Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Psychological stress and risk of hormone
Document related concepts
Transcript
UNIVERSITY OF CALIFORNIA Los Angeles Psychological stress and risk of hormone-dependent cancers A dissertation submitted in partial satisfaction of the requirements for the degree Doctor of Philosophy in Epidemiology by Naja Rod Nielsen 2007 The dissertation of Naja Rod Nielsen is approved Abdelmonem A. Afifi Morten Grønbæk Beate Ritz Barbara Visscher Zuo-Feng Zhang, Committee Chair University of California, Los Angeles, 2007 TABLE OF CONTENTS I. Introduction_______________________________________________________ 1 Objective and specific aims____________________________________________ 2 Causal hypothesis ___________________________________________________ 4 Outline ___________________________________________________________ 4 Definition of stress __________________________________________________ 5 Incidence and prevalence of hormone-dependent cancers _____________________ 7 II. Material and methods _____________________________________________ 14 The Copenhagen City Heart Study _____________________________________ 14 Measurement of perceived stress_______________________________________ 15 Assessment of endpoints by linkage to national registers_____________________ 16 Statistical methods _________________________________________________ 16 III. Stress and breast cancer: a current overview of prospective studies ________ 23 Introduction ______________________________________________________ 23 Assessment methods ________________________________________________ 24 Results __________________________________________________________ 26 Discussion _______________________________________________________ 29 Conclusion _______________________________________________________ 33 IV. Stress and other hormone-dependent cancers: current knowledge _________ 39 Endometrial cancer _________________________________________________ 39 Colorectal cancer __________________________________________________ 40 Prostate cancer ____________________________________________________ 42 V. Self-reported stress and risk of breast cancer: a prospective cohort study ____ 47 Introduction ______________________________________________________ 47 Methods _________________________________________________________ 47 Results __________________________________________________________ 49 Discussion _______________________________________________________ 50 Conclusions ______________________________________________________ 53 VI. Self-reported stress and risk of endometrial cancer: a prospective cohort study __________________________________________________________________ 61 Introduction ______________________________________________________ 61 Methods _________________________________________________________ 62 Results __________________________________________________________ 64 Discussion _______________________________________________________ 66 ii VII. Perceived stress and risk of colorectal cancer: a prospective cohort study __ 75 Introduction ______________________________________________________ 75 Methods _________________________________________________________ 76 Results __________________________________________________________ 78 Discussion _______________________________________________________ 80 VIII. Sociodemographic status, stress and risk of prostate cancer: a prospective cohort study _______________________________________________________ 92 Introduction ______________________________________________________ 92 Methods _________________________________________________________ 93 Results __________________________________________________________ 95 Discussion _______________________________________________________ 97 Conclusion ______________________________________________________ 100 IX. Perceived stress and sex steroid hormones: a cross-sectional study ________ 106 Introduction _____________________________________________________ 106 Methods ________________________________________________________ 107 Results _________________________________________________________ 109 Discussion ______________________________________________________ 109 X. Stress and hormone-dependent cancers: a discussion of the evidence_______ 117 Is there a relation between stress and hormone-dependent cancers?____________ 117 The strength of the Copenhagen City Heart Study_________________________ 120 Selection bias and external validity ____________________________________ 121 Misclassification of perceived stress ___________________________________ 121 Misclassification of outcome measures _________________________________ 123 Confounding_____________________________________________________ 124 Random error ____________________________________________________ 127 XI. Conclusion_____________________________________________________ 128 XII. Public health implications________________________________________ 130 Is stress a public health problem? _____________________________________ 130 Future population studies on stress and risk of hormone-dependent cancers _____ 132 XIII. Bibliography _________________________________________________ 136 iii LIST OF TABLES Chapter I Table 1-1. Overview of major stress theories and definition .............................................. 9 Table 1-2. Estimates of the incidence and mortality rate per 100,000 as well of the oneyear prevalence in the total population of hormone-dependent cancers in Denmark, USA, and the World............................................................................................................. 10 Chapter II Table 2-1. Baseline characteristics of the 12,698 men and women who participated in the second examination of the Copenhagen City Heart Study in 1981-83............................. 20 Table 2-2. Number of primary site-specific hormone-dependent cancer cases occurring during follow-up (year 1981-2000) in the Copenhagen City Heart Study ....................... 21 Chapter III Table 3-1. Summary of prospective studies on stress and breast cancer incidence ......... 35 Table 3-2. Summary of prospective studies on stress and breast cancer relapse ............. 37 Chapter IV Table 4-1. Summary of previous studies on stress and endometrial cancer..................... 44 Table 4-2. Summary of previous studies on stress and colorectal cancer ........................ 45 Chapter V Table 5-1. Baseline characteristics of women who participated in the second examination of the Copenhagen City heart study in 1981-3. Values are numbers (percentages) unless stated otherwise ................................................................................. 55 Table 5-2. Incidence and hazard ratio of primary breast cancer associated with intensity and frequency of stress among 6689 Danish women participating in the second examination of the Copenhagen City heart study in 1981-3 ............................................. 56 Table 5-3. Incidence and hazard ratio of primary breast cancer associated with stress score among 6689 Danish women participating in the Copenhagen City heart study in 1981-3 ................................................................................................................................... 57 Table 5-4.Incidence and hazard ratio of primary breast cancer associated with categorized stress score among 6689 Danish women participating in the Copenhagen City heart study in 1981-3, according to time period of follow-up .................................. 58 iv Table 5-5.Hazard ratio of primary breast cancer associated with stress score among 6689 Danish women participating in the Copenhagen City heart study in 1981-3, in subgroups of hormone therapy .............................................................................................................. 59 Chapter VI Table 6-1. Baseline characteristics of 6,760 women who participated in the second examination of the Copenhagen City Heart Study in 1981-83.......................................... 70 Table 6-2. Incidence, hazard ratio (HR), and 95 % confidence interval (CI) for first-time incidence of primary endometrial cancer associated with perceived stress among 6,760 women who participated in the Copenhagen City Heart Study in 1981-83 ..................... 71 Table 6-3. Incidence, hazard ratio (HR), and 95 % confidence interval (CI) for first-time incidence of primary endometrial cancer associated with perceived stress among 6,760 women who participated in the Copenhagen City Heart Study in 1981-83 by hormone status...................................................................................................................................... 72 Table 6-4. Incidence, hazard ratio (HR), and 95 % confidence interval (CI) for first-time incidence of primary endometrial cancer associated with perceived stress among 6,760 women who participated in the Copenhagen City Heart Study in 1981-83 by body mass index...................................................................................................................................... 73 Chapter VII Table 7-1. Baseline characteristics of the 6,488 women and 5,426 men who participated in the second examination of the Copenhagen City Heart Study in 1981-83................... 85 Table 7-2. Women. Incidence, hazard ratio (HR), and 95 % confidence interval (CI) for first-time incidence of primary colorectal, colon, and rectal cancer associated with perceived stress among 6,488 women who participated in the Copenhagen City Heart Study in 1981-83 .................................................................................................................. 86 Table 7-3. Hazard ratio (HR), and 95 % confidence interval (CI) for first-time incidence of primary colorectal, colon, and rectal cancer associated with a seven-unit stress score among 1,716 pre-menopausal and 4,772 post-menopausal women who participated in the Copenhagen City Heart Study in 1981-83.................................................................... 88 v Table 7-4. Men. Incidence, hazard ratio (HR), and 95 % confidence interval (CI) for first-time incidence of primary colorectal, colon, and rectal cancer associated with perceived stress among 5,426 men who participated in the Copenhagen City Heart Study in 1981-83............................................................................................................................. 89 Chapter VIII Table 8-1. Baseline characteristics of the 5,496 men who participated in the second examination of the Copenhagen City Heart Study in 1981-83........................................101 Table 8-2. Risk of primary prostate cancer associated with sociodemographic variables among 5,496 men who participated in the Copenhagen City Heart Study in 1981-83 .102 Table 8-3. Risk of primary prostate cancer associated with perceived stress among 5,496 men who participated in the Copenhagen City Heart Study in 1981-83 ........................103 Table 8-4. Risk of primary prostate cancer associated with perceived stress and sociodemographic variables according to period of follow-up .......................................104 Chapter IX Table 9-1. Median and relative difference in geometric mean of 17β-estradiol by category of stress among 832 postmenopausal women who did not use hormones ......112 Table 9-2. Median and relative difference in geometric mean of free testosterone by category of stress among 840 postmenopausal women who did not use hormones ......113 Table 9-3. Median and relative difference in geometric mean of cortisol by category of stress among 840 postmenopausal women who did not use hormones ..........................114 Chapter XII Table 12-1. Estimates of the number of ischemic heart disease and breast cancer there can be attributed/ prevented by perceived stress among women in the Copenhagen City Heart Study. ........................................................................................................................134 vi LIST OF FIGURES Chapter I Figure 1-1. Causal hypothesis for the relation between perceived stress and risk of hormone-dependent cancers ................................................................................................ 11 Figure 1-2. Different approaches to stress research and their relation to hormonedependent cancers ................................................................................................................ 12 Figure 1-3. Age-standardized incidence and mortality rates of site-specific cancers in Denmark in year 2002.......................................................................................................... 13 Chapter II Figure 2-1. Formation of the stress-score based on an additive approach. The numbers indicates the score on the seven-point stress-score and the numbers in parentheses indicates the proportion of the total study sample in each category. The lines indicate the categorization of the seven-point stress-score into low, medium, and high stress........... 22 Chapter III Figure 3-1. Flow diagram of the data collection process................................................... 38 Chapter V Figure 5-1. Causal diagram of the relation between perceived stress and risk of breast cancer .................................................................................................................................... 60 Chapter VI Figure 6-1. Causal diagram for the relation between perceived stress and risk of endometrial cancer ............................................................................................................... 74 Chapter VII Figure 7-1. Causal diagram for the relation between perceived stress and risk of colorectal cancer................................................................................................................... 91 Chapter VIII Figure 8-1. Causal diagram for the relation between socio-economic status, marital status, perceived stress, and risk of prostate cancer.........................................................105 vii Chapter IX Figure 9-1. Flow diagram of the data selection process ..................................................115 Figure 9-2. Circadian pattern for 17β-estradiol, free testosterone, and cortisol.............116 Chapter XII Figure 12-1. Questionnaire used for the Perceived Stress Scale.....................................135 viii LIST OF ABBREVATIONS AND DEFINITIONS Allostasis: The process by which we actively adjust ourselves to predictable and unpredictable events, and thereby preserve the homeostatic stability of vital body systems. The primary mediators of allostasis are the sympathetic nervous system and hormones of the hypothalamus-pituitary-adrenal axis. CI: Confidence interval Distress: The mental and bodily reaction generated by a stressor. Such reactions may include neurobiological changes in cortisol and other hormones, physical changes in blood pressure, heart rate, and muscle tension, as well as varying mental states (e.g. anxiety, nervousness, depression). HR: Hazard ratio Hormone-dependent cancers: Cancers that have a hormonal component, so that their development to some degree depends on levels of sex steroid hormones. HPA-axis: Hypothalamic-Pituitary-Adrenal axis HPG-axis: Hypothalamic-Pituitary-Gonadal axis Perceived stress: An individual’s appraisal of an imbalance between demands (stressors) and his or hers resources to cope with it. Stressor: An external stimuli that may increase the demand of the individual, for example adverse life events (e.g. divorce, death within the family), daily hassles (e.g. computer problems, being late for the bus), and characteristics of the job situation (e.g., demands, control, rewards). ix AKNOWLEDGEMENTS Writing this dissertation has provided me with an opportunity to comprehensively address the relation between stress and cancer and combine it with an exploration of conventional and upcoming epidemiologic methods. Epidemiologic studies at the Department of Epidemiology at University of California, Los Angeles has inspired me to use alternative epidemiologic methods and has led me to see some of the flaws sometimes contained in conventional approaches. I thank my committee members, Drs. Zhang, Ritz, Visscher, Afifi, and Grønbæk for their support and guidance throughout my doctoral studies. I especially want to thank my advisor Dr. Zhang for encouraging me to do my doctoral studies at UCLA and for being a supportive and inspiring mentor. I had the pleasure of being able to write the dissertation at the Danish National Institute of Public Health surrounded by great colleagues. I am grateful for Dr. Grønbæk’s openness to and sincere interest in my research as well as his detailed guidance during the writing process. I also want to thank Dr. Søndergaard Kristensen for his constructive comments and for sharing with me his broad knowledge on stress research, and Dr. Hansen for her help and support with the laboratory analyses. I am grateful that my good friends and colleagues Majken Karoline Jensen and Katrine Strandberg-Larsen took the time to comment on my dissertation, and I highly value the statistical comments and philosophical discussions with my father Jens Nielsen. Thanks to my friends and family for their support and love and a special thanks to Morten Hulvej Jørgensen for making it all worthwhile. I thank the steering committee and the staff of the Copenhagen City Heart Study for letting me use the data and for provided me with valuable help. Chapter 3 is a version of Nielsen NR, Grønbæk M. Stress and breast cancer: A systematic update on the current knowledge. Nat Clin Pract Oncol 2006; 3:612-620. Chapter 5 is a version of Nielsen NR, Zhang ZF, Kristensen TS, Netterstrøm B, Schnohr P, Grønbæk M. Selfreported stress and risk of breast cancer. BMJ 2005;331:548. This study was supported by funds from the Health Insurance Foundation. Chapter 6 is a version of Nielsen NR, Strandberg-Larsen K, Grønbæk M, Kristensen TS, Schnohr P, Zhang ZF. Is perceived stress associated with lower risk of endometrial cancer? A prospective cohort study. Psychosom Med (in press). The study was supported by funds from the Health Insurance x Foundation and the Lundbeck Foundation. Chapter 7 is a version of Nielsen NR, Kristensen TS, Strandberg Larsen K, Zhang ZF, Schnohr P, Grønbæk M. Perceived stress and risk of colorectal cancer in men and women: a prospective cohort study (submitted). The study was supported by funds from the Lundbeck Foundation and the Danish Cancer Society. Chapter 8 is a version of Nielsen NR, Kristensen TS, Zhang ZF, Strandberg-Larsen K, Schnohr P, Grønbæk G. Sociodemographic status, stress and risk of prostate cancer: a prospective cohort study. Ann Epidemiol (in press). The study was supported by funds from the Health Insurance Foundation and the Lundbeck Foundation. Drs. Åse Marie Hansen, Jens Nielsen, and Morten Grønbæk contributed to the study on stress and sex steroid hormones presented in chapter 9. xi VITA November 17, 1977 Born, Odense, Denmark 2001 Research assistant The Bandim Health Project Guinea-Bissau, West Africa 2001-2002 Research assistant Institute of Preventive Medicine Copenhagen, Denmark 2002 B.S., Public Health Science University of Copenhagen Copenhagen, Denmark 2002-2006 Graduate research assistant National Institute of Public Health Copenhagen, Denmark 2004 M.S., Public Health Science University of Copenhagen Copenhagen, Denmark 2006 Raymond D. Goodman Scholarship Award UCLA School of Public Health Los Angeles, California 2006 Teaching assistant Department of Social Medicine University of Copenhagen Copenhagen, Denmark xii PUBLICATIONS AND PRESENTATIONS Nielsen NR (May 2003). Is the effect of alcohol on risk of stroke confined to highly stressed persons? Oral presentation at the 12th European Stroke Conference, Valencia, Spain Nielsen NR, Schnohr P, Jensen G, Grønbæk M. (2004). Is the relationship between type of alcohol and mortality influenced by socio-economic status? J Intern Med, 255:280-8. Nielsen NR, Kjøller M, Jørgensen FK, Grønbæk M. (2004) [Stress among working population of Danes]. Ugeskr Laeger, 166:4155-4160. Nielsen NR (April, 2004). Stress reporting in a representative health interview study of Danes. Oral presentation at The 1 st Danish Stress Conference, Copenhagen, Denmark Nielsen NR, Thygesen LC, Johansen D, Jensen G, Grønbæk M. (2005). The influence of duration of follow-up on the interpretation of estimates from prospective studies. Alcohol and mortality in the Copenhagen City Heart Study. Ann Epidemiol, 15:44-55 Nielsen NR, Truelsen T, Barefoot J, Johnsen SP, Overvad K, Schnohr P, Grønbæk M. (2005). Is the association between alcohol and stroke modified by self-reported stress? Neuroepidemiology, 25:105-113 Nielsen NR, Zhang ZF, Kristensen TS, Netterstrøm B, Schnohr P, Grønbæk M. (2005). Self-reported stress and risk of breast cancer. BMJ, 331:548. Nielsen NR (June, 2005). Self-reported stress and risk of breast cancer. Oral presentation at the Society of Epidemiologic Research annual meeting, Toronto, Canada xiii Nielsen NR (November, 2005). Perceived stress and risk of ischemic heart disease: Causation or bias? Oral presentation at the 2nd Danish Stress Conference, Copenhagen, Denmark Nielsen NR, Kristensen TS, Prescott E, Strandberg Larsen K, Schnohr P, Grønbæk M. (2006). Perceived stress and risk of ischemic heart disease: Causation or bias? Epidemiology, 17:391-397 Nielsen NR (June, 2006). Interactions between intakes of alcohol and postmenopausal hormones on risk of breast cancer. Oral presentation at the 2 nd American Congress of Epidemiology, Seattle, Washington Nielsen NR (June, 2006). Is perceived stress associated with lower risk of endometrial cancer? Poster presentation at the IEA-EEF European Congress of Epidemiology, Utrecht, Holland Nielsen NR, Grønbæk M. (2006). Stress and breast cancer: A systematic update on the current knowledge. Nat Clin Pract Oncol, 3:612-620. Nielsen NR (November, 2006). Perceived stress and risk of colorectal cancer in men and women. Poster presentation at the International Congress of Behavioral Medicine, Bangkok, Thailand Nielsen NR, Strandberg Larsen K, Grønbæk M, Kristensen TS, Schnohr P, Zhang ZF (2007). Is perceived stress associated with lower risk of endometrial cancer? A prospective cohort study. Psychosom Med (in press) Nielsen NR, Kristensen TS, Zhang ZF, Strandberg-Larsen K, Schnohr P, Grønbæk M (2007). Sociodemographic status, stress and risk of prostate cancer. A prospective cohort study. Ann Epidemiol (in press) xiv Strandberg Larsen K, Nybo Andersen A, Olsen J, Nielsen NR, Grønbæk M. (2006). Do women give the same information on binge drinking during pregnancy when asked repeatedly? Eur J Clin Nutr, 60: 1294-1298 Truelsen T, Nielsen NR, Boysen G, Grønbæk M (2003). Self-reported stress and risk of stroke. Stroke, 34(4):856-62. xv ABSTRACT OF THE DISSERTATION Psychological stress and risk of hormone-dependent cancers by Naja Rod Nielsen Doctor of Philosophy in Epidemiology University of California, Los Angeles, 2007 Professor Zuo-Feng Zhang, Chair Background and objective: Psychological stress is an increasing public health problem and may play a role in the etiology of hormone-dependent cancers by impairing the body’s synthesis of and sensitivity towards sex steroid hormones. The objective of the dissertation is to address a potential relation between perceived stress and risk of primary breast, endometrial, colorectal, and prostate cancers as well as to address if perceived stress affects endogenous levels of sex steroid hormones. Material and methods: The 12,698 men and women who participated in the longitudinal Copenhagen City Heart Study were asked about their stress level in 198183. The participants were followed in a cancer registry until year 2000 and first-time incidence of primary hormone-dependent cancers were identified. Less than 0.1 % was lost to follow-up. Causal diagrams were used to visualize the assumed causal model and Cox proportional hazard models were used to analyze data. Plasma levels of sex steroid hormones were assayed in a sub-sample of postmenopausal women. Results: A one-unit change in perceived stress measured on a seven-point stress scale was consistently associated with lower risk of breast (hazard ratio: 0.92; 95 % confidence intervals: 0.85-0.99), endometrial (0.88; 0.76-1.01), and colon cancer (0.89; 0.81-0.99) in a linear dose-response manner in women. Some sub-groups were more sensitive to the effect of stress than others, especially those women who received xvi hormone therapy. However, no differences in endogenous sex steroid hormones were observed among postmenopausal women with varying stress levels. There were no evidence of a relation between perceived stress and risk of colorectal (0.99; 0.90-1.09) or prostate cancer (0.99; 0.90-1.09) in men. Discussion: We found perceived stress to be associated with a lower risk of hormonedependent cancers in women. However, it is important to emphasize that stress is not a healthy response and that the total burden of disease attributable to perceived stress most likely may exceed the few cases of hormone-dependent cancers that may be prevented by stress. xvii I. Introduction Psychological stress is a public health problem in the Western world. Taking Denmark as an example, about one in every ten Danish adults report high levels of stress in their daily life, and the number is increasing.1 This may imply adverse health consequences for the individual as well as lead to expenses at the societal level. Psychological stress may affect the immune system and the hormone system, as well as lead to changes in health behavior and may thereby affect the risk of cancer.2 More than half of all incident cancers in women and more than one third of all incident cancers in men in the Western world seem to be related to endogenous levels of sex steroid hormones.3 This includes common cancers such as breast, colorectal, and prostate cancers.4-9 Each of these cancers constitute a major public health problem and known risk factors can only partly explain their incidence. In order to prevent these diseases, modifiable risk factors, which affect the cancer process directly via initiation or promotion of the carcinogenesis or indirectly through changes in hormone levels, should therefore be identified. The role of psychological stress in the etiology of hormone-dependent cancers has been an area of emerging interest, in part because of its suggested ability to alter endogenous levels of sex steroid hormones.10-12 The sympathetic nervous system and the hypothalamic-pituitary-adrenal (HPA) axis are the main mediators of the stress response.13 We use this system to adapt ourselves to unexpected and stressful situations. Stress hormones released by the HPA-axis might suppress the synthesis and metabolism of sex steroid hormones, 10-12 which are important risk factors for hormone-dependent cancers. This hypothesised stress-induced suppression of sex steroid hormones may lead to a lower risk of hormone-dependent cancers. Conversely, persistent activation of the HPA-axis also seem capable of suppressing the function of the immune system, which may result in a reduced ability to recognize and destroy neoplastic cell growth.2;14 This mechanism may increase the risk of cancer and thereby oppose the stress-induced suppression of sex steroid hormones. The importance of each of these mechanisms probably depends on the population and the type of cancer in question. Three distinct phenomena have been studied as aspects of the stress concept, that is, external stressors, perceived stress, and individual states of stress (distress). The 1 majority of stress research in relation to hormone-dependent cancers has focused on external stressors and especially stressful life events as a potential risk factor for breast cancer. 15 Perceived stress, on the other hand, arises when the individual finds there to be an imbalance between external stressors and his or her ability to cope with them.16 This will often lead to distress, which is an individual state of stress ranging in severity from modest stress-reactions and sleep-problems to fatigue, vital exhaustion, burn out, and depression. Few studies have addressed the impact of perceived stress on risk of tumors of the mammary gland, the male and female reproductive system organs, or colorectal cancer. The physiological stress response is highly dependent on the type and the timing of the stressor, which makes it vital to distinguish between different types of stress. Several studies have reported that the physiological stress response resulting from acute stress differs markedly from the one resulting from chronic stress.13;17 The physiological effects of acute stressors are in most cases reversible due to the remarkable ability of the human organism to re-establish homeostasis. In an acute stress situation, the body is prepared to use all its energy on survival and re-establishment of homeostasis, which is an appropriate response. The problems mainly arise if the stress-response is prolonged and becomes chronic in nature, which may result in permanent disturbances in the homeostasis of vital body systems.17 My hypothesis is that perceived stress can lead to chronic disturbances in the homeostasis of sex steroid hormones and thereby be related to risk of hormone-dependent cancers. Objective and specific aims The main objective of the dissertation is to address a potential relation between perceived stress and risk of primary breast, endometrial, colorectal, and prostate cancers as well as to address if perceived stress affects endogenous levels of sex steroid hormones. The objectives are addressed in studies based on data from the Copenhagen City Heart Study, which is a cohort of 12,698 Danish men and women prospectively followed-up for 20 years. An approach combining the information obtained from the present studies with current biological, psychological, and epidemiological knowledge 2 will be used to address the posed questions in a valid manner. The dissertation has five specific aims: Specific aim 1. To examine the main effects of self-reported measures of stress intensity and stress frequency on first time incidence of primary breast cancer and to investigate whether the main effects of stress are altered by menopause and postmenopausal hormone therapy among the women who participated in the Copenhagen City Heart Study. Cox proportional hazard models are used to analyse data. Specific aim 2. To examine the main effects of self-reported measures of stress intensity and stress frequency on first time incidence of endometrial cancer and to investigate whether the main effects of stress are altered by menopause and postmenopausal hormone therapy among the women who participated in the Copenhagen City Heart Study. Cox proportional hazard models are used to analyse data. Specific aim 3. To examine the main effects of self-reported measures of stress intensity and stress frequency on incidence of first time primary colon and rectal cancers and to investigate whether the main effects of stress are altered by sex among the participants in the Copenhagen City Heart Study. Cox proportional hazard models are used to analyse data. Specific aim 4. To examine the main effects of socio-economic status, marital status and self-reported measures of stress intensity and stress frequency on first time incidence of primary prostate cancer among the men who participated in the Copenhagen City Heart Study. Cox proportional hazard models are used to analyse data. Specific aim 5. To assess if self-reported stress intensity and stress frequency were associated with plasma levels of estrogen, testosterone, and cortisol in a subset of 1,150 postmenopausal women randomly sampled from all the postmenopausal women who participated in the Copenhagen City Heart Study. Linear regression models are used to analyse data. 3 Causal hypothesis The underlying causal hypothesis is that perceived stress will affect the risk of hormonedependent cancers by direct pathophysical mechanisms and indirect behavioral mechanisms (figure 1-1). Perceived stress may directly lead to prolonged activation of the sympathetic nervous system and the hypothalamus-pituitary-adrenal axis, which through cortisol and other stress hormones can affect the synthesis and metabolism of sex steroid hormones as well as impair the immune system and thereby, depending on the etiology of each specific type of cancer, result in an altered risk of hormonedependent cancers. Further, high levels of perceived stress may lead to changes in health-related behavior, whereby stress contributes to a higher frequency of adverse health behaviors (e.g., smoking, high fat diet, physical inactivity, and high alcohol consumption), which may indirectly affect the risk of hormone-dependent cancers. Outline The dissertation is based on a review paper and four epidemiologic studies, which are either published in or submitted to peer reviewed journals (please refer to acknowledgements for references). The dissertation is divided into twelve chapters. The theoretical basis of the dissertation is presented in the current chapter. Chapter two is a detailed description of the material and methods used in the dissertation. The following chapter is a systematic review of the evidence from prospective studies on stress as a risk factor for breast cancer incidence or relapse. The fourth chapter is an overview over the relatively few studies that have assessed the association between stress and other hormone-dependent cancers. The following four chapters (five to eight) each include a prospective cohort study that addresses the association between perceived stress and risk of breast, endometrial, colorectal, and prostate cancer, respectively. A cross-sectional study on the association between perceived stress and sex steroid hormones is described in chapter nine. Chapter ten includes a critical discussion of the evidence of a relation between perceived stress and risk of hormone-dependent cancers. The degree to which the estimated quantitative measure of the association between perceived stress and risk of hormone-dependent cancers deviates from a possible causal relation by systematic and random processes is also addressed. The empirical evidence of a relation between 4 perceived stress and risk of hormone-dependent cancer and its causal implications is summarized in the conclusion in chapter eleven. The conclusion is put into a public health perspective in the final chapter, where the individual and social consequences of a relation between stress and risk of hormone-dependent cancers and other health outcomes are briefly addressed, and future studies on stress and hormone-dependent cancers are discussed. Definition of stress Theoretical background There is no clear consensus on how to define stress, and many different definitions have been suggested. Three distinct phenomena of the stress concept have primarily been studied, that is, external stressors, perceived stress, and distress. Stress theories, which provide the theoretical basis for the different definitions of stress, can at the very general level be divided into three major and quite distinct approaches: stimulus-oriented, response-oriented, and interactional theories.18 The stimulus-oriented and responseoriented theories provide the theoretical background for the study of external stressors and the stress reaction (distress), respectively, while the interactional theories emphasize perceived stress (figure 1-2). The main characteristics of the different theories are summarized in table 1-1. The stimulus-oriented theories define stress in terms of external stimuli that may increase the demand of the individual (stressors). The main focus of the stimulusoriented theories is the number and amount of external stressors, such as adverse life events (e.g., divorce, death within the family), daily hassles (e.g., computer problems, being late for the bus), and characteristics of the job situation (e.g., demands, control, rewards). Response-oriented theories, on the other hand, define stress as the mental and bodily reaction generated by stressful stimuli.18 Such reactions may include neurobiological changes in cortisol levels, physical changes in blood pressure, heart rate and muscle tension, as well as varying degrees of distress (e.g., anxiety, nervousness, depression). By emphasizing the individual as an important mediator between external stressors and the resulting stress response, the interactional theories provide a newer approach