each of the tumour characteristic samples.... 264 Appendix 8: Food patterns and food items loading on them in MSI, Site, and Stage tumour characteristic samples.... 266 Appendix 9: Micro- and macro-nutrient patterns and nutrients loading on them in MSI, Site, and Stage tumour characteristic samples.... 270 Appendix 10: Correlation of principal component scores between the largest sample vs. tumour characteristic subsamples... 274 Appendix 11: Correlation of dietary principal component scores among different tumour characteristic samples... 275 Appendix 12: Statistically significant correlations among lifestyle and personal factors in the four tumour characteristic samples.... 276 Appendix 13: Association of education with life style factors in the four tumour characteristic samples.... 279 xv

Appendix 14: Association of BMI with education, lifestyle and personal factors in the four tumour characteristic samples.... 281 Appendix 15: Association of the variables with tumour grade in standard analysis.... 290 Appendix 16: Association of the variables in with tumour MSI in standard analysis.... 295 Appendix 17: Association of the variables with tumour site in standard analysis.... 299 Appendix 18: Association of the variables with tumour stage in the standard analysis.... 303 Appendix 19: Association of the variables in the models including tumour grade in the path analysis.... 307 Appendix 20: Association of the variables in the model including tumour MSI in the path analysis.... 313 Appendix 21: Association of the variables in the model including tumour site in the path analysis.... 318 Appendix 22: Association of the variables in the model including tumour stage in the path analysis.... 322 Appendix 23: Summary and comparison of the results from standard analysis and path analysis.... 327 Appendix 24: Correlations (r: Pearson correlation coefficient) among food and nutrient patterns in four samples.... 349 Appendix 25: Distribution of variables by sex and Bethesda status in Grade sample (N=926). 351 Appendix 26: Scatter plot of BMI versus alcohol consumption per month in grams in the grade sample.... 353 Appendix 27: The association of tumour grade with BMI categorized as >=25 or <25, among men and women... 354 xvi

Appendix 28: The association of tumour grade with physical activity, among Bethesda positive and Bethesda negative... 355 Appendix 29: The association of tumour stage with healthy food pattern, among Bethesda positive and Bethesda negative... 356 xvii

1.1 Background Chapter 1. Introduction Colorectal cancer (CRC) is the second leading cause of cancer death in Canada [1]. It is estimated that in 2012, 23,300 Canadians will be diagnosed with CRC (an age-standardized incidence rate of about 50 per 100,000 Canadian population) and 9,200 will die of it (an agestandardized mortality rate of about 20 per 100,000 Canadian population) [1]. Among men, one in 13 is expected to develop CRC during their lifetime and one in 28 will die of it; among women, one in 16 is expected to develop CRC during their lifetime and one in 32 will die of it [1]. Moreover, the prognosis of CRC is poor when compared to some other common cancer diagnoses in Canada, e.g. breast and prostate cancer, with only about 63% 5-year relative survival rate for colorectal cancer compared to 88% and 96% for breast and prostate cancer respectively [1] and [2]. Much of the research has focused on recognizing and addressing risk factors of CRC and this is very valuable. Several genetic and environmental factors (including diet, alcohol, smoking, physical activity, and obesity) have been associated with CRC risk. However, given the poor prognosis of CRC, there is a need to better understand the modifiable factors associated with prognosis of CRC. Factors have been identified which are associated with the prognosis of CRC. Factors associated with prognosis include both biologic characteristics of tumours (grade, stage, and site of the tumour, and presence of microsatellite instability (MSI) [3]-[9]); and lifestyle and personal characteristics (aspects of diet, physical activity, alcohol, smoking, obesity, screening, SES, sex, family history, and age at diagnosis, [10]-[16]). This investigation will determine whether these prognostic biologic tumour characteristics are associated with lifestyle and personal characteristics, and will investigate for the first time pathways through which they are associated. 1.2 Research Problem Where survival data are available, factors associated with prognosis are best identified through studies of their association with survival from CRC. However, prognosis with CRC is strongly strong associated with biologic tumour characteristics (stage. grade, MSI, and tumour site) [17]. Therefore, where survival data are not available, an alternative approach is to study factors 1

