Abstract
Background. Poor social support and small social networks have been associated with increased risks for conditions such as coronary heart disease as well as with overall mortality. We investigated the association between social ties and risk for cancer. Material and methods. The study sample consisted of 8 548 Danes who had been examined in 1991–1994 within the Copenhagen City Heart Study. The median length of follow-up was 9.3 years (range, 0–11.2 years). Social ties were measured from answers to a questionnaire on social networks. Regression analyses for cancers at the most frequent sites (breast, lung, prostate and colon and rectum) were conducted with the Cox proportional hazards model, with adjustment for a number of well-known risk factors for cancer. Results. While we found no significant association between social ties and risk for cancer in men, women with high social network scores had an increased risk for lung cancer of borderline significance (HR, 2.16; 95% CI, 1.02–4.60). The risks for breast cancer and colorectal cancers were not significantly increased in the same group of women. Discussion. The results of this study do not support the hypothesis that social network size is associated with a decreased risk for cancer
Scientific interest in an association between social ties and health is primarily focused on the protective effect of ‘significant others’ Citation[1]. Previous research on the subject addressed the role of both social support and social networks. ‘Social support’ is defined mainly by the quality of social contacts. It can be divided into the subcategories “instrumental support”, “informational support” and “emotional support” Citation[2]. ‘Social network’ describes the existence and quantity of social contacts and thus focuses on the structural aspects of social support, namely on social ties that provide support.
It has been reported that people with a limited social network and poor social support are at increased risk for several adverse health outcomes Citation[3], such as overall mortality Citation[4–7] and the incidence of coronary heart disease Citation[8–10]. Studies in which measures of both social networks and social support were included showed that both are independent risk factors for coronary disease and mortality Citation[11–13].
An association between limited social support and poor health has been proposed to be due to differences in exposure to well-known lifestyle risk factors for chronic disease Citation[14], possibly through an interaction between social connections and stress Citation[15]. People characterised as having limited social networks and poor social support may live in stressful conditions, smoke more, consume more alcohol, exercise less and have a less healthy diet than persons with more social support, as social support may influence the choice of lifestyle Citation[1]. Lack of social support and small social networks could thus, by promoting a disadvantageous life-style and health behaviour act as a promoter of a pre-existing genetic damage which may in turn lead to the development of cancer.
Few studies have investigated whether social networks and social support are associated with the risk for cancer. One prospective cohort study on social connections and risk for cancer observed no such association for men but showed that socially isolated women were at increased risk for death from cancer and especially smoking-related cancers Citation[16]. Other studies showed no association between measures of social network and cancer risk Citation[6], Citation[17] or found an association only in specific populations: In an American study, Kinney, Bloor, Dudley, Milikan, Marshall, Martin et al. Citation[18] found an association between limited emotional support and risk for colon cancer among blacks but not among whites. In Japan, Iwasaki et al. Citation[19] found an increased risk for cancer only in women with small social networks who lived in urban areas.
We investigated whether there is an association between social ties and the risk for cancer in a prospective cohort study, focusing on the most frequent sites in men and women (breast cancer, lung cancer, prostate cancer, colorectal cancers). The aim of the study was to analyse the hypothesis that a small social network is associated with an increased risk for cancer. We hypothesised that poor social support and network increase the cancer risk by change in health behaviours.
Material and methods
Study sample
Data were obtained from the Copenhagen City Heart Study, which has been described in detail elsewhere Citation[20], Citation[21]. In this study, which was initiated in 1976, 19 698 persons aged 20 and older at the date of inclusion were invited to participate. The participants were selected randomly within age and sex strata from two districts of Copenhagen, Denmark. At inclusion, they completed a self-administered questionnaire and underwent a health examination. A second survey was conducted as a follow-up in 1981–1983, when 500 additional participants aged 20–24 years were included. The data for the present publication stem from a further follow-up of this cohort, in 1991–1994, when a further 3 000 persons, aged 20–49 years, where invited to join the study.
A total of 10 135 persons completed the questionnaire and underwent the health examination in the third survey (response rate, 61%). Because of a cancer diagnosis before the date of entry, 589 persons (6%) were excluded from the analysis. A further 998 (10%) were excluded because data were missing on at least one of the behavioural or socioeconomic confounding variables (n = 808) included in the analysis or on one of the social network variables (n = 190), leaving 8 548 persons (84%) eligible for the data analyses.