to 5 stress research.18 Each individual has different capacity and ways of handling stressful situations, and the same external stressor may therefore result in different levels of stress depending on the individual’s mental attitude and personal resources.13 Early learning in concert with psychosocial and cultural influences may be important in determining the individual’s attitude toward a potential stressful situation.19 Not only the attitude toward stressful situations but also external resources play an important part in coping with stressful situations. While some individuals have a strong social network, which may act as a buffer against a detrimental behavioral or physiological stress response, others lack this kind of support. An individual’s social network and personal resources seem to greatly influence the way the individual adapts to and resolves an encounter with a stressful situation.20 The interactional theories emphasize the importance of taking individual differences in appraisal and coping into account when measuring stress. The theory on stress, appraisal and emotion developed by Richard S. Lazarus has been a cornerstone in the development of the interactional theories.16 According to Lazarus, neither the stimulus- nor the response-oriented approaches succeed in explaining the stress experience. The traditional stimulus-oriented approach fails to explain why and how we respond differently to an external stimulus, which makes the stimulus alone insufficient to define stress.16 Individual differences in stress response becomes more pronounced when addressing less acute stressors, like everyday stress, making the stimulus-approach even more insufficient for this type of research. The responseoriented approach does not provide any important additional information beyond that of the stimulus-approach because it is built on a circular reasoning, where the stressor is defined by the fact that it leads to a stress response, while the stress response, in turn, is defined in reference back to the stressor.16 Further, some response-oriented theories have been used to try to describe the phenomenon (stress) by describing its parts (e.g. elevated blood pressure). A problem with this approach is that we cannot necessarily explain what happens on one level of analysis by referring to another level. The interactional theories are putting the person back in the equation by emphasizing the importance of the personal cognitive appraisal of how threatening an external stimulus seems to the specific person’s values, beliefs and goals, and whether he or she will be able to successfully cope with it. 16 In the dissertation, stress will 6 therefore, in accordance with Lazarus, be defined as the individual’s appraisal of an imbalance between demands and the individual’s resources to cope with it.16 Acute and chronic stress The concept of allostasis will be used to illuminate the subtle differences between acute and chronic stress in their effect on pathologies such as hormone-dependent cancers. Allostasis is central to the stress response and to survival in general, because it is the process by which we actively adjust ourselves to predictable and unpredictable events, and thereby preserve the homeostatic stability of vital body systems.17 The primary mediators of allostasis are the sympathetic nervous system and hormones of the hypothalamus-pituitary-adrenal (HPA) axis. These mediators are altered in response to changing environments or other challenges to the individual and are therefore essential to the stress response. Most allostatic mediators have a biphasic role with protective effects in the short run and damaging effects over longer time periods.17 Glucocorticoids are good examples of this biphasic role in that they can convert proteins and lipids to carbohydrates and thereby replenish energy reserves in an acute stress situation. At the same time, they can result in obesity and insulin resistance if they are chronically elevated in response to sustained stress.17 Sustained stress is therefore important in terms of disease development and this type of stress will be the main focus of this dissertation. In sum, stress should ideally be assessed as an individual perception of prolonged stress. Incidence and prevalence of hormone-dependent cancers Breast cancer is the most common cancer in terms of incidence and the second most common cancer in terms of mortality among Danish women (figure 1-3). Colorectal cancer is the second most common cancer in both men and women in Denmark, closely followed by prostate cancer in terms of both incidence and mortality among Danish men. Other hormone-dependent cancers such as endometrial, ovarian, cervix, and testicular cancers are relatively rare both in terms of incidence and mortality. Fortunately, the incidence rates of even the most common hormone-dependent cancers are low, and the studies included in the dissertation will therefore be confined to the study of stress as a risk factor for the three most common hormone-dependent cancers, 7 namely breast-, colorectal-, and prostate cancer. Despite a low incidence rate, a potential relation between stress and risk of endometrial cancer will also be addressed, because endometrial cancer, together with breast cancer, shows the strongest dependency on sex steroid hormones, especially estrogens. Estimates of crude and age standardized incidence and mortality rates as well as the one-year prevalence of breast, endometrial, colorectal, and prostate cancer are shown in table 1-2. The world standardized incidence rates of all of these cancers are markedly higher in both Denmark and the U.S. compared to the world average. The age standardized incidence rates of the cancers in question were slightly lower in Denmark compared to the U.S., but the mortality rates were in general higher. Among the Danish population of about five million people, 3,792 women had prevalent breast cancer, 1,401 women had colorectal cancer, and 610 women had endometrial cancer in year 2002.3 These three types of cancer were estimated to account for 47.3 % of all new cancer cases among Danish women in the same year.3 Among Danish men, 1,636 had prostate cancer and 1,405 had colorectal cancer, and these two types of cancer were estimated to account for 31.2 % of all new cancer cases in Danish men in 2002.3 Thus, hormone-dependent cancers seem to be a major public health issue. 8 Table 1-1. Overview of major stress theories and definition Stimulus-oriented theories Main focus Measurement method Examples of measuring tools Perceived stress Stress is the individual’s appraisal of an imbalance between demands and the individual’s resources to cope with it. The stress response Stress is the mental and bodily reaction generated by stressful stimuli Stressful life events Daily hassles Job strain Effort-reward imbalances Harmful work conditions Perceived disparity between threats and coping resources Physical: Level of stress hormones, blood pressure reactivity, heart rate reactivity, muscle tension Degree of distress: Nervousness, anxiety, depression Self-rated Assignment of mean score Independent count of external stressors Self-rated subjective appraisal Measurement of physical indicators of stress Measurement of stress-related behavior and emotions The social readjustment rating scale The job content questionnaire The Perceived Stress Scale Well-developed and validated Responsive to individual differences in appraisal of external stimuli Failure to account for the individual differences in appraisal of and coping with external stressors A question on perceived stress may be understood differently by different individuals, leading to possible misclassification Beck Depression Inventory The SCL-90-R (a multidimensional self-report symptom inventory developed to assess symptomatic psychological distress) Standardized ‘objective’ biological and physical measures. Validated scales The reasoning is circular, in that the stress stimuli is defined by the fact that is lead to a stress response, while the stress response, in turn, is defined by referring back to the stimuli. Strengths Weaknesses Response-oriented theories External stressors Stress is environmental stimuli Stress definition Measures Interactional theories Table 1-2. Estimates of the incidence and mortality rate per 100,000 as well of the one-year prevalence in the total population of hormonedependent cancers in Denmark, USA, and the World Denmark Female breast cancer Crude incidence rate Age-standardized incidence rate Crude mortality rate Age-standardized mortality rate 1-year prevalence in the population Endometrial cancer Crude incidence rate Age-standardized incidence rate Crude mortality rate Age-standardized mortality rate 1-year prevalence in the population USA World 144.2 88.7 51.7 27.8 3 792 143.8 101.1 29.4 19.0 211 400 37.4 37.4 13.3 13.2 1 060 042 24.0 13.3 6.2 2.9 610 32.8 22.8 4.0 2.6 46 696 6.5 6.5 1.6 1.6 183 528 Colorectal cancer Female Crude incidence rate Age-standardized incidence rate Crude mortality rate in the population Age-standardized mortality rate 1-year prevalence in the population Male Crude incidence rate Age-standardized incidence rate Crude mortality rate Age-standardized mortality rate 1-year prevalence in the population 66.9 33.0 41.4 19.2 1401 55.1 33.1 20.3 11.6 69 939 15.4 14.6 8.1 7.6 362 911 69.3 41.0 40.1 23.3 1405 60.0 44.6 20.9 15.2 75 481 17.6 20.1 8.9 10.2 423 416 Prostate cancer Crude incidence rate Age-standardized incidence rate Crude mortality rate Age-standardized mortality rate 1-year prevalence in the population 69.8 39.3 40.7 22.6 1636 168.9 124.8 22.8 15.8 233 926 21.7 25.3 7.1 8.2 604 506 Source: Globocan 2002, the International Agency for Research on Cancer Age-standardized rate: a summary measure of the rate that a population would have had if it had had the age structure of the world. Figure 1-1. Causal hypothesis for the relation between perceived stress and risk of hormone-dependent cancers Perceived stress Changes in health-related behavior Physical stress response High alcohol consumption, high fat diet, physical inactivity, and smoking Prolonged activation of the HPA-axis and the sympathetic nervous system Stress hormones Cortisol Hormone system Immune system Decreased synthesis of sex steroid hormones Impaired function Hormone-dependent cancers Figure 1-2. Different approaches to stress research and their relation to hormone-dependent cancers Physiological stress response External stressors Perceived stress Cancer Behavioral changes Stimulus-oriented theories Interactional theories Response-oriented theories Figure 1-3. Age-standardized incidence and mortality rates of site-specific cancers in Denmark in year 2002 13 II. Material and methods The Copenhagen City Heart Study The study objective will be addressed using data from the Copenhagen City Heart Study. The Copenhagen City Heart Study is a longitudinal study initiated in 1976.21 An agestratified sample of 19,698 men and women aged 20 to 93 years who lived in the Copenhagen area were randomly drawn from the Central Population Registry and invited by letter to participate in the study. A physical examination was performed and participants were asked to fill in a questionnaire regarding various risk factors. A blood sample was drawn during the examination. In 1981-83 the study population was supplemented with 500 men and women aged 20 to 29 years, and additional study assessments were performed for both new and continuing study participants. The study participants were asked about their stress level only at this second examination, which is therefore used as baseline for the studies included in the dissertation. The 12,698 women and men who participated in the second examination constituted a response proportion of 70 %. The vast majority of the participants were Caucasians and all participants gave written informed consent. The Danish ethics committee for the City of Copenhagen and Frederiksberg approved the Copenhagen City Heart Study (# 01-144/01). Table 2-1 shows the baseline characteristics of the participants in the second examination of the Copenhagen City Heart Study. There are more women than men and the average age was 57 for women and 56 for men. About 28 percent of the women and 20 percent of the men reported moderate to high stress intensity, and about the same percentages experienced stress on a weekly or daily basis. A relatively high proportion of the population had low education and low income. The male participants generally had a higher weekly alcohol intake compared to their female counterparts. More than half of the population was current smokers at baseline, while less than 20 percent were physically inactive. The mean body mass index was in the normal range and few participants had diabetes mellitus. Most women had gone trough menopause at baseline, and only about one fifth of these women used postmenopausal hormone therapy. 14 Measurement of perceived stress The study participants in the Copenhagen Heart Study were asked about their level of stress in terms of intensity and frequency at baseline in 1981-83. In the questionnaire, stress was exemplified as the sensation of tension, nervousness, impatience, anxiety, or sleeplessness and no time frame was specified. To assess stress intensity, the participants were asked: “Do you feel stressed?” The response categories were: (0) none, (1) light, (2) moderate, or (3) high. To measure stress frequency the participants were asked: “How often do you feel stressed?” The response categories were: (0) never/hardly ever, (1) monthly, (2) weekly, or (3) daily. In order to combine the two dimensions of stress intensity and frequency, we chose to use an additive approach, where the values of the two questions were added and combined into a seven-point stress score ranging from 0 (indicating low stress) to 6 (indicating high daily stress). How the score was formed is shown in figure 2-1. For example, if a woman reported moderate, daily stress this woman would be given a stress score of 5 (2 points for moderate at the intensity dimension plus 3 points for daily at the frequency dimension). This stress-score was categorized into low (0-1 points), medium (2-4 points), and high (5-6 points) stress. An obvious alternative would have been to combine stress intensity and frequency analogous to methods used in nutritional epidemiology, where the participants are often asked about the average intake of a specific nutrient or food (intensity), which is then multiplied with the frequency of the intake. If stress intensity was given the values 0 (none), 1 (low), 2 (medium), 3 (high), and stress frequency was measured as number of days per month, so that monthly stress had the value 1, weekly stress had the value 4, and daily stress had the value 30, using the nutritional approach to create an alternative stress-score, we would multiply the values of stress intensity and stress frequency. This scale would give considerably more weight to frequency than intensity (even with higher values given to stress intensity), where the additive approach weighs them equally. An intuitive problem with the multiplicative approach is that weekly high stress would be given a score of 12, which is less than half the score of 30, which would be given to daily low stress. However, most people would find it more straining to be exposed to weekly high stress than daily low stress. This would be better resembled in the additive approach, where weekly high stress has a higher score than daily low stress. We 15 therefore chose to use the additive instead of the multiplicative approach in the studies included in the dissertation. Assessment of endpoints by linkage to national registers The participants of the Copenhagen City Heart Study were followed in national registers from date of entry into the study till date of first diagnosis of a hormone-dependent cancer, death, loss to follow-up, emigration, or end of follow-up between 2000 and 2003, depending on the study in question. In Denmark, every newborn citizen is assigned a unique civil registry number, which is a ten-digit number consisting of the date of birth and four unique digits. Using the civil registry number a complete hospital discharge history was established for each individual and unambiguous record linkage could be performed. Identification of site-specific cancer cases was obtained through linkage to the Danish National Cancer Registry. Reporting of new cancer cases to the registry is compulsory in Denmark and the Danish National Cancer Registry contains, according to the National Board of Health, data on more than 95 percent of all cancer diagnoses in Denmark. The Cancer Registry has consistently used the International Classification of Disease codes revision seven (ICD-7). The vital status of the study population was followed in the Civil Registration System. Assessment of endpoints in central disease registries allowed for nearly complete follow-up (less than 0.1 percent were lost to follow-up). The number of cancer cases that occurred at each site is shown in table 2-2. Statistical methods Building a conceptual model Causal diagrams will be used to clarify the assumed causal models that constitute the basis for the statistical analyses in each of the included studies. The use of causal diagrams in epidemiology was suggested by Greenland, Pearl, and Robins as a useful alternative to conventional statistical models, and as a way of elucidating assumptions about the web of causation without incorporating strong parametric assumptions.22 In a 16 causal diagram, one’s assumptions about causal relations are graphically shown in a diagram, where the variables are connected with arrows that represent assumed causal relations. These assumed relations should always be based on the most updated scientific knowledge. The main advantage of using causal diagrams is their ability to visualize model assumptions and to include both measured and unmeasured variables. As opposed to the conventional statistical methods, in which the underlying causal model is based primarily on statistical associations assessed in a particular data material at hand, causal diagrams have the advantage of making use of prior knowledge of causal relations. The diagrams will be used in each study as a way to identify a set of potential confounders that may create a spurious association between perceived stress and the hormonedependent cancer in question. The stated diagrams should not be taken as comprehensive causal models, and should instead be understood as a graphical presentation of the assumed causal model underlying the statistical model. Every statistical model is based on an assumed causal model, whether stated or not, and by clearly stating our assumption in causal diagrams we make the assumptions open to debate. The Cox proportional hazard model Cox proportional hazard models will be used to estimate the hazard ratios of hormonedependent cancers associated with perceived stress in the dissertation. The Cox proportional hazard model is a log linear model where the log hazard function at time t is given by log(h(t)| X=x)) = log(h0(t)) + bx where h0(t) is the baseline hazard function and X=x is the exposure level for a given subpopulation. The regression coefficient b can be interpreted as the log relative hazard associated with one unit increase in X. Thus, if X denotes the continuous stress score with seven levels, the hazard ratio associated with a one-level increase in stress score would be given by exp(b). The above model only includes one covariate, namely X, but it can easily be extended to include multiple covariates log(h(t)| X1=x1,…., Xk=xk) = log(h0(t)) + b1x1 + … + b kxk 17 If X1 is the continuous stress score, then exp(b1) will be the hazard ratio associated with one unit increase in the stress score, adjusting for the other covariates. The effect of stress intensity and stress frequently will be assessed separately as well as combined in a stress score with seven levels. Stress intensity and stress frequency were assessed in a multiple-choice form and are therefore naturally ordered into four categories each. These categories will be added to the model, so that each level of stress intensity is compared to the no stress category and each level of stress frequency is compared to the never/hardly ever stress category. The combined stress score will be included as a continuous variable and in categories of low, medium, and high stress. By including the stress score as a continuous variable we maximize the utility of the data available. By doing this however, we also assume that the hazard ratio associated with one unit increase in stress score is constant. This is not necessarily a valid assumption, because the hazard ratio associated with going from low stress on a monthly basis to low stress on a weekly basis may be very different from the hazard ratio associated with going from high stress on a weekly basis to high stress on a daily basis. The stress score will therefore be included both as a continuous and as a categorical variable in order to evaluate how these assumptions affect the results. Age will be used as the underlying time scale in order to soundly adjust for confounding by this variable. The proportional hazards assumption, stating that the proportional hazards will be constant over the time-scale, will therefore apply to this age-scale. The validity of this assumption will be evaluated graphically as well as by testing an age-covariate interaction in data. Linear regression models Linear regression models will be used to address the association between perceived stress and plasma levels of estrogen, testosterone, and cortisol. A simple linear regression model is given by: E(y│X=x) = a + bx + ε 18 where y denotes the dependent variable and X=x is the exposure level for a given subpopulation. a and b are the parameter estimates and ε (the error term) is the unexplained and unpredicted variance of y. The regression coefficient b can be interpreted as the expected change in y with a one-unit increase in X. Thus, if X denotes the continuous stress score with seven levels, b is the expected change in plasma levels of, for example, estrogens associated with a one-level increase in stress score. The above model only includes one covariate, namely X, but it can easily be extended to include multiple covariates E(y | X1=x1,…., Xk=xk) = a + b1x1 + … + b kxk + ε If X1 is the continuous stress score then b 1 will be the expected change in y associated with a one-unit increase in the stress score, adjusted for the other covariates. A linear regression model is based on several assumptions: (1) The outcome is continuous; (2) The effect of exposure is measured as differences in the mean of the outcome variable; (3) The conditional mean of the outcome given the independent variables included in the model are assumed to be a linear function of the independent variables; (4) The outcome is assumed to be normally distributed for each of the combinations of the independent variables; (5) Variance homogeneity. We used plasma levels of estrogens, testosterone, and cortisol as outcome measures which are continuous in nature (assumption 1). The distributions of the outcome measures were (skewed) nonnormal and we therefore log-transformed the outcome measures in order to make assumptions 2 and 4 reasonable. We either assumed that the conditional mean of the outcome was a linear function of the independent variables or we included the independent variables as categorical variables (assumption 3). We plotted the residuals (observed value minus expected value) against the fitted values to address variance homogeneity (assumption 5). In a well-fitted model there should be no pattern to the residuals plotted against the fitted values. 19 Table 2-1. Baseline characteristics of the 12,698 men and women who participated in the second examination of the Copenhagen City Heart Study in 1981-83 Women Men Study population, n 7,018 5,680 Mean age, y (range) 57 (21-91) 56 (21-98) Moderate to high stress intensity (%) 1,971 (28) 1,163 (20) Weekly or daily stress (%) 1,933 (28) 1,163 (20) <8 years of education (%) 3,249 (46) 2,539 (45) Income of <$1000/month (%) 2,570 (37) 1,437 (25) 4 (7) 14 (16) 3,734 (53) 3,622 (64) 1,270 (18) 922 (16) 25 (5) 26 (4) Diabetes mellitus (%) 125 (2) 197 (3) Oral contraceptive use (%) 258 (4) - Menopause (%) 5,167 (74) - Hormone replacement therapy (%) 1,062 (15) - Mean alcohol consumption, drinks/wk (SD) Current smoker (%) Physical inactive (%) 2 Mean body mass index, kg/m (SD) Table 2-2. Number of primary site-specific hormone-dependent cancer cases occurring during follow-up (year 1981-2000) in the Copenhagen City Heart Study Site-specific cancers No. of primary cases Breast 251 Female reproductive system organs Ovary Cervix Endometrial Other female genital organs 190 82 31 72 5 Male reproductive system organs Prostate Testis Other male genital organs 157 157 0 0 Colorectal Colon Rectum 328 236 92 Figure 2-1. Formation of the stress-score based on an additive approach. The numbers indicates the score on the seven-point stress-score and the numbers in parentheses indicates the proportion of the total study sample in each category. The lines indicate the categorization of the sevenpoint stress-score into low, medium, and high stress Intensity Frequency No Light Medium High Never 0 (38 %) 1 (1 %) 2 (0 %) 3 (0 %) Monthly 1 (9 %) 2 (21 %) 3 (5 %) 4 (1 %) Weekly 2 (1 %) 3 (5 %) 4 (9 %) 5 (3 %) Daily 3 (0 %) 4 (0 %) 5 (1 %) 6 (4 %) III. Stress and breast cancer: a current overview of prospective studies Introduction Breast cancer is the most common cancer among women both in terms of incidence and prevalence.23 Psychological stress may affect the immune system and the hormonal system, and may also lead to changes in risk behavior, which may again play a role in breast cancer etiology and prognosis.2;10;12 Considerable research has been carried out in an attempt to establish the type and strength of a relation between psychological stress and breast cancer.15;24-29 The heterogeneity of the results reported in the individual studies are reflected in the conflicting conclusions of previous reviews. 15;24-29 Stress may also render the individual more susceptible to the progression or recurrence of breast cancer either by disturbing recovery or by affecting treatment compliance. The association between stress and breast cancer relapse has gained less attention than the association with breast cancer incidence and the results are conflicting.30 We have chosen to address breast cancer incidence and relapse as two distinct entities in order to assess if stress affects development and progression of breast cancer differently. Previous reviews assessing the relationship of stress and breast cancer have included a range of studies based on different designs, which may explain some of the heterogeneity in the results. In a recent meta-analysis of stressful life events and risk of breast cancer, Duijts and colleagues found that the estimates from retrospective studies differed from those of prospective ones.15 Diagnosis of cancer is stressful, which might increase the likelihood for recall bias in retrospective studies. Therefore, we have chosen to only include prospective studies in this review, where stress is measured before breast cancer incidence and therefore is not affected by the diagnosis. The major limitation is that prospective studies are often expensive and time-consuming, making only few prospective studies on stress and breast cancer available. How stress is defined and measured depends on the theoretical background applied. Some researchers define stress in terms of external demands or stressors, while others emphasize the importance of the individual’s appraisal of the demands, depending 23 on his or hers coping capacities (perceived stress). The majority of research on stress and breast cancer has been concerned with major life events such as death of a spouse or near relative.15 Life events are measures of major external demands (stressors) that have gained popularity partly because of their objectivity and the fact that they can relatively easily be measured retrospectively in case-control studies.31 Work-related stress, often defined as a work situation with high demands and low control, is another popular measure of external demands.32 However, a person’s appraisal of, and reaction to, a major life event or a demanding work-situation might be just as crucial for breast cancer onset and prognosis as the event per se. Several investigators have emphasized the importance of also including subjective measures of stress, such as measures of perceived stress, in addition to the more objective measures of life events and workrelated stress.16;24;33;34 Thus, we will include measures of stressful life events, workrelated stress, and measures of perceived stress in the present review to accommodate these different theoretical views. The purpose of this review is to systematically address if various measures of stress are risk factors for breast cancer incidence or breast cancer relapse in prospective studies, and to identify sources of potential heterogeneity in the results. Assessment methods Criteria for assessment We used systematic and explicit methods to identify, select, and critically appraise relevant studies as recommended by the Cochrane Collaboration.35 The articles included in this analysis had to describe the relation between some measure of stress and breast cancer incidence or breast cancer relapse. Furthermore, the articles had to describe prospective studies, defined as studies in which stress was measured before disease occurrence. Only the prospective parts of studies that included both retrospective and prospective elements were included in this review. The association between stressful life events has also been evaluated in so called limited prospective designs, where women who undergo biopsies are interviewed about life events prior to the final diagnosis.36-42 These women may both be emotionally affected by the uncertainty of their final diagnosis, and may also have a fairly good prediagnostic forecast of their diagnosis.43 24 Results from limited prospective studies might therefore be affected by recall bias and cannot necessarily be applied to the general population. Therefore, such studies were excluded from this review. Articles describing the association between stress and breast cancer case-fatality or breast cancer mortality, and studies on stress arising in those diagnosed with breast cancer were also excluded. Figure 3-1 shows a flow diagram of the data collection process. The search strategy for this review was based on the following five steps: Step 1. The MEDLINE and PsycINFO databases were searched for publications until November 2005, with no language restriction. The following terms were used in this systematic search: the articles were selected if “stress” or “ life event” or “job strain” or “effort reward” appeared in the Title; OR “stress” or “life event” or ” job strain” or “effort reward” appeared in a keywords search, AND “stress” or “life event” or “job strain” or “effort reward”, and “breast” or “mamma”, and “cancer” or “carcinoma” or “neoplasms” appeared in the abstract. This search resulted in 258 potentially relevant studies. Step 2. After reading the abstracts of the 258 studies, studies that did not describe the relation between some measure of stress and breast cancer incidence or breast cancer relapse were excluded, which meant that 35 studies remained. Non-English language studies were classified according to their title and abstract written in English. Step 3. A total of 23 studies not describing prospective studies were excluded. Study design was not used to limit the search, because not all studies are indexed by study design. Step 4. Assessment of reference lists of all relevant studies and previous reviews identified eight potentially relevant prospective studies.44-51 The majority of these studies were not identified in the general search because they focused primarily on overall cancer risk and only included subgroup analyses on breast cancer. Seven of these studies were included in our analysis, while the last study was excluded because of methodological flaws.50 25 Step 5. Two of the papers included in this search reported on the association between job strain and breast cancer incidence in the Nurses’ Health Study with varying followup.52;53 Only the paper with the longer follow-up was included in the review.53 Altogether, 18 prospective studies were included in our analysis. Data extraction Information on population, follow-up, exposure, outcome, confounding, and results were extracted from each study. The analyses of the association between stress and breast cancer incidence should ideally be adjusted for the following behavioral and lifestyle factors: alcohol, physical activity, body mass index, postmenopausal hormone use; and the following biological factors: family history of breast cancer, benign breast disease, age at menarche, age at menopause, age at first birth, parity; as well as some measures of socio-economic status (i.e., income and education). Because of the heterogeneity in the measures of stress applied in the different studies, calculating a summary measure of effect was neither feasible nor reasonable. Instead, we grouped the studies according to the type of stress measure applied. We interpreted the evidence based on an evaluation of both the point estimates and the statistical variability of the estimates, and the assessment was therefore not restricted to an evaluation of statistical significance. Whether a risk estimate is statistically significant depends both on the strength of the association and the power of the analysis, and because some of the studies had low statistical power we found it inappropriate to base our evaluation solely on statistical significance. Results Study characteristics Thirteen studies on stress and breast cancer incidence44-49;53-59 and five studies on stress and breast cancer relapse were included in the review.51;60-63 The studies were either from Europe, 44-47;49;54;56;57;59-63 North America,51;53;55;58 or Israel,48 and all of them were published in English. These 18 studies included 15 prospective cohort studies44;47;48;51;53- 26 63 and three prospective case-control studies.45;46;49 Six of the studies were record linkage studies based on national registers.44-49 Stress and breast cancer incidence The association between stress and breast cancer incidence was addressed in ten prospective cohort studies and three prospective case-control studies (table 3-1). The association between ‘stressful life events’ and risk of breast cancer was evaluated in five cohort studies and three nested case-control studies. 44-49;58;59 Six of these studies were registry-linkage studies, which were based solely on information from national registries. 44-49 The following events were used as measures of major life events: losing a husband by death or divorce,45;49 parental death in childhood,58 death of a child,44;46 having a child with cancer,47 losing an adult son in war or accident,48 or the combination of a range of stressful life events such as divorce, death of a husband, or death of a close relative.59 Three registry-linkage studies assessed the risk of developing breast cancer among parents who had a child with cancer or who had lost a child, and none of these studies found a clear association between these major stress factors and breast cancer incidence.44;46;47 A registry-linkage study from Denmark reported a slightly lower risk of breast cancer among women who had lost a husband by death.49 This lower risk was not supported in a Norwegian registry-linkage study, where loss of a husband by death was not associated with breast cancer risk.45 Instead, loss of a husband by divorce was associated with a slightly lower risk of breast cancer in this study. In another registrylinkage study Levav et al. reported a slightly higher risk of breast cancer incidence among Israeli women who had experienced losing an adult child, but the risk estimates were rather unstable.48 In contrast to the null results of the registry-linkage studies, the death of ones mother, but not ones father, in childhood was associated with a doubling of breast cancer risk in the Baltimore Epidemiologic Catchments Area Study.58 This association was based on a relatively small sample of 1,213 women followed-up for 15 years during which only 29 cases of breast cancer occurred. An increased risk of breast cancer associated with major life events was also reported in the larger Finnish Twin Cohort Study, which included 10,808 female twins followed for 15 years.59 All of the studies on stressful life events and risk of breast cancer lacked information on postmenopausal hormone therapy and several of the other risk factors for breast cancer. 