2 associated with poor prognostic biologic characteristics of CRC. This study will adopt the latter approach, as survival data were not available. The importance of this study to understanding of pathways 1 among factors influencing survival, rests on evidence, supported by detailed review in Chapter 2, that prognostic tumour characteristics in this investigation are associated with survival from CRC [17] and on the assumption that, as shown in Figure 1-1, these prognostic characteristics are mediators 2 of the association between lifestyle/personal factors and survival (solid lines in Figure 1-1). However, as also shown in Figure 1-1 (dotted line), it is possible that an association of modifiable lifestyle and personal factors to survival occurs through other pathways which do not involve prognostic tumour characteristics. Modifiable lifestyle and personal factors Prognostic tumour characteristics CRC survival Figure 1-1: Assumed pathways of associations between the modifiable lifestyle and personal factors and survival from CRC. This hypothesized model shows both direct (dotted lines) and mediated (solid lines) pathways from the predictor variables to tumour characteristics. The associations described in the figure could ideally be investigated in a prospective cohort study of cancer-free individuals with a sample size that would allow sufficient incident cancer with various clinical characteristics to occur. However, a cost-effective design would be a prospective study of cases followed to assess survival. In such a study, pre-diagnostic lifestyle would need to be recalled by the cases at the time of recruitment and participants would then be 1 Seauqence of associations generally for considering the intermediate variables that exist in the association of an exposure to an outcome variable, e.g. the pathway from education to BMI may include physical activity and healthy diet. It means that education may affect body weight through its association with physical ctivity and a healthy diet. 2 Mediation refers to a causal pathway in which the independent variable affects the mediator which in turn affects the outcome.

3 followed to assess their survival. The cases recruited into the Ontario Familial Cancer Registry (OFCCR) form such a cohort of cases. Therefore, the present study is conceived as a cohort study of cases to evaluate the first step in a pathway between lifestyle and survival, mediated by prognostic clinical tumour characteristics. It therefore evaluates the association of lifestyle and other factors to prognostic characteristics. Because no information on survival was available when the investigation began, it is not possible to examine the association of clinical characteristics to survival, the second step in the mediated pathway shown in the figure, or any direct associations of lifestyle and other factors to survival. The choices for study of prognostic clinical characteristics (grade, stage, microsatellite instability (MSI), and tumour site) and of the lifestyle and other factors (education, diet, smoking habit, alcohol use, physical activity, body mass index (BMI) and colorectal cancer screening) are outlined below and are fully justified in the literature review in Chapter 2. To our knowledge, the study is innovative in four ways. It is the first study to examine the association of lifestyle with grade. It is also the first study to investigate causal pathways among the concepts under study, and to investigate, where appropriate, the modification of these pathways by sex and family history of cancer. Although some previous studies have considered dietary patterns, rather than individual foods and nutrients, in investigations of CRC risk and one study has considered the association of dietary patterns and CRC survival, none has considered the association of patterns of food and nutrient intake with measures of tumour aggression. Finally, this study has compared in practical circumstances the results obtained through two types of analysis for the investigation of mediated pathways, as well as comparing these with the findings of conventional analysis in which mediated pathways cannot be detected. The choice of prognostic clinical characteristics for this study is based on the evidence (reviewed in Chapter 2) that higher tumour stage, higher tumour grade, microsatellite stable tumours, and more distal (distal colon and rectal) tumours in CRC are associated with worse survival from CRC. These prognostic characteristics reflect different aspects of tumour biology, but each is associated with prognosis. Tumour stage reflects the spread of the disease and hence indicates a more advanced disease [17]. Tumor grade indicates a biologically aggressive tumour [17]. Rectal cancers have poorer prognosis than colon cancer [2]; they may also have different risk factors and are less commonly associated with genetic and familial factors, compared to proximal