Cancer cases and follow-up
Follow-up for cancer occurrence was begun on the date of the examination for the third survey in the period October 10, 1991 to September 16, 1994 and ended on the date of diagnosis of a first primary cancer, emigration, death or 30 November 2002, whichever came first. As non-melanoma skin cancer is common and has a good prognosis, it was not considered to be a first primary cancer, and any cancer occurring subsequently was regarded as the first primary cancer.
Cancer cases were ascertained by data linkage with the Danish Cancer Registry and the Danish Population Register (CPR). Since April 1, 1968 the CPR has assigned all residents of Denmark a unique ten-digit personal identification number, which allows for complicated register linkage procedures, which is used for administrative and research reasons. The continuously updated vital status of all residents includes status as alive, dead, emigrated or disappeared. This information was obtained for each study participant. In general linkage between the administrative and health as well as disease registries may be conducted after permission obtained in The Danish Data Protection Board. Research activities, which do not imply contact between participants and researchers, do not, in general, require permission from the ethical committee system as a prerequisite.
The Danish Cancer Registry has documented all cases of malignant neoplasms in Denmark in a population-based registration since 1942 Citation[22]. The registration of cancer cases is based on a modified Danish version of the International Classification of Diseases, 7th Revision (ICD-7). In persons with two or more cancers, the first cancer was considered for the analysis according to the date of diagnosis as documented in the Cancer Registry.
Variables of interest
Behavioural and lifestyle factors
Nine variables for behavioural and lifestyle factors known to influence cancer risk were included. Information on physical activity during leisure time was collected in categories on three levels: sedentary, moderate activity ≤4 h/week and activity >4 h/week. Tobacco consumption was classified into five categories: never smoker, former smoker and current smoker of 1–14, 15–24 or >24 g of tobacco per day, calculated on the basis that one cigarette equals 1 g, one cheroot 3 g and one cigar 5 g of tobacco. Drinking status was included as a dichotomous variable (non-drinkers vs. consumers), and daily alcohol intake (drinks per day) was included as a linear variable. Body mass index was also included as a linear variable.
Years of schooling were divided into three categories: <8 years (completed primary school), 8–11 years and >11 years. The annual household income at date of inclusion was registered in six categories (<100 000, 100 000–149 000, 150 000–199 000, 200 000–299 000, 300 000–399 000 and ≥400 000 DKK). Cohabitation was considered as a dichotomous variable (living alone vs. not living alone).
Social network variables
Social network score (quantity of social contacts)
Participants were asked to indicate the frequency of their contacts with eight classes of person on a six-point scale, the categories being parents, children, other family members, spouse or partner, colleagues (after work), neighbours, childhood friends, other friends and home help. For the analyses, the categories 'childhood friends’ and 'other friends‘ were combined, and the 'home help‘ category was omitted because of the likelihood that this is an indicator of disability. There were thus six classes of contact persons.
For the analyses of social contact variables, the answer categories 'not available‘ and 'never‘ were combined and coded as 0, 'rarely‘ and 'monthly‘ were combined and coded as 1, and 'weekly‘ and 'daily‘ were combined and coded as 2. The social network score was then calculated as the sum score of the frequency of contact with the six remaining classes of contact persons, resulting in a possible range of 0 to 12. For the analyses of groups with different extents of social contact, the network score was categorised into sample tertiles (0–4, 5–6 and 7–12 points).
Quality of social contacts
To measure qualitative aspects of the social contacts, analyses were adjusted for satisfaction with contacts from answers to the question, “How satisfied are you with these contacts?”. This variable was included as a binary variable (“very satisfied” vs. “somewhat satisfied” or “not satisfied at all” combined) because of the skewness of the distribution (very satisfied: 68%, somewhat satisfied: 30.5%, not satisfied: 1.5%).
In order to account for possible confounding between social contact and the need for help due to illness or disability, the analyses were further adjusted for perceived health status, measured from answers to the question “How would you assess your health during the past year?” and included in the analyses as a variable with three categories: “extraordinarily good”, “good”, and “bad” or “very bad” combined. As in the satisfaction variable, the last categories were combined due to the skewness of the distribution (extraordinarily good: 8.7%, good: 65.1%, bad: 26.6%, very bad: 2.5%).