27 This may have hindered proper control of potential confounding from these factors. In sum, major stressors such as death of a child or losing a husband by death or divorce did not seem to be associated with higher incidence of breast cancer in large-scale registrylinkage studies. An accumulation of stressful life events may, however, increase the risk of breast cancer. The association between ‘work-related stress’ and risk of breast cancer has only been prospectively assessed in the Nurses’ Health Study.53 Schernhammer and colleagues reported a slightly lower risk of breast cancer among women with high strain jobs, defined as jobs with high demands and low influence, compared to women with low strain jobs in a questionnaire-based study, where job strain was measured in the 27item job content questionnaire.53 However, there may have been insufficient variation in the exposure to work-related stressors among nurses because most of them work in very similar jobs. To validly address the effect of work-related stress on breast cancer risk, more prospective studies that include different types of professions are needed. Also, there was no information on current night-shift work, which may both have led to higher job strain as well as have disrupted the circadian rhythm of the hormonal system and thereby predicted a higher risk of breast cancer.64 Such confounding would, however, have resulted in an artificial increased risk of breast cancer contrary to what was observed in the study by Schernhammer et al. Four studies assessed the association between ‘perceived stress’ and incidence of breast cancer.54-57 A Swedish study, which included 1,462 women followed for 24 years, reported an increased risk of breast cancer among women who had experienced stress occasionally or more often during the five years preceding baseline based on a simple question describing the extent to which the women had previously experienced stress.54 These results contrast with those of the three other studies that have addressed the effect of perceived stress, all of which included more women and more cases than the Swedish study. In a Finish study including 10,519 female twins followed for 21 years, no association between stress of daily activities and breast cancer incidence was found.56 A different form of stress, namely stress from adult caregiving, was associated with a slightly lower risk of breast cancer incidence in the Nurses’ Health Study after appropriate adjustment for confounding.55 This inverse association was supported by the fact that women with high levels of stress from caregiving also had lower levels of sex 28 steroid hormones.55 An inverse dose-response association between perceived stress and incidence of breast cancer has also been found in a recent Danish study.57 Despite the inconsistent results, perceived global stress does not seem to be associated with higher risk of breast cancer. Whether perceived stress reduces the risk of breast cancer remains to be confirmed and the underlying mechanisms need to be elaborated. Stress and breast cancer relapse Five prospective studies have evaluated the association between stress and breast cancer recurrence (table 3-2). Four of these studies were conducted in Europe and one in Canada. Breast cancer patients were followed from the date of diagnosis to first-time recurrence. All studies included some measure of stress one year prior to diagnosis, while only two studies included measures of the life events that occurred during followup.60;63 Breast cancer patients who had experienced any severe event or difficulty one year prior to diagnosis were at lower risk of relapse in an English study with 3.5 years of follow-up,60 while no association was reported in another English study of a similar size but with five years of follow-up.63 Consistently, no association between relapse and stress from recent life events prior to diagnosis was reported in the Canadian study.51 In a smaller study from Finland, experience of life events 12 months prior to diagnosis was associated with a markedly higher risk of relapse after eight years of follow-up.62 Chronic stress prior to diagnosis was also associated with a slightly higher risk of relapse in a small study from Belgium.61 Experience of severe life events during followup was not associated with higher risk of breast cancer relapse in either of the two English studies.60;63 In fact, one of the studies reported a lower risk of relapse five years after diagnosis.63 All studies controlled for axillary-node involvement, which is one of the main prognostic factors for relapse. Most of the studies also controlled for other prognostic factors such as grade and stage of the tumor. However, none of the studies had information on treatment history or treatment compliance. Discussion The relation between stress and breast cancer incidence and relapse were evaluated using qualitative data from 18 prospective studies. All studies were conducted in Westernized 29 countries and the results were inconsistent. No higher risk of breast cancer was found in large-scale registry linkage studies that addressed the effect of a major life event such as death of a child or divorce. An accumulation of stressful life events may, however, be associated with higher risk of breast cancer. Work-related stress and perceived stress were either not associated with breast cancer risk or were associated with a slightly lower risk of breast cancer. In general, a higher risk of breast cancer incidence was mainly seen in the smaller studies, which could indicate some degree of publication bias; we conclude that stress does not seem to be an important risk factor for breast cancer incidence. By contrast, stress experienced one year prior to diagnosis of breast cancer was associated with a higher risk of relapse in some, but not all, studies.51;60-63 Only two studies addressed whether experience of life events after diagnosis affected the risk of breast cancer relapse, and neither study reported a changed risk.60;63 Burke and Goodkin suggested a range of criteria when addressing the evidence of a link between stress and cancer.33 First, they emphasized the importance of distinguishing between stressors and stress. A stressor is an external stimulus that may potentially cause stress depending on the individual’s perception of the threat the stressor poses. Second, the association between stress and cancer should be assessed prospectively to avoid recall bias from either attribution or emotional repression. Third, according to Burke and Goodkin, a fair test of the association of cancer risk and stress should examine a specific type of cancer, because cancer is a heterogeneous group of diseases with multiple etiologies.33 In the present review we distinguished between stressors (stressful life events and work-related stress) and stress (perceived stress) and we only included prospective studies that addressed a specific type of cancer, namely breast cancer. Furthermore, we used systematic methods to identify the literature. In spite of this we found inconsistent results. Sources of heterogeneity The heterogeneity of the results observed in the present review could be ascribed to several factors. First, a broad range of different exposures has been regarded to as measures of stress, though they might not all express dimensions of the same underlying concept. We have tried to categorize the different measures of stress into stressful life events, work-related stress, and perceived stress in order to make the exposure groups 30 more homogenous. However, this resulted in few studies in each category, and even within the categories different measures of stress were applied in the different studies. Perceived stress of daily life is not necessarily the same as stress of daily activities. Similarly, the death of one’s mother in childhood is different from losing a child later in life. Thus, the heterogeneity in results might be a reflection of the variety in the measures of stress as well as the lack for sufficient prospective studies to address the effect of each stress measure on breast cancer risk and progression. Also, most of the studies included only a single measure of stress assessed at baseline. Each of these measures may therefore only account for a fraction of the individual stress burden. We cannot exclude that an ideal study, in which the full stress burden for each individual could have been comprehensively measured, might have come up with a different result. Second, the physiological stress response is highly dependent on the type and the timing of the stressor,17 which makes it important to distinguish between different types of stress. Thus, by categorizing stress into stressful life events, work-related stress, and perceived stress we implicitly assumed that these measures would affect the risk of breast cancer differently. The results partly support this assumption in that they indicate some differences in the effect of major acute life events and the more chronic exposures to stress of everyday life. A lower risk of breast cancer was especially observed in the studies that applied measures of perceived stress. Chronic impairment of estrogen synthesis has been set forth as a possible explanation to this observed lower risk.55;57 The reason why the lower risk is not also consistently observed in studies that use stressful life events as measures of stress may be that perceived stress is simply a better measure of prolonged exposure to stress. Alternatively, women who report high levels of stress may be more accurately and adaptively handling their stress and thereby have fewer health consequences. Third, some of the studies lacked control for important risk factors for breast cancer, which may have biased the results. Several of the registry-linkage studies on stressful life events and breast cancer were not appropriately adjusted for potential confounding. However, in order to affect the results, these risk factors should also be associated with the stressful life event in question. For example, for an unmeasured covariate to have distorted the association between the loss of a child and breast cancer, this major life event would have to be associated with family history of breast cancer, 31 age at menarche and menopause, age at first birth, etc. Although there may be some associations between the major life events and other risk factors, they are not expected to be substantial enough to seriously have distorted the risk estimates in the large linkage studies. Future studies that combine information from registries with information from survey designs may be needed to fully adjust for potential confounding. Also, some of the studies may have adjusted for factors on the causal path from stress to breast cancer. An example; stress may effect alcohol consumption that may in turn affect the risk of breast cancer. Adjustment for alcohol consumption would lead to an underestimate of the effect of stress on breast cancer mediated through alcohol. This could possibly explain the null findings in some of the studies, although not in the registry linkage studies where no such adjustments were performed. Fourth, even after restricting the review to only include prospective studies, the methodology of the included studies might still vary to a degree that could affect the results. Some of the studies used self-reported measures of breast cancer incidence, while others identified breast cancer cases in validated cancer registries. This may have led to differences in outcome assessment. Publication bias might also have affected the results. Small studies with positive results are often more likely to be published than those with negative results, either because of the reluctance to publish studies with no significant results or because authors choose not to submit such studies. Interestingly, only the two studies with the smallest number of participants found a markedly higher risk of breast cancer incidence associated with stress.54;58 This could indicate some degree of publication bias in the included studies. Further, the studies on breast cancer relapse had insufficient sample sizes to warrant firm conclusions. These differences in methodology might, to some extent, explain the heterogeneity in the results. Finally, the length of follow-up differed between the studies; most studies only included a baseline measure of stress, which may have changed over time in a manner that is most likely independent of subsequent incidence of cancer. Thus, non-differential exposure misclassification may have reduced the ability to detect a potential relation between the different measures of stress and risk of breast cancer in studies with a long follow-up. 32 Causal pathways The human organism is in a state of dynamic equilibrium, whereby stability is maintained throughout change, also known as allostasis.17 Allostasis is central to the stress response and to survival in general in that it is the process by which we actively adjust ourselves to predictable and unpredictable events, and thereby preserve the homeostatic stability of vital body systems.17 The stress response is initiated when external and internal forces, the stressors, challenge this allostasis. The sympathetic nervous system and the Hypothalamic-Pituitary-Adrenal (HPA) axis are the main mediators of the stress response.13 We use this system to adapt ourselves to unexpected and stressful situations. This review does not support stress as an important etiologic factor in breast cancer etiology, which may be due to the organism’s amazing ability to re-establish allostasis. Stress hormones released by the HPA-axis might suppress the synthesis and metabolism of estrogen, which is an important risk factor for breast cancer. This biological mechanism may be consistent with the lower risk of breast cancer incidence associated with different measures of stress observed in several of the included studies. Conversely, persistent activation of the HPA-axis and some of the mediators released by the HPA-axis, such as corticosteroids and catecholamines, also seem capable of suppressing immune function resulting in a reduced ability to recognize and destroy neoplastic cell growth.2;14 This mechanism may increase the risk of breast cancer and thereby oppose the stress-induced suppression of estrogens. Thus, stress may affect the risk of breast cancer directly through biological processes or indirectly by affecting health-related behavior, and the relation between stress and breast cancer is probably a result of a complex system with different mechanisms working in opposite directions. The importance of each mechanism probably depends on the population and the type of breast cancer in question. Conclusion In spite of the heterogeneity of the results, the combined evidence from prospective studies does not indicate that stress is an important risk factor for breast cancer incidence. It is still unclear if stress prior to a breast cancer diagnosis may render women 33 more susceptible to an accelerated progression of the disease. Studies with more homogenous measures of stress, more thorough adjustments for confounding, and larger studies on stress and breast cancer relapse are required to pursue this issue further. 34 Table 3-1. Summary of prospective studies on stress and breast cancer incidence First author, Population and follow-up N (no. of cases) Exposure year 32 Ewertz M Linkage study based on 1782 cases and 1738 Divorce or loss of a Denmark, national registries. Nested controls husband by death 1986 case-control study. Age<70 yrs Jacobs JR41 The Baltimore 1213 (29) Parental death in USA, 2000 Epidemiologic Catchment childhood Area Study. Mean age 43. Follow-up=15 yrs Johansen C30 Linkage study based on 5716 exposed women Cancer in a child Denmark, national registries. Mean (198 cases in the 1997 age? Follow-up=25 yrs exposed) Kvikstad A28 Linkage study based on 4547 cases and 34 470 Divorce or loss of a Norway, 1994 national registries. Nested controls husband by death case-control study. Mean age? Kvikstad A29 Linkage study based on 4340 cases and 29 750 Death of a child Norway, 1996 national registries. Nested controls case-control study. Mean age? Levav I31 Linkage study based on 3299 exposed women Loss of an adult son Israel, 2000 national registries. Mean (96 cases in the in war or accident age 55 yrs. Follow-up=20 exposed) yrs 27 Li J Linkage study based on 168 138 (?) Death of a child Denmark, national registries. Mean 2002 age 33. Mean followup=10.5 yrs Lillberg K42 The Finnish Twin Cohort 10 808 (180) Stressful life events Finland, 2003 Study. Mean age 41. Follow-up=15 yrs Confounding* Results No adjustment Divorce: OR=0.9 (0.7-1.2) Loss of husband by death: OR=0.8 (0.7-1.0) No association between death of father and breast cancer. Death of a mother in childhood: HR=2.13 (1.02-4.55) Mothers whose child has cancer compared to background population: SIR=1.0 (0.9-1.1) Divorce: OR=0.84 (0.76-0.92) Loss of a husband by death: OR=1.15 (0.95-1.39) Mothers who lost a child: OR=0.93 (0.79-1.10) Age, AL, HIS, SES No adjustment BIR, PAR Age, PAR Age Age, PAR, SES Age, AL, PA, BMI, BIR, PAR, SES Loss of a son in war: OR=1.32 (0.86-2.02) Loss of a son in accident: OR=1.16 (0.92-1.46) Mothers who lost a child: HR=1.10 (0.89-1.35) One-event increase in stressful life events: HR=1.07 (1.00-1.15 One-event increase in major events: HR=1.35 (1.09-1.67) Work-related stress Schernhammer Nurses’ Health Study. 37 562 (1030) ES 36 Mean age 55. Follow-up=8 USA, 2004 yrs Job strain Age, AL, PA, BMI, HT, HIS, BD, MA, MO, BIR, PAR, SES High strain jobs versus low strain jobs RR= 0.87 (0.73-1.04) Stress during the last five years Age, AL, BMI, HIS, MA, MO, BIR, PAR, SES Stress versus no stress: HR=2.0 (1.1-3.5) Perceived global stress Helgesson Ö37 The Prospective Population 1462 (47) Sweden, 2003 Study of Women in Gothenburg. Age 38-60. Follow-up=24 yrs Kroenke CH38 Nurses’ Health Study. 69 886 (1700) USA, 2004 Aged 46-71 yrs. Followup=8 yrs Lillberg K39 The Finnish Twin Cohort 10 519 (205) Finland, 2001 Study. Aged >18 yrs. Follow-up=21 yrs Nielsen NR40 The Copenhagen City 6 689 (251) Denmark, Heart Study. Mean age 57. 2005 Follow-up=18 yrs Hours and stress of Age, AL, PA, informal caregiving BMI, HT, HIS, BD, MA, MO, BIR, PAR, SES Perceived stress of Age, AL, PA, daily activities BMI, BIR, PAR, SES Perceived stress Age, AL, PA, BMI, HT, MO, PAR, SES ≥15 hrs to child: HR=1.19 (0.871.62) ≥15 hrs to adult: HR=0.87 (0.661.16) Stress (child caregiving): HR=1.12 (0.92-1.36) Stress (adult caregiving): HR=0.82 (0.68-1.00) No stress: 1 (reference) Some stress: 1.11 (0.78-1.57) Severe stress: 0.96 (0.53-1.73) Low stress: 1 (reference) Medium stress: 0.80 (0.62-1.04) High stress: 0.60 (0.37-0.97) * Behavioural factors: alcohol (AL), physical activity (PA), body mass index (BMI), postmenopausal hormone therapy (HT); Biological factors: family history of breast cancer (HIS), benign breast disease (BD), age at menarche (MA), age at menopause (MO), age at first birth (BIR), parity (PAR); Socio-economic status: Some measure of income or education Table 3-2. Summary of prospective studies on stress and breast cancer relapse First author, N (no. of Population and follow-up Exposure Confounding year cases) Barraclough Women recruited from 204 (47) Adverse life Age, axillary lymph node J43 breast cancer clinics in events and social involvement UK, 1992 Southampton and difficulties (not Portsmouth. Mean age 54. incl. own health) Follow-up=3 ½ yrs Brabander BD 44 Belgium, 1999 Women admitted to Free 44 (10) University hospital of Brussels. Median age 58 yrs. Follow-up=3 ½ yrs Chronic stress 1 y prior to diagnosis / Acute stress of diagnosis Forsén A45 Finland, 1991 Breast cancer patients from 86 (41) Finnish hospitals. Mean age 51 yrs. Followup=8 yrs Stressful life events 12 months before diagnosis Graham J46 UK, 2002 NHS breast clinic, London. 202 (54) Age < 60. Follow-up= 5 yrs Severe life experiences 1 yr prior and 5 yrs after diagnosis Hislop TG34 Women referred to the 124 (38) Canada, 1987 AMEC clinic in Vancouver. Age <55 yrs. Follow-up=4 yrs Stress from recent life events prior to diagnosis Results Any severe event or difficulty 1 yr before diagnosis HR=0.43 (0.20-0.93) Any severe event or difficulty during follow-up HR=0.88 (0.48-1.64) Age, tumor size, lymph node Acute stress: N/S in a forward invasion, estrogen receptor stepwise model positive, progesterone receptor Chronic stress: positive, expression of neuOR=1.34 (1.00-1.80) oncogen, expected result of biopsy Age, radiotherapy, axillary 2 SD increase in life events measured node status, chemotherapy, by the Social Readjustment Rating estrogen receptor status, Scale: histological type of tumor, HR=3.48 (1.72-7.04) clinical stage Tumor size, tumor histology, Severe life experiences 1 yr prior axillary nodes with tumor (y/n): infiltration, age, social class, HR=1.01 (0.58-1.74) marital status Severe life experiences 5 yrs after (y/n): HR=0.52 (0.29-0.95) Age, clinical stage, pathologic Stress from recent life events nodal status, histologic grade, Low: HR=1 (reference) estrogen receptor status Medium: HR=0.83 P-value=0.82 High: HR=0.74 Figure 3-1. Flow diagram of the data collection process 258 studies from the search 223 studies did not address the association between stress and breast cancer incidence or breast cancer relapse 35 studies on breast cancer incidence or relapse 23 studies were not prospective in design (8 were limited prospective and 15 were retrospective) 4 studies on stress and breast cancer relapse 8 studies on stress and breast cancer incidence 7 studies were included from scanning the reference lists of included studies and previous reviews 5 studies on stress and breast cancer relapse 14 studies on stress and breast cancer incidence 2 studies described the same association in the same population 5 studies on stress and breast cancer relapse 13 studies on stress and breast cancer incidence 38 IV. Stress and other hormone-dependent cancers: current knowledge Most of the previous studies on stress and hormone-dependent cancers have been confined to the study of breast cancer. Fewer studies have addressed the association between stress and endometrial, colorectal, and prostate cancers. The evidence from these studies will be summarized in this chapter. Only prospective studies were included in the review on stress and breast cancer presented in the previous chapter, but the limited number of studies on stress and other hormone-dependent cancers does not allow for such a restriction. Furthermore, varying study designs may make it even harder to compare the results from the different studies directly. Endometrial cancer Four studies have previously assessed the relation between stress and risk of endometrial cancer (table 4-1).45-48 They have all studied the effect of a major stressful life event and they were solely based on information from national registries. In a nested case-control study based on information from national registries, Kvikstad and colleagues found a slightly lower risk of endometrial cancer among women who had been divorced compared to married women, but they found no association between loss of a husband by death and risk of endometrial cancer.45 In another registry-linkage study, Kvikstad and Vatten found no evidence of a higher risk of endometrial cancer among women who had lost a child compared to those who had not experienced this severe stressor.46 Johansen and Olsen hypothesized that learning that one’s child has cancer is one of the most severe stressors a parent can be exposed to.47 They compared the observed incidence rate of endometrial cancer and a range of other malignancies among women who had a child with cancer to the incidence rate that would have been expected based on the incidence of this malignancy in the general population, and they found no association between having a child with cancer and risk of endometrial cancer. In a study from Israel also based on information from national registries, Levav and colleagues found no association between loosing an adult child in war and risk of 39 uterine/ ovarian cancers.48 However, they found more than a doubling of the risk of these cancers among women who had lost an adult son in an accident. Uterine and ovarian cancers were combined as one endpoint in this study and we can therefore not sort out the separate effect of loss of an adult son on risk of endometrial cancer from this study. The advantages of using registry-based information are that the exposure can be assessed independently of the cancer diagnosis and that a large number of cases can be included into the study. The disadvantage is that each stressful life event is seen as a separate entity and that stress arising in daily life is not incorporated. This may lead to serious underestimation of the total stress burden experienced by the individual. Also, the individual differences in reactions to the life event are not taken into account. The aim of the study presented in chapter five of this dissertation is to assess a potential relation between more stress in daily life and risk of endometrial cancer. Colorectal cancer Eight previous studies have assessed the association between stress and colorectal cancer (Table 4-2).45;46;48;65-69 The study designs included four case-control studies,65-68 three registry linkage studies,45;46;48 and one cohort study.69 Different measures of stress were applied in these studies including measures of perceived stress and stressful life events. The case-control and the registry-linkage studies had colorectal cancer incidence as the endpoint, while the cohort study assessed the risk of colorectal cancer mortality. Five of the studies addressed the association between stressful life events and colorectal cancer.45;46;48;65;67 In a population-based case-control study from Sweden, Courtney et al. addressed the association between stressful life events and risk of colorectal cancer in 569 cases and 510 controls.67 Stressful life events were assessed retrospectively during the last 10 years in both cases and controls. They found a relatively strong positive association between work-related problems or changes in residence and risk of colorectal cancer. They also found a moderately increased risk of colorectal cancer associated with death of a spouse, but they found no clear associations with other stressful life events such as death of a child, divorce, or financial problems. Kune and colleagues found recent life changes within five years prior to diagnosis as 40 well as the reporting of being extremely upset by these changes to be important risk factors for colorectal cancer in a population-based case-control study including 715 cases and 727 age- and sex-matched controls.65The associations were strongest for men as opposed to women and for colon cancer as opposed to rectal cancer. In contrast to these results, Kvikstad et al. found a lower risk of colorectal cancer among divorced women, but no association between the loss of a husband by death and risk of colorectal cancer in a large registry-linkage study.45 Through the national registries, they obtained information on marital status prior to diagnosis. In another registry-linkage study, Kvikstad and Vatten report a slightly lower risk of colorectal cancer among women who have lost a child.46 In a study from Israel also based on information from national registries, Levav and colleagues found some evidence of a slightly higher risk of colorectal cancer among fathers and mothers who had lost a son in an accident, but there were few cases and the confidence intervals were broad.48 In the same study no association was found between the loss of a son in war and colorectal cancer. In sum, the evidence from studies on stressful life events and risk of colorectal cancer are conflicting. In general, stress from acute severe stressors such as the loss of a child or a spouse did not seem to markedly increase the risk of colorectal cancer, while more prolonged stress such as major work-related problems may be associated with higher risk of colorectal cancer. Two studies have addressed the potential relation between work-related stress and risk of colorectal cancer.66;68 In a population-based case-control study that included 744 matched case-control pairs, Courtney et al. assessed the association between perceived job control, demands, and social support five and 30 years before the interview and the risk of colon cancer.66 Control, demands, and social support in jobs held one year prior to the interview were imputed for occupational categories. They found a weak positive association between a self-reported lack of job control and risk of colon cancer, but a similar association was not found for the imputed measures of job control. No consistent association was found between job demands and colon cancer, and lower levels of social support seemed to be linked to a lower risk of colon cancer. Spiegelman and colleagues used data from the Third National Cancer Survey to assess the relation between workrelated stress and risk of colorectal cancer.68 They included 343 male cases and compared them to 643 controls with cancers they assumed to be unrelated to 41 occupational exposures. They had no direct measure of stress and the measure of job stress was therefore imputed from occupational categories. Based on this measure of stress, they found men with high demands and low control to have more than twice the risk of colorectal cancer than those with low demands and high control. Their use of other cancer cases as controls as well as their imputation of individual measures of stress from occupational categories that did not account for the within-occupation variance in stress exposure may be problematic. Thus, there is no clear evidence of an association between work-related stress and colorectal cancer, and prospective studies with individual measures of work-related stress are needed to pursue this issue further. Only one previous study has assessed the association between perceived stress in daily life and colorectal cancer.69 In a cohort of 78,007 Japanese men and women, Kojima and colleagues found women who reported high levels of daily stress to have a 1.64-fold higher risk of colon cancer mortality, while they reported no association between stress and colorectal cancer in men or stress and rectal cancer in women. Colorectal cancer mortality was assessed as the endpoint, which does not allow for a distinction between etiologic and prognostic factors. The aim of the cohort study presented in chapter six of this dissertation is to address the association between perceived stress of daily life and incidence of colorectal cancer. Prostate cancer Only one study has previously addressed the association between stress and prostate cancer.47 In a registry-linkage study from Denmark, Johansen and Olsen assessed the relative incidence of a range of malignancies, one of them being prostate cancer, among men who had been exposed to the severe stressor of having a child with cancer compared to the incidence rate in the general population.47 They found no evidence of an association between this severe stressor and risk of prostate cancer (Standardized incidence rate =1.0; 95% CI: 0.8-1.2). Although cancer in a child is a severe stressor, it may still just represent one dimension of the total stress burden experienced by the individual. Fortunately, having a child with cancer is a rare event and the association between the much more prevalent experience of stress in daily life and risk of prostate 42 cancer remains to be studied. The study presented in chapter seven of this dissertation is aimed at addressing this question. 