4 tumors [18]. Proximal colon cancers have the best prognosis compared to those in other locations in the large bowel [2]. Microsatellite stable tumours are associated with worse prognosis than microsatellite instable tumours [7]. The underlying association being investigated is that unhealthy lifestyle factors may be associated with more biologically aggressive tumours which are associated with lower rates of survival. Although there is a large body of literature on the association between lifestyle and personal factors and risk of CRC, e.g. [19], few studies have examined the association of these factors with survival from CRC [21]-[26] and their results have not been conclusive. There are no studies to our knowledge that have examined pathway of associations with prognostic characteristics of colorectal tumours, although association of individual risk factors have been studied for tumour stage, e.g. [27], [28], tumour location e.g. [18], and MSI [29]-[34]. Moreover, most dietary studies have used single dietary factors or nutrients rather than dietary pattern. Since people consume different foods and nutrients in combination, it is difficult and sometimes impossible to measure the effect of an individual food or nutrient when its consumption is highly correlated with other foods or nutrients. Therefore, studying the food or nutrient patterns in different individuals should better reflect the effect of such combinations when they happen to be very closely related. Obesity is a risk factor for CRC which is associated with each of diet, alcohol, smoking and physical activity [23], [19]. Given that obesity is known to be associated with survival from CRC, e.g. [24], it may mediate an association between these lifestyle risk factors and survival. Moreover, obesity may affect survival through its association with prognostic tumour characteristics. To our knowledge, no studies have examined the association between obesity and prognostic characteristics of colorectal tumours. Socioeconomic status (SES) is another factor that has been associated with survival from CRC, with low SES being associated with poorer survival, e.g. [25]. This may occur because SES is associated with lifestyle factors, such as diet, alcohol intake, smoking and physical activity [35]- [38], factors which, as explained above, may themselves be associated with survival. SES is also associated with screening, with those of lower SES less likely to be screened than those of higher

5 SES [39]. Lack of screening results in higher stage at diagnosis and lower survival [26]. These observations suggest a mediating role for lifestyle and screening in the association between SES and CRC survival. Thus low SES may be associated with high-risk lifestyle and lack of screening, and these lifestyle and personal factors could reduce survival. SES is often considered as a potential confounder of associations between risk factors and disease [40] but this does not allow a full understanding of the pathways to disease. To my knowledge, full sequential associations or pathways from SES to lifestyle and other factors, to disease have never been investigated. This study will investigate such pathways among colorectal cancer patients, but will consider whether pathways seen among cases are likely to apply in the general population. Although SES is not itself a modifiable factor, its correlates, high-risk lifestyle and lack of screening, are modifiable. Therefore, identifying a role for SES in CRC prognosis, together with the pathways involved, could assist in focusing public health interventions to the vulnerable populations if appropriate and in targeting key lifestyle and personal factors. In addition to modifiable factors which may influence both the risk and prognosis of CRC, there are factors which may modify 3 the associations of these modifiable factors with risk/prognosis. This study will focus on sex and on family history of disease. Sex influences the association of lifestyle factors and risk of developing colorectal cancer [19], as well as survival after the disease has occurred [18]. For example, the report on Food, Nutrition, Physical Activity and the Prevention of Cancer by the World Cancer Research Fund/American Institute for Cancer Research (WCRF/AICR) suggests that some lifestyle factors such as alcohol have stronger association with CRC risk among men compared to women [19] or another study showed a diet high on processed meat is associated with lower survival from CRC more strongly among women than men [41]. Therefore, sex may also modify the relationships between CRC risk and prognosis and lifestyle and personal factors. 3 Effect modification in this context refers to difference in associations of an exposure and outcome depending on the level of a third variable, an effect modifier variable. An example is the difference in associations found among men and women.

