Obesity and hormone-dependent tumors: Cohort and co-twin control studies based on the Swedish Twin Registry
Article first published online: 30 MAY 2003
Copyright © 2003 Wiley-Liss, Inc.
International Journal of Cancer
Volume 106, Issue 4, pages 594–599, 10 September 2003
How to Cite
Jonsson, F., Wolk, A., Pedersen, N. L., Lichtenstein, P., Terry, P., Ahlbom, A. and Feychting, M. (2003), Obesity and hormone-dependent tumors: Cohort and co-twin control studies based on the Swedish Twin Registry. Int. J. Cancer, 106: 594–599. doi: 10.1002/ijc.11266
- Issue published online: 1 JUL 2003
- Article first published online: 30 MAY 2003
- Manuscript Accepted: 14 APR 2003
- Manuscript Revised: 10 APR 2003
- Manuscript Received: 12 NOV 2002
- Swedish Cancer Society
- John D. and Catherine T. MacArthur Foundation
- Swedish Council for Planning and Coordination of Research
- hormone-dependent tumor;
- body mass index;
Obesity increases the risk of certain cancer types, e.g., cancer of the endometrium, colon and gallbladder. For some other cancer forms, e.g., prostate cancer, the association is less clear. We examined the association between body mass index (BMI) and hormone-dependent tumors, utilizing a cohort of 21,884 Swedish twins born during 1886–1925. Information about BMI at different ages and potential confounding factors was collected prospectively. The Swedish Cancer Registry was used to identify cases of cancer in the prostate (n = 666), breast (n = 607), corpus uteri (n = 150) and ovary (n = 118) during 1969–1997. The material was analyzed as a traditional cohort and with co-twin control analyses that allow for control of genetic influences. Obesity (BMI ≥30 kg/m2) at baseline was positively associated with cancer in the corpus uteri [relative risk (RR) = 3.03, 95% confidence interval (CI) 1.82–5.03], as was BMI at age 25, independently of BMI at baseline. Increased risk was also found for breast cancer but only in older women (≥70 years). Overweight at age 25 was associated with decreased risk of breast cancer (RR = 0.51, 95% CI 0.33–0.78). No association was found for prostate cancer. We conclude that age is an important effect modifier of cancer risk associated with obesity and that obesity and overweight in young adult life may affect cancer risk also later in life. © 2003 Wiley-Liss, Inc.
Obesity has become an increasing health concern in many countries and is related to an increased risk of Type 2 diabetes, coronary heart disease, respiratory complications and certain forms of cancer.1 The prevalence of overweight, defined as body mass index (BMI, weight/height2) of 25–29 kg/m,2 and obesity, defined as BMI ≥30 kg/m,2, has increased rapidly over the last decades; and studies have found a prevalence of obesity of 10–25% in Europe and the United States.2 Overweight and obesity are increasing also among young people3 and is generally more common in women than in men.2
Several studies have found an association between obesity and hormone-dependent cancers. The strongest positive associations and most consistent results have been observed for endometrial cancer and, to some extent but with weaker effects, for postmenopausal breast cancer, while results for other hormone-dependent tumors are less consistent.4, 5, 6, 7, 8 Most studies have focused on recent BMI, which mainly means obesity late in life, as the studied cancer types are more common at older ages. However, overweight and obesity early in life have been associated with reduced risk of premenopausal breast cancer, also with adjustment for recent BMI.9 Studies on the effect of obesity early in life on endometrial and prostate cancer risk have provided inconsistent results.5, 6, 10, 11, 12, 13
Obesity occurs when the energy balance is disturbed and the expenditure is less than the intake. Intake and expenditure can be modified by diet (e.g., high-fat diets) and physical activity (or inactivity). Influence on the energy balance is multifactorial and includes genetic, environmental and psychosocial factors.1 Genetic factors account for a large proportion of the variation in BMI, through susceptibility genes.1, 2, 14 Differences in genetic susceptibility within a population determine who is the most likely to become obese, but the environment determines the opportunity for exposure to external factors affecting energy balance (e.g., diet and physical activity).
Our aim was to further explore the association between BMI at different ages and the risk of developing hormone-dependent tumors, e.g., cancer of the breast, corpus uteri, ovary and prostate. We utilized the Swedish Twin Registry with prospectively collected data on BMI and lifestyle factors such as physical activity and smoking. The registry was analyzed both as a cohort and through co-twin control analyses that take into account early environmental and genetic influences.