43 Table 4-1. Summary of previous studies on stress and endometrial cancer First author, year Study design Population Measure of stress Endpoints Johansen C47 Denmark, 1997 Registry-linkage 5716 exposed women Cancer in a child study with 25 years compared to the of follow-up general population Kvikstad A45 Norway, 1994 Registry-linkage 576 cases and 34 470 study. Nested case- controls control study Divorce and loss of Endometrial a husband by death cancer incidence Kvikstad A46 Norway, 1996 Registry-linkage study. Nested casecontrol study Registry-linkage study with 10.5 years of follow-up Death of a child Levav I48 Israel, 2000 488 cases and 29 750 controls 3299 exposed and 524 916 unexposed Endometrial cancer incidence (N=46) Adjustment Age Age, parity Endometrial Age, parity cancer incidence Loss of an adult son Uterine/ Age, year of in war or accident ovarian cancer birth, region incidence of origin, (N=8235) period of immigration Results Mothers whose child had cancer compared with the background population: SIR= 1.0 (95% CI: 0.71.3) Divorce: OR= 0.64 (95% CI: 0.480.85) Loss of a husband by death: OR= 1.08 (95% CI: 0.661.80) Death of a child: OR= 0.84 (95% CI: 0.511.39) Loss of son in war: OR= 1.16 (95% CI: 0.821.64) Loss of son in accident: OR= 2.19 (95% CI: 1.323.63) Table 4-2. Summary of previous studies on stress and colorectal cancer First author, Study Population Measure of stress Endpoints year design 67 Courtney JG Population- 569 cases and 510 Stressful life Colorectal cancer Sweden, 1993 based case- controls. Controls events within the incidence control study were selected from last 10 years the general population Courtney JG66 USA, 1996 Kojima M69 Japan, 2005 Adjustment Results Age, sex, diet, body Death of a spouse: mass index, physical OR=1.5 (95% CI: 1.1-2.4) activity, alcohol Serious occupational problems: OR=5.3 (95% CI: 2.2-13.0) Change of residence of >200 km: OR=3.2 (95% CI: 1.1-8.3) No association with death of a child/friend/relative, divorce, unemployment, or financial problems Population- 744 matched case- Perceived job Colon cancer Age, sex, Low vs. high control based case- control pairs. demand, control, incidence neighborhood, OR=1.3 (95% CI: 1.0-1.6) control study Controls were and social support physical activity, High vs. low demands selected from the family history, OR=1.0 (95% CI: 0.8-1.3 general population employment status, Low vs. high social support weight, alcohol, diet, OR=0.6 (95% CI: 0.4-1.0) pregnancies No sex differences Cohort study 78,007 men and Perceived stress of Colorectal cancer Body mass index, High vs. low stress with an women participating daily life deaths (N=320) family history, Colon cancer: average of in the Japan smoking, alcohol, Men: HR=0.95 (95% CI: 0.559.6 years of Collaborative sleep duration, intake 1.66) follow-up Cohort Study for of green leafy Women: HR=1.63 (95 % CI: Evaluation of vegetables, walking, 1.00-2.64) Cancer Risk constipation, Rectal cancer: education, marital Men: HR=0.96 (95% CI: 0.53status, children, 1.73) employment Women: HR=1.27 (95% CI: 0.58-2.81) Continues on next page Kune S 65 Population- 715 cases and 727 Stressful life Australia, 1991 based case- controls. Controls events with the control study were selected from last 5 years the general population Colorectal cancer Sex, age, diet, beer, incidence family history, age at birth of first child, number of children Kvikstad A45 Registry918 cases and 34 Norway, 1994 linkage 470 controls. study. Females only Nested casecontrol study Kvikstad A46 Registry861 cases and 29 Norway, 1996 linkage 750 controls. study. Females only Nested casecontrol study Levav I48 Registry6284 exposed and Israel, 2000 linkage 1 019 255 study with unexposed 10.5 years of follow-up Divorce and loss of a husband by death Colorectal cancer Age incidence Death of a child Colorectal cancer Age incidence Loss of an adult son in war or accident Loss of son in war Colorectal cancer Age, year of birth, Fathers: OR=1.06 (0.81-1.38) incidence (N=22 region of origin, 158) period of immigration Mothers: OR= 1.12 (0.82-1.54) Spiegelman D68 Case-control 343 cases and 626 USA, 1985 study controls. Controls were cancer cases with no confirmed occupational association. Males only Major family illness: OR=1.22 (95% CI: 1.01-1.47) Major family problem: OR=1.32 (95% CI: 1.08-1.61) Major work problem: OR=2.54 (95% CI: 1.74-3.72) Extremely upsetting event: OR=1.68 (95% CI: 1.30-2.17) Divorce: OR= 0.66 (95% CI: 0.53-0.83) Loss of a husband by death: OR= 1.01 (95% CI: 0.68-1.51) Death of a child: OR= 0.70 (95% CI: 0.48-1.02) Loss of son in accident Fathers: OR= 1.39 (0.90-2.16) Mothers: OR= 1.62 (0.85-1.54) High demands and low control vs. low demands and high control: Imputed measure Colorectal cancer Age, race, marital of job stress based incidence status, region, on occupation ponderosity, socioeconomic status, diet OR= 2.51 (P-value:0.03) V. Self-reported stress and risk of breast cancer: a prospective cohort study Introduction Breast cancer is the most common cancer in women in terms of both incidence and prevalence.23 It is a hormone-dependent disease with a clear positive relation to high endogenous concentrations of estrogen.70 The role of stress in the etiology of breast cancer has been an area of emerging interest, partly because stress seem to affect the hormonal system and especially estrogen synthesis.10-12;71 A potential relation between stress and risk of breast cancer has been examined in studies with different designs and with conflicting results.15;41;47;52;54-56;59 Different measures of stress, as applied in these studies, may well have different physiological and psychological impacts. Death of a spouse or near relative is a major acute stressor, whereas stress experienced in daily life is more moderate and chronic in nature. The risk of breast cancer associated with the acute stress of major life events has been assessed in several studies,15;41;47;59 but less attention has been given to the effect of perceived daily stress.54-56 Prolonged low key stress in everyday life results in a persistent activation of stress hormones, which may impair estrogen synthesis,72 and may thereby be related to a lower risk of breast cancer. Everyday stress may also indirectly affect the risk of breast cancer through changes in health-related behavior. The objective of this study is to explore the impact of everyday stress on the long term risk of first time incidence of primary breast cancer among 6,689 women prospectively followed up for 18 years. Methods Study population Information from the second examination of the prospective Copenhagen City Heart Study in 1981-83 was used to address the objective. The second examination of the study included 7,018 women and we excluded women with breast cancer before baseline 47 (n=120) or lacking information on stress or other covariates (n=209), leaving 6,689 women. Twenty-six (<0.1%) women were lost to follow-up. Perceived stress was assessed at baseline and included two questions on stress intensity and stress frequency. A seven-point stress score was created based on these two questions. Please refer to the material and methods chapter for a detailed description of the Copenhagen City Heart study and the measurement of perceived stress. Covariates We considered the following variables as potential confounders for the analyses based assumptions shown in the causal diagram in figure 5-1: current oral contraceptive use (yes/no), other hormone therapy (yes/no), menopause at baseline (yes/no), body mass index (continuous), number of children (0, 1-2, ≥3), physical activity in leisure time (low, medium, high), alcohol consumption (0 drinks/week, 1-14 drinks/week, >14 drinks/week), and education (<8 years, 8-11 years, ≥12 years). All variables were measured at baseline. Early menarche, age at first birth, family history of breast cancer, ionizing radiation, benign breast cancer, mammographically dense breasts and mutations in BRCA1/2 or p53 and contraceptive use were not expected to be strongly associated with stress and were therefore also not included in the analyses. Follow-up We followed participants from the date of the second examination until the date of first diagnosis of primary breast cancer (n=251), death (n=2,224), emigration (n=26), or the end of follow-up on 31 December 1999 (n=4,188). We used the civil registry number, which is unique to every Danish citizen, to identify primary breast cancer events through linkage to the Danish National Cancer Registry, which contains data on all cancer diagnoses in Denmark. We used ICD-7 codes170.0-170.5, 470.0-470.5, and 870.0-870.2 to identify cases of primary invasive breast cancer. We followed the vital status of the study population in the Central Death Registry. Information on diagnosis of breast cancer was updated until 1999, making it possible to follow the participants from the second examination for 16-18 years for a primary diagnosis of breast cancer. 48 Statistical methods We used Cox regression models (SAS/STAT software version 8.2) to analyse data with age as the time scale. All included variables met the assumption of proportional hazards. Initially, we estimated the age adjusted hazard ratio of primary breast cancer associated with stress intensity, stress frequency, and stress score (continuous and in categories of low, medium, and high stress). Subsequently, we fitted a multivariate Cox regression model to adjust for potential confounding from baseline covariates. We used trend analyses to assess dose-response relations between stress and breast cancer. A variable was only included in the analyses as a continuous variable if a linear trend could be confirmed in a likelihood ratio test. To estimate the effect of prolonged follow-up, we assessed the association in the first and last nine years of follow-up. We also used these analyses to address if the introduction of mammography screening for women aged 5069 years in 1991 in the Copenhagen area affected the results. Finally, we did subgroup analyses to assess potential effect modification of the ratio measure. Results Baseline characteristics Mean age at baseline was 57 (range 21-91) years. Ten percent of the women reported high levels of stress (table 5-1). Mean age, number of children, and body mass index seemed to be similar at the different stress levels. A lower proportion of highly stressed women than less stressed women were premenopausal and used oral contraceptives. A higher proportion of the women in the high stress group received hormone therapy, had low education and high alcohol intake, and was physically inactive in their leisure time compared with women with lower levels of stress. During follow-up, 251 cases of primary breast cancer occurred. A higher percentage of women in the high stress group (n=261, 39.3 %) than in the medium (n=972, 30.4 %) or low stress groups (n= 991, 35.1 %) died during follow-up. 49 Stress intensity, stress frequency, and risk of breast cancer Seven percent of the women reported high stress intensity, and 10 % reported high frequency of stress. The adjusted hazard ratio of primary breast cancer seemed to be inversely associated with both stress intensity (test for trend, P=0.02) and stress frequency (test for trend, P=0.06) (table 5-2). Stress score and risk of breast cancer After adjustment for potential confounders, an 8% lower risk (hazard ratio 0.92, 95% confidence interval 0.85 to 0.99) of primary breast cancer occurred for each increase in stress level on the six-point stress scale (table 5-3). Higher stress was inversely associated with incidence of primary breast cancer (test for trend, P=0.02), and high stress was associated with a hazard ratio of 0.60 (0.37 to 0.97) for breast cancer compared with low stress. Subgroup analyses One hundred and fourteen first time primary breast cancers occurred in the first nine years of follow-up and 137 cases occurred in the last nine years. The relative effect of stress on incidence of breast cancer seemed to be similar in the two periods of follow-up (table 5-4). Sixteen percent (n=1045) of the women were receiving hormone therapy at baseline, and the effect of stress on risk of breast cancer seemed to be mainly confined to these women (table 5-5). The P-value for statistical interaction was 0.09. Among women receiving hormone therapy, the hazard ratio for primary breast cancer was 0.83 (0.72 to 0.97) for each increase in stress level on the six-point stress scale. No notable effect modification occurred in subgroups of menopausal status, physical activity, alcohol consumption, oral contraceptive use, education, and number of children (data not shown). Discussion Among 6,689 women followed-up for an average of 18 years, higher self-reported everyday stress was associated with lower risk of breast cancer. Our results are similar to 50 those from the Nurses’ Health Study, in which Kroenke et al found self-reported stress from adult caregiving to be associated with lower incidence of breast cancer.55 However, our results contrast with the results of a Finnish cohort study, which found no association between stress of daily activities and breast cancer, and a Swedish study in which severe mental stress was associated with higher incidence of breast cancer.54;56 Some of the discrepancy may be explained by the fact that different measures of stress were applied and that the Finnish and Swedish studies included all incident cases of breast cancer, whereas we confined our analyses to first time incidence of primary breast cancer. Apart from these studies, the main focus has been on stressful life events. However, the nature of sustained everyday stress is different from stressful life events, and an increased risk of breast cancer associated with stressful life events is not necessarily in contrast with a lower risk of breast cancer associated with daily stress. Strengths and weaknesses The prospective design of the Copenhagen City Heart Study ensured temporality between stress and incidence of breast cancer. The cohort is a large random sample of the general population of Copenhagen. Furthermore, linkage of civil registry numbers to nationwide population based registers enabled identification of virtually all cases of breast cancer and allowed for nearly complete long-term follow-up. Information on several important risk factors for breast cancer, such as family history of breast cancer, age at menarche, and age at first full time pregnancy, was not obtained. However, to confound the results these factors should also be related to stress. We cannot exclude that having experienced breast cancer in a near family member may act as a stressor and thereby lead to higher levels of stress, which would result in a spurious positive association between stress and breast cancer. This is opposite to the inverse association we observe in our study and therefore cannot explain it. Late age of menarche would have to be relatively strongly related to high stress in order to explain our results, but we know of no empirical evidence or biological explanation that would support such an expectation. Late age at first full term pregnancy is a well-established risk factor for breast cancer. More women with high stress had low education (55% versus 47% in the low stress group), and women with low education tend to have their first pregnancy earlier than other women. Thus, we cannot exclude that some of the 51 observed inverse association is due to uncontrolled confounding by age at first pregnancy. However, adjustment for number of children, which also tends to be correlated with age at first pregnancy, as well as adjustment for education only changed the estimates slightly. Mammographic screening for women aged 50-69 years was introduced in Copenhagen in 1991.73;74 Stressed women may have been less likely to participate in the screening program and may therefore have had invasive breast cancer diagnosed at a later stage. According to Olsen et al., an expected increase in incidence of invasive breast cancer occurred after the first screening round, but the incidence dropped to the pre-screening level in the following screening rounds, indicating that over-diagnosis of breast cancer was not a major problem.74 Furthermore, screening only took place in women aged 50-69 years and in the last nine years of our study. The effect estimates were similar in the first and last nine years of follow-up (table 5-4), so bias by screening is unlikely to explain our results. How stress is to be defined and measured remains a point of debate. So far the literature has focused on external stressors with less emphasis given to how they are perceived by the individual. However, each person has different capacity and ways of coping with stressful situations, and the same external stressor may result in different levels of perceived stress. A measure of perceived stress will therefore provide a better measure of the actual level of stress experienced by the individual as opposed to a count of potential stressful situations defined by the researcher. More women in the high stress group (39.3%) than in the medium (30.4%) and low stress (35.1%) groups died during follow-up. Although this indicates no systematic difference, it may raise concern about how censoring has influenced the results. We assumed, in the statistical model, that censoring was independent of breast cancer risk within each stratum of stress. Competing causes, such as death from cardiovascular disease, could, however, be associated with risk of breast cancer within strata of stress owing to other common risk factors. Some of the common risk factors, such as socioeconomic status, have opposite effects on the two diseases, whereas other risk factors, such as low physical activity, increase the risk of both diseases.70;75 On average, we would expect the bias to level out and the average risk of breast cancer among women not censored to be similar to what it would have been if no such censoring had 52 occurred. Thus, although our results may have been influenced by bias from nonindependent censoring, we find it unlikely that this could completely explain them. Possible causal pathways between perceived stress and breast cancer In most cases, the physiological effects of acute stressors are reversible owing to the amazing ability of the human organism to re-establish allostasis. The problems arise when the stress response becomes chronic and results in permanent disturbances. In a normally functioning female reproductive system, the hypothalamus releases luteinizing hormone releasing hormone, which stimulates the pituitary gland to release luteinizing hormone and follicle stimulation hormone. Luteinizing hormone stimulates the ovaries to synthesize estrogens, whereas follicle stimulation hormone stimulates the ovaries to release eggs. This is called the hypothalamic-pituitary-gonadal axis. Stress can affect the signals of this axis by activating the hypothalamic-pituitary-adrenal axis and the sympathetic nervous system. Several studies in mammals have found that activation of the hypothalamic-pituitary-adrenal axis inhibits the function of the hypothalamicpituitary-gonadal axis and thereby decreases estrogen synthesis, but data on humans are sparse.10-12;72 Still, in a recent study in which caregiving was used as an indicator of chronic stress, significantly lower concentrations of bioavailable estradiol were found among female caregivers than among non-caregivers.55 In essence, stress induced suppression of estrogen secretion could explain a reduced risk of breast cancer. An imbalance in the allostatic concentration of reproductive hormones may also result in other reproductive disturbances and mood swings and initiate depression in susceptible women.76;77 Some women may be more sensitive to hormonal disturbances and therefore be more likely to receive hormone therapy to lessen their symptoms. Hormone sensitive women are more likely to be susceptible to stress induced changes in estrogen synthesis, which could explain why stress mainly seems to be associated with lower risk of breast cancer among women receiving hormone therapy. Conclusions It is biologically plausible that the lower risk of breast cancer associated with stress observed in this long term prospective cohort study might be due to stress induced 53 imbalances in normal concentrations of estrogens. However, stress induced disturbances of allostasis cannot be considered a healthy response, and prolonged stress may have harmful effects on a range of other diseases, especially cardiovascular diseases. 54 Table 5-1. Baseline characteristics of women who participated in the second examination of the Copenhagen City heart study in 1981-3. Values are numbers (percentages) unless stated otherwise Stress* Study population Participants Low Medium High 6689 2823 (42) 3201 (48) 665 (10) 57 (12) 58 (12) 55 (12) 58 (12) 1766 (26) 672 (24) 978 (31) 116 (17) 250 (4) 99 (4) 143 (4) 8 (1) 1045 (16) 328 (12) 568 (18) 149 (22) 1.6 (1.3) 1.6 (1.4) 1.6 (1.3) 1.7 (1.4) 25 (5) 25 (5) 25 (4) 25 (5) 449 (7) 152 (5) 230 (7) 67 (10) < 8 years of formal education 3096 (46) 1340 (47) 1388 (43) 368 (55) Physically inactive 1179 (18) 442 (16) 520 (16) 217 (33) Mean (SD) age (years) Premenopausal Current oral contraceptive users Other hormone therapy Mean (SD) No. of children 2 Mean (SD) body mass index (kg/m ) High alcohol consumption *Participants reported stress intensity and frequency on a standard questionnaire with four multiple-choice categories (0-3 points) for each stress measure. The scores of the two questions were added and combined into a continuous stress score from 0 to 6. This stress score was categorized into low (0-1 points), medium (2-4 points), and high (5-6 points). Table 5-2. Incidence and hazard ratio of primary breast cancer associated with intensity and frequency of stress among 6689 Danish women participating in the second examination of the Copenhagen City heart study in 1981-3 No. of breast Incidence per 100 000 Age adjusted hazard ratio Multi-adjusted* hazard ratio cancers person years (95% CI) (95% CI) None (n=2214) 96 292 1 (reference) 1 (reference) Light (n=2608) 97 243 0.89 (0.67 to 1.18) 0.85 (0.64 to 1.13) Moderate (n=1384) 44 210 0.74 (0.52 to 1.05) 0.68 (0.47 to 0.98) High (n=483) 14 203 0.65 (0.37 to 1.13) 0.61 (0.35 to 1.07) 0.04 0.02 Stress intensity P value for trend Stress frequency Never (n=2854) 119 282 1 (reference) 1 (reference) Monthly (n=1994) 71 228 0.90 (0.67 to 1.20) 0.85 (0.64 to 1.15) Weekly (n=1168) 40 227 0.83 (0.58 to 1.19) 0.78 (0.55 to 1.13) Daily (n=673) 21 213 0.70 (0.44 to 1.11) 0.67 (0.42 to 1.07) 0.09 0.06 P value for trend *Adjusted for age, current oral contraceptives use, other hormone therapy, menopausal status, number of children, body mass index, alcohol consumption, physical activity in leisure time, and education. Table 5-3. Incidence and hazard ratio of primary breast cancer associated with stress score among 6689 Danish women participating in the Copenhagen City heart study in 1981-3 Stress score No. of breast Incidence per Age adjusted hazard ratio Multi-adjusted* hazard ratio cancers 100 000 person years (95% CI) (95% CI) 0.93 (0.87 to 1.00) 0.92 (0.85 to 0.99) Continuous Categorised: Low stress (n=2823) 120 285 1 (reference) 1 (reference) Medium stress (n=3201) 112 229 0.84 (0.65 to 1.09) 0.80 (0.62 to 1.04) High stress (n=665) 19 194 0.63 (0.39 to 1.02) 0.60 (0.37 to 0.97) 0.04 0.02 P value for trend *Adjusted for age, current oral contraceptives use, other hormone therapy, menopausal status, number of children, body mass index, alcohol consumption, physical activity in leisure time, and education. Table 5-4. Incidence and hazard ratio of primary breast cancer associated with categorized stress score among 6689 Danish women participating in the Copenhagen City heart study in 1981-3, according to time period of follow-up First nine years of follow-up Last nine years of follow-up Incidence per 100 000 person years Multiple adjusted hazard ratio* (95% CI) Incidence per 100 000 person years Multiple adjusted hazard ratio* (95% CI) Low stress 239 1 (reference) 339 1 (reference) Medium stress 190 0.82 (0.55 to 1.21) 272 0.78 (0.55 to 1.11) High stress 167 0.63 (0.31 to 1.28) 227 0.60 (0.30 to 1.16) P-value for trend 0.14 0.07 *Adjusted for age, current oral contraceptives use, other hormone therapy, menopausal status, number of children, body mass index, alcohol consumption, physical activity in leisure time, and education. Table 5-5. Hazard ratio of primary breast cancer associated with stress score among 6689 Danish women participating in the Copenhagen City heart study in 1981-3, in subgroups of hormone therapy Stress score No hormone therapy (n=5644) Hormone therapy (n=1045) Age adjusted hazard ratio Multi-adjusted* hazard ratio (95% CI) (95% CI) 184 0.95 (0.87 to 1.04) 0.96 (0.88 to 1.05) 67 0.82 (0.71 to 0.95) 0.83 (0.72 to 0.97) No. of breast cancers *Adjusted for age, current oral contraceptive use, menopausal status, number of children, body mass index, alcohol consumption, physical activity in leisure time, and education. Figure 5-1. Causal diagram of the relation between perceived stress and risk of breast cancer Oral contraceptive use Family history of breast cancer Ionizing radiation Menopause Time of menarche Benign breast cancer Estrogens Postmenopausal hormone theraphy Perceived stress Breast cancer Alcohol Body mass index Physical activity Mammographically dense breasts Mutations in p53 Number of children Age at first birth Mutations in BRCA1/2 Socio-economic status VI. Self-reported stress and risk of endometrial cancer: a prospective cohort study Introduction Endometrial cancer is the fourth most common cancer in Europe and North America.78 Some of the best known risk factors for endometrial cancer are endogenous concentrations of estrogens and exposure to unopposed exogenous estrogens.79 Psychological stress may play a role in endometrial carcinogenesis by affecting synthesis and metabolism of estrogens as well as by decreasing the sensitivity of the uterus toward estrogen stimulation.10;80;81 Estradiol, which is the most active estrogen metabolite, can bind to the estrogen receptors, activate gene expression, and thereby increase cell proliferation. By this pathway, estrogens may promote growth of already initiated cells. There is also some evidence that estrogens can be metabolized into more reactive intermediates that can, in turn, lead to structural changes in the DNA and thereby act as initiating agents.82 Chronic stress from caregiving has been associated with lower levels of bioavailable estradiol among female nurses.55 If high levels of stress hormones indeed result in lower levels of endogenous estradiol, we would expect psychological stress to lower the incidence of endometrial cancer. However, a possible relation between stress and endometrial cancer remains to be studied. Stress may affect the risk of endometrial and breast cancer through the same hormonal pathway, namely, by impairment of estrogen synthesis. For breast cancer risk, several large registry-linkage studies have found no association with major life events such as death of a spouse, divorce, or having a child with cancer,44-47;49 whereas some cohort studies have suggested a lower risk in women exposed to chronic stress in everyday life.53;55;57 In contrast, a higher risk of breast cancer associated with measures of stress has also been reported in some prospective studies.48;54;58;59 However, these studies were smaller than the ones reporting no association or a protective effect. Thus, the combined evidence showed no increased or maybe even a decreased risk of breast cancer associated with different measures of stress. 61 On the other hand, combined evidence from experimental and animal studies have suggested that stress may promote the initiation and progression of cancer by impairment of the elements involved in the immune surveillance of the cell.2 Stress could, by this pathway, potentially increase the risk of cancer. However, cancer is a heterogeneous group of diseases with multiple causes and the involvement of the immune system may therefore depend on the specific cancer in question. We aim to address the association between stress and endometrial cancer in a prospective cohort study including 6,760 women prospectively followed up for 18 years. Our hypothesis is that perceived stress may suppress synthesis of and sensitivity towards estrogens and thereby be related to lower risk of endometrial cancer. Methods Study population Information from the second examination of the prospective Copenhagen City Heart Study in 1981-83 was used to address the objective. The second examination of the study included 7,018 women, and we excluded women with endometrial cancer before baseline (n=38) or with lacking information on stress (n=25) or other covariates (n=195), leaving 6,760 women for the analyses. Perceived stress was assessed at baseline and included two questions on stress intensity and stress frequency. A sevenpoint stress score was created based on these two questions. Please refer to the material and methods chapter for a detailed description of the Copenhagen City Heart study and the measurement of perceived stress. Covariates The confounder identification was based on a causal diagram, in which we included the most current prior knowledge about causal relations between perceived stress, covariates, and endometrial cancer (Figure 6-1).22 This served as a way to visualize and explicitly elaborate the assumptions about the web of causation for the relation between stress and endometrial cancer and to identify variables that must be measured and controlled to obtain unconfounded estimates. According to the diagram the analyses should be adjusted for the following covariates: age (continuous), education (less than 8 62 years, 8-11 years, or 12 or more years), physical activity in leisure time (none or very little activity; 2-4 hours of light activity per week; more than 4 hours of light activity or 2-4 hours of high level activity; and competitional level or more than 4 hours of hard level activity per week), body mass index (continuous), tobacco smoking (never-smoker, ex-smoker, smokers of 1-14 grams per day, 15-24 grams per day, and more than 24 grams per day), diabetes mellitus (yes/no), number of children (0, 1-2, 3 or more), menopause (yes/no), and hormone therapy (yes/no). Alcohol intake, dietary patterns, hypertension, late menopause, and stress hormones are possible intermediates on the pathway from perceived stress and endometrial cancer and will therefore not be included in the analyses. Also, early menarche and contraceptive use are not included in the analyses, as they were not assumed to be strongly associated with stress. Follow-up The women were followed from date of the second examination until date of first diagnosis of primary endometrial cancer (n=72), death (n=2,418), emigration out of Denmark (n=28), or end of follow-up on December 31, 2000 (n=4,242). Thus, fewer than 0.1% were lost to follow-up due to emigration. Using the civil registry number, which is unique to every Danish citizen, primary endometrial cancer events were identified through linkage to the Danish National Cancer Registry, which contains data on all cancer diagnoses in Denmark since 1942. The following ICD7-codes were used to identify primary invasive endometrial cancer cases: 172.0-172.2. The vital status of the women was followed in the Central Death Registry. Information on diagnosis of endometrial cancer has been updated until December 31, 2000, making it possible to follow the women for 17 to 19 years. Statistical methods Data were analyzed by means of Cox proportional hazards models with age as the time variable. Stress intensity, stress frequency, and the combined stress score all met the assumption of proportional hazards. Initially, we estimated the age-adjusted hazard ratio of endometrial cancer associated with stress intensity, stress frequency, and the combined stress score in separate models. By including age as the time variable the estimates were soundly adjusted for age. Secondly, a multivariate Cox proportional 63 hazards model was fitted to adjust for potential confounding from baseline covariates. Trend tests were used to test for linear dose-response trends in the associations between stress and endometrial cancer. Thirdly, statistical interactions between stress and all variables included in the multivariate model were addressed and subgroup analyses were conducted for interactions with a P-value below 0.20. Thus, subgroup analyses were done for hormone therapy (P-value for interaction: 0.08) and body mass index (P-value for interaction: 0.09). No cases of endometrial cancer occurred among underweight women (BMI≤ 18.5) so the analyses were only stratified into normal weight (BMI≤ 25) and overweight (BMI>25). Finally, separate analyses for the first and last nine years of follow-up were done in order to address the possible effect of prolonged follow-up. Results Baseline characteristics The median age at baseline was 58 years ranging from 21 to 91 years. Ten percent of the women reported high levels of stress (table 6-1). Women reporting medium stress were slightly younger and had lower body mass index than women reporting both low and high levels of stress. A higher proportion of women who reported high stress used hormone therapy, were current smokers, were physically inactive in their leisure time, and had low education compared to women with lower levels of stress. A lower proportion of highly stressed women were pre-menopausal and were nulliparous. The percentage with diabetes mellitus was similar at the different stress levels. Perceived stress and incidence of endometrial cancer At the end of follow-up, 72 incident cases of primary endometrial cancer had occurred. The adjusted hazard ratios of endometrial cancer were 0.52 (95 % CI: 0.26-1.04) and 0.72 (95 % CI: 0.28-1.89) for women reporting medium and high stress intensity, respectively, compared to women who did not perceive their life as stressful (table 6-2). In terms of stress frequency, women reporting daily stress had a hazard ratio of 0.40 (95 % CI: 0.14-1.15) compared to women who reported to never experience stress. There appeared to be an inverse dose-response association between stress frequency and risk of endometrial cancer (P-value for trend: 0.07). For the combined stress score, the hazards 64 ratio for endometrial for each unit change on the seven-point stress-scale was 0.88 (95 % CI: 0.76-1.01). Adjustment for potential confounders made the associations slightly stronger. Stratified analyses Fifteen percent (n=1,041) of the women received hormone therapy at baseline and these women had a more than four-fold higher risk of endometrial cancer (HR=4.15; 95 % CI: 2.54 - 6.78) compared to women who did not receive such treatment (data not shown). The association between stress and endometrial cancer was only apparent in women who received hormone therapy (table 6-3). Among women receiving hormone therapy, stress intensity (P-value for trend: 0.06) and stress frequency (P-value for trend: 0.01) were both inversely associated with risk of endometrial cancer, and the hazard ratio for primary endometrial cancer was 0.77 (0.61 to 0.96) for each unit increase in stress level on the seven-point stress scale among these women. No associations between perceived stress and risk of endometrial cancer were observed among women who did not receive hormone therapy. A body mass index of more than 25 kg/m2 was associated with a hazard ratio of 1.27 (95 % CI: 0.79 – 2.04) compared to those with a lower body mass index (data not shown). Among the 4,004 women with a body mass index of less than or equal to 25, there was a clear inverse association between both stress intensity (P-value for trend: 0.01) and stress frequency (P-value for trend: 0.005) and risk of endometrial cancer (Table 6-4). The hazard ratio for primary endometrial cancer was 0.73 (0.58 to 0.91) for each unit increase in stress level on the seven-point stress scale among these women. No clear association between stress and risk of endometrial cancer was noted among women with a BMI above 25 kg/m2. No notable effect modification of the ratio measure occurred in subgroups of menopausal status, physical activity, smoking, education, and number of children (data not shown). Half the cases of endometrial cancer (n=36) occurred in the first nine years of follow-up, while the other 36 cases occurred in the last nine years of follow-up. The association between stress score and endometrial cancer was stronger in the first nine years (HR=0.85, 95 % CI: 0.70-1.04) compared to the last nine years (0.91, 0.75-1.10) of follow-up. 65 Discussion Among the 6,760 women followed up for 18 years, perceived stress appeared to be associated with lower risk of primary endometrial cancer in an inverse dose-response manner. The association was confined to women who received hormone therapy and women of normal weight. No studies have, to our knowledge, previously addressed the relation between stress and endometrial cancer. Both breast and endometrial cancers are estrogen-dependent cancers with a clear relation to endogenous concentrations of estrogens.70;79 If stress in fact affects the risk of endometrial cancer through a hormonal pathway, then our results are in accordance with the lower risk of breast cancer previously observed among stressed women in the same cohort.57 Other studies that have addressed the effect of chronic stress, such as that arising from caregiving or from high strain jobs, have also reported a lower risk of breast cancer among stressed women,53;55 In studies that have applied other measures of stress, such as stressful life events, the results have been more conflicting.15 Strengths and weaknesses The prospective design of the Copenhagen City Heart Study ensured temporality between self-reported stress and incidence of endometrial cancer. Linkage of civil registry numbers to nationwide population-based registers enabled identification of virtually all cases of diagnosed endometrial cancer and allowed for nearly complete long-term follow-up. How stress is defined and measured remains a point of debate. In this study, stress was defined as an individual state of high arousal and displeasure, sometimes referred to as distress.83 By using a measure of perceived stress we accounted for the fact that each individual has different capacities and ways to cope with stressful situations.16 Perceived stress was assessed by combining two questions on stress intensity and stress frequency asked at baseline. Since these two questions are not yet validated against a more extensive scale, such as the Perceived Stress Scale,84 we cannot fully determine the magnitude of the misclassification and an even stronger relation between perceived stress and risk of endometrial may have been blurred. Perceived stress was only assessed 66 at baseline and may have changed over time in a manner that is most likely independent of subsequent development of endometrial cancer. In another Danish cohort study85 that included a question on perceived stress, the majority of the participants (62%) reported the same level of stress in 1994 as in 2000 (Nielsen, unpublished data, 2006). Although this finding indicates that a measure of perceived stress may be relatively stable over time, a considerable minority changed stress levels. The association between stress and risk of endometrial cancer was strongest in the first nine years compared to the last nine years of follow-up. This may either be due to changes in stress levels over time and thus exposure misclassification, or it may be due to a differential effect of stress depending on whether it affects the multistage carcinogenesis early as an initiating agent or later as a promoting agent. Women with breast cancer are often treated with tamoxifen, a non-steroidal compound with partially estrogenic and anti-estrogenic effects depending on the target tissue. Such treatment seems to increase the risk of second primary endometrial cancer among women with breast cancer.79 Women who had breast cancer before baseline may also report higher levels of stress; thus, confounding from history of breast cancer could arise. However, even though 126 of the women included in this study were diagnosed with breast cancer before baseline, none of them developed primary endometrial cancer during follow-up, and confounding from breast cancer history is therefore unlikely. An inverse association between tobacco smoking and risk of endometrial cancer has previously been reported.86 More women with high levels of stress also engaged in tobacco smoking in this study, which may raise concern about residual confounding from smoking. We adjusted our analyses for smoking in five categories and such adjustment only slightly changed the risk estimates. We assumed that hypertension was a possible intermediate on the pathway between stress and endometrial cancer. We also assumed that use of oral contraceptives was unrelated to stress. Neither of these variables was therefore included in the statistical analyses. However, it could be argued that these factors may have been associated with stress because of unmeasured confounding and thus should have been included in the analyses. However, adjustment of hypertension and oral contraceptive use did not change the risk estimates (data not shown). 