6 Two specific types of familial clusters with a high risk of CRC have been recognized, Familial Adenomatous Polyposis (FAP) and Hereditary Non-Polyposis CRC (HNPCC) [46], [47] for which associated genes have been identified. FAP is an inherited mutation in APC gene which accounts for less than 1% of CRCs in North America [124]. The mutation leads to the formation of thousands of colorectal polyps and causes a high risk of CRC (nearly 100% by age 35-40) [124]. It is not considered further in this thesis because of its low prevalence and strong genetic etiology, as well as because FAP was an exclusion criterion for cases in the Ontario Familial Colorectal Cancer Registry on which this study is based. HNPCC is a syndrome of familial susceptibility to CRC and certain other cancers (endometrium, ovary, kidney, brain, stomach, lymph, pancreas, liver [48]) thought to be inherited in an autosomal dominant pattern and result from inherited mutated mismatch repair genes [106]. It accounts for about 5% of colorectal cancer cases [43]. Despite the high risk of CRC in HPCC families, prognosis of tumours in these families is better than that in the general population [44] and it is possible that modifiable lifestyle factors may affect prognosis in people with HNPCC genes differently than in people without them, a so-called gene-environment interaction [52]. A recent study [51] has identified additional familial clusters of CRC and other cancers. The newly identified familial clusters observed co-occurrence of extra-colonic cancer sites not previously considered in HNPCC and suggested that the nature of cancers seen in familial clusters may differ for families with predominantly 1) colon and rectal cancers, and 2) cancers of the left and right colon. In addition, there were cases who reported a family history but in whom no particular patterns of cancer occurrence were observed. Taken together, these clusters of familial susceptibility indicate the potential for genes or environmental factors which cluster in families to be associated with risk, as well as the possibility of an interaction of genetic and environmental factors in determining cancer risk or prognosis. In addition, because members of families with a history of colorectal cancer may wish to reduce their risk, it is possible that family membership may influence behaviour, such as lifestyle choices or the use of colorectal cancer screening [42], and that such behaviour may cause the presence of a family history to modify associations seen between risk factors and CRC risk and/or prognosis.

7 1.3 Study Objectives Primary objective (Figure 1-2 Conceptual Model): 1. To examine among colorectal cancer patients the pathways of association between education, lifestyle (diet, physical activity, smoking and drinking) and personal factors (screening and obesity) and prognostic tumour characteristics in CRC, controlling for any confounding effect of age, and with consideration of the possible effect modification by sex and the presence of a family history of colorectal cancer satisfying Bethesda criteria. The specific hypotheses being tested in Primary Objective 1 include evaluation of whether among colorectal cancer patients: (i) Education (as an indicator of SES) is associated with lifestyle factors including physical activity, diet, smoking, and drinking and personal factors including screening and obesity. (ii) Lifestyle factors including physical activity, diet, smoking, and drinking and personal factors including screening are associated with obesity. (iii) Lifestyle factors including physical activity, diet, smoking, and drinking and personal factors including screening and obesity, as well as education, are associated with prognostic CRC tumour characteristics (grade, MSI, stage, and site). (iv) Lifestyle factors including physical activity, diet, smoking, and drinking and personal factors including screening mediate the association between education and obesity in this patient population. (v) Obesity mediates the association between life style factors including physical activity, diet, smoking, and drinking and personal factors including screening and CRC prognostic tumour characteristics (grade, MSI, stage, and site).

8 (vi) Lifestyle factors including physical activity, diet, smoking, and drinking and personal factors including screening and obesity mediate the association between education and prognostic CRC tumour characteristics (grade, MSI, stage, and site). (vii) Sex or family history satisfying Bethesda criteria for familial CRC modify associations that exist with prognostic CRC tumour characteristics (grade, MSI, stage, and site). In order to achieve these objectives, measures of diet are required. The following additional descriptive objective is therefore part of Primary Objective 1. viii) To identify patterns of consumption of foods and patterns of consumption of nutrients among cases in this study. Secondary Objective: 1. To compare in a practical situation with real data the results of two methods of analysis for investigating mediated pathways, with their associated software. The approaches are a) a standard approach to analysis using SPSS software, (here referred to as standard method) in which defined criteria of mediation are evaluated through a series of independent regression analyses; b) Path analysis using the specifically designed software for this type of analysis (Mplus) in which all pathways are estimated simultaneously. 2. To compare findings from these analyses not only with each other but also with conventional analysis which cannot detect mediated pathways because it treats relevant variables as confounders. Tertiary Objective: The final objective is hypothesis generating: 3. To determine whether newly recognized patterns of familial clusters of cancer modify associations between diet, obesity and prognostic characteristics of colorectal cancer.