MATERIAL AND METHODS
Our study was conducted within a prospective population-based cohort of same-sexed twin pairs, born 1886–1925 and both still living in Sweden in 1961, when the Swedish Twin Registry was established. Self-administered questionnaires regarding lifestyle factors were mailed to the cohort during the years 1961, 1963 and 1967.15 We included the 21,884 subjects who completed the first questionnaire with at least minimal information. This corresponds to approximately 85% of the 25,778 subjects in the original cohort. The questionnaires included information on a wide range of areas, e.g., demographic, social, medical and lifestyle factors. Twin zygosity was determined using the following question: “Were you as children as alike as two peas in a pod?” When both twins answered affirmatively, they were defined as monozygotic. This method has been demonstrated to have sensitivity of 99% and specificity of 92%.16 Questions about weight and height were completed by 94% of subjects.
Information about cancer incidence and date of death was obtained through record linkage to the Swedish Cancer Registry and Cause of Death Registry, both maintained by the National Board of Health and Welfare. The computerized record linkage was made possible by a 10-digit civil registration number, unique for each Swedish citizen. The cohort was followed from 1969 to the end of 1997, cancer diagnosis or death, whichever came first, and included 20,596 subjects, 8,998 males and 11,598 females. Excluded were 1,288 individuals who had died before 1969. Furthermore, for each cancer site-specific analysis, prevalent cancer at baseline was excluded. The study focused on hormone-dependent tumors and included tumors of the breast [International Classification of Diseases, 7th revision (ICD7 170, 607 cases), corpus uteri (ICD7 172, 150 cases), ovary (ICD7 175, 118 cases) and prostate (ICD7 177, 666 cases)]. The classification used by the Swedish Cancer Registry did not allow identification of endometrial cancer until the latest 5 years of the study period; therefore, we used the broader category corpus uteri.
BMI (kg/m2) was used as a measure of relative body weight and calculated from self-reported weight and height. Subjects were asked to report their weight at the time they answered the questionnaire and at ages 25 and 40 years. We categorized BMI into 4 groups according to the WHO criteria for thinness and overweight.17 The categories are BMI <18.5, 18.5–24.9, 25–29.9 and ≥30. The referent category was set to BMI = 18.50–24.9. Approximately 6% (n = 1,222) of subjects lacked information about baseline weight or height, leaving 19,374 subjects for analysis. The numbers of subjects without information about weight at ages 25 and 40 were higher, 32% and 22%, respectively. Thus, 14,131 subjects were included in the analyses of BMI at age 25 and 16,035 subjects in the analyses of BMI at age 40.
Weight change and height
Analyses were also made of the risk related to weight change between age 25 and baseline and height. Weight change was divided into the categories weight loss, no weight increase to 5 kg (reference), 6–10 kg increase, 11–20 kg increase and >21 kg increase. Height was categorized into quartiles, with the upper quartile divided into 2 categories at the 90th percentile.
Confounding factors and effect modifiers
All results were adjusted for age at enrollment. Further confounders that were included as covariates in the statistical models were level of education, smoking habits (nonsmokers, former smokers and current smokers), alcohol consumption (g/month), physical activity at work and during leisure hours and, for women, whether they had children or not. Control of these confounders did not change the results in any material way, but limited the number of subjects available for analyses due to internal nonresponse. Therefore, they were not included in the final analyses. Analyses of BMI at ages 25 and 40 were adjusted for BMI at baseline. To take menopause into account, we made separate analyses of women over age 55 (i.e., we started follow-up after age 55 in these analyses). We further analyzed a possible modifying effect of age through separate analyses of individuals older than 70. Age 70 was chosen as a cut-off point because of the distribution toward older ages during the follow-up period.
We calculated person-years from 1 January 1969 until year and month of death or diagnosis or the end of the study, 31 December 1997. The association between BMI and cancer risk was analyzed using 2 different methods. First, the twin cohort was used as a population-based cohort without considering twin status. The relative risk (RR) of cancer was estimated through Cox proportional hazards modeling using the SAS program PHREG (SAS Institute, Cary, NC). RR estimates are presented with 95% confidence intervals (CIs). All risk estimates are adjusted for age (as a continuous variable). To ensure that 95% CIs were not erroneously narrowed due to dependence within twin pairs, we performed analyses that adjusted variance estimates for correlated outcomes. We accomplished this through the use of an SAS macro that stems from the same theoretical background18, 19, 20 and yields the same results as the published Fortran program of Dr. D.Y. Lin.21 In simple terms, variance estimates are increased in magnitude proportional to the degree of extra correlation within twin pairs. Thus, adjusted 95% CIs are more conservative than unadjusted ones. If correlations within twin pairs are not different from what is observed between unrelated individuals in the cohort with respect to cancer risk, adjusted and unadjusted variance estimates are identical. RR estimates are not altered by this procedure.