Statistical analysis
The analyses of the relation between cancer incidence rates and social network variables were based on sex-stratified Cox proportional hazard models, with age as the time scale to ensure that the estimation was based on comparisons of individuals at the same age Citation[23]. The analyses were corrected for delayed entry (left truncation). Time under study was included as a time-dependent variable and was modelled by a linear spline with boundaries at 1 year, 2 years and 3 years after entry into the study. A linear spline was used because this allows the hazard to be nonlinear Citation[24].
All models were adjusted for physical activity, drinking status, alcohol intake, smoking habits, duration of smoking, body mass index, schooling, household income and cohabitation status. Whenever possible, variables were entered linearly into the Cox model, as this is biologically more reasonable than the step functions corresponding to categorisation and, furthermore, increases the power of the analyses Citation[25]. For variables not categorised in the questionnaire (bmi, alcohol consumption and social network score), the linearity of the associations was evaluated graphically by linear splines with three boundaries Citation[24] placed at the quartiles among cases. We found no departures from linearity.
We also performed unadjusted analyses, including persons with missing data in one of the confounding variables or one of the social support variables. These analyses did not produce any additional significant risk estimate. One significant risk estimate (see results section) could not be reproduced with the unadjusted analyses. This was considered a chance finding and thus the overall results and conclusions did not change when analyses were carried out unadjusted (data not shown).
Associations were estimated for the two sexes separately and for the two sexes combined. In the combined analysis, we found significant differences in social network variables between the two sexes with the Wald test; therefore, analyses were carried out separately for each sex. Regression analyses were conducted for cancers at the most frequent sites in men and women (breast cancer, lung cancer, prostate cancer, colorectal cancers).
Two-sided 95% confidence intervals (CIs) for the hazard ratio (HR) were calculated on the basis of Wald′s test of the Cox regression parameter on the log rate ratio scale. The procedure PHREG in SAS (release 8.2; SAS Institute, Inc., Cary, North Carolina) on the TextPad platform was used for statistical analyses.
Results
The median length of follow-up was 9.3 years (range, 0–11.2 years). During follow-up, cancer was diagnosed in 987 persons (12% of the sample). The most frequent cancer sites were breast cancer in women and lung cancer in men ().
Table I. Cancer cases (n = 987) in women and men in the third survey of the Copenhagen City Heart Study (1991–2002)
Both men and women with high social network scores were younger and reported more physical activity, a higher household income, longer schooling and more cohabitation than men and women with low social contact scores (). A higher percentage of men with high social network scores had never smoked than among those with lower social contact scores.
Table II. Sample characteristics at date of entry of 4710 women and 3838 men in the third survey of the Copenhagen City Heart Study (1991–2002), Denmark
Regression analyses of the association between social network size and risk for cancer were carried out separately for men and women. All analyses were adjusted for lifestyle factors, health and satisfaction with social contacts. Overall, the analyses revealed that women with higher social network scores were at slightly increased risk for cancers at the sites investigated. Even after adjustment for lifestyle factors, health and satisfaction with social contacts, women with the highest social network scores had an increased risk for lung cancer of borderline significance (HR, 2.16; 95% CI, 1.02–4.60). The risks for breast cancer and colorectal cancers among women with medium and high social network scores were not significantly increased ().
Table III. Cancer risk by social contact score in 4 710 women in the third survey of the Copenhagen City Heart Study (1991–2002), Denmark
Men with the highest social network scores had slightly, but not significantly decreased risks for prostate cancer and colorectal cancers and a slightly, but not significantly increased risk for lung cancer. As for women, adjustment for lifestyle factors, health and satisfaction with social contacts had little effect on the risk estimates ().