9 Figure 1-2 depicts our hypothesized associations. However, we acknowledge that other pathways may exist that are not the focus of this thesis. All these other pathways will be included in the direct path. Dietary Patterns Education Alcohol Smoking BMI Tumour characteristic Screening Physical activity Family history and sex Figure 1-2 Conceptual Model This hypothesized model shows both direct (dotted lines) and mediated (solid lines) pathways from the predictor variables to tumour characteristics. All pathways, except those between SES and lifestyle, are potentially modified by family history and sex. All correlations among lifestyle factors are hypothesized but have not been shown in this graph for ease of demonstration. 1.4 Importance of this Study To our knowledge, no published studies have examined the sequence of pathways of associations between education, diet and other lifestyle and personal factors, obesity, and the prognostic

10 tumour characteristics in CRC, or how such associations may be modified by family history of cancer and sex. Understanding the association between modifiable lifestyle factors and tumour characteristics which affect survival from CRC, could lead to consideration of possible social and biological mechanisms through which lifestyle may be associated with survival from CRC. The results could have important implications in improving survival from colorectal cancer which is currently poor. The opportunity to perform this study is provided by OFCCR, a large population-based colon cancer familial registry with sufficient sample size to test multiple effects and interactions. Also, the wealth of information collected by this registry provides a rare opportunity for research in this field. The investigation of nutrient and food patterns in dietary data available through the OFCCR gives us the opportunity to identify the importance of combinations of foods or nutrients reflected by eating patterns. The use of pathway analysis allows identification of the pathways through which risk factors such as diet may affect tumour characteristics. This is a newly emerging area in the field of epidemiology which examines the pathways rather than controlling for different factors that may mediate causal pathways as confounders and thereby avoids potentially misleading conclusions (such as concluding absence of an association after controlling for a mediating variable when in fact there is a mediated association). We also have the advantage of using the information on newly identified familial clusters [51] to generate new hypotheses. These new familial clusters among colorectal cancer and other cancer sites raise the possibility of a role in colorectal carcinogenesis for presently unknown genetic factors and for environmental factors that are shared among these families. Therefore, this study could provide important clues to gene-environment interaction in predicting prognostic tumour characteristics.

2.1 Introduction Chapter 2. Literature Review In this chapter, I have reviewed the literature on the following topics and have identified gaps in the literature that justify the importance of this thesis. The topics include: A review of tumour characteristics related to CRC survival A review of familial patterns related to the occurrence of CRC A review of personal and lifestyle factors associated with incidence and survival from CRC Measurement issues with the variables in this study Pathways to CRC Methods of analysis of pathways 2.2 Tumour characteristics related to CRC survival Two groups of tumour characteristics have been recognized as important predictors of survival from CRC: clinicopathologic and genetic factors. The strong association between these factors and survival from CRC shows the importance of studies designed to identify the predictors of these tumour characteristics. 2.2.1 Clinicopathologic tumour characteristics 2.2.1.1 Stage Stage reflects the degree of spread of a cancer in adjacent tissues and the body [67]. Stage, including clinicopathologic and pathologic stage, is a generally accepted prognostic factor in 11

12 CRC [68]. The local extension of the tumour and the involvement of the surgically removed regional lymph nodes are generally determined through pathologic studies. These results are referred to as pathologic stage [69]. However, distant metastasis of the tumour is usually suspected clinically and confirmed through radiological studies [70]. Dukes introduced a staging system in 1932 [70] which has been modified over time and is still the most commonly used clinicopathologic staging system in CRC. The TNM (Tumour, Node, Metastasis) system is another classification tool for different cancers which uses tumour, lymph node, and metastasis status to define the stage of a cancer [71]. Table 2.1 shows the association between Dukes system and TNM classification [72]. Table 2.1 Dukes system and its association with TNM system [72] DUKES STAGING SYSTEM CORRELATED WITH TNM SYSTEM Dukes A T1, N0, M0 (stage I) T2, N0, M0 (stage I) Dukes B T3, N0, M0 (stage II) T4, N0, M0 (stage II) Dukes C T (any), N1, M0; T (any), N2, M0 (stage III) Dukes D T (any), N (any), M1 (stage IV) T1: cancer grown through muscularis mucosa and extends into submucosa; T2: cancer grown through submucosa, and extends into muscularis propria; T3: cancer grown completely through muscularis propria into subserosa but not neighbouring organs or tissues; T4: cancer spread completely through wall of the colon or rectum into nearby tissues or organs; N0: No lymph node involvement; N1: 1-3 nearby lymph nodes involved. N2: >=4 nearby lymph nodes involved. M0: No distant spread; M1: Distant spread. It has been repeatedly shown that pathologic stage is an important predictor of recurrence, [73], and clinicopathologic stage is an important predictor of survival [17] in CRC patients. The College of American Pathologists Consensus Statement of 1999 recognized pathologic stage as a proven prognostic factor in CRC [74]. The median 5-year survival rates in different studies are as follows: for Dukes A and B, 89.9%; for Dukes C, 69.6%; and for Dukes D 11.9% [76].