67 Possible pathways between stress and endometrial cancer A stress-induced distortion of estrogen synthesis may explain the lower incidence of endometrial cancer among women with high levels of stress. The hypothalamicpituitary-gonadal (HPG) axis regulates the synthesis of estrogens in a normally functioning female reproductive system. Stress can affect the signals of this axis by activating the hypothalamic-pituitary-adrenal (HPA) axis and the sympathetic nervous system. Several experimental studies have found that the activation of the HPA axis can inhibit the function of the HPG axis and thereby decrease the synthesis of estrogen.1012;72;81 Although human evidence is sparse, significantly lower levels of bioavailable estradiol was also found among female caregivers compared to non-caregivers in a recent epidemiologic study.55 These mechanisms may explain the lower risk of endometrial cancer observed among normal-weight women. In overweight women, peripheral conversion of androgens to estrogens in adipose tissue is another important source of estrogens.87 No evidence indicates that stress hormones have an effect on this extra-gonadal synthesis of estrogens. Also, overweight women seem to be in a proinflammatory state that facilitates carcinogenesis by creating an environment susceptible to tumor initiation and promotion as well as increased estrogen production.88 This may counteract the stress-induced suppression of estrogens and thus explain the less pronounced effect of stress on endometrial cancer among overweight women. Because of the long delay between the development and the detection of the disease and the fact that we were only able to follow the study participants in registries, we cannot determine whether stress, through its effect on estrogen synthesis and metabolism, hinders initiating and/or interrupts promotion of already initiated cells. Experimental evidence shows that stress may also produce a direct effect on the uterus by changing the response of its structures toward estrogen.89-91 It is well known that unopposed estrogens increase the proliferative activity of uterine tissues as well as lead to morphogenetic changes, which may increase the risk of developing tumors in these tissues. Chronic exposure to glucocorticoids, a major end-point in a physiologic stress response, has been shown to produce a complex anti-estrogenic effect in the uterus of mice by turning the estrogen-dependent changes from the direction of pre-cancerous, atypical hyperplasia formation to the more favorable development of simple and cystic hyperplasia.89 Acute stress and administration of glucocorticoids in doses normally 68 experienced during a stress response have both been reported to decrease the uterus’ sensitivity toward estrogen stimulation and decrease estrogen receptor concentration in ovariectomized rats treated with estrogen.90;91 In these studies, the effects of stress on the uterine structures depended on the presence of estrogens. The majority of the women in the Copenhagen City Heart Study was post-menopausal at baseline and therefore had a relatively low ovarian synthesis of estrogens. If the effect of stress on the sensitivity of the uterine structures to estrogens depends on the presence of estrogen, then it may explain the stronger effect of stress observed among women receiving hormone therapy in the present study. This is the first prospective cohort study to address an association between stress and endometrial cancer. Perceived stress seemed to be associated with lower incidence of endometrial cancer in women receiving hormone therapy and in women of normal weight. These results are biologically plausible and are in agreement with the lower risk of breast cancer previously observed in the same cohort. 69 Table 6-1. Baseline characteristics of 6,760 women who participated in the second examination of the Copenhagen City Heart Study in 1981-83 Stress* Study population Low Medium High P-value Persons, n (% of study sample) 6760 2848 (42) 3234 (48) 678 (10) 57 (12) 58 (12) 55 (12) 58 (11) <0.001 Pre-menopausal, n (%) 1775 (26) 676 (24) 979 (30) 120 (18) <0.001 Hormone therapy, n (%) 1041 (15) 330 (12) 564 (17) 147 (22) <0.001 1661 (25) 741 (26) 766 (24) 154 (23) 0.07 Mean body mass index, kg/m (SD) 24.9 (4.6) 25.2 (4.6) 24.6 (4.4) 24.8 (5.1) <0.001 Current smokers, n (%) 3616 (53) 1398 (49) 1796 (56) 422 (62) <0.001 Low education, n (%) 3116 (46) 1348 (47) 1396 (43) 372 (55) <0.001 Physically inactive, n (%) 1200 (18) 446 (16) 527 (16) 227 (33) <0.001 119 (2) 49 (2) 59 (2) 11 (2) 0.92 Mean age (SD) Nulliparous, n (%) 2 Diabetes mellitus, n (%) * Participants were asked to report their stress intensity and frequency in a standard questionnaire with four multiple-choice categories (0-3 points) for each stress measure. The scores of the two questions were added and combined into a continuous stress-score from 0 to 6. This stressscore was categorized into low (0-1 points), medium (2-4 points), and high (5-6 points) stress. Table 6-2. Incidence, hazard ratio (HR), and 95 % confidence interval (CI) for first-time incidence of primary endometrial cancer associated with perceived stress among 6,760 women who participated in the Copenhagen City Heart Study in 1981-83 Cancers, n Incidence per 100 000 years Age-adjusted HR (95 % CI) Multi-adjusted HR* (95 % CI) None (n=2,233) 30 86 1 (reference) 1 (reference) Light (n=2,627) 26 61 0.79 (0.47-1.34) 0.71 (0.42-1.21) Moderate (n=1,406) 11 49 0.61 (0.31-1.23) 0.52 (0.26-1.04) High (n=494) 5 68 0.78 (0.30-2.00) 0.72 (0.28-1.89) 0.22 0.10 Stress intensity P-value for trend Stress frequency Never (n=2,879) 36 81 1 (reference) 1 (reference) Monthly (n=2,010) 20 60 0.90 (0.52-1.55) 0.77 (0.44-1.34) Weekly (n=1,190) 12 63 0.86 (0.45-1.66) 0.73 (0.38-1.42) Daily (n=681) 4 38 0.47 (0.17-1.32) 0.40 (0.14-1.15) 0.18 0.07 0.91 (0.80-1.04) 0.88 (0.76-1.01) P-value for trend Stress score (continuous) 72 * Adjusted for age, education, physical activity during leisure time, body mass index, tobacco smoking, diabetes mellitus, number of children, menopause at baseline, and hormone therapy. Table 6-3. Incidence, hazard ratio (HR), and 95 % confidence interval (CI) for first-time incidence of primary endometrial cancer associated with perceived stress among 6,760 women who participated in the Copenhagen City Heart Study in 1981-83 by hormone status Women receiving hormone therapy at baseline Cancers, n Incidence per 100 000 years Age-adjusted HR (95 % CI) Multi-adjusted HR* (95 % CI) Stress intensity None (n=256) Light (n=412) Moderate (n=274) High (n=99) P-value for trend 13 11 5 2 312 165 112 121 1 (reference) 0.53 (0.24-1.18) 0.37 (0.13-1.03) 0.41 (0.09-1.80) 0.05 1 (reference) 0.54 (0.24-1.21) 0.36 (0.13-1.02) 0.44 (0.10-2.00) 0.06 Stress frequency Never (n=338) Monthly (n=335) Weekly (n=224) Daily (n=144) P-value for trend 16 10 4 1 303 178 109 42 1 (reference) 0.62 (0.28-1.37) 0.39 (0.13-1.17) 0.14 (0.02-1.07) 0.01 1 (reference) 0.58 (0.26-1.30) 0.42 (0.14-1.26) 0.14 (0.02-1.09) 0.01 0.76 (0.61-0.95) 0.77 (0.61-0.96) Stress score (continuous) 31 Women not receiving hormone therapy at baseline Cancers, n Incidence per 100,000 years Age-adjusted HR (95 % CI) Multi-adjusted HR* (95 % CI) Stress intensity None (n=1,977) Light (n=2,215) Moderate (n=1,132) High (n=395) P-value for trend 17 15 6 3 55 42 33 52 1 (reference) 0.88 (0.44-1.77) 0.67 (0.27-1.71) 0.95 (0.28-3.23) 0.56 1 (reference) 0.89 (0.44-1.80) 0.68 (0.27-1.75) 1.09 (0.31-3.81) 0.66 Stress frequency Never (n=2,541) Monthly (n=1,675) Weekly (n=966) Daily (n=537) P-value for trend 20 10 8 3 51 36 52 37 1 (reference) 0.91 (0.43-1.96) 1.17 (0.52-2.67) 0.74 (0.22-2.47) 0.86 1 (reference) 0.91 (0.42-1.97) 1.23 (0.54-2.83) 0.82 (0.24-2.80) 0.98 0.97 (0.81-1.15) 0.98 (0.82-1.17) Stress score (continuous) 41 * Adjusted for age, education, physical activity during leisure time, body mass index, tobacco smoking, diabetes mellitus, number of children, and menopause at baseline. 72 Table 6-4. Incidence, hazard ratio (HR), and 95 % confidence interval (CI) for first-time incidence of primary endometrial cancer associated with perceived stress among 6,760 women who participated in the Copenhagen City Heart Study in 1981-83 by body mass index Body mass index ≤ 25 kg/m2 at baseline Cancers, n Incidence per 100,000 years Age-adjusted HR (95 % CI) Multi-adjusted HR* (95 % CI) Stress intensity None (n=1,218) Light (n=1,638) Moderate (n=844) High (n=304) P-value for trend 16 14 4 2 83 52 29 43 1 (reference) 0.70 (0.34-1.43) 0.37 (0.12-1.10) 0.49 (0.11-2.14) 0.07 1 (reference) 0.54 (0.26-1.13) 0.26 (0.09-0.80) 0.37 (0.08-1.63) 0.01 Stress frequency Never (n=1,607) Monthly (n=1,250) Weekly (n=727) Daily (n=420) P-value for trend 20 12 3 1 79 57 26 15 1 (reference) 0.87 (0.42-1.78) 0.35 (0.10-1.19) 0.18 (0.02-1.36) 0.02 1 (reference) 0.64 (0.31-1.34) 0.29 (0.08-0.97) 0.14 (0.02-1.05) 0.005 0.79 (0.64-0.97) 0.73 (0.58-0.91) Stress score (continuous) 36 Body mass index > 25 kg/m2 at baseline Cancers, n Incidence per 100,000 years Age-adjusted HR (95 % CI) Multi-adjusted HR* (95 % CI) Stress intensity None (n=1,015) Light (n=989) Moderate (n=562) High (n=190) P-value for trend 14 12 7 3 90 77 80 109 1 (reference) 0.92 (0.43-2.00) 0.98 (0.39-2.42) 1.26 (0.36-4.41) 0.85 1 (reference) 0.89 (0.41-1.94) 0.90 (0.36-2.26) 1.37 (0.38-4.89) 0.89 Stress frequency Never (n=1,272) Monthly (n=760) Weekly (n=463) Daily (n=261) P-value for trend 16 8 9 3 82 66 123 78 1 (reference) 0.93 (0.40-2.19) 1.66 (0.73-3.77) 0.98 (0.28-3.36) 0.53 1 (reference) 0.86 (0.36-2.04) 1.56 (0.68-3.57) 0.93 (0.26-3.25) 0.63 1.04 (0.87-1.25) 1.03 (0.86-1.24) Stress score (continuous) 36 * Adjusted for age, education, physical activity during leisure time, body mass index, tobacco smoking, diabetes mellitus, number of children, and menopause at baseline. 73 Figure 6-1. Causal diagram for the relation between perceived stress and risk of endometrial cancer History of breast cancer Tamoxifen Problematic menopause Postmenopausal hormone therapy Late menopause Nulliparity Early menarche Stress hormones Estrogens Combined estrogen-progestin contraceptive Perceived stress Endometrial cancer Cigarette smoking Hypertension Diet high in fat and low in fiber and vegetables Diabetes Obesity Physical activity Education Alcohol Age VII. Perceived stress and risk of colorectal cancer: a prospective cohort study Introduction Colorectal cancer is the third most common cancer worldwide and the age-adjusted incidence rate of colorectal cancer for Danish men and women are more than two times as high as the world average.3;92 Psychological stress may directly affect the risk of colorectal cancer by suppressing immune function,2 which might lead to increased neoplastic growth. Stress may also indirectly affect the risk of colorectal cancer by altering physical activity levels and dietary habits, which are some of the major risk factors for colorectal cancer.93;94 In addition, colorectal cell lines express both functional estrogen receptors that mediate the proliferative activity of estradiol and enzymes capable to synthesize and metabolize estrogens, suggesting that sex steroid hormones may also increase the risk of colorectal cancer.8;95-98 This is supported by the fact that women with primary breast cancer, especially those receiving tamoxifen treatment, as well as men with primary testicular cancer are at higher risk of developing colorectal cancer.99;100 Persistent activation of stress hormones seems to impair the gonadal synthesis of estrogens and thereby reduce bioavailable estradiol and testosterone.10 Stress-induced impairment of estrogen synthesis and metabolism may therefore protect against colorectal cancer. Only a few studies have addressed a potential relation between stress and colorectal cancer and a majority of these studies have focused on stressful life events or work-related stress and have been applied in case-control designs. 65-68 The physiological reaction to acute stress of stressful life events may be very different from the reaction to a more chronic state of stress in everyday life. Thus, the health consequences of these two measures are not directly comparable. Further, it is stressful to be diagnosed with cancer, which may render it difficult for cancer patients to recall their stress level prior to the diagnosis without recall bias, and thereby to validly assess the association between stress and risk of colorectal cancer in a case-control design. Only one previous prospective cohort study has addressed the association between stress 75 and colorectal cancer.69 However, they used colorectal cancer mortality as their outcome measure, which does not allow for a distinction between causal and prognostic factors. We aim to address the association between perceived stress of everyday life and incidence of primary colorectal cancer in a prospective design. Methods Study population Information from the second examination of the prospective Copenhagen City Heart Study in 1981-83 was used to address the objective. The second examination of the study included 12,698 participants. Individuals with colorectal cancer before baseline (n=49) or with lacking information on stress (n=42) or other covariates (n=605) were excluded. Women with lacking information on reproductive and hormonal factors (n=88) were also excluded, leaving 6,488 women and 5,426 men for the analyses. Perceived stress was assessed at baseline and included two questions on stress intensity and stress frequency. A seven-point stress score was created based on these two questions. Please refer to the material and methods chapter for a detailed description of the Copenhagen City Heart study and the measurement of perceived stress. Covariates The confounder identification was based on a causal diagram (figure 7-1) in which we included the most current prior knowledge about causal relations between perceived stress, covariates, and colorectal cancer:22 According to the diagram, the analyses should be controlled for: age (continuous), education (less than 8 years, 8-11 years, or 12 or more years), income (<$1,000, $1,000 to $2,500, or more than $2,500), cohabitation (living with partner and/or children, or living alone), physical activity in leisure time (none or very little activity, 2-4 hours of light activity per week, more than 4 hours of light activity or 2-4 hours of high level activity, and competitional level or more than 4 hours of hard level activity per week), alcohol consumption (<1, 1-14, >14 drinks/week), tobacco smoking (never-smoker, ex-smoker, smokers of 1-14 grams per day, 15-24 grams per day, and more than 24 grams per day), body mass index (< 18.5, 18.5-30, >30 kg/m2) and diabetes mellitus (yes/no). Analyses conducted in women were also adjusted 76 for current oral contraceptive use (yes/no), hormone therapy (yes/no), menopause (yes/no) and number of children (0, 1-2, 3 or more). All variables were measured at baseline in 1981-83. Fruit and vegetable intake, dietary fibers, folate, vitamins, and stress hormones were assumed to be intermediates on the pathway between stress and colorectal cancer and were therefore not included in the analyses. Family history of colorectal cancer, red meat consumption, calcium intake, use of NSAIDS or aspirins, and inflammatory bowls disease or Chronhn’s disease were assumed not to be strongly associated with stress and were therefore not included in the analyses. Follow-up Participants were followed from date of the second examination till date of first diagnosis of primary colorectal cancer (162 in women and 166 in men), death from other causes (n=4,816), emigration (n=55), or end of follow-up on December 31, 2000 (n=6,715). Thus, less than 0.1 % of the participants were lost to follow-up due to emigration. Using the civil registry number, which is unique to every Danish citizen, diagnoses of primary colorectal cancer were identified through linkage to the Danish National Cancer Registry, which contains data on all cancer diagnoses in Denmark since 1942. The following ICD7-codes were used to identify primary invasive colon cancer cases: 153.0, 153.4,153.5,154.9, 253.0-253.4, 453.0-453.5,453.8, 454.9, 853.0-853.5, and 854.9. The following ICD7-codes were used to identify primary invasive rectal cancer cases: 154.0, 454.0, and 854.0. The vital status of the study population was followed in the Central Death Registry. Statistical methods Data were analyzed by means of proportional hazards regression models with age as the time scale using SAS/STAT software version 8.2. Stress intensity, stress frequency, and the stress score all met the assumption of proportional hazards. We estimated the ageadjusted and multi-adjusted hazard ratio of colorectal cancer associated with stress intensity, stress frequency, and the stress-score. By including age as the time scale, the estimates were soundly adjusted for confounding by age. The analyses were also done separately for colon and rectal cancer. Trend analyses were used to address doseresponse associations between stress and colorectal cancer. There appeared to be sex- 77 differences in the associations and all analyses were therefore conducted separately for men and women. The associations between stress and risk of colorectal cancer in women were also assessed in subgroups according to menopausal status at baseline. Finally, to address the possible effect of prolonged follow-up we conducted the analyses between stress score and colorectal cancer separately for the first and last nine years of follow-up. Results Baseline characteristics The median age at baseline was 58 years for women and 57 years for men ranging from 21 to 98 years. Ten percent of the women and six percent of the men reported high levels of stress (Table 7-1). Individuals with medium stress were slightly younger than individuals with both low and high stress levels. Mean body mass index and percentage with diabetes mellitus were similar at the different stress levels in both men and women. A higher proportion of men and women at the high stress level had low education, low income, high alcohol intake, were current smokers, and were physically inactive in their leisure time compared to men and women with low levels of stress. A lower proportion of highly stressed women were nulliparous, pre-menopausal, and used oral contraceptives compared to less stressed women, while a higher proportion received hormone therapy. Perceived stress and risk of colorectal cancer in women During follow-up, 162 cases of primary colorectal cancer (125 colon and 37 rectal cancers) occurred in women. Women reporting moderate and high stress intensity had a hazard ratio of 0.60 (95 % CI: 0.37-0.98) and 0.52 (0.23-1.14) for colorectal cancer, respectively, compared to women who reported no stress intensity (table 7-2), and there appeared to be a linear trend in the dose-response association (p-value for trend: 0.02). For stress frequency, monthly and weekly stress were not associated with risk of colorectal cancer, but women reporting daily stress were at lower risk of developing colorectal cancer (HR=0.28, 95 % CI: 0.11-0.67) compared to women who never experienced stress. For each unit increase in the combined seven-point stress score, women were 10 percent less likely to develop colorectal cancer (HR=0.90, 95 % CI: 78 0.82-0.98). Dividing colorectal cancer into colon and rectal cancer made it clear that the association between high stress and lower risk of colorectal cancer was strongest for colon cancer. Stress intensity was associated with lower risk of colon cancer in an inverse doseresponse manner (test for trend: p= 0.03). Daily stress was also associated with lower risk of colon cancer (HR=0.23, 95% CI: 0.07-0.73), while monthly and weekly stress were not associated with colon cancer. The association between perceived stress and risk of rectal cancer in women was based on few cases. Even though stress frequency and the stress score seemed to be associated with lower risk of rectal cancer, no clear picture emerged. In general, adjustment for potential confounders only changed the risk estimates slightly. One hundred and forty-six colorectal cancer cases occurred in postmenopausal women, while only 33 cases occurred among pre-menopausal women. The inverse association between stress and risk of colon cancer appeared to be strongest in postmenopausal women, while a stronger association between stress and rectal cancer was suggested in pre-menopausal women (table 7-3). Sixty-five cases of colorectal cancer occurred in women during the first nine years of follow-up and 97 cases occurred during the last nine years of follow-up. The association between the stress score and colorectal cancer was stronger in the first nine years (HR=0.84, 95 % CI: 0.72-0.97) than in the last nine years (HR=0.94, 95 % CI: 0.83-1.05) of follow-up (data not shown). Perceived stress and risk of colorectal cancer in men During follow-up, 166 cases of primary colorectal cancer (111 colon and 55 rectum cancers) occurred in men. High stress intensity was associated with a higher risk of colorectal cancer among men (HR=1.96, 95 % CI: 1.03-3.74) while moderate stress intensity appeared to be associated with lower risk of colorectal cancer (HR=0.64, 95 % CI: 0.37-1.10) compared to no stress (table 7-4). There was no notable association between neither stress frequency nor the combined stress score and risk of colorectal cancer in men. A similar picture emerged when studying colon cancer alone. High stress intensity was associated with a hazard ratio of 2.94 (95 % CI: 1.09-7.93) for rectal cancer compared to no stress, while there was no association between neither stress frequency nor stress score and risk of rectal cancer. The risk estimates for rectal cancer 79 were rather unstable because of the low number of cases. Sixty-three cases of colorectal cancer in men occurred during the first nine years of follow-up, while the other 103 cases occurred during the last nine years of follow-up. The impact of a one-unit change in stress score on colorectal cancer risk was quite similar in the first nine years (HR=0.97, 95 % CI: 0.82-1.14) and the last nine years (HR=0.99, 95 % CI: 0.88-1-12) of follow-up (data not shown). Discussion We found sex-differences in the associations between perceived stress and incidence of colorectal cancer among 11,914 men and women prospectively followed for 18 years. In women, higher stress intensity and daily stress were associated with lower incidence of colon cancer in particular, while there was no clear association between these measures of stress and rectal cancer. Although high stress intensity appeared to be associated with a higher incidence of rectal cancer in men, this result was based on few cases and there was no clear evidence of a relation between stress and colorectal cancer. While we addressed incidence of colon and rectal cancer, sex-differences in the effect of stress has previously been noted for colorectal cancer mortality in a cohort of Japanese men and women prospectively followed-up for nine years.69 Contrary to our results, perceived psychological stress, assessed by the question “Do you feel stressed in your daily life?” with two response categories, was weakly associated with higher colon cancer mortality in women, but not in men, while stress was not associated with rectal cancer mortality in neither women nor men. There were no clear trends in these associations and the analyses were based on mortality alone, which as already mentioned does not allow for a distinction between causal and prognostic factors. Other studies have primarily focused on stressful life events or stressors at work and have been conducted as case-control studies.65-68 Different measures of stress have been applied in these studies and the results are conflicting. Strengths and weaknesses of the study The prospective design of the Copenhagen City Heart Study ensured temporality between self-reported stress and incidence of colorectal cancer, and linkage of civil 80 registry numbers to population-based registers with nationwide coverage enabled identification of virtually all cases of colorectal cancer and allowed for nearly complete long-term follow-up. Stress was defined as an individual state of high arousal and displeasure,83 and by using a measure of perceived stress, we accounted for the fact that each individual has different capacity and ways to cope with stressful situations.16 Stress was assessed by combining two questions on stress intensity and stress frequency asked at baseline. Using these two questions instead of a more extensive scale, such as the Perceived Stress Scale,84 may have blurred an even stronger relation between perceived stress and risk of colorectal cancer. However, in a recent study, two single-item measures on stress were found to be just as reliable and valid as three fully validated multi-item measures on perceived stress.101 This may provide some assurance that the single-item measurements used in the present study actually provided a reasonably valid measure of stress. Also, the same measure of stress have previously been found to be predictive of a range of other diseases in the same cohort.57;102;103 In the questionnaire, stress was exemplified as impatience, anxiety, and sleeplessness. Based on the two simple questions of stress applied in this study, we cannot fully exclude the possibility that we have also partly measured the effect of depressive symptoms or personality traits, which may be closely related to stress as it is measured in this study. Stress was only assessed at baseline and the participants may have changed stress level over time in a manner that is most likely independent of subsequent development of colorectal cancer. The association between stress and risk of colorectal cancer in women was strongest in the first nine years compared to the last nine years of follow-up, while there was virtually no difference for men. Thus, the long follow-up period may partly have blurred the associations, especially in women. In another Danish cohort study85 that included a question on perceived stress, the majority of the participants (62%) reported the same level of stress in 1994 as in 2000 (Nielsen, unpublished data, 2006). Although this finding indicates that a measure of perceived stress may be relatively stable over time, a large minority changed stress levels over time. Thus, we cannot exclude the possibility that the relation between stress and risk of colorectal cancer may have been partly blurred by nondifferential misclassification from changes in stress levels over time. 81 Information on some important risk factors for colorectal cancer such as family history of colorectal cancer and diet were not obtained in the Copenhagen City Heart Study. However, to confound the results, these factors would also have to be related to stress. We cannot exclude that having experienced colorectal cancer in the near family may act as a stressor and thereby lead to higher levels of stress, which would result in a spurious positive association between stress and colorectal cancer. This is opposite to the inverse association between perceived stress and colon cancer observed among women, but it may explain some of the positive association between stress and rectal cancer in men. In order to determine potential confounding from dietary factors, it is important to determine whether dietary habits are a cause or consequence of stress. Some evidence indicates that during chronic stress, individuals tend to eat smaller, but more frequent, amounts of energy-dense, fatty, low-protein foods.93 Thus, diet is more likely to be an intermediate factor in the causal path from stress to colorectal cancer than a confounder. Men and women with high levels of stress smoked more, drank more alcohol, and did less exercise. Also, they had lower income and education than those with lower levels of stress. We adjusted our analyses for baseline differences in these factors to avoid confounding, but one may still be concerned about residual confounding. The amount of residual confounding is, however, proportional to the confounding originally present,104 and adjustment for confounding only slightly changed the risk estimates, making residual confounding less of a concern. The age range of the study participants was relatively broad and one may be concerned that the perceptions of stress in the general adult population may be different from that of the very old. To address this question, we reanalyzed data after excluding all persons above the age of 75 at baseline. This only led to minor changes in the risk estimates (data not shown). More women in the high stress group (42 %) than in the medium (31 %) and low (37 %) stress group died during follow-up. The pattern was similar in men and the proportions that died during follow-up were 58 %, 41 %, 51 % for high, medium, and low stress, respectively. Although this indicates no systematic difference, it may raise concern about how censoring has influenced the results. We assumed, in the statistical model, that censoring was independent of colorectal cancer risk within strata of stress. However, it is possible that competing causes, such as death of cardiovascular disease, 82 could be associated with risk of colorectal cancer within strata of stress due to other common risk factors. Some of the common risk factors, such as alcohol consumption, have opposite effects on the two diseases, while other risk factors, such as low physical activity, increase the risk of both diseases. To some degree, we would expect the bias to level out and the average risk of colorectal cancer among women not censored to be similar to what it would have been if no such censoring had occurred. Also, both a higher proportion of men and of women in the high stress group died during follow-up, and we therefore find it unlikely to explain the sex-differences observed in the present study. Thus, while our results may have been influenced by bias from non-independent censoring, we find it unlikely to fully explain them. Possible pathway from perceived stress to colorectal cancer Stress may affect the risk of colorectal cancer directly through biological processes or indirectly by affecting health-related behavior. Several possible pathways between perceived stress and incidence of colorectal cancer may be identified. The importance of each of these pathways may be sex-specific and explain the observed sex-differences in the present study. First, the hypothalamic-pituitary-gonadal (HPG) axis regulates the synthesis of sex steroid hormones in a normally functioning reproductive system. Stress can affect the signals of this axis by activating the hypothalamic-pituitary-adrenal (HPA) axis and the sympathetic nervous system. Several experimental studies have found that the activation of the HPA axis inhibits the function of the HPG axis and thereby decrease estrogen and testosterone synthesis.10-12;72 In a recent epidemiologic study, significantly lower levels of bioavailable estradiol was also found among female caregivers compared to noncaregivers.55 In vitro treatment of a human colon cancer cell line with estradiol was found to rapidly stimulate intermediates in a signal transduction pathway that is known to trigger cell proliferation.96 Further, the intensity of the stimulatory effect of estradiol in human colon cancer cells is similar to the estradiol responsiveness of human mammary cancer cell lines, which are classical estradiol sensitive cells.96 A stressinduced distortion of estrogen synthesis may therefore explain the lower incidence of colon cancer among women with high levels of stress. 83 Secondly, long-term stress may lead to a persistent activation of the HPA-axis. Some of the mediators released by the HPA-axis, like corticosteroids and catecholamines, seem to be capable of suppressing the immune function and are thereby reduce its ability to recognize and destroy neoplastic cell growth.2;105 A decrease in cytotoxic T-cell and natural-killer-cell activity and in the general cellular immune response has been found in laboratory animals exposed to stress.2 Thus, by affecting the immune system and the DNA repair system, perceived stress may promote the initiation and progression of colorectal cancer. Men seem to respond to psychological stress with greater increases in cortisol than women,106 which may explain the higher incidence of colorectal cancer observed among men reporting high levels of stress. Thirdly, there is growing evidence that chronic stress may lead to the development of obesity and insulin resistance either by a behavioral pathway with altered dietary and activity patterns or through an abnormal diurnal cortisol rhythm induced by the activation of the HPA-axis.107 Obesity, insulin resistance, and insulin growth factor are all potential metabolic mediators for tumor progression and may thereby increase the risk of colorectal cancer.107 We adjusted the analyses for body mass index at baseline, but high levels of stress may still have led to weight changes in some individuals during follow-up. Experimental studies have also suggested that stress may alter bowel movement,93;94;108 and thereby indirectly affect the risk of colorectal cancer. Finally, stress may affect health-related behavior. The participants with high levels of perceived stress were more likely to also have a high alcohol intake, be current smokers, and be physically inactive at baseline. We adjusted our analyses for such differences in health-related behavior at baseline in order to control for confounding, but we would also expect some of the effect of stress on colorectal cancer risk to be mediated through such changes in health-related behavior occurring during follow-up. In addition, some of the effect of stress on risk of colorectal cancer may be mediated through mental processes such as depression or burnout. In conclusion, the relation between perceived stress and risk of colorectal cancer seems to be a result of a complex system with different mechanisms working in opposite directions. In this cohort study we found high stress to be associated with lower risk of colon cancer in women, while there is no clear evidence of a relation between stress and risk of colorectal cancer in men. 84 Table 7-1. Baseline characteristics of the 6,488 women and 5,426 men who participated in the second examination of the Copenhagen City Heart Study in 1981-83 Stress score Study population Low Medium High Women Persons, N (% of study sample) Mean age (SD) Low education (%) Low income (%) Physically inactive (%) High alcohol consumption (%) Mean body mass index, kg/m2 (SD) Current smokers (%) Diabetes mellitus (%) Nulliparous (%) Hormone replacement therapy, % Pre-menopausal, % Current oral contraceptive users, % 6,488 57 (12) 2,961 (45) 2,431 (37) 1,124 (17) 445 (7) 25 (5) 3,468 (53) 114 (2) 1,608 (25) 995 (15) 1,716 (26) 249 (4) 2,745 (42) 58 (13) 1,286 (47) 1,080 (39) 417 (15) 150 (5) 25 (5) 1,343 (49) 48 (2) 726 (26) 315 (11) 655 (24) 99 (4) 3,101 (48) 55 (12) 1,322 (43) 996 (32) 494 (16) 228 (7) 25 (4) 1,727 (56) 56 (2) 739 (24) 539 (17) 949 (31) 143 (5) 642 (10) 58 (11) 353 (55) 355 (55) 213 (33) 67 (10) 25 (5) 398 (62) 10 (2) 143 (22) 141 (22) 112 (17) 7 (1) Men Persons, N (% of study sample) Mean age (SD) Low education (%) Low income (%) Physically inactive (%) High alcohol consumption (%) Mean body mass index, kg/m2 (SD) Current smokers (%) Diabetes mellitus (%) 5,426 56 (13) 2,122 (45) 1,345 (25) 854 (16) 1,791 (33) 26 (4) 3,466 (64) 187 (3) 2,984 (55) 58 (13) 1,405 (47) 774 (26) 425 (14) 911 (31) 26 (4) 1,870 (63) 99 (3) 2,984 (39) 53 (12) 845 (40) 439 (21) 323 (15) 760 (36) 26 (4) 1,362 (64) 77 (4) 329 (6) 57 (11) 172 (52) 132 (40) 106 (32) 120 (36) 26 (4) 234 (71) 11 (3) Table 7-2. Women. Incidence, hazard ratio (HR), and 95 % confidence interval (CI) for first-time incidence of primary colorectal, colon, and rectal cancer associated with perceived stress among 6,488 women who participated in the Copenhagen City Heart Study in 1981-83 Cancers, n Incidence per 100,000 years Age- and sex-adjusted HR (95 % CI) Multi-adjusted HR* (95 % CI) Colorectal cancer Stress intensity None (n=2144) Light (n=2535) Moderate (n=1345) High (n=464) P-value for trend 62 71 22 7 185 173 103 101 1 (reference) 1.08 (0.77-1.52) 0.62 (0.38-1.02) 0.55 (0.25-1.20) 0.03 1 (reference) 1.06 (0.75-1.49) 0.60 (0.37-0.98) 0.52 (0.23-1.14) 0.02 Stress frequency Never (n=2772) Monthly (n=1947) Weekly (n=1122) Daily (n=647) P-value for trend 77 52 28 5 180 163 158 50 1 (reference) 1.16 (0.81-1.65) 1.01 (0.66-1.56) 0.29 (0.12-0.71) 0.05 1 (reference) 1.14 (0.80-1.63) 0.98 (0.63-1.51) 0.28 (0.11-0.67) 0.04 0.91 (0.83-0.99) 0.90 (0.82-0.98) Stress score (continuous) 162 Colon cancer Stress intensity None (n=2144) Light (n=2535) Moderate (n=1345) High (n=464) P-value for trend 50 53 18 4 150 130 84 58 1 (reference) 1.01 (0.69-1.49) 0.64 (0.38-1.10) 0.39 (0.14-1.09) 0.03 1 (reference) 0.99 (0.67-1.47) 0.63 (0.37-1.09) 0.39 (0.14-1.08) 0.03 Stress frequency Never (n=2772) Monthly (n=1947) Weekly (n=1122) Daily (n=647) P-value for trend 59 41 22 3 137 128 124 30 1 (reference) 1.22 (0.82-1.82) 1.06 (0.65-1.73) 0.23 (0.07-0.73) 0.09 1 (reference) 1.21 (0.81-1.81) 1.04 (0.64-1.71) 0.23 (0.07-0.73) 0.09 0.90 (0.81-1.00) 0.89 (0.81-0.99) Stress score (continuous) 125 Continues on next page 86 Rectal cancer Stress intensity None (n=2144) Light (n=2535) Moderate (n=1345) High (n=464) P-value for trend 12 18 4 3 36 44 19 43 1 (reference) 1.35 (0.65-2.81) 0.56 (0.18-1.74) 1.17 (0.33-4.14) 0.68 1 (reference) 1.29 (0.62-2.71) 0.49 (0.16-1.52) 0.94 (0.26-3.41) 0.43 Stress frequency Never (n=2772) Monthly (n=1947) Weekly (n=1122) Daily (n=647) P-value for trend 18 11 6 2 42 34 34 20 1 (reference) 0.97 (0.45-2.05) 0.87 (0.34-2.19) 0.47 (0.11-2.02) 0.36 1 (reference) 0.95 (0.44-2.02) 0.77 (0.30-1.96) 0.38 (0.09-1.68) 0.21 37 0.93 (0.78-1.12) 0.90 (0.75-1.09) Stress score (continuous) * Adjusted for age, education, income, physical activity in leisure time, alcohol consumption, body mass index, tobacco smoking, diabetes mellitus, cohabitation, number of children, hormone replacement therapy, menopause at baseline, and oral contraceptive use. Women with lacking information on any of these variables were excluded from the analyses. 