Second, the same material was also analyzed using the co-twin control method. These analyses include only disease discordant twin pairs. In co-twin control analyses, the material is analyzed as a matched case-control study where a twin pair constitutes the case and matched control. We estimated RRs through conditional logistic regression. Monozygotic and dizygotic twins were not separated in the analyses because the number of monozygotic twin pairs was too low to allow separate analyses. Because the 2 twins in a pair share their childhood environment, co-twin control analyses control for unmeasured childhood environmental effects. Further, because monozygotic twins share all of their genes and dizygotic twins share 50% of their segregating genes, we also partially controlled for genetic factors.
The median age at baseline was 56 years (range 44–83). The cohort was followed for 29 years, with a median follow-up time of 22 years for men and 26 years for women. Table I presents the results of the cohort analyses. BMI ≥30 was associated with a risk estimate of 3.2 (95% CI 2.0–5.2) for cancer of the corpus uteri. For tumors in the breast, we found a small but unstable risk increase (RR = 1.2, 95% CI 0.8–1.6). Risk estimates for prostate cancer were close to unity (RR = 1.0, 95% CI 0.6–1.5). Further adjustment for education, smoking, alcohol consumption, number of children and physical activity at work or leisure time did not change the results (data not shown).
|Number||RR1||95% CI||Number||RR||Number||RR||95% CI||Number||RR||95% CI|
|Breast (170) (n = 580)||11||1.0||0.6–1.8||321||1.0||208||1.1||1.0–1.4||40||1.2||0.8–1.6|
|Corpus uteri (172) (n = 137)||1||0.4||0.1–3.1||69||1.0||46||1.3||0.9–1.9||21||3.2||2.0–5.2|
|Ovary (175) (n = 111)||1||0.4||0.1–3.1||69||1.0||39||1.0||0.7–1.5||2||0.3||0.1–1.1|
|Prostate (177) (n = 631)||6||1.4||0.6–3.1||355||1.0||248||1.0||0.8–1.2||22||1.0||0.6–1.5|
Co-twin control analyses were based on a smaller number of subjects, and consequently, the risk estimates are more unstable (Table II). The results confirm the finding in the cohort analysis for cancer of the corpus uteri, but the risk estimate has wide 95% CIs. For breast cancer the point estimate was higher than in the cohort analysis, and for prostate cancer a slightly increased risk was found for BMI ≥30; but again, the effect estimate is unstable.
|Number of cases||Number of controls||RR||95% CI||Number of cases||Number of controls||RR||Number of cases||Number of controls||RR||95% CI||Number of cases||Number of controls||RR||95% CI|
|Breast (170) (n = 402)||6||9||0.6||0.2–1.9||232||244||1.0||134||125||1.3||0.9–1.8||30||24||1.5||0.8–3.0|
|Corpus uteri (172) (n = 109)||1||1||1.0||0.1–16.0||60||58||1.0||30||40||0.8||0.4–1.6||18||10||2.5||0.7–8.8|
|Ovary (175) (n = 95)||1||2||0.5||0.0–5.5||59||62||1.0||33||27||1.3||0.6–2.7||2||4||0.02||0.0–2|
|Prostate (177) (n = 331)||3||2||2.0||0.2–22.1||190||185||1.0||125||136||0.9||0.6–1.3||13||8||1.6||0.6–4.1|
The results for postmenopausal cancer (defined as cancer occurring after age 55) were virtually the same as those shown in Table I (data not shown). The number of premenopausal cancers was too small for separate analyses [only 38 (7%) breast cancer cases, 18 (13%) corpus uteri cancer cases and 7 (6%) ovarian cancer cases].
For women who developed cancer after the age of 70 years, a BMI at baseline between 25 and 29.9 was associated with a 2-fold increased risk of cancer of the corpus uteri, and BMI ≥30 was associated with a risk estimate of 4.9 (95% CI 2.5–9.8) (Table III). A 60% risk increase was found for breast cancer among women in the highest BMI category (95% CI 1.1–2.4). Results were close to unity for prostate cancer after the age of 70.