Table IV. Cancer risk by social contact score in 3 838 men in the third survey of the Copenhagen City Heart Study (1991–2002), Denmark
Discussion
The results of this study reveal differences between the sexes in the association between social ties and cancer incidence. Among women, there seemed to be a dose–response relation between social contact score and risk for cancer, in the opposite direction from that expected; however, most of the hazard ratios were not statistically significant. The estimates for cancer risk tended to be somewhat higher in women than in men, but in the models with full adjustment for confounders (socioeconomic and behavioural factors, health and satisfaction with social contacts) only the risk for lung cancer among women with a high score in the social contact measure reached borderline significance, based on 20 cancer cases only. The small numbers, the multiple comparisons and the borderline significance make us interpret this finding as due to chance. The risks for breast cancer and colorectal cancers were not significantly increased in women with high social network scores.
Adjustment for behavioural factors surprisingly did not change the overall results, indicating that the association is not mediated by lifestyle variables. As frequent social contacts can be an indicator of poor health (e.g., contact with health-care providers or caring family members), all analyses were also adjusted for perceived health status and satisfaction with social contacts. These adjustments also did not change the results.
Among men, we found no dose–response pattern with regard to social contact scores and cancer risk. For men with high scores in the social contact measure, the risk estimates for prostate cancer and colorectal cancers were below 1.0, while those for lung cancer were slightly, but not significantly, increased. Adjustment for lifestyle factors, perceived health status and satisfaction with social contacts also did not change the results for men.
Our finding of a clear social gradient in lifestyle is in line with the results of previous studies that low social support is associated with poor health behaviour Citation[26], Citation[27]. Social support may buffer the adverse effects of stress, or social ties may provide support for advantageous health behaviour, such as drinking less and reducing or quitting smoking Citation[1], Citation[28].
Our findings do not uniformly support the results of studies that found that women with poor social support have generally poor health Citation[29]. Furthermore, our results do not clearly support those of Reynolds & Kaplan Citation[16], who reported that strong social support was associated with a lower cancer risk in women. A previously published analysis of mortality in the cohort investigated in this study showed that persons with several types of contact had decreased all-cause mortality Citation[30].
Various limitations of the study with regard to measurement of social network variables must be taken into consideration. First, although we measured the quantity of social contacts, we had access to only a very broad measure of the quality of the contacts. Measurement of both quantity and quality of social contacts is a prerequisite for meaningful analyses and interpretation of any association of social ties and health outcomes Citation[31], which was fulfilled in our analyses. But as the single item concerning satisfaction with social contacts could be included only as a dichotomous variable, the adjustment for quality of contacts in our study was poor. Secondly, social contacts were not measured with a standard instrument, making it difficult to compare our results with those of others. This is, however, a general problem in research on social networks and health. A wide variety of measures of social network and social support have been used, and several reviews point out that the lack of a gold standard or at least common instruments is a major problem in coming to a conclusion about the influence of social support variables on cancer risk or health in general Citation[32–35]. Thirdly, social network variables were measured only once, because there was only one measurement of these variables available in the original study. We therefore cannot discount the possibility that we may have measured a temporary state of unusually high or low social contacts for some persons. Repeated measurements would have been preferable for a more precise assessment of social support and social network, because this would have provided more reliable information on the stability of the social network and general satisfaction with the contacts. A study on changes in social ties over time (two measurements) reported that a decrease in social ties was associated with increased mortality risk in a population of elderly persons Citation[36]. However, this association was eliminated after adjustment for confounding sociodemographic and health-related variables Citation[36], most of which were also taken into consideration in the present study.
In their study of social networks and risk for cancer, Reynolds & Kaplan Citation[16] concluded that emotional support is probably more crucial than instrumental support. The questionnaire used in our study unfortunately did not allow for differentiation between emotional and instrumental support. We can assume, however, that instrumental support is reflected in the size of the social network, while emotional support is related to satisfaction with social contacts. We had limited means to assess the quality of support and thus focused on social network size. We adjusted our analyses for satisfaction with social contacts, but the binary-coded measure may have lacked accuracy. The focus on social network size rather than on the quality of contacts and social support might to some extent explain our results. Small networks with high quality of support may be more beneficial than larger networks with medium or low quality of support.
Since the response rate of the underlying third survey of the Copenhagen City Heart study was only 61% there might also be some incalculable influence of selection bias. But given that there was no information on support variables or health for the non-responders available this issue remains hypothetical. However, we also do not have any reason to assume that the variables in question would be distributed differentially in the non-responder sample.