87 Table 7-3. Hazard ratio (HR), and 95 % confidence interval (CI) for first-time incidence of primary colorectal, colon, and rectal cancer associated with a seven-unit stress score among 1,716 pre-menopausal and 4,772 post-menopausal women who participated in the Copenhagen City Heart Study in 1981-83 Colorectal cancer Cancers, n Multi-adjusted HR* (95 % CI) Colon cancer Cancers, n Multi-adjusted HR* (95 % CI) Rectal cancer Cancers, n Multi-adjusted HR* (95 % CI) Pre-menopausal Stress score (per one-unit change) 16 0.97 (0.70-1.35) 12 1.12 (0.78-1.62) 4 0.56 (0.21-1.46) 146 0.89 (0.81-0.98) 113 0.88 (0.79-0.98) 33 0.92 (0.76-1.12) Post-menopausal Stress score (per one-unit change) * Adjusted for age, education, income, physical activity in leisure time, alcohol consumption, body mass index, tobacco smoking, diabetes mellitus, cohabitation, number of children, hormone replacement therapy, and oral contraceptive use. Table 7-4. Men. Incidence, hazard ratio (HR), and 95 % confidence interval (CI) for first-time incidence of primary colorectal, colon, and rectal cancer associated with perceived stress among 5,426 men who participated in the Copenhagen City Heart Study in 1981-83 Cancers, n Incidence per 100 000 years Age- and sex-adjusted HR (95 % CI) Multi-adjusted HR* (95 % CI) Colorectal cancer Stress intensity None (n=2559) Light (n=1779) Moderate (n=892) High (n=196) P-value for trend 91 48 16 11 258 181 123 472 1 (reference) 0.94 (0.66-1.34) 0.67 (0.39-1.14) 2.13 (1.14-3.99) 0.84 1 (reference) 0.93 (0.65-1.32) 0.64 (0.37-1.10) 1.96 (1.03-3.74) 0.96 Stress frequency Never (n=2984) Monthly (n=1352) Weekly (n=747) Daily (n=343) P-value for trend 106 31 19 10 256 150 176 234 1 (reference) 0.82 (0.55-1.24) 0.98 (0.60-1.61) 1.03 (0.54-1.96) 0.85 1 (reference) 0.81 (0.54-1.22) 0.94 (0.57-1.55) 0.98 (0.51-1.89) 0.69 1.00 (0.91-1.10) 0.99 (0.90-1.09) Stress score (continuous) 166 Colon cancer Stress intensity None (n=2559) Light (n=1779) Moderate (n=892) High (n=196) P-value for trend 65 30 10 6 184 113 77 258 1 (reference) 0.84 (0.54-1.30) 0.60 (0.30-1.16) 1.65 (0.71-3.81) 0.54 1 (reference) 0.82 (0.53-1.27) 0.57 (0.29-1.13) 1.51 (0.64-3.56) 0.42 Stress frequency Never (n=2984) Monthly (n=1352) Weekly (n=747) Daily (n=343) P-value for trend 70 24 12 5 169 116 111 117 1 (reference) 1.00 (0.62-1.59) 0.97 (0.53-1.81) 0.79 (0.32-1.96) 0.69 1 (reference) 0.97 (0.60-1.55) 0.94 (0.50-1.75) 0.76 (0.30-1.90) 0.59 0.97 (0.86-1.09) 0.96 (0.85-1.08) Stress score (continuous) 111 Continues on next page 89 Rectal cancer Stress intensity None (n=2559) Light (n=1779) Moderate (n=892) High (n=196) 26 18 6 5 74 68 46 215 36 7 7 5 87 34 65 117 1 (reference) 1.18 (0.64-2.16) 0.84 (0.34-2.05) 3.27 (1.25-8.55) 0.23 P-value for trend 1 (reference) 1.19 (0.65-2.19) 0.80 (0.32-1.95) 2.94 (1.09-7.93) 0.32 Stress frequency Never (n=2984) Monthly (n=1352) Weekly (n=747) Daily (n=343) P-value for trend 1 (reference) 0.51 (0.23-1.16) 0.99 (0.44-2.26) 1.45 (0.57-3.69) 0.84 1 (reference) 0.52 (0.23-1.18) 0.92 (0.40-2.11) 1.34 (0.51-3.49) 0.99 55 1.06 (0.91-1.24) 1.04 (0.89-1.22) Stress score (continuous) * Adjusted for age, education, income, physical activity in leisure time, alcohol consumption, body mass index, tobacco smoking, diabetes mellitus, and cohabitation. 90 Figure 7-1. Causal diagram for the relation between perceived stress and risk of colorectal cancer Family history of colorectal cancer Oral contraceptive use Body mass index Reproductive history Red meat consumption Stress hormones Diabetes mellitus Estrogen Problematic menopause Postmenopausal hormone therapy Perceived stress Colorectal cancer Age Smoking Calcium Social network Physical activity Alcohol Dietary fibers Fruits and vegetables NSAIDS and aspirin Infection Folate Vitamins Socio-economic status Immune system Inflammatory Bowls diseases / Chronhn’s disease VIII. Sociodemographic status, stress and risk of prostate cancer: a prospective cohort study Introduction Prostate cancer is among the most common malignancies in men and the incidence of the disease is increasing worldwide.109 Little is known about the etiology of prostate cancer and modifiable risk factors for the disease are essentially unknown. There are remarkable racial and geographic variations in the incidence of and mortality from prostate cancer,109 indicating the existence of both genetic and environmental risk factors. Prostate cancer is also a disease of striking social disparities. Men of higher socio-economic status seem to have an elevated incidence of prostate cancer compared to those of lower socio-economic status.110-112 For prostate cancer mortality, the picture is reversed with lower prostate cancer mortality rates among men with higher socioeconomic status.112 These differences in incidence and mortality patterns have mainly been ascribed to differential access to medical care and thereby also to prostate-specific antigen (PSA) testing in the different socio-economic and racial groups. Socio-economic disadvantages can lead to psychological stress. Animal models have provided evidence that stress affects tumor growth and development,2 but few epidemiologic studies have assessed the relation between stress and risk of prostate cancer.47 In a large registry linkage study, no increased risk of prostate cancer was observed among men who had experienced losing a child to cancer.47 However, the manner in which a stressor is perceived and coped with may be important to the magnitude of the stress response and thereby also to the resulting health consequences.16 In addition, the bodily reactions to acute stress from major life events such as losing a child or a spouse may be very different from those resulting from the chronic “low-key” stress of everyday life. Repeated episodes of stress or chronic stress may weaken the immune system and render the individual susceptible to malignancies.2 Marital status, as an indicator of social support, may be an important way of coping with socio-economic disadvantages or stress and may thereby also affect prostate cancer risk. One study found that divorced and separated men had a higher risk of 92 prostate cancer than married men.113 In the present study, we include information on socio-economic status, perceived stress of everyday life, and marital status. We aim to assess if the social disparity in prostate cancer incidence prevail in a racially homogenous population of Caucasians with free access to medical care. We also aim to assess if perceived stress of everyday life is associated with risk of prostate cancer, and to address differences in prostate cancer incidence according to marital status in a prospective study with 20 years of follow-up. Methods Study population Information from the second examination of the prospective Copenhagen City Heart Study in 1981-83 was used to address the objective. Of the 12,698 participants in the second examination 5,680 were men, and men with prostate cancer before baseline (n=19) and men who lacked information on stress (n=15), educational level (n=12), income (n=47), marital status (n=5), or other covariates (n=86) were excluded, leaving 5,496 men for the analyses. Perceived stress was assessed at baseline and included two questions on stress intensity and stress frequency. A seven-point stress score was created based on these two questions. Please refer to the material and methods chapter for a detailed description of the Copenhagen City Heart study and the measurement of perceived stress. Sociodemographic variables All sociodemographic variables were based on self-reported information at baseline in 1981-83. Educational level was reported in categories of: low (less than 8 years), medium (8-11 years), and high (12 or more years) education. Income was divided into categories of: low (<$1,000 per month), medium ($1,000-2,500 per month), and high (>$2,500 per month) income. Marital status categories were: married, unmarried, divorced/ separated, and widowed. Covariates Potential confounders was identified according to the methods of causal diagrams developed by Greenland, Pearl, and Robins.22 The causal diagram that we have specified 93 is shown in figure 8-1 and it served as a way to visualize and explicitly elaborate our assumptions about the web of causation for the relation of interest as well as to help identify variables that must be controlled for to obtain unconfounded estimates. According to the diagram, age (continuous) was the main possible confounder of the associations of interest. The associations between the sociodemographic variables and risk of prostate cancer will therefore only be adjusted for age. Socio-economic status and marital status were also possible confounders for the relation between stress and risk of prostate cancer and the association will be adjusted for these factors. Physical activity, body mass index, and alcohol intake were either possible intermediates on the pathway between stress and risk of prostate cancer (as they are presented in the diagram), but they may also be potential confounders for the association between stress and prostate cancer. The analysis of stress will therefore be presented both with adjustment for age alone and with adjustment for age as well as sociodemographic factors, physical activity in leisure time (none or very little activity; 2-4 hours of light activity per week; more than 4 hours of light activity or 2-4 hours of high level activity; and competitional level or more than 4 hours of high level activity per week), body mass index (continuous), and alcohol intake (<1, 1-7, 8-14, 15-21, 22+ drinks/wk). Lucopene, fish intake, selenium/vitamin E, and foods high in animal fat are possible intermediates on the pathway from stress to prostate cancer and will therefore not be included in the analysis. Insulin-like growth factor (IGF-1), BRCA1/2 mutations, family history of prostate cancer, and sexually transmitted diseases were also not included in the analysis, as they were assumed not to be strongly associated with the socio-demographic factors or stress. Follow-up The participants were followed from the date of the second examination until the date of first diagnosis of primary prostate cancer (n=157), death (n=2,889), emigration out of Denmark (n=31), or end of follow-up on December 31, 2002 (n=2,419). Thus, less than 0.1% were lost to follow-up due to emigration. Using the civil registry number, which is unique to every Danish citizen, primary prostate cancer events were identified through linkage to the Danish National Cancer Registry, which contains data on all cancer diagnoses in Denmark since 1942. The following ICD7-codes were used to identify 94 primary invasive prostate cancer cases: 177.0, 477.0, and 877.0. The vital status of the participants was followed in the Central Person Registry. Statistical methods Data were analyzed by means of Cox proportional hazards models with age as the time variable. Stress and the sociodemographic variables met the assumption of proportional hazards. Initially, we estimated the age-adjusted hazard ratio of prostate cancer associated with the sociodemographic variables and stress. By including age as the time variable, the estimates were adjusted for confounding by age. A trend test was used to test for linear dose-response trends in the associations between the sociodemographic variables or stress and prostate cancer. Secondly, a multivariate Cox proportional hazards model was fitted to adjust the association between stress and prostate cancer for potential confounding from the sociodemographic variables as well as for alcohol, physical activity in leisure time, and body mass index. Finally, PSA screening is less common in Denmark than in the US and Canada, but introduction of PSA testing in the late 1980s may still have affected the diagnosis pattern in this population. Thus, we divided the follow-up period into two periods, before and after 1990, and conducted separate analyses for these periods in order to address whether the introduction of PSA testing affected our estimates. Results Baseline characteristics The median age at baseline was 57 years ranging from 21 to 98 years. Forty-five percent of the study participants had low education (table 8-1). In the total study population, 25 % had low income, 22 % were unmarried or divorced, and six percent reported high levels of stress. As expected, income and education were correlated, so that men with low education were more likely to also have low income. Men with high education appeared to be somewhat more likely to be unmarried or divorced, while a slightly higher proportion of men with low education reported high levels of stress. A higher proportion of men with low education were physically inactive than observed among 95 men with higher educational levels. Mean body mass index and mean alcohol consumption were similar over the different educational levels. Sociodemographic status and risk of prostate cancer With an incidence rate of 317 per 100,000 years, men with low income had markedly higher incidence of prostate cancer than men with higher income (table 8-2). This difference in risk could, in large part, be explained by age, so that no clear association between income and risk of prostate cancer remained after adjusting for age. Men with more than 12 years of formal education seemed to be at a slightly higher risk of prostate cancer (HR=1.22; 95 % CI: 0.76-1.96), but the estimates were unstable and there was no dose-response effect (P-value for trend: 0.42). Widowed men had a markedly higher incidence of prostate cancer than any other marital group, but this could again be explained by age, as there seemed to be no association between marital status and prostate cancer risk after adjusting for age. Stress and risk of prostate cancer The incidence of prostate cancer was highest among men who reported no stress intensity and was lowest among men with high stress (table 8-3). High stress intensity seemed to be associated with a lower risk of prostate cancer (HR=0.38; 95 % CI: 0.091.57), but the estimate was only based on two prostate cancer cases and there was no trend in the dose-response effect (P-value for trend: 0.84). The association between stress frequency and prostate cancer risk was similar to that of stress intensity. A oneunit change in the stress score was not associated with risk of prostate cancer (HR=0.99; 95 % CI: 0.90-1.09). Different periods of follow-up During the 1980s, 41 men developed prostate cancer, while 116 men developed prostate cancer from 1990 and until the end of the study in 2002 (table 8-4). Men with high income, with high education, or who were widowed all seemed to have a slightly higher incidence of prostate cancer in the first follow-up period. Stress seemed to be weakly associated with prostate cancer in an inverse manner in the first follow-up period (HR=0.92, 95 % CI: 0.75-1.14 per one unit increase in the seven-point stress score). 96 However, the associations during this period were based on few cases and no clear trends were found. Prostate cancer incidence did not seem to be associated with neither stress nor any of the sociodemographic variables in the last period of follow-up. Discussion In this racially homogenous population of Caucasians with free access to health care, we found no evidence of social disparities in the incidence of prostate cancer, nor that the risk of prostate cancer was associated with marital status or perceived stress. The fact that we found no social gradient in the incidence of prostate cancer is in contrast to the results from several other large studies.111 In a registry linkage study from Finland, Pukkala and Weiderpass found a higher incidence of prostate cancer among the higher social classes than the lower. The results were consistent over time and most pronounced for localized prostate cancers.114 A social gradient was also found in another large registry-based study from the UK, in which aggregated data on socio-economic factors were used to calculate a deprivation score.112 The authors found that men in the most affluent group had an about 50 percent higher incidence of prostate cancer compared to men in the most deprived group, while prostate cancer survival displayed a negative association with deprivation. In a study of 11,896 prostate cancer cases derived from a cancer surveillance system in Detroit, Schwartz and colleagues found that cases from the highest socio-economic group were more likely to have local stage disease than those from the lowest socio-economic group.110 Access to PSA testing became widely available in most westernized countries in the late 1980s. Socio-economic status distinguishes subgroups within a population from each other by economic opportunities and health awareness, which may affect the access and utilization of the health care system including PSA testing. Dutta and colleagues found that the incidence trend between socio-economic groups primarily started to diverge after PSA testing became widely available.112 Health insurance is strongly associated with undergoing prostate cancer screening. A plausible reason for the lack of a social gradient in the observed prostate cancer incidence in the present study is that Denmark has a free access health care system accessible to all Danes. This supports the hypothesis that the social gradient in prostate cancer observed in some populations is 97 due to differential access and utilization of the health care system, and is not a result of socio-economic status per se. We found no association between stress and risk of prostate cancer in spite of the fact that prolonged stress may affect both the immune2 and hormonal systems,10 both of which could be expected to alter the risk of prostate cancer. Although we used a very different measure of stress, our results are in agreement with the null result of the previously mentioned registry-based study on prostate cancer risk among men who had lost a child to cancer.47 Both breast and prostate cancers are assumed to be hormonedependent diseases, and the findings from breast cancer studies may therefore also be relevant to prostate cancer. In general, major life events such as the death of a child or a divorce have not been associated with a higher risk of breast cancer in large-scale registry linkage studies.44;46;47;49 More chronic exposures to stress, such as work-related stress and perceived stress, have either not been associated with breast cancer or have been associated with a slightly lower risk of breast cancer.53;55;56 We have previously found that the same measure of perceived stress, as the one applied in this study, was inversely associated with the risk of breast cancer among the female participants in the Copenhagen City Heart Study.57 Strengths and weaknesses The prospective design of the Copenhagen City Heart Study ensured temporality between assessment of the exposures of interest and incidence of prostate cancer. Linkage of civil registry numbers to nationwide population-based registers enabled the identification of virtually all cases of diagnosed prostate cancer and allowed for nearly complete long-term follow-up. Several previous studies on social disparities in prostate cancer risk have used ecological measures of socio-economic status derived from census-registries.110;112 This becomes problematic if an individual is assigned to a socio-economic group different from his actual socio-economic position. We used individual measures to avoid this problem. We used self-reported measures of income and education as proxy measures of socioeconomic status, and one may argue that these are only crude measures of a much more complex concept. We agree that a more comprehensive measure of socioeconomic status would be required to reveal a modest social gradient in prostate cancer 98 incidence. However, we still expect that the measures of socio-economic status in this study are sufficient to reveal any major social disparities in prostate cancer incidence. In this study, stress was defined as an individual state of high arousal and displeasure, sometimes referred to as distress.83 By using a measure of perceived stress, we accounted for the fact that each individual has different capacities and ways to cope with stressful situations.16 We assessed perceived stress by combining a question on stress intensity and a question on stress frequency asked at baseline. Using these two questions instead of a more extensive scale may have resulted in some conceptual exposure misclassification. Since the two questions we used as a measure of perceived stress has not yet been validated against a more extensive scale such as the Perceived Stress Scale,84 we cannot fully determine the magnitude of this conceptual misclassification. A disparity between the ideal and the operational measure will have reduced our ability to address the relation between perceived stress and prostate cancer and could possibly explain why we found no such association. However, the same measure of perceived stress has previously been found to be associated with breast cancer57 as well as cardiovascular disease,103 making it less likely that the null finding was due to exposure misclassification. The sociodemographic variables and stress were measured at baseline, and we can therefore not exclude that the association between these variables and risk of prostate cancer may have been partly blurred by non-differential misclassification from changes in socio-economic position, marital status, or stress levels over time. We lacked information on family history of prostate cancer, which is an important risk factor for prostate cancer. However, in order to confound the associations between the exposures of interest and prostate cancer, family history of prostate cancer would have to also affect the person’s sociodemographic position or stress level. We find it unlikely that a family history of prostate cancer would affect a person’s education, income, or marital status, but the awareness of the hereditability of the disease could possibly increase the person’s stress level. Such confounding would create an artificial positive association between stress and prostate cancer, which cannot explain the lack of an association observed in the present study. Even though prostate cancer is the most common cancer in men, only 157 of the 5,496 men developed prostate cancer during the 20 years of follow-up. Lacking power 99 of the statistical analyses may have resulted in unstable risk estimates and could be a possible explanation to the null findings. However, the risk estimates did not indicate any trends or strong associations, and we would therefore not expect more statistical power to have changed our conclusions considerably. Some studies have found that higher socio-economic status was associated with lower stages and grades of prostate cancers at the time of diagnosis.111 We derived information on prostate cancer cases from a national cancer registry and information on the histopathological characteristics of the disease at the time of diagnosis was not available. Thus, we cannot exclude a possible social inequality in the stage of the disease at the time of diagnosis. Conclusion We found no evidence of a social gradient in the incidence of prostate cancer in a population with free access to medical care. Thus, the social gradient observed in other studies is most likely a result of differential usage of PSA testing. If early detection of prostate cancer indeed has an effect on prostate cancer mortality, breaking down barriers for health care utilization and PSA screening may be a major public health issue in populations without free access to medical care. Also, prostate cancer risk did not seem to be associated with either perceived stress of everyday life or with marital status. 100 Table 8-1. Baseline characteristics of the 5,496 men who participated in the second examination of the Copenhagen City Heart Study in 1981-83 Education Study population Low Medium High Persons, n (% of study sample) 5496 2462 (45) 2175 (40) 859 (16) 56 (13) 59 (11) 55 (12) 50 (15) Low income, n (%) 1366 (25) 887 (36) 351 (16) 128 (15) Divorced or unmarried, n (%) 1224 (22) 538 (23) 454 (21) 232 (27) 338 (6) 175 (7) 117 (5) 46 (5) 26 (4) 27 (4) 26 (4) 25 (3) 872 (16) 518 (21) 269 (12) 85 (10) 14 (16) 13 (16) 14 (16) 13 (15) Mean age (SD) High stress, n (%) 2 Mean body mass index, kg/m (SD) Physically inactive, n (%) Mean alcohol consumption, drinks/wk (SD) Table 8-2. Risk of primary prostate cancer associated with sociodemographic variables among 5,496 men who participated in the Copenhagen City Heart Study in 1981-83 Cancers, n Incidence per 100 000 years Age-adjusted HR (95 % CI) Low (n=1366) 51 317 1 (reference) Medium (n=2719) 60 139 0.72 (0.49-1.05) High (n=1411) 46 193 1.17 (0.78-1.76) Income P-value for trend 0.51 Education Low (n=2462) 71 207 1 (reference) Medium (n=2175) 63 183 1.07 (0.76-1.50) High (n=859) 23 160 1.22 (0.76-1.96) P-value for trend 0.42 Marital status Married (n=4035) 122 197 1 (reference) Unmarried (n=639) 12 119 1.03 (0.57-1.86) Divorced/ separated (n=585) 12 145 0.92 (0.51-1.67) Widowed (n=237) 11 419 1.09 (0.59-2.03) Table 8-3. Risk of primary prostate cancer associated with perceived stress among 5,496 men who participated in the Copenhagen City Heart Study in 1981-83 Cancers, n Incidence per 100 000 years Age-adjusted HR (95 % CI) Multi-adjusted HR* (95 % CI) None (n=2586) 88 233 1 (reference) 1 (reference) Light (n=1800) 38 132 0.80 (0.55-1.18) 0.78 (0.53-1.15) Moderate (n=909) 29 206 1.33 (0.87-2.04) 1.30 (0.85-2.00) 2 79 0.41 (0.10-1.68) 0.38 (0.09-1.57) 0.99 0.84 Stress intensity High (n=201) P-value for trend Stress frequency Never (n=3020) 99 223 1 (reference) 1 (reference) Monthly (n=1366) 29 130 0.88 (0.58-1.34) 0.85 (0.56-1.30) Weekly (n=759) 24 205 1.45 (0.92-2.27) 1.41 (0.89-2.23) Daily (n=351) 5 108 0.57 (0.23-1.40) 0.54 (0.22-1.33) 0.97 0.81 1.00 (0.91-1.10) 0.99 (0.90-1.09) P-value for trend Stress score (continuous) 157 * Adjusted for age, income, educational level, marital status, alcohol intake, physical activity in leisure time, and body mass index. Table 8-4. Risk of primary prostate cancer associated with perceived stress and sociodemographic variables according to period of follow-up Baseline -1989 1990-2002 Cancers, n Multi-adjusted HR*(95 % CI) Cancers, n Multi-adjusted HR*(95 % CI) Low 19 1 (reference) 32 1 (reference) Medium 10 0.55 (0.25-1.23) 50 0.69 (0.43-1.09) High 12 1.49 (0.70-3.18) 34 0.96 (0.58-1.59) Income P-value for trend 0.46 0.98 Education Low 20 1 (reference) 51 1 (reference) Medium 15 1.07 (0.55-2.08) 48 1.05 (0.71-1.55) 6 1.37 (0.55-3.41) 17 1.15 (0.66-1.99) High P-value for trend 0.55 0.63 Marital status Married 30 1 (reference) 92 1 (reference) Unmarried 3 1.14 (0.35-3.73) 9 0.97 (0.49-1.92) Divorced/ separated 3 0.90 (0.27-2.96) 9 0.92 (0.46-1.82) Widowed 5 1.36 (0.52-3.56) 6 1.07 (0.46-2.46) 41 0.92 (0.75-1.14) 116 1.00 (0.89-1.12) Stress score (continuous) * The associations between income, education, or marital status and prostate cancer are adjusted for age. The association between stress and prostate cancer is adjusted for age, income, educational level, marital status, alcohol intake, physical activity in leisure time, and body mass index. Figure 8-1. Causal diagram for the relation between socio-economic status, marital status, perceived stress, and risk of prostate cancer Age BRCA1/2 mutations Lucopene Fish intake IGF-1 Selenium/ Vitamin E Socio-economic status Perceived stress Prostate cancer Androgens Marital status Physical activity Alcohol Body mass index Foods high in animal fat Family history of prostate cancer Sexually transmitted diseases IX. Perceived stress and sex steroid hormones: a cross-sectional study Introduction Estrogens and other reproductive hormones play an important role in the etiology of several human cancers.4-6;70;97;115-117 Estrogen-induced proliferation appears to be partly responsible for tumor formation in organs that express estrogen receptors, such as those found in the reproductive system organs and the mammary gland.21;116 Estradiol, the most active estrogen metabolite, can bind to the estrogen receptors, activate gene expression, and thereby increase cell proliferation. By this pathway, estrogens promote growth of already initiated cells. There is also some evidence that estrogens can act as initiating agents themselves by being metabolized into more reactive intermediates that, in turn, may lead to structural changes in the DNA.82 The role of psychological stress in the etiology of hormone-dependent cancers has been an area of emerging interest, in part, because of its suggested ability to alter endogenous levels of sex steroid hormones.10-12 Prolonged low-key stress of everyday life results in a persistent activation of stress-hormones such as glucocorticoids, which have been suggested to participate in the down-regulation of the synthesis of endogenous sex steroid hormones.72 Our hypothesis is that stress leads to lower levels of sex steroid hormones and thereby explain the lower risk of hormone-dependent cancers previously observed among women with chronic stress in some studies.53;55;57 The objective of the present study is to examine a potential association between perceived stress and plasma levels of sex steroid hormones in postmenopausal women. We also aim to assess if a questionnaire-based measure of perceived stress is associated with cortisol, which is often referred to as one of the major stress hormones. 106 Methods Study population Information from the second examination of the prospective Copenhagen City Heart Study in 1981-83 was used to address the objective. Hormone levels were assayed in a random sample of women (n=1,150) drawn from the base population in 1981-1983. A flow diagram of the data selection process is shown in figure 9-1. Women with hormone-dependent cancer prior to baseline, women who took exogenous hormones, and pre-menopausal women were excluded from the analyses. The final analyses included 832 women for the 17β-estradiol analysis and 840 women for the analyses of free testosterone and cortisol. Perceived stress was assessed at baseline and included two questions on stress intensity and stress frequency. A seven-point stress score was created based on these two questions. Please refer to the material and methods chapter for a detailed description of the Copenhagen City Heart study and the measurement of perceived stress. Laboratory analyses The hormone analyses were conducted at a laboratory at the Danish National Research Center for Working Environment during 2006. All blood samples were taken after 30 minutes rest and the date and time for each blood sample were recorded. All samples for the hormone analysis were stored frozen at –20°C. Endogenous hormone levels were measured in serum and all hormones were measured in duplicate and the mean of the two measurements was used in the statistical analyses. The Radio Immune Assays (RIA) used for determination of free testosterone and 17ß-estradiol in serum was Coat-a-count kits purchased from Diagnostic Products Corporation (DPC, Los Angeles, CA). The analyses were carried out according to the manufacturer’s specifications. A 1470 Wizard gamma counter (Wallac, Turku, Finland) was used for measurement of radioactivity. A method evaluation of free testosterone and 17ß-estradiol was performed by use of Con6 Immunoassay Tri-level Controls from DPC at three levels. The method evaluation for free testosterone showed no bias of the method, recovery 97.7 [CI95%: 87.0%;108.3%]. Limit of detection (LOD) was 1.2 pmol/l. The method evaluation for 17ß-estradiol showed no bias of the method, recovery 103.7 [CI95%: 99.0%;108.3%]. Limit of detection (LOD) was 9.8 pmol/l. The assay used for the determination of cortisol in serum was a 107 competitive Radio Immuno Assay (RIA) (Spectria Cortisol Coated Tube RIA) purchased from Orion Diagnostica, Espoo, Finland. A method evaluation of certified reference material in water showed a bias of the method, recovery was 88.8% [CI95%: 79.1%; 98.6%]. LOD was 4.7 nmol/l. To show equivalence between the different runs, reference materials were analysed together with the samples.118;119 Westgard control charts were used to document that the analytical method remained in analytical and statistical control, that is, the trueness and the precision of the analytical methods remained stable. Con6 Immunoassay Tri-level Controls from DPC were used for cortisol, free testosterone, and 17ß-estradiol. Statistical analyses The association between perceived stress and levels of endogenous sex steroid hormones (17β-estradiol and free testosterone) and cortisol was addressed in a cross-sectional study design using blood samples and data on perceived stress from the second examination of the Copenhagen City Heart Study in 1981-1983. Cortisol and the sex steroid hormones are known to have a circadian rhythm, making it important to control for the time of day at which the blood is drawn. We graphically examined the circadian pattern of each hormone and included both a linear and a quadratic term of time of day for the blood draw to adjust for the observed patterns. First, we assessed the median and the 10th and 90th percentiles for each of the hormones for categories of stress intensity, stress frequency, and the combined stress score. Secondly, we regressed levels of sex steroid hormones and cortisol on categories of stress in a linear regression model. Due to non-normal (skewed) distributions and heteroscedastic variances (proportional to level of measurements) in the biomarker data, all concentrations of the biomarkers were analysed on logarithmic scales. The regression models were adjusted for age, body mass index, physical activity, alcohol consumption, tobacco smoking, education, and time of day for blood draw. We included age as linear splines with two knots at ages 55 and 70 in order to adjust for changing associations between age and hormone levels. 