13 2.2.1.2 Grade Grade is another important tumour characteristic that reflects the degree of differentiation of tumour cells and is determined through pathologic examination of the tumour [67]. Several studies have found that poor differentiation is an indicator of poor prognosis in CRC, [77]-[80]. In 1999, the College of American Pathologists Consensus Statement suggested that tumour grade is a valid prognostic indicator for CRC [74]. Most systems divide tumours into 4 categories: grade 1 (well differentiated), grade 2 (moderately differentiated), grade 3 (poorly differentiated), and grade 4 (undifferentiated) and many studies collapse this into low grade (Grade 1 and 2) and high grade (Grade 3 and 4) [71]. The College of American Pathologists, however, recommends a two-tiered grading system in order to increase consistency among studies [74]. This suggestion was supported by a large Korean study of 2230 patients which showed that the 5-year survival rates of different tumour grades, after controlling for other important prognostic factors in CRC, were as follows: grade 1, 66.5%; grade 2, 62.5%; grade 3, 50%; and grade 4, 51% [83]. 2.2.1.3 Site Site of the tumour is another prognostic factor in CRC, though not as strongly predictive of survival as stage and grade. Several studies have shown that overall, rectal cancer has a worse prognosis than colon cancer [77], [82]-[84]. However, individual studies vary in their findings of which site has the best/worst prognosis. One study found that left colon cancer had the best prognosis and rectal cancer had the worst, with right colon being in the middle, [82], while another found that right colon cancer had the best prognosis, followed by left colon and rectal cancer [84]. What is more consistently shown in studies is the worse prognosis of rectal cancer [77]. However, the lower survival rate of rectal cancer compared to proximal colon cancer seems to have disappeared in more recent years, with a significantly lower 5-year relative survival rate for rectal cancer compared to proximal colon cancer in 1994-1996 reaching almost same survival rates in 2004-2006 [1]. This has been attributed to decreasing gap between poor prognosis Microsattellite Stable (MSS) tumours (which are more often distal) compared to generally better prognosis MSI-H tumours (which are usually proximal) due to a better response of MSS tumours to new adjuvant chemotherapeutic agents like 5-fluorouracil [123].

14 Location of tumour differs by sex. Among women, the most common site is proximal colon or right-sided cancers (42% of all cases) followed by rectal and distal colon or left-sided cancers (28%) and (22%), respectively. Among men, the most common site is rectal cancer (38% of all cases), followed by proximal (31%) and distal colon cancer (25%) [76]. Proximal location in this report was defined as all colon cancers from cecum up to the end of splenic flexure; distal colon was defined as after splenic flexure up to the end of sigmoid colon; and rectal included all cancers of rectum and rectosigmoid junction [76]. Table 2.2 Distribution of colon and rectal cancer [76]. Site Women (%) Men (%) Proximal colon 42 31 Distal colon 22 25 Rectum 28 38 In addition to female sex, older age and black non-hispanic race have been associated with a greater likelihood of developing CRC in a proximal location [86]. As these factors are associated with participation in screening programs [89], it is possible that screening may account for some of the shift from distal to more proximal locations in the colon which has been seen in the past two decades [85]-[88]. The incidence of rectal tumours has decreased by about 0.9% per year from 2000 to 2007, while the incidence of proximal tumours increased on average by 0.3% per year[1]. Therefore, the increased relative prevalence of proximal colon tumours may be associated with participation in screening programs and removal of the distal adenomas through proctosigmoidoscopy[76]. As screening may be important in determining tumour location as well as grade and stage it will be important to consider its role in studies designed to investigate the relationship between lifestyle and clinical factors related to prognosis. In summary, studies that have looked at the relative importance of clinicopathologic prognostic factors in CRC have found that stage has the largest effect on survival followed by grade and tumour location, respectively [76].