Regardless of the above-mentioned limitations, the study has certain strengths which ensure that the results are methodologically sound. First, the analyses were based on a large, well-defined cohort with almost complete follow-up. Secondly, the study design was prospective, avoiding methodological flaws in the measurement of exposure, such as recall bias. Thirdly, the data on exposure were obtained independently of the outcome. The outcomes in the present study were classified by linking the identification numbers of study participants to the Danish Cancer Registry, a nationwide, population-based register of all cancer cases in Denmark which has been in operation since 1943. Comprehensive evaluation has shown that the Danish Cancer Registry is 95–99% complete and valid Citation[37], Citation[38].
Finally, all analyses were adjusted for a number of lifestyle factors known to be associated with risks for certain cancer types Citation[39–41]. This adjustment ensures that our results are devoid of any mediating effects due to the potential association between social support and behavioural factors.
Conclusion
The results of this study indicate that the relation between social ties and cancer risk may differ for men and women. The results do not support the hypothesis that social network size is associated with cancer risk.
Acknowledgements
This work has been supported by a UICC American Cancer Society International Fellowship for Beginning Investigators and the Sharon D. Lund Foundation, the Institute of Medical Psychology of the University Clinic Hamburg, the Danish Cancer Society and the Danish Heart Foundation. Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.
References
- Adler N, Matthews K. Health psychology: Why do some people get sick and some stay well?. Annu Rev Psychol 1994; 45: 229–59
- Krishnasamy M. Social support and the patient with cancer: A consideration of the literature. J Adv Nurs 1996; 23: 757–62
- Orth-Gomer K, Johnson JV. Social network interaction and mortality. A six year follow-up study of a random sample of the Swedish population. J Chronic Dis 1987; 40: 949–57
- Berkman LF, Syme SL. Social networks, host resistance, and mortality: A nine-year follow-up study of Alameda County residents. Am J Epidemiol 1979; 109: 186–204
- Seeman TE, Kaplan GA, Knudsen L, Cohen R, Guralnik J. Social network ties and mortality among the elderly in the Alameda County Study. Am J Epidemiol 1987; 126: 714–23
- Vogt TM, Mullooly JP, Ernst D, Pope CR, Hollis JF. Social networks as predictors of ischemic heart disease, cancer, stroke and hypertension: Incidence, survival and mortality. J Clin Epidemiol 1992; 45: 659–66
- Berkman LF. The role of social relations in health promotion. Psychosom Med 1995; 57: 245–54
- Hemingway H, Marmot M. Evidence based cardiology: Psychosocial factors in the aetiology and prognosis of coronary heart disease. Systematic review of prospective cohort studies. BMJ 1999; 318: 1460–7
- 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
- Rutledge T, Reis SE, Olson M, Owens J, Kelsey SF, Pepine CJ, et al. Social networks are associated with lower mortality rates among women with suspected coronary disease: The National Heart, Lung, and Blood Institute-Sponsored Women's Ischemia Syndrome Evaluation study. Psychosom Med 2004; 66: 882–8
- Hanson BS, Isacsson SO, Janzon L, Lindell SE. Social network and social support influence mortality in elderly men. The prospective population study of “Men born in 1914,” Malmo, Sweden. Am J Epidemiol 1989; 130: 100–11
- Orth-Gomer K, Rosengren A, Wilhelmsen L. Lack of social support and incidence of coronary heart disease in middle-aged Swedish men. Psychosom Med 1993; 55: 37–43
- Rosengren A, Wilhelmsen L, Orth-Gomer K. Coronary disease in relation to social support and social class in Swedish men. A 15 year follow-up in the study of men born in 1933. Eur Heart J 2004; 25: 56–63
- Steptoe A, Wardle J, Pollard TM, Canaan L, Davies GJ. Stress, social support and health-related behavior: A study of smoking, alcohol consumption and physical exercise. J Psychosom Res 1996; 41: 171–80