108 Results The majority of the women had their blood drawn in the morning between 8 AM and noon (76 %). The median 17β-estradiol value was 24.9 pmol/L with 10th and 90th percentiles of 7.4 pmol/L and 122.5 pmol/L. Estradiol did not show any clear circadian pattern in this study (figure 9-2). The median value of 17β-estradiol was similar in the different categories of stress intensity, stress frequency, and the stress score (table 9-1). After adjustment for age and time of blood draw, women who reported no stress intensity or to never experience stress had the highest level of 17β-estradiol, while women who reported light or monthly stress had the lowest levels. There were no doseresponse associations and the differences were most likely due to chance. Adjustment for a range of potential confounders only slightly changed the risk estimates. The median value of free testosterone was 4.9 pmol/L with 10th and 90th percentiles of 2.3 pmol/L and 9.0 pmol/L and there were also no clear circadian patterns (figure 9-2). Mean levels of free testosterone were not dependent on categories of stress intensity, stress frequency, or the combined stress score (table 9-2). The median cortisol value was 239 nmol/L with 10th and 90 th percentiles of 134 nmol/L and 392 nmol/L. Cortisol showed a circadian pattern with declining values over time during the morning after which it levelled off (figure 9-2). The analyses of stress and cortisol are controlled for this circadian pattern. After adjustment for a range of potential confounders, there were no clear associations between any of the stress measures and cortisol levels (table 9-3). Discussion We found no associations between perceived stress in daily life and endogenous levels of sex steroid hormones in postmenopausal women. Also, perceived stress did not seem to be associated with cortisol, which is hypothesized to be one of the main neuroendocrine mediators of the stress response.13 In contrast to our results, Kroenke and colleagues used caregiving as an indicator of chronic stress, and they found lower levels of estradiol and testosterone among postmenopausal women who provided more than 15 hours of caregiving to an adult per week.55 However, in line with the results from the present study, they also found no relation between self-reported stress and 109 levels of endogenous sex steroid hormones. We have previously found perceived stress to be associated with lower risk of breast cancer in the Copenhagen City Heart Study and we suggested lower levels of endogenous sex steroid hormones among stressed women to a be possible explanation.57 This mechanistic explanation is not supported by the present study. The ovarian synthesis of estrogens is low among postmenopausal women, which may be one reason why we do not observe an association between stress and estrogen levels in this study. If stress indeed impairs estrogen synthesis, it may have been more relevant to address this hypothesis in premenopausal women who have a much higher ovarian synthesis of estrogens. However, it would have been difficult to disentangle the effects of stress from the large fluctuations in estrogen levels during the menstrual cycle among these women. Even though we find no associations between stress and sex steroid hormones, stress hormones may still affect the responsiveness of the target tissues toward estrogen stimulation so that similar levels of estrogen may lead to different rates of cell proliferation depending on the level of stress hormones. For example, experimental evidence has shown that stress may produce a direct effect on the uterus by changing the response of its structures to estrogen.89-91 This hypothesis cannot be addressed in the present study, but may be important to keep in mind when addressing the relation between stress and hormone-dependent cancers. Estradiol, testosterone, and cortisol all have intrinsic biological variability and cortisol also shows a clear circadian pattern. This may have decreased the reliability of single measurements of hormonal levels, as the one included in this study, and may have blurred a possible weak association between stress and hormonal levels. Several measures of hormones at different times of the day in each individual would have been preferable. On the other hand, such an approach would have been much more time consuming and expensive and would possibly not have been feasible to accomplish in a large population study such as the one at hand. In support of the validity of the results, we found body mass index to be associated with higher levels of estradiol levels (data not shown), similar to what would be expected based on results from other studies.120 The blood samples were first assayed more than 20 years after collection. During this period they have been stored at -20°C. Steroid hormones, in general, seem to be 110 relatively stable over time at this temperature,121 and we therefore do not expect this time span to have seriously distorted our internal comparisons of hormone levels between stress categories. However, due to the long-term storing, a higher proportion of the sex steroid hormones maybe in an unbound form compared to other studies.122 Perceived stress was assessed by two questions on stress intensity and stress frequency measured at baseline. This may not be the optimal way to assess a complex phenomenon such as stress and we cannot exclude the possibility that other measures of stress could be associated with sex steroid hormones. However, in a recent study, two single-item questions on stress were found to be just as valid as three fully validated multi-item measures on perceived stress.101 Also, perceived stress has been related to other health outcomes in the present study population.57;102;103 However, perceived stress was not associated with cortisol levels in this study. One reason may be that we only had a single measure of cortisol, which is not necessarily a valid presentation of an individual’s cortisol profile. If stress for examples leads to a blunted circadian cortisol rhythm, this will not be captured by a single measure. Hormone levels were only assayed in a random sample of the total study population, and one may be concerned that lacking statistical power is a possible explanation to the null findings. Both the estradiol and testosterone analyses indicated lower levels of these hormones among stressed women. However, there were no doseresponse trends and the observed differences were only modest. In conclusion, we found no evidence of a relation between perceived stress and levels of sex steroid hormones among postmenopausal women. In spite of measurement problems, which may have blurred a relation between stress and hormone levels, stressinduced impairment of estrogen synthesis is not supported as a possible etiologic link between psychological stress and risk of hormone-dependent cancers in this study. 111 Table 9-1. Median and relative difference in geometric mean of 17β-estradiol by category of stress among 832 postmenopausal women who did not use hormones Age-and-time adjusted relative difference Multi-adjusted relative difference in No. Median 17β-estradiol, in geometric mean of 17β-estradiol, geometric mean of 17β-estradiol*, (10 -90 percentiles) % (95% CI) % (95% CI) pmol/L th th Stress intensity None 312 26 (8; 150) Ref Ref Light 283 22 (8; 114) -17 (-31; -1) -18 (-32; -1) Moderate 179 25 (7; 126) -12 (-28; 9) -12 (-29; 9) 58 24 (8; 89) -11 (-35; 23) -12 (-36; 22) Never 403 26 (8; 113) Ref Ref Monthly 198 22 (8; 128) -13 (-29; 6) -17 (-32; 1) Weekly 147 25 (7; 142) -8 (-26; 14) -11 (-29; 10) 84 26 (6; 124) -3 (-26; 27) -5 (-27; 25) Low 388 26 (8; 113) Ref Ref Medium 360 24 (7; 129) -9 (-23; 8) -11 (-25; 5) 84 25 (7; 123) -3 (-26; 28) -4 (-27; 26) High Stress frequency Daily Stress score High * Adjusted for age, body mass index, physical activity, alcohol consumption, tobacco smoking, education, and time of day for blood draw Table 9-2. Median and relative difference in geometric mean of free testosterone by category of stress among 840 postmenopausal women who did not use hormones Age-and-time adjusted relative difference Multi-adjusted relative difference in No. Median free testosterone, in geometric mean of free testosterone, geometric mean of free testosterone*, (10 -90 percentiles) % (95% CI) % (95% CI) pmol/L th th Stress intensity None 315 4.9 (2.4; 9.4) Ref Ref Light 284 4.9 (2.4; 9.2) -2 (-11; 9) -3 (-12; 9) Moderate 179 4.6 (2.1; 8.7) -9 (-19; 3) -11 (-21; 0) 62 5.0 (2.5; 7.5) -6 (-21; 12) -9 (-23; 8) Never 405 4.8 (2.3; 9.0) Ref Ref Monthly 199 5.0 (2.5; 9.0) 1 (-10; 13) 0 (-10; 12) Weekly 149 4.6 (2.2; 8.8) -6 (-17; 6) -7 (-17, 5) 87 5.0 (2.5; 9.3) 2 (-12; 18) -1 (-14; 15) Low 391 4.8 (2.3; 9.0) Ref Ref Medium 361 4.8 (2.2; 8.8) -4 (-12; 5) -5 (-13; 4) 88 5.0 (2.5; 9.3) 0 (-14; 16) -2 (-16; 13) High Stress frequency Daily Stress score High * Adjusted for age, body mass index, physical activity, alcohol consumption, tobacco smoking, education, and time of day for blood draw Table 9-3. Median and relative difference in geometric mean of cortisol by category of stress among 840 postmenopausal women who did not use hormones Age-and-time adjusted relative difference Multi-adjusted relative difference in No. Median cortisol, nmol/L (10 th-90 th percentiles) in geometric mean of cortisol, geometric mean of cortisol*, % (95% CI) % (95% CI) Stress intensity None 315 230 (135; 389) Ref Ref Light 285 258 (149; 403) 7 (0; 14) 5 (-1; 13) Moderate 180 231 (124; 382) -2 (-9; 6) -3 (-9; 5) 60 234 (132; 387) -1 (-11; 11) -1 (-11; 11) Never 404 230 (137; 392) Ref Ref Monthly 200 248 (139; 392) 4 (-4; 11) 2 (-5; 10) Weekly 150 262 (131; 383) 5 (-3; 13) 4 (-4; 12) 86 234 (140; 406) 3 (-7; 13) 2 (-7; 12) Low 391 234 (139; 392) Ref Ref Medium 362 248 (133; 391) 2 (-4; 8) 1 (-5; 7) 87 234 (138; 395) -1 (-10; 8) -2 (-11; 8) High Stress frequency Daily Stress score High * Adjusted for age, body mass index, physical activity, alcohol consumption, tobacco smoking, education, and time of day for blood draw Figure 9-1. Flow diagram of the data selection process 12698 participants in the Copenhagen City Heart Study 7018 women - 207 women with hormone-dependent cancer prior to baseline 6811 women - 1716 premenopausal women 5095 women - 86 women with no blood samples 5009 women - 127 women without information on stress or other covariates 4882 women Random sample of 1150 women - 86 blood samples not found 1064 women - 214 women who used hormone theraphy 850 women - 10 (-18) because ofinsuffient material for the analyses Estradiol: 832 women Testosterone: 840 women 115 Cortisol: 840 women Figure 9-2. Circadian pattern for 17β-estradiol, free testosterone, and cortisol 0 Fitted values/S_Estradiol 100 200 300 400 500 Estradiol 5 6 7 8 9 10 11 noon 1 2 Time of day, hour Fitted values 3 4 5 6 7 4 5 6 7 4 5 6 7 S_Estradiol 0 Fitted values/S_testosteron 5 10 15 20 25 Free testosterone 5 6 7 8 9 10 11 noon 1 2 Time of day, hour Fitted values 3 S_testosteron 0 Fitted values/S_kortisolnmolL 200 400 600 800 Cortisol 5 6 7 8 9 10 11 noon 1 2 Time of day, hour Fitted values 116 3 S_kortisolnmolL X. Stress and hormone-dependent cancers: a discussion of the evidence Is there a relation between stress and hormone-dependent cancers? Perceived stress and breast cancer Previous prospective studies on stress and breast cancer have reported inconsistent results. In general, exposure to major stressors such as death of a child or divorce was not associated with a higher risk of breast cancer in large-scale registry-linkage studies.44-49 It is unclear whether an accumulation of stressful life events is associated with higher risk of breast cancer.59 Previous prospective studies that addressed the effect of work-related stress or perceived stress either did not find an association with breast cancer risk or found indications of a lower risk of breast cancer among stressed women.53;55;56 In the Copenhagen City Heart Study, we found high levels of perceived stress to be associated with a lower risk of breast cancer in an inverse dose-response manner. We found that women who reported high levels of stress in their daily life were almost half as likely to develop breast cancer compared to women who reported no stress. It was hypothesized that stress could lead to a suppression of estrogen synthesis and thereby reduce the risk of breast cancer, which is highly dependent on endogenous levels of estrogens.70 Given that this hypothesis is correct, we would expect to also see a lower risk of other hormone-dependent cancers among stressed women. Perceived stress and endometrial cancer Endometrial cancer is another clearly estrogen-dependent cancer in women. Previous studies on stress and endometrial cancer have only addressed the effect of major stressful life events and these studies have solely been based on information from national registries. None of the studies found any clear evidence of an association between stressful life events and risk of endometrial cancer. 45-48 As argued in the introduction chapter, the health consequences of acute severe stress from stressful life events can differ considerably from the health consequences of the more chronic stress 117 experienced in daily life. We therefore found it relevant to address the association between perceived stress and risk of endometrial cancer in the Copenhagen City Heart Study. Although few women in the cohort developed endometrial cancer during followup, we still found an inverse dose-response association between stress and risk of endometrial cancer similar to the one observed for breast cancer. This association was particularly strong among women who received hormone therapy and in women of normal weight. Some experimental studies have shown that stress hormones may impair both estrogen synthesis, as hypothesized in the breast cancer study, as well as make the target organs less responsive to estrogen stimulation.10;81 In this cohort of middle-aged women where most were postmenopausal and therefore had a low ovarian synthesis of estrogens, the lower responsiveness of the target organs toward estrogen stimulation may explain the stronger effect of stress in women who received hormone therapy. Overweight women, on the other hand, may be in a pro-inflammatory state that facilitates carcinogenesis. This alternative pathway would counteract the stress-induced suppression of estrogens and thereby explain the less pronounced effect of stress among overweight women. The similar results found for breast and endometrial cancer provide some support to the hormone hypothesis stating that impairment of estrogen synthesis and metabolism may be an etiological explanation to the lower risk of these two malignancies observed among stressed women. Perceived stress and colorectal cancer Colorectal cancer seems to also have a hormonal component, although less pronounced than in breast and endometrial cancers. In previous studies on stress and colorectal cancer, stress from acute severe stressors such as the loss of a child or a spouse did not markedly increase the risk of colorectal cancer, while there were some evidence to support that more prolonged stressors such as major work-related problems was associated with higher risk. 45;46;48;65;67 The association between perceived stress and colorectal cancer has only been addressed in a cohort study once before.69 This earlier study assessed colorectal cancer mortality as the endpoint, which did not allow for a distinction between etiologic and prognostic factors. We addressed the association between perceived stress and risk of colorectal cancer in the Copenhagen City Heart Study and found sex-differences in the results. In women, higher stress was associated 118 with a lower risk of colon cancer in particular, while there was no clear relation between perceived stress and colon or rectal cancer in men. This may indicate different etiologic pathways for colorectal cancer in men and women, and it may again lend some support to the hormone hypothesis in women. Such sex-differences have, however, not previously been reported and the analyses should be replicated in other cohorts before conclusions can be drawn. Perceived stress and prostate cancer Prostate cancer is assumed to be the primary hormone-dependent cancer in men, although a clear relation between testosterone levels and prostate cancer risk still remains to be established.9 One previous registry-linkage study found no increased risk of prostate cancer among men who had experienced having a child with cancer.47 Similarly, we did not find any support of a relation between stress and risk of prostate cancer in the Copenhagen City Heart Study. Perceived stress and endogenous sex steroid hormones In the Copenhagen City Heart Study, perceived stress was associated with lower risk of breast, endometrial, and colon cancer in women, while there was no clear evidence of a relation between perceived stress and colorectal or prostate cancer in men. This lower risk of hormone-dependent cancers consistently observed among stressed women provide some support to the hypothesis that stress hormones may impair estrogen synthesis as well as make the target organs less sensitive to estrogen stimulation. Following in the line of this hypothesis, we investigated whether women who reported high levels of stress also had lower endogenous levels of estrogens or other sex steroid hormones at baseline compared to less stressed women. We therefore measured estradiol and testosterone in a sub-sample of the postmenopausal women from the Copenhagen City Heart Study. Women with high levels of stress did not have lower endogenous levels of sex steroid hormones. This may either be due to methodological flaws in the measurement or it may question the underlying hormone hypothesis. If stress indeed impairs estrogen synthesis it may have been more relevant to address the hormone hypothesis in premenopausal women. However, this would have led to difficulties in disentangling the effect of stress from the large fluctuations in estrogen levels during the 119 menstrual cycle, especially as only one measurement of estrogens was available in the present study population. Summary Perceived stress seems to be associated with lower risk of hormone-dependent cancers among women in the Copenhagen City Heart Study, but the findings remain to be confirmed in other cohort studies and the underlying mechanisms need to be further explored. The strength of the Copenhagen City Heart Study The Copenhagen City Heart Study is one of few cohort studies that contains both information on perceived stress and has a sufficient follow-up time to address specific cancer types separately. In addition, there were blood samples available on the majority of the participants. The prospective design of the Copenhagen City Heart Study also ensured temporality between self-reported stress and incidence of hormone-related cancers, which is very important in stress research where recall problems may constitute a major problem. The cohort is a large random sample of the general population of Copenhagen, and the attendance rate was relatively high. Linkage of civil registry numbers to nationwide population-based registers enabled identification of virtually all hormone-dependent cancer cases and allowed for nearly complete long-term follow-up. This made the Copenhagen City Heart Study a valuable data-source for the study of stress and risk of hormone-dependent cancers. In this cohort, we found perceived stress to be associated with lower risk of hormone-dependent cancer in women, but not in men. Despite the strength of the Copenhagen City Heart Study, the observed associations may still deviate from the causal relations of interest by systematic processes, such as selection bias, misclassification of exposure or disease, and confounding, as well as by random processes (random errors). The goal was to measure the associations as accurately as possible, but errors in estimation are unavoidable and will therefore be given attention in the following sections. 120 Selection bias and external validity In cohort studies, as the ones applied in the dissertation, loss to follow-up is the main source of selection bias, whereas non-participation in the study at time of initiation constitutes a question of external validity. A small proportion (less than 0.1 %) of the women were lost to follow-up, and strong selection bias is therefore unlikely. Seventy percent of the originally invited cohort participated in the second examination of the Copenhagen City Heart Study. A decision not to participate could well be related to stress level, since high stress may limit the time available for participating in surveys. Thus, the results of the included studies on stress and hormone-dependent cancers cannot necessarily be generalized to groups with extremely high levels of stress, as these individuals may not be represented in the study population. Also the results cannot necessarily be generalized to younger age groups or to other ethnical groups. We studied the relation between perceived stress and risk of hormone-dependent cancers in the Copenhagen City Heart Study. The associations observed in this cohort may differ from what might have been observed in another cohort. We tried to integrate evidence from other cohorts before drawing conclusions in each of the studies, but we cannot completely refuse that some of the results may be specific to this cohort. There is a need to replicate the studies in other study populations. Misclassification of perceived stress The impact and magnitude of misclassification of perceived stress will be addressed by various means. None of these can fully determine the exact amount of exposure misclassification, but they can, in concert, provide a picture of the magnitude and direction of this type of bias. There are two ways by which the ideal and the operational measure of exposure can deviate from one another.104 The first, conceptual exposure misclassification arises when the operational measure addresses something different from what is intended. That is, the ascertained measure does not resemble what is of etiological interest (the ideal measure). The other way is more often addressed in traditional approaches to evaluate exposure misclassification, and it deals with errors in implementing the chosen operational measure. This includes concerns about recall bias, 121 underreporting, technical imprecision, etc. Both types of misclassification need equal attention. Perceived stress has been the construct of etiological interest (the ideal measure) in the dissertation. Every disparity between this ideal measure and the way it is operationalized and measured may have biased the assessment of the causal relations of interest. The question is how widespread this kind of bias has been. Two different measures of perceived stress (intensity and frequency) were obtained in the Copenhagen City Heart Study, both resembling different aspects of the concept of interest. Neither of the two measures can be said to be better than the other, but by combining them into one stress-score we hoped to create a measure with less conceptual error. On the other hand, one of the measures may be measured with less precision than the other, which would result in the combined score being less precise than at least one of the measures alone. In each of the studies we therefore chose to both present the results for each of the measures alone as well as for the combined score. Using the two questions on stress intensity and stress frequency instead of a more extensive scale may have resulted in some conceptual exposure misclassification. The question is whether the two questions on stress intensity and stress frequency did capture the intended ideal measure of perceived stress to an acceptable degree. Since the two questions used as an operational measure of perceived stress has not yet been validated against a more extensive scale such as the Perceived Stress Scale,84 we cannot fully determine the magnitude of this conceptual misclassification. A disparity between the ideal and the operational measure will have reduced our ability to address the hypothesized etiologic relations between perceived stress and hormone-dependent cancers, and by using only two measures of stress intensity and stress frequency instead of a more extensive scale an even stronger relation between perceived stress and risk of hormone-dependent cancers may have been blurred. The ability to retrieve known health consequences of perceived stress in the Copenhagen City Heart Study would give some idea about the degree of exposure misclassification. Perceived stress is a relatively new measure of interest in stress research, and known health consequences of perceived stress therefore remain to be established. However, we have previously found perceived stress to be associated with higher risk of stroke and ischemic heart disease in the same cohort,102;103 which provide 122 some modest confidence that the operational measure applied in this cohort resembled the ideal measure of perceived stress. Further, a dose-response gradient may support a causal relation between the operational measure and disease and thereby support a relation between the ideal measure of etiological interest and the applied operational measure, while the opposite is not necessarily the case.104 An inverse dose-response relation was found for the associations between perceived stress and risk of breast cancer, endometrial cancer, and colon cancer in women. One also has to identify the etiologically relevant time window for exposure, because inclusion of irrelevant periods of exposure would also constitute exposure misclassification.104 Perceived stress was the measure of etiologic interest in this study, but whether this includes stress experienced until shortly before diagnosis or only prolonged stress, for example ten years prior to diagnosis, is unclear. In order to address this question, we have assessed the relation between perceived stress and risk of hormone-dependent cancer in different intervals of follow-up in each of the sub-studies. In most of the studies the associations were strongest in first nine years of follow-up. One of two explanations, or a combination of both, may apply. The etiologically relevant time frame for stress exposure is primarily within the nine years prior to diagnosis. Alternatively, perceived stress was only assessed at baseline and may have changed over time in a manner that is most likely independent of subsequent incidence of cancer. Such non-differential exposure misclassification may have reduced our ability to detect a potential relation between stress and risk of hormone-dependent cancers and thereby explain the attenuated associations observed in the last period of follow-up. Which explanation that applies cannot be clarified from the present data, but it is probably a combination of both. Misclassification of outcome measures In countries with nationwide morbidity and mortality registries, like the Danish, it is often assumed that disease misclassification is a minor problem. It may indeed be a much smaller problem than in studies where disease status is obtained by, for example, self-report, but it may still exist. Sources of disease misclassification may, analogue to exposure misclassification, be divided into being conceptual or practical in nature. 123 Conceptual error may arise in the very definition of the disease, while errors in registration and in the actual process by which individuals come to be identified as cases arises in the implementation of the case definition.104 We have chosen to address each of the site-specific types of hormone-dependent cancers separately in order not to combine subsets of disease with different etiologies. This approach has hopefully reduced the problem of conceptual disease misclassification. Reporting of new cancer cases to the registry is compulsory in Denmark and, according to the Danish National Board of Health, the Danish National Cancer Registry contains data on more than 95 percent of all cancer diagnoses in Denmark.123 Disease misclassification is therefore unlikely to have severely affected our results. Estradiol, testosterone, and cortisol all have an intrinsic biological variability, which may have decreased the reliability of single measurements of hormonal levels, as the one applied in the study that assessed the association between stress and sex steroid hormones. This may also be a likely explanation to the finding of no association between stress and sex steroid hormones in this study. Several measures of hormones at different times of the day in each individual should be the aim of future studies on this relation. Confounding In this and most other epidemiologic studies, an unexposed group is chosen to provide an estimate of what the experience of the exposed group would have been, had they not been exposed. Confounding is said to be present whenever the disease experience of the unexposed group differs from what it would have been in the exposed group, had they not been exposed.104 Thus, a central question in the present study is: Did individuals with low levels of stress (the reference group) have the risk of cancer that individuals with higher levels of stress would have had, if their stress level had been low? In other words, is the observed baseline risk of hormone-dependent cancers in the unexposed group exchangeable with the hypothetical counterfactual baseline risk of hormonedependent cancers in the exposed groups? If not, the measure of association comparing the exposed and unexposed groups would not have been a valid estimate of the causal effect of interest. In order to address this possible non-exchangeability, we tried to identify covariates that could serve as markers of this non-exchangeability (potential 124 confounders) in each of the studies.104 A covariate cannot be a confounder unless 1) it can causally affect disease risk within exposure groups, 2) it is distributed differentially among the compared groups, and 3) it is not an intermediate variable in the causal pathway between exposure and disease.124 These necessary conditions for a confounder apply to a source population of individuals at risk of becoming cases. Identification of potential confounders, using these necessary conditions, is commonly achieved by a stepwise selection procedure, where covariate-disease associations are observed in data. This is not automatically a sound approach because we cannot be sure that the observed associations themselves are not confounded or otherwise biased. According to Greenland and Morgenstern, we should therefore rely on prior knowledge of causal effects when identifying confounders in a study.124 There are various ways by which to control for potential confounding. Confounding can either be controlled directly in the study design (e.g., restriction, randomization) or in the statistical analysis (e.g., stratification, using regression models). The last approach was applied in the present studies, where the estimated associations between perceived stress and risk of hormone-dependent cancers were controlled for confounding effects of other covariates in multivariate regression models. The statement that the adjusted measure of association is a valid estimate of the causal effect depends on two assumptions; 1) that the variables available for the analysis were sufficient, and 2) that they were adequately measured and categorized.124 Neither of the assumptions can be proven to be true, but we used sensitivity analyses and evaluation of residual confounding to evaluate the impact of these assumptions in each of the studies. A set of confounders was identified based on causal diagrams in each of the studies. However, the proposed set of potential confounders would not be adequate if the assumed causal model was wrong. It is, unfortunately, never possible to be completely certain about the assumed model, but by explicitly stating the assumptions made either in a causal diagram or by verbalizing them, it is possible for other investigators to acknowledge and discuss the assumptions. Further, alternative assumptions and their effect on the relation of interest can be addressed in sensitivity analyses and thereby provide important information on the impact of the assumptions. Such sensitivity analyses were performed in several of the studies in order to address the impact of our assumptions. As an example, we assumed that hypertension was a possible intermediate 125 on the pathway from perceived stress to endometrial cancer, and hence we did not adjust our analyses for this variable. Some investigators may not agree with this assumption, and in order to address the impact of this assumption we performed a sensitivity analysis where we adjusted for hypertension. Such adjustment did not change the risk estimates, which suggests that the results were not very sensitive to this particular assumption. Information on several important risk factors for hormone-dependent cancers was not obtained in the Copenhagen City Heart Study. The fact that we did not have information on family history of the specific hormone-dependent cancers in first-degree relatives may be especially of concern. However, we would not expect perceived stress to be strongly associated with family history of hormone-dependent cancers, and strong confounding from this important risk factor is therefore unlikely. We also lacked information on other important risk factors for the site-specific hormone-related cancers. We have discussed the impact of this lacking information in each of the studies and in general, we concluded that confounding from unmeasured covariates were unlikely to explain the inverse relations between perceived stress and hormone-dependent cancers observed in women. Misclassification of confounders may result in incomplete statistical control from covariates included in the analysis and thereby lead to residual confounding.104 Some variables are more prone to this kind of misclassification than others. Day of birth was derived from the nationwide Central Person Registry, and a variable like age is therefore unlikely to be subject to a great amount of misclassification. The same is evident for sex. On the contrary, a concept like socio-economic status is more prone to both conceptual and measurement error. We used self-reported education and income as proxy-measures for socio-economic status. Undoubtedly, these proxy-measures did not fully capture the underlying concept of etiologic interest, and residual confounding was inevitable. Socioeconomic status could just as well have included measures of wealth, position, etc., and omission of such aspects of the concept makes the assessment incomplete. The amount of residual confounding from a variable is, however, often proportional to the amount of confounding originally present by the specific factor in question, and evaluating the change in effect estimates after adjustment for the incomplete operational measure can give us some idea of the magnitude of distortion.104 Adjustment for socio-economic 126 status did not substantially change the results in any of the studies and thereby make residual confounding from this variable less of a concern. Random error Even in as large a cohort study as the Copenhagen City Heart Study where the participants were followed for almost two decades, the number of hormone-dependent cancer cases was limited. We found inverse dose-response associations between perceived stress and risk of breast, endometrial, and colon cancers in women. This may provide some comfort that the studies were sufficiently powerful to address our main hypothesis, namely that the risk of hormone-dependent cancers differed between stress groups. However, power problems arose as soon as we wanted to conduct subgroup analyses in order to identify susceptible subgroups, and the results of these analyses should therefore be interpreted with caution. 