15 2.2.2 Genetic tumour characteristics CRCs are assumed to develop from adenomas [90], [91]. This development takes place over several years and involves accumulation of genetic alterations [90], [92], [93]. Two major pathways are involved. 2.2.2.1 Chromosomal Instability pathway (CIN) or microsatellite stable (MSS) Approximately 80-85% of CRCs result from allelic losses (such as 1p and 8p deletions), chromosomal amplification and translocation (such as loss of heterozygosity of 17p and 18p)[4], [96]. This pathway is called chromosomal instability pathway (CIN) or the microsatellite stable pathway [91]. In this pathway, the genes that have so far been discovered are k-ras (an oncogene), APC, DCC, and p53 (tumour-suppressor genes) [97]-[100], however, more genes are yet to be recognized. MSS tumours have a worse prognosis (average 5-year survival of 54%) than MSI (Microsatellite unstable) tumours (average 5-year survival of 76%) [101], [102]. 2.2.2.2 Microsatellite Instability (MSI) The other 10-15% of CRCs observed in large population-based studies [101], [103] occur through a second pathway called microsatellite instability pathway (MSI) [104]. In this pathway, genetic destabilizations results from a loss in DNA mismatch repair function caused by mutation or epigenetic silencing in mismatch repair genes [91]. During cell proliferation, the mismatch occurs most commonly in parts of the DNA with short simple repeated sequences or microsatellites, thus, tumours of the individuals who have lost mismatch repair function have Microsatellite Instability or MSI [91]. MSI is a hallmark of tumours from Hereditary Non- Polyposis CRC (HNPCC families) where some, but not all, of the mutations involved have been identified [44]. In most of the HNPCC families, the genetic basis of the disease is an inherited mutation in mismatch repair (MMR) genes [105]. The mutation in mismatch repair genes is called mutator phenotype [91]. MMR genes are involved in the repair of the mutations that normally occur during DNA replication, and therefore, the mutation in these genes results in persistence of replication errors [106]. Such replication errors, when they occur in oncogenes or tumour suppressor genes, can result in uncontrolled replication of the cells and tumourgenesis [106].

16 Five of these genes have been most frequently associated with HNPCC when they are inherited in mutated form; these are hmlh1, hmsh2, hmsh6, hpms1, and hpms2 [91], [107], [108], [109]. Mutation in hmlh1 and hmsh2 account for more than 90% of mutations found in HNPCC families [91]. However, in subjects who meet the familial criteria for HNPCC, genetic mutations are not always identified. Indeed, MLH1 and MSH2 were detected by PCR in only about 50% of the CRCs [110]. These findings raise the possibility that other genes might be involved which have not yet been recognized. In fact, there is evidence suggesting that among those who meet Amsterdam I criteria but in whom no gene mutations have been detected may be different from the recognized HNPCC families [111]. In such cases, it has been shown that the members of these families are at increased for colorectal cancer but not for other HNPCC related cancer sites and therefore, these cases were named Syndrome X as opposed to HNPPCC-Lynch Syndrome [111]. Since 20% of the sporadic CRC tumours contain the same genetic defects, these are considered to happen as a result of epigenetic silencing. The silencing of the mismatch repair genes, usually MLH1, occurs as a result of biallelic or hemiallelic hypermethylation of DNA (referred to as methylator phenotype) [91]. Therefore, there are three distinct groups of genetic alterations in MSI pathway: 1. Mutation in MMR genes among HNPCC families; 2. Mutation in unidentified genes among HNPCC families; 3. Hypermethylation of DNA in MMR genes among sporadic cases. Regardless of whether genetic mutation or epigenetic silencing of MMR genes (mutator or methylator phenotypes) or mutation in unidentified genes are responsible for MSI-H status of the tumour, MSI-H is associated with an overall better prognosis compared to tumours with similar stage and grade but with MSS or MSI-L status [102], [44]. However, it is possible that the pathways from lifestyle factors to MSI-H tumours may be different for familial versus sporadic cases because those who already have mutated mismatch repair genes may develop CRC independent of environment when there is no gene-environment interaction; while those without a previous genetic mutation are more likely to have developed the genetic changes as a result of environment. That is the reason it is important to examine the pathways to MSI status of the tumours considering an indicator of familial susceptibility.