- Falk A, Hanson BS, Isacsson SO, Ostergren PO. Job strain and mortality in elderly men: Social network, support, and influence as buffers. Am J Public Health 1992; 82: 1136–9
- Reynolds P, Kaplan GA. Social connections and risk for cancer: Prospective evidence from the Alameda County Study. Behav Med 1990; 16: 101–10
- Eng PM, Rimm EB, Fitzmaurice G, Kawachi I. Social ties and change in social ties in relation to subsequent total and cause-specific mortality and coronary heart disease incidence in men. Am J Epidemiol 2002; 155: 700–9
- Kinney AY, Bloor LE, Dudley WN, Millikan RC, Marshall E, Martin C, et al. Roles of religious involvement and social support in the risk of colon cancer among Blacks and Whites. Am J Epidemiol 2003; 158: 1097–107
- Iwasaki M, Otani T, Sunaga R, Miyazaki H, Xiao L, Wang N, et al. Social networks and mortality based on the Komo-Ise cohort study in Japan. Int J Epidemiol 2002; 31: 1208–18
- Appleyard M, Hansen T, Schnohr P, Jensen G, Nyboe J. The Copenhagen City Heart Study, Osterbroundersogelsen: 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
- Schnohr P, Jensen P, Lange H, Scharling H, Appleyard M. The Copenhagen City heart study. Osterbroundersogelsen: Tables with data from the third examination 1991–1994. Eur Heart J 2001; 3: H1–H83
- Jensen OM, Storm HH, Jensen HS. Cancer registration in Denmark and the study of multiple primary cancers, 1943–80. Natl Cancer Inst Monogr 1985; 68: 245–51
- Korn EL, Graubard BI, Midthune D. Time-to-event analysis of longitudinal follow-up of a survey: Choice of a time-scale. Am J Epidem 1997; 145: 72–80
- Greenland S. Avoiding power loss associated with categorization and ordinal scores in dose-response and trend analysis. Epidemiology 1995; 6: 450–4
- Greenland S. Dose-response and trend analysis in epidemiology: Alternatives to categorical analysis. Epidemiology 1995; 6: 356–65
- Allgower A, Wardle J, Steptoe A. Depressive symptoms, social support, and personal health behaviors in young men and women. Health Psychol 2001; 20: 223–7
- Steptoe A, Wardle J. Health behaviour, risk awareness and emotional well-being in students from Eastern Europe and Western Europe. Soc Sci Med 2001; 53: 1621–30
- Broman CL. Social relationships and health-related behavior. J Behav Med 1993; 15: 335–50
- Krantz G, Ostergren PO. Common symptoms in middle aged women: Their relation to employment status, psychosocial work conditions and social support in a Swedish setting. J Epidemiol Community Health 2000; 54: 192–9
- Barefoot JC, Gronbaek M, Jensen G, Schnohr P, Prescott E. Social network diversity and risks of ischemic heart disease and total mortality: Findings from the Copenhagen City Heart Study. Am J Epidemiol 2005; 161: 960–7
- Seeman TE. Social ties and health: The benefits of social integration. Ann Epidemiol 1996; 6: 442–51
- Fox, BH. The role of psychological factors in cancer incidence and prognosis. Oncology (Williston Park) 1995;9:245–53; Discussion 253–6.
- Ell K. Social networks, social support and coping with serious illness: The family connection. Soc Sci Med 1996; 42: 173–83
- Garssen B. Psycho-oncology and cancer: Linking psychosocial factors with cancer development. Ann Oncol 2002; 13(Suppl 4)171–5
- Garssen B. Psychological factors and cancer development: Evidence after 30 years of research. Clin Psychol Rev 2004; 24: 315–38
- Cerhan JR, Wallace RB. Change in social ties and subsequent mortality in rural elders. Epidemiology 1997; 8: 475–81
- Storm HH, Michelsen EV, Clemmensen IH, Pihl J. The Danish Cancer Registry–history, content, quality and use. Dan Med Bull 1997; 44: 535–9
- Kjaergaard J, Clemmensen IH, Storm HH. Validity and completeness of registration of surgically treated malignant gynaecological diseases in the Danish National Hospital Registry. J Epidemiol Biostat 2001; 6: 387–92
- Olsen JH, Andersen A, Dreyer L, Pukkala E, Tryggvadottir L, Gerhardsson de Verdier M, et al. Summary of avoidable cancers in the Nordic countries. APMIS Supplementum 1997; 76: 141–6
- BW Stewart, Kleihues, P. World Cancer Report. Lyon, IARC Press, 2003.
- Gotay CC. Behavior and cancer prevention. J Clin Oncol 2005; 23: 301–10