127 XI. Conclusion Perceived stress was consistently associated with lower risk of breast, endometrial, and colon cancer in a dose-response manner among women participating in the Copenhagen City Heart Study. This supports the existence of a common causal pathway between perceived stress and these hormone-dependent cancers. We proposed a hormone hypothesis, where stress impairs the synthesis of estrogens, as a possible explanation. This hypothesis was not supported in a cross-sectional study conducted in a subset of the women, where we found no differences in endogenous levels of sex steroid hormones among women with different stress levels. We could not determine if the lacking association between stress and sex steroid hormones was in fact valid or a consequence of an insufficient measurement of the hormone levels. Alternatively, stress may also render the target tissues less sensitive toward estrogen stimulation. This part of the hormone hypothesis was not tested in the dissertation. Further, there were no clear evidence of a relation between perceived stress and risk of colorectal or prostate cancers in men. Prior biological, psychological, and epidemiological knowledge was used to identify causal models for the statistical analysis. Such an approach ensured transparency of the applied assumptions and increased the efficiency of the statistical model by making it possible to identify a minimum sufficient set of confounders based on the stated causal model. Due to almost complete follow-up, selection bias was unlikely to have distorted the results of the studies. However, misclassification of perceived stress may be of concern. The applied measure of stress intensity and stress frequency may not have fully captured the concept of etiologic interest, namely perceived stress. This misclassification of perceived stress is likely to have resulted in an underestimation of possibly stronger relations between perceived stress and risk of hormone-dependent cancers. Further, the magnitude of the effect was diluted by the long period of follow-up and possible changes in stress levels during follow-up. Only a few participants developed hormone-dependent cancer during follow-up, which resulted in the lack of sufficient statistical power in some of the studies and especially in the subgroup analyses. Drawing firm causal conclusions from epidemiological studies are problematic because the empirical results will always be on a continuum of uncertainty. 128 We tried to integrate the empirical evidence from this study with prior scientific knowledge about the relation between stress and risk of hormone-dependent cancer. Based on this approach we can conclude that perceived stress seems to affect the risk of hormone-dependent cancers in women, but the findings from the Copenhagen City Heart Study remain to be confirmed in other cohort studies and the underlying causal mechanisms need to be further explored. 129 XII. Public health implications Is stress a public health problem? Perceived stress is quite prevalent in most western societies and even a relatively small change in risk could therefore have a large public health impact at the population level. We found perceived stress to be associated with lower risk of hormone-dependent cancers in women. Hopefully, these results may prevent some women from blaming their own stressful life style if they are diagnosed with cancer. However, it is also important to emphasize that stress cannot be considered a healthy response. Stress is not a desirable state to be in neither mentally nor physically and it may lead to the development of diseases other than hormone-dependent cancers, such as cardiovascular diseases. Whether stress affects the risk of non-hormone-dependent cancers is still an area of debate. There is considerable experimental evidence indicating that stress plays a role in the development, course, and outcome of tumors in animals.125 An increased secretion of glucocorticoids in an acute stress situation is found to suppress the function of the immune system and thereby reduce its ability to recognize and destroy neoplastic cell growth.14 The evidence of a relation between stress and cancer is less consistent in humans,125 which may partly be due to different mechanisms working in opposite directions. Apart from cancer, stress in its varying kinds may have a range of other health consequences. An association between stress and risk of cardiovascular diseases is both biologically plausible and empirically supported in a range of observational studies.105;126-128 In a recent large case-control study that included cases and controls from 52 countries, a higher risk of myocardial infarction was reported among individuals who experienced stress at home or at work, were under severe financial stress, had experienced stressful life events in the past year, or were depressed.126 Stress seems to activate the sympathetic nervous system, with various metabolic effects: increased blood pressure, pulse rate, and platelet aggregation; reduction in insulin sensitivity; and promotion of endothelial dysfunction.105 Stress may also lead to changes in smoking, 130 diet, alcohol consumption, and level of physical activity, and thereby indirectly influence the risk of cardiovascular diseases.93;129-131 It is well established that stress measured by both self-report and objective life events is associated with increased susceptibility to infectious diseases.132 This risk is mediated by both direct modulation of functional and enumerative aspects of immunity as well as by indirect changes in health-related behavior. However, a marked variability among individuals in immune response to stress has been noted. This has primarily been ascribed to the fact that each individual has different abilities to cope with stressors, which again makes stress appraisal more predictive than external stressors.132 Stress is also associated with a higher risk of depression.133;134 Chronic stress decreases the brains sensitivity to circulating glucocorticoids and thereby its normal role in down-regulating increased levels. The neurotransmitters serotonin, norepinephrine, and dopamine are all suspected to play a part in the etiology of depression. Glucocorticoids can alter the synthesis, the breaking-down process, and the number of receptors for each of these neurotransmitters, which may explain a link between prolonged stress and depression. By itself, depression is a major public health problem because of its severity, prolonged impact on quality of life, and relatively high prevalence in the population. Further, epidemiologic studies evaluating the relation between depression and ischemic heart disease have consistently found a positive association between major depression episodes and incidence of cardiac events.135 In sum, stress is associated with an increased risk of common diseases such as colds and other infectious diseases, depression, cardiovascular disease, and probably some types of non-hormone-dependent cancers. Even if stress is only related to a modestly increased risk of each of these diseases, the impact of stress on these may more than counteract the possible protective effect of stress on hormone-dependent cancers among women. To illustrate this, I calculated the absolute impact of perceived stress on ischemic heart disease and breast cancer, respectively, among women in the Copenhagen City Heart Study (table 12-1). I found that perceived stress prevented about 14 percent of all the breast cancer cases that would have occurred among women in the study and that only seven percent of all cases of ischemic heart disease were attributable to perceived stress. However, since ischemic heart disease is a more common disease than breast cancer, this amounts to about 71 cases of ischemic heart disease being attributable 131 to perceived stress, while only about 35 cases of breast cancer were prevented by perceived stress among women in the Copenhagen City Heart Study. These calculations are based on the assumptions that the effect estimates represented valid estimates of causal effects, and that removing exposure to stress would not have affected the size of the population at risk. These assumptions are seldom completely valid and the example should therefore only be used to illustrate that a modestly increased risk of a common disease like ischemic heart disease may more than counteract a markedly lower risk of a relatively seldom disease like breast cancer, when we address the actual caseload. In general, I would expect the total burden of disease attributable to perceived stress to far exceed the relatively few cases of hormone-dependent cancers that may be prevented by stress. Superimposed on the health consequences of stress come economic considerations such as days lost through sickness, lost earnings, and hospital expenditures. This makes stress a major public health problem and emphasizes the importance of public health initiatives to remedy the problem. Future population studies on stress and risk of hormone-dependent cancers In the Introduction section, stress was defined as the individual’s appraisal of an imbalance between demands and the individual’s resources to cope with it.16 This measure of etiologic interest was probably not fully captured by the self-reported information on stress intensity and stress frequency in the Copenhagen City Heart Study. Using only two questions may have resulted in more conceptual misclassification than if a more extensive scale of stress perception had been applied. One scale that would be relevant to include in future studies of the relation between perceived stress and risk of hormone-dependent cancer is the Perceived Stress Scale (PSS). Cohen and colleagues proposed this scale in 1983 as an attempt to develop a psychometrically valid measure of perceived stress.84 The scale is aimed at measuring the degree to which situations in one’s life are perceived as stressful. Items are designed to measure how unpredictable, uncontrollable, and overloaded respondents find their life.84 The original scale included 14 items, but a shorter version including 10 items has been validated in a probability sample of the US population and appears to provide at least as good a measure as does the longer one.136 The questions included in the PSS10 are shown in figure 12-1. The 132 PSS10 has good construct validity and high internal reliability (Cronbach’s alpha coefficient = 0.78). It can be administered in few minutes and is easy to score. The total score is just the sum of the score of each question with positive questions scored reversed. Further, the Perceived Stress Scale provided better predictions of psychological and physical symptoms than did life-event scales in the validation study conducted by Cohen and Williamson.136 One problem with using this scale in future prospective studies is that level of perceived stress will be influenced by changes in daily hassles, major life events, and changes in coping resources, and that the predictive value of the scale is therefore expected to fall over time.136 This is, however, also a problem if one measures “objective” stressors, such as daily hassles, and should therefore be considered an inherited problem in stress research in general and not a problem specific to measuring perceived stress. One way to address this problem would be to use repeated measures of perceived stress during follow-up. In order to get a more accurate assessment of the relation between perceived stress and risk of hormonedependent cancers, future studies should ideally be prospective in design, be large enough to ensure sufficient statistical power, include repeated measures of perceived stress using the Perceived Stress Scale or another comprehensive stress measure, and include a better measurement of the sex steroid hormones in order to address a possible hormonal pathway between stress and hormone-dependent cancers. In an ideal study with information on stressors and perceived stress at different time-points in life, it would also be interesting to address the health consequences of stress in a life course perspective. Some of the studies indicated that the effect of perceived stress might be confined to specific groups and these differences should be more directly addressed in future studies with more statistical power. This dissertation has mainly focused on stress as a potential risk factor for hormone-dependent cancers, but in future studies it may also be interesting to assess whether stress affects the prognosis of these cancers. 133 Table 12-1. Estimates of the number of ischemic heart disease and breast cancer there can be attributed/ prevented by perceived stress among women in the Copenhagen City Heart Study. Breast cancer** Ischemic heart disease* Cases Multi-adjusted HR (95 % CI) Cases Multi-adjusted HR (95 % CI) Low stress 447 1 (reference) 120 1 (reference) Medium stress 435 1.08 (0.94-1.23) 112 0.80 (0.62-1.04) High stress 129 1.41 (1.16-1.72) 19 0.60 (0.37-0.97) Attributable/ prevented fraction in the population Attributable/ prevented number AF = 1 – ((447/1011)/1 + (435/1011)/1.08 + PF = 1 – (1/((120/251)/1 + (112/251)/0.80 + (129/1011)/1.41) = 0.07 (19/251)/0.60) = 0.14 a = 0.07 (1011) = 71 a(0) = 0.14 (251) = 35 * The risk estimates are taken from a study on perceived stress and ischemic heart disease in the Copenhagen City Heart Study, which has previously been published. Reference: Nielsen NR, Kristensen TS, Prescott E, Strandberg Larsen K, Schnohr P, Grønbæk M. Perceived stress and risk of ischemic heart disease: Causation or bias? Epidemiology 2006;17:391-397 ** The risk estimates are taken from the study on perceived stress and breast cancer in the Copenhagen City Heart Study, which is included in the dissertation. Reference: Nielsen NR, Zhang ZF, Kristensen TS, Netterstrøm B, Schnohr P, Grønbæk M. Self-reported stress and risk of breast cancer. BMJ 2005;331:548. Figure 12-1. Questionnaire used for the Perceived Stress Scale The questions in this scale ask you about your feelings and thoughts during the last month. In each case, you will be asked to indicate by circling how often you felt or thought a certain way. 0 = Never 1 = Almost Never 2 = Sometimes 3 = Fairly Often 4 = Very Often 1. In the last month, how often have you been upset because of something that happened unexpectedly? ................................... 0 1 2 3 4 2. In the last month, how often have you felt that you were unable to control the important things in your life?..................................................0 1 2 3 4 3. In the last month, how often have you felt nervous and “stressed”? ........... 0 1 2 3 4 4. In the last month, how often have you felt confident about your ability to handle your personal problems? ............................................................... 0 1 2 3 4 5. In the last month, how often have you felt that things were going your way?.................................................................................... 0 1 2 3 4 6. In the last month, how often have you found that you could not cope with all the things that you had to do? .......................................................... 0 1 2 3 4 7. In the last month, how often have you been able to control irritations in your life?................................................................... 0 1 2 3 4 8. In the last month, how often have you felt that you were on top of things?.. 0 1 2 3 4 9. In the last month, how often have you been angered because of things that were outside of your control? .................................... 0 1 2 3 4 10. In the last month, how often have you felt difficulties were piling up so high that you could not overcome them?......................... 0 1 2 3 4 135 XIII. Bibliography 1. Nielsen NR, Kjoller M, Kamper-Jorgensen F, Gronbaek MN. [Stress among working population of Danes]. Ugeskr Laeger 2004;166:4155-60. 2. Reiche EM, Nunes SO, Morimoto HK. Stress, depression, the immune system, and cancer. Lancet Oncol 2004;5:617-25. 3. Globocan 2002 [database on the internet]. Lyon: International Agency for Research on Cancer [Cited 2006 October 10]. Available from: www.iarc.fr 4. Lukanova A, Lundin E, Micheli A, Arslan A, Ferrari P, Rinaldi S et al. Circulating levels of sex steroid hormones and risk of endometrial cancer in postmenopausal women. Int J Cancer 2004;108:425-32. 5. Kaaks R, Berrino F, Key T, Rinaldi S, Dossus L, Biessy C et al. Serum sex steroids in premenopausal women and breast cancer risk within the European Prospective Investigation into Cancer and Nutrition (EPIC). J Natl Cancer Inst 2005;97:755-65. 6. Missmer SA, Eliassen AH, Barbieri RL, Hankinson SE. Endogenous estrogen, androgen, and progesterone concentrations and breast cancer risk among postmenopausal women. J Natl Cancer Inst 2004;96:1856-65. 7. Singh S, Sheppard MC, Langman MJ. Sex differences in the incidence of colorectal cancer: an exploration of oestrogen and progesterone receptors. Gut 1993;34:611-5. 8. Talamini R, Franceschi S, Dal Maso L, Negri E, Conti E, Filiberti R et al. The influence of reproductive and hormonal factors on the risk of colon and rectal cancer in women. Eur J Cancer 1998;34:1070-6. 9. Severi G, Morris HA, MacInnis RJ, English DR, Tilley W, Hopper JL et al. Circulating steroid hormones and the risk of prostate cancer. Cancer Epidemiol Biomarkers Prev 2006;15:86-91. 10. Rivier C. Luteinizing-hormone-releasing hormone, gonadotropins, and gonadal steroids in stress. Ann N Y Acad Sci 1995;771:187-91. 11. Negro-Vilar A. Stress and other environmental factors affecting fertility in men and women: overview. Environ Health Perspect 1993;101:59-64. 136 12. Breen KM, Karsch FJ. Does cortisol inhibit pulsatile luteinizing hormone secretion at the hypothalamic or pituitary level? Endocrinology 2004;145:692-8. 13. Plante GE. Vascular response to stress in health and disease. Metabolism 2002;51:25-30. 14. Luecken LJ, Compas BE. Stress, coping, and immune function in breast cancer. Ann Behav Med 2002;24:336-44. 15. Duijts SF, Zeegers MP, Borne BV. The association between stressful life events and breast cancer risk: a meta-analysis. Int J Cancer 2003;107:1023-9. 16. Lazarus R. Stress and emotion. A new synthesis. London: Free Association Books; 1999. 17. McEwen BS, Wingfield JC. The concept of allostasis in biology and biomedicine. Horm Behav 2003;43:2-15. 18. Derogatis L. Self-reported measures of stress. In: Goldberger L, Breznitz S, editors. Handbook on stress. Theoretical and clinical aspects. New York: The Free Press; 1993. p. 200-33. 19. Soufer R, Arrighi JA, Burg MM. Brain, behavior, mental stress, and the neurocardiac interaction. J Nucl Cardiol. 2002;9:650-62. 20. Moos R, Schaefer J. Coping resources and processes: current concepts and measures. In: Goldberger L, Breznitz S, editors. Handbook of stress. Theoretical and clinical aspects. New York: The Free Press; 1993. p. 234-57. 21. Appleyard M, Hansen AT, Schnohr P, Jensen G, Nyboe J. The Copenhagen City Heart Study. A book of tables with data from the first examination (1976-78) and a five year follow-up (1981-83). Scand J Soc Med 1989;170:1-160. 22. Greenland S, Pearl J, Robins JM. Causal diagrams for epidemiologic research. Epidemiology 1999;10:37-48. 23. Parkin DM, Bray FI, Devesa SS. Cancer burden in the year 2000. The global picture. Eur J Cancer 2001;37:S4-66. 24. Dalton SO, Boesen EH, Ross L, Schapiro IR, Johansen C. Mind and cancer. Do psychological factors cause cancer? Eur J Cancer 2002;38:1313-23. 137 25. Bleiker EM, van der Ploeg HM. Psychosocial factors in the etiology of breast cancer: review of a popular link. Patient Educ Couns 1999;37:201-14. 26. Gerits P. Life events, coping and breast cancer: state of the art. Biomed Pharmacother 2000;54:229-33. 27. Petticrew M, Fraser J, Regan M. Adverse life-events and risk of breast cancer: A meta-analysis. Br J Health Psychol 1999;4:1-17. 28. Hilakivi-Clarke L, Rowland J, Clarke R, Lippman ME. Psychosocial factors in the development and progression of breast cancer. Breast Cancer Res Treat. 1994;29:141-60. 29. Garssen B. Psychological factors and cancer development: evidence after 30 years of research. Clin Psychol Rev 2004;24:315-38. 30. De Boer MF, Ryckman RM, Pruyn JF, Van den Borne HW. Psychosocial correlates of cancer relapse and survival: a literature review. Patient Educ Couns 1999;37:215-30. 31. Holmes TH, Rahe RH. The Social Readjustment Rating Scale. J Psychosom Res 1967;11:213-8. 32. Karasek RA. Job demands, job decision latitude, and mental strain: Implications for job redesign. Adm Sci Q 1979;24:285-308. 33. Burke MA, Goodkin K. Stress and the development of breast cancer: a persistent and popular link despite contrary evidence. Cancer 1997;79:1055-9. 34. Cox T, Mackay C. Psychosocial factors and psychophysiological mechanisms in the aetiology and development of cancers. Soc Sci Med 1982;16:381-96. 35. Jackson N, Waters E. Guidelines for systematic reviews in health promotion and public health taskforce. Health Promot Int 2005;20:367-74. 36. Geyer S. Life events, chronic difficulties and vulnerability factors preceding breast cancer. Soc Sci Med 1993;37:1545-55. 37. Geyer S. Life events prior to manifestation of breast cancer: a limited prospective study covering eight years before diagnosis. J Psychosom Res 1991;35:355-63. 138 38. Chen CC, David AS, Nunnerley H, Michell M, Dawson JL, Berry H et al. Adverse life events and breast cancer: case-control study. BMJ 1995;311:152730. 39. Price MA, Tennant CC, Butow PN, Smith RC, Kennedy SJ, Kossoff MB et al. The role of psychosocial factors in the development of breast carcinoma: Part II. Life event stressors, social support, defense style, and emotional control and their interactions. Cancer 2001;91:686-97. 40. Ollonen P, Lehtonen J, Eskelinen M. Stressful and adverse life experiences in patients with breast symptoms; a prospective case-control study in Kuopio, Finland. Anticancer Res 2005;25:531-6. 41. Protheroe D, Turvey K, Horgan K, Benson E, Bowers D, House A. Stressful life events and difficulties and onset of breast cancer: case-control study. BMJ 1999;319:1027-30. 42. Cooper CL, Faragher EB. Psychosocial stress and breast cancer: the interrelationship between stress events, coping strategies and personality. Psychol Med 1993;23:653-62. 43. Schwarz R, Geyer S. Social and psychological differences between cancer and noncancer patients: cause or consequence of the disease? Psychother Psychosom. 1984;41:195-9. 44. Li J, Johansen C, Hansen D, Olsen J. Cancer incidence in parents who lost a child: a nationwide study in Denmark. Cancer 2002;95:2237-42. 45. Kvikstad A, Vatten LJ, Tretli S, Kvinnsland S. Widowhood and divorce related to cancer risk in middle-aged women. A nested case-control study among Norwegian women born between 1935 and 1954. Int J Cancer 1994;58:512-6. 46. Kvikstad A,.Vatten LJ. Risk and prognosis of cancer in middle-aged women who have experienced the death of a child. Int J Cancer 1996;67:165-9. 47. Johansen C, Olsen JH. Psychological stress, cancer incidence and mortality from non-malignant diseases. Br J Cancer 1997;75:144-8. 48. Levav I, Kohn R, Iscovich J, Abramson JH, Tsai WY, Vigdorovich D. Cancer incidence and survival following bereavement. Am J Public Health 2000;90:1601-7. 49. Ewertz M. Bereavement and breast cancer. Br J Cancer 1986;53:701-3. 139 50. Jones DR, Goldblatt PO, Leon DA. Bereavement and cancer: some data on deaths of spouses from the longitudinal study of Office of Population Censuses and Surveys. BMJ (Clin Res Ed) 1984;289:461-4. 51. Hislop TG, Waxler NE, Coldman AJ, Elwood JM, Kan L. The prognostic significance of psychosocial factors in women with breast cancer. J Chronic Dis 1987;40:729-35. 52. Achat H, Kawachi I, Byrne C, Hankinson S, Colditz G. A prospective study of job strain and risk of breast cancer. Int J Epidemiol 2000;29:622-8. 53. Schernhammer ES, Hankinson SE, Rosner B, Kroenke CH, Willett WC, Colditz GA et al. Job stress and breast cancer risk: the nurses' health study. Am J Epidemiol 2004;160:1079-86. 54. Helgesson O, Cabrera C, Lapidus L, Bengtsson C, Lissner L. Self-reported stress levels predict subsequent breast cancer in a cohort of Swedish women. Eur J Cancer Prev 2003;12:377-81. 55. Kroenke CH, Hankinson SE, Schernhammer ES, Colditz GA, Kawachi I, Holmes MD. Caregiving stress, endogenous sex steroid hormone levels, and breast cancer incidence. Am J Epidemiol 2004;159:1019-27. 56. Lillberg K, Verkasalo PK, Kaprio J, Teppo L, Helenius H, Koskenvuo M. Stress of daily activities and risk of breast cancer: a prospective cohort study in Finland. Int J Cancer 2001;91:888-93. 57. Nielsen NR, Zhang ZF, Kristensen TS, Netterstrom B, Schnohr P, Gronbaek M. Self reported stress and risk of breast cancer: prospective cohort study. BMJ 2005;331:548. 58. Jacobs JR, Bovasso GB. Early and chronic stress and their relation to breast cancer. Psychol Med 2000;30:669-78. 59. Lillberg K, Verkasalo PK, Kaprio J, Teppo L, Helenius H, Koskenvuo M. Stressful life events and risk of breast cancer in 10,808 women: a cohort study. Am J Epidemiol 2003;157:415-23. 60. Barraclough J, Pinder P, Cruddas M, Osmond C, Taylor I, Perry M. Life events and breast cancer prognosis. BMJ 1992;304:1078-81. 61. De Brabander B, Gerits P. Chronic and acute stress as predictors of relapse in primary breast cancer patients. Patient Educ Couns 1999;37:265-72. 140 62. Forsen A. Psychosocial stress as a risk for breast cancer. Psychother Psychosom 1991;55:176-85. 63. Graham J, Ramirez A, Love S, Richards M, Burgess C. Stressful life experiences and risk of relapse of breast cancer: observational cohort study. BMJ 2002;324:1420. 64. Schernhammer ES, Laden F, Speizer FE, Willett WC, Hunter DJ, Kawachi I et al. Rotating night shifts and risk of breast cancer in women participating in the nurses' health study. J Natl Cancer Inst 2001;93:1563-8. 65. Kune S, Kune GA, Watson LF, Rahe RH. Recent life change and large bowel cancer. Data from the Melbourne Colorectal Cancer Study. J Clin Epidemiol 1991;44:57-68. 66. Courtney JG, Longnecker MP, Peters RK. Psychosocial aspects of work and the risk of colon cancer. Epidemiology 1996;7:175-81. 67. Courtney JG, Longnecker MP, Theorell T, Gerhardsson V. Stressful life events and the risk of colorectal cancer. Epidemiology 1993;4:407-14. 68. Spiegelman D, Wegman DH. Occupation-related risks for colorectal cancer. J Natl Cancer Inst 1985;75:813-21. 69. Kojima M, Wakai K, Tokudome S, Tamakoshi K, Toyoshima H, Watanabe Y et al. Perceived psychologic stress and colorectal cancer mortality: findings from the Japan Collaborative Cohort Study. Psychosom Med 2005;67:72-7. 70. Hankinson S, Hunter D. Breast Cancer. In: Adami H, Hunter D, Trichopoulos D, editors. Textbook of cancer epidemiology. New York: Oxford University Press; 2002. p. 301-39. 71. McEwen BS. Protective and damaging effects of stress mediators. N Engl J Med 1998;338:171-9. 72. Ferin M. Clinical review 105: Stress and the reproductive cycle. J Clin Endocrinol Metab 1999;84:1768-74. 73. Olsen AH, Jensen A, Njor SH, Villadsen E, Schwartz W, Vejborg I et al. Breast cancer incidence after the start of mammography screening in Denmark. Br J Cancer 2003;88:362-5. 141 74. Olsen AH, Njor SH, Vejborg I, Schwartz W, Dalgaard P, Jensen MB et al. Breast cancer mortality in Copenhagen after introduction of mammography screening: cohort study. BMJ 2005;330:220. 75. The World Health Organization MONICA Project (monitoring trends and determinants in cardiovascular disease): a major international collaboration. WHO MONICA Project Principal Investigators. J Clin Epidemiol 1988;41:10514. 76. Halbreich U. Role of estrogen in postmenopausal depression. Neurology 1997;48:S16-S19. 77. Birkhauser M. Depression, menopause and estrogens: is there a correlation? Maturitas 2002;41:S3-S8. 78. Amant F, Moerman P, Neven P, Timmerman D, Van Limbergen E, Vergote I. Endometrial cancer. Lancet 2005;366:491-505. 79. Persson I, Adami H. Endometrial cancer. In: Adami H, Hunder D, Trichopoulos D, editors. Textbook of cancer epidemiology. New York: Oxford University Press; 2002. p. 359-77. 80. Gunin AG. Effect of chronic stress on estradiol action in the uterus of ovariectomized rats. Eur J Obstet Gynecol Reprod Biol 1996;66:169-74. 81. Kam K, Park Y, Cheon M, Son GH, Kim K, Ryu K. Effects of immobilization stress on estrogen-induced surges of luteinizing hormone and prolactin in ovariectomized rats. Endocrine 2000;12:279-87. 82. Pitot H, Dragan Y. Chemical carcinogenesis. In: Klaassen C, editor. Toxicology: the basic science of poisons. New York: McGraw-Hill; 2001. p. 241-320. 83. Warr P. The measurement of well-being and other aspects of mental-health. J Occup Health 1990;63:193-210. 84. Cohen S, Kamarck T, Mermelstein R. A global measure of perceived stress. J Health Soc Behav. 1983;24:385-96. 85. Helweg-Larsen K, Kjoller M, Davidsen M, Rasmussen NK, Madsen M. The Danish National Cohort Study (DANCOS). Dan Med Bull 2003;50:177-80. 142 86. Viswanathan AN, Feskanich D, De Vivo I, Hunter DJ, Barbieri RL, Rosner B et al. Smoking and the risk of endometrial cancer: results from the Nurses' Health Study. Int J Cancer 2005;114:996-1001. 87. Kaaks R, Lukanova A, Kurzer MS. Obesity, endogenous hormones, and endometrial cancer risk: a synthetic review. Cancer Epidemiol Biomarkers Prev 2002;11:1531-43. 88. Modugno F, Ness RB, Chen C, Weiss NS. Inflammation and endometrial cancer: a hypothesis. Cancer Epidemiol Biomarkers Prev 2005;14:2840-7. 89. Gunin AG, Mashin IN, Zakharov DA. Proliferation, mitosis orientation and morphogenetic changes in the uterus of mice following chronic treatment with both estrogen and glucocorticoid hormones. J Endocrinol 2001;169:23-31. 90. Gunin A. Realization of estradiol effects in the uterus of ovariectomized rats under acute stress. Eur J Obstet Gynecol Reprod Biol 1995;60:69-74. 91. Rabin DS, Johnson EO, Brandon DD, Liapi C, Chrousos GP. Glucocorticoids inhibit estradiol-mediated uterine growth: possible role of the uterine estradiol receptor. Biol Reprod 1990;42:74-80. 92. Weitz J, Koch M, Debus J, Hohler T, Galle PR, Buchler MW. Colorectal cancer. Lancet 2005;365:153-65. 93. Wardle J, Gibson EL. Impact of stress on diet: processes and implications. In: Stansfeld S, Marmot M, editors. Stress and the heart. Psychosocial pathways to coronary heart disease. London: BMJ Books; 2002. p. 124-49. 94. Pohorecky LA. Stress and alcohol interaction: an update of human research. Alcohol Clin Exp Res 1991;15:438-59. 95. Singh S, Sheppard MC, Langman MJ. Sex differences in the incidence of colorectal cancer: an exploration of oestrogen and progesterone receptors. Gut 1993;34:611-5. 96. Di Domenico M, Castoria G, Bilancio A, Migliaccio A, Auricchio F. Estradiol activation of human colon carcinoma-derived Caco-2 cell growth. Cancer Res 1996;56:4516-21. 97. Fiorelli G, Picariello L, Martineti V, Tonelli F, Brandi ML. Functional estrogen receptor beta in colon cancer cells. Biochem Biophys Res Commun 1999;261:521-7. 143 98. Singh S, Langman MJ. Oestrogen and colonic epithelial cell growth. Gut 1995;37:737-9. 99. Newcomb PA, Solomon C, White E. Tamoxifen and risk of large bowel cancer in women with breast cancer. Breast Cancer Res Treat 1999;53:271-7. 100. Travis LB, Curtis RE, Storm H, Hall P, Holowaty E, Van Leeuwen FE et al. Risk of second malignant neoplasms among long-term survivors of testicular cancer. J Natl Cancer Inst 1997;89:1429-39. 101. Littman AJ, White E, Satia JA, Bowen DJ, Kristal AR. Reliability and validity of 2 single-item measures of psychosocial stress. Epidemiology 2006;17:398-403. 102. Truelsen T, Nielsen N, Boysen G, Gronbaek M. Self-reported stress and risk of stroke: the Copenhagen City Heart Study. Stroke 2003;34:856-62. 103. Nielsen NR, Kristensen TS, Prescott E, Larsen KS, Schnohr P, Gronbaek M. Perceived Stress and Risk of Ischemic Heart Disease: Causation or Bias? Epidemiology 2006;17:391-7. 104. Savitz D. Interpreting epidemiologic evidence. Strategies for study design and analysis. New York: Oxford University Press; 2003. 105. Stratakis CA, Chrousos GP. Neuroendocrinology and pathophysiology of the stress system. Ann N Y Acad Sci 1995;771:1-18. 106. Kudielka BM, Kirschbaum C. Sex differences in HPA axis responses to stress: a review. Biol Psychol 2005;69:113-32. 107. Boyd DB. Insulin and cancer. Integr Cancer Ther 2003;2:315-29. 108. Monnikes H, Tebbe JJ, Hildebrandt M, Arck P, Osmanoglou E, Rose M et al. Role of stress in functional gastrointestinal disorders. Evidence for stressinduced alterations in gastrointestinal motility and sensitivity. Dig Dis 2001;19:201-11. 109. Hsing AW, Tsao L, Devesa SS. International trends and patterns of prostate cancer incidence and mortality. Int J Cancer 2000;85:60-7. 110. Schwartz KL, Crossley-May H, Vigneau FD, Brown K, Banerjee M. Race, socioeconomic status and stage at diagnosis for five common malignancies. Cancer Causes Control 2003;14:761-6. 144 111. Gilligan T. Social disparities and prostate cancer: mapping the gaps in our knowledge. Cancer Causes Control 2005;16:45-53. 112. Dutta RS, Philip J, Javle P. Trends in prostate cancer incidence and survival in various socioeconomic classes: a population-based study. Int J Urol 2005;12:644-53. 113. Lund Nilsen TI, Johnsen R, Vatten LJ. Socio-economic and lifestyle factors associated with the risk of prostate cancer. Br J Cancer 2000;82:1358-63. 114. Pukkala E, Weiderpass E. Socio-economic differences in incidence rates of cancers of the male genital organs in Finland, 1971-95. Int J Cancer 2002;102:643-8. 115. Riman T, Nilsson S, Persson IR. Review of epidemiological evidence for reproductive and hormonal factors in relation to the risk of epithelial ovarian malignancies. Acta Obstet.Gynecol.Scand. 2004;83:783-95. 116. Thomas M, Thomas J. Toxic responses of the reproductive system. In: Klaassen C, editor. Toxicology: The basic science of poisons. New York: McGraw-Hill; 2001. p. 673-710. 117. Di Leo A, Messa C, Cavallini A, Linsalata M. Estrogens and colorectal cancer. Curr Drug Targets Immune Endocr Metabol Disord 2001;1:1-12. 118. Christensen S, Anglov J, Christensen J, Olsen E, Poulsen O. Application of a new AMIQAS computer program for integrated quality control, method evaluation and proficiency testing. Fresen J Anal Chem 1993;345:343-50. 119. Westgard JO, Barry PL, Hunt MR, Groth T. A multi-rule Shewhart chart for quality control in clinical chemistry. Clin Chem 1981;27:493-501. 120. Hankinson SE, Willett WC, Manson JE, Hunter DJ, Colditz GA, Stampfer MJ et al. Alcohol, height, and adiposity in relation to estrogen and prolactin levels in postmenopausal women. J Natl Cancer Inst 1995;87:1297-302. 121. Kley HK, Schlaghecke R, Kruskemper HL. Stability of steroids in plasma over a 10-year period. J Clin Chem Clin Biochem 1985;23:875-8. 122. Bolelli G, Muti P, Micheli A, Sciajno R, Franceschetti F, Krogh V et al. Validity for epidemiological studies of long-term cryoconservation of steroid and protein hormones in serum and plasma. Cancer Epidemiol Biomarkers Prev 1995;4:50913. 145 123. Cancer Registry [homepage on the internet]. Copenhagen: Danish National Board of Health [cited 2007 January 15]. Available from: www.sst.dk 124. Greenland S, Morgenstern H. Confounding in health research. Annu Rev Public Health 2001;22:189-212. 125. Stein M, Miller A. Stress, the immune system, and health and illness. In Goldberger L, Breznitz S, eds. Handbook on stress. Theoretical and clinical aspects, pp 127-41. New York: The Free Press, 1993. 126. Rosengren A, Hawken S, Ounpuu S, Sliwa K, Zubaid M, Almahmeed WA et al. Association of psychosocial risk factors with risk of acute myocardial infarction in 11119 cases and 13648 controls from 52 countries (the INTERHEART study): case-control study. Lancet 2004;364:953-62. 127. Rosengren A, Tibblin G, Wilhelmsen L. Self-perceived psychological stress and incidence of coronary artery disease in middle-aged men. Am J Cardiol 1991;68:1171-5. 128. Iso H, Date C, Yamamoto A, Toyoshima H, Tanabe N, Kikuchi S et al. Perceived mental stress and mortality from cardiovascular disease among Japanese men and women: the Japan Collaborative Cohort Study for Evaluation of Cancer Risk Sponsored by Monbusho (JACC Study). Circulation 2002;106:1229-36. 129. Pohorecky LA. Interaction of alcohol and stress at the cardiovascular level. Alcohol 1990;7:537-46. 130. Bhui K. Physical activity and stress. In Stansfeld S, Marmot M, eds. Stress and the heart. Psychosocial pathways to coronary heart disease, pp 158-67. London: BMJ books, 2002. 131. Jarvis M. Smoking and stress. In Stansfeld S, Marmot M, eds. Stress and the heart. Psychosocial pathways to coronary heart disease, pp 150-7. London: BMJ books, 2002. 132. Marsland A, Bachen E, Cohen S, Manuck S. Stress, immunity, and susceptibility to infectious disease. In: Baum A, Revenson T, Singer J, editors. Handbook of health psychology. Mahwah, NJ: Erlbaum; 2001. 133. Rugulies R, Bultmann U, Aust B, Burr H. Psychosocial Work Environment and Incidence of Severe Depressive Symptoms: Prospective Findings from a 5-Year 146 Follow-up of the Danish Work Environment Cohort Study. Am J Epidemiol 2006;16:877-87 134. Wang J, Patten SB. Perceived work stress and major depression in the Canadian employed population, 20-49 years old. J Occup Health Psychol 2001;6:283-9. 135. Rozanski A, Blumenthal JA, Kaplan J. Impact of psychological factors on the pathogenesis of cardiovascular disease and implications for therapy. Circulation 1999;99:2192-217. 136. Cohen S, Williamson GM. Perceived stress in a probability sample of the United States. In: Spacapan OS, editor. The social psychology of health. Newbury Park: Sage; 1988. 147