|Number||RR1||95% CI||Number||RR||Number||RR||95% CI||Number||RR||95% CI|
|Breast (170) (n = 266)||3||0.5||0.2–1.6||165||1.0||91||1.2||0.9–1.5||7||0.5||0.2–1.1|
|Corpus uteri (172) (n = 78)||1||0.6||0.1–4.3||47||1.0||21||0.9||0.6–1.6||9||2.3||1.1–4.5|
|Ovary (175) (n = 58)||0||—||—||40||1.0||17||0.9||0.5–1.6||1||0.3||0.0–2.2|
|Prostate (177) (n = 151)||1||0.7||0.1–4.5||88||1.0||56||1.0||0.7–1.4||6||1.1||0.5–2.5|
|Breast (170) (n = 314)||8||1.4||0.7–2.9||156||1.0||117||1.2||0.9–1.5||33||1.6||1.1–2.4|
|Corpus uteri (172) (n = 59)||0||—||—||22||1.0||25||1.9||1.1–3.4||12||4.9||2.5–9.8|
|Ovary (175) (n = 53)||1||0.9||0.1–7.2||29||1.0||22||1.2||0.7–2.0||1||0.3||0.0–1.9|
|Prostate (177) (n = 480)||5||1.5||0.6–3.5||267||1.0||192||1.0||0.8–1.2||16||0.9||0.6–1.5|
Table IV shows the results for BMI at ages 25 and 40 years, adjusted for BMI at baseline. The numbers of subjects in these analyses are smaller because the nonresponse was higher on these questions. Therefore, results are shown only for cancer in the breast, corpus uteri and prostate. Very few subjects had a BMI ≥30 at age 25; therefore, results are shown only for BMI ≥25. For cancer of the corpus uteri, a high BMI at age 25 as well as at age 40 was associated with increased risk independent of baseline BMI (RR = 1.9, 95% CI 1.2–3.0 and RR = 2.0, 95% CI 0.9–4.4, respectively). For breast cancer, an inverse association was observed for BMI ≥25, most evident for BMI at age 25 (RR = 0.5, 95% CI 0.3–0.8). An increased risk of prostate cancer was found for subjects with a BMI <18.5 at age 40 (RR = 2.5, 95% CI 1.1–5.5).
|Diagnosis (ICD7)||Breast (170)||Corpus uteri (172)||Prostate (177)|
|Number||RR1||RR2||95% CI||Number||RR||RR||95% CI||Number||RR||RR||95% CI|
|BMI at age 25||(n = 421)||(n = 107)||(n = 504)|
|BMI at age 40||(n = 462)||(n = 116)||(n = 542)|
Weight gain was related to increased risk of breast cancer, with risk estimates that increased with increasing weight gain. For subjects who had gained 21 kg or more in weight between age 25 years and baseline, the risk was estimated to be 2.1 (95% CI 1.3–3.3) (Table V). The corresponding risk estimate for cancer of the corpus uteri was 2.5 (95% CI 1.1–5.4).
|Diagnosis (ICD7)||<0 kg||0–5 kg2||6–10 kg||11–20 kg||≥21 kg|
|Number||RR1||95% CI||Number||RR||Number||RR||95% CI||Number||RR||95% CI||Number||RR||95% CI|
|Corpus uteri (172)||18||0.8||0.4–1.6||32||1.0||23||1.0||0.6–1.8||24||1.2||0.7–2.2||10||2.5||1.1–5.4|
The tallest 10% of women had an increased risk of breast cancer (RR = 1.5, 95% CI 1.1–2.0). For ovarian cancer, an increased risk was indicated among women with above median height, with further increased risk in the 2 upper exposure categories (RR = 2.8, 95% CI 1.4–5.7 and RR = 2.4, 95% CI 1.1–5.0, respectively). All risk estimates for prostate cancer related to height were at unity (Table VI).
|Diagnosis (ICD7)||1st quartile1||2nd quartile13||3rd quartile1||76–89%1||90th percentile1|
|Number||RR2||95% CI||Number||RR||Number||RR||95% CI||Number||RR||95% CI||Number||RR||95% CI|
|Corpus uteri (172)||24||0.8||0.5–1.4||28||1.0||22||1.0||0.5–1.7||17||1.4||0.7–2.5||16||1.3||0.7–2.5|
Our results confirm the findings in previous studies of an increased risk of endometrial cancer related to overweight and obesity. The highest risk increase was found for endometrial cancer at older ages (>70 years). Overweight and obesity at age 25 and age 40 were associated with increased risk of endometrial cancer independently of current BMI, and weight gain between age 25 years and baseline age also increased risk. The increase of endometrial cancer risk was also evident in co-twin control analyses, where control for early environmental and genetic factors was possible. However, these results were based on a small sample, and results were unstable. We observed a dual effect of obesity on breast cancer risk: an inverse association with overweight and obesity at age 25 and a modest positive association between BMI at baseline and cancer risk in older women (>70 years). Weight gain was associated with increased risk of breast cancer, as was height. Height was also associated with an increased risk of ovarian cancer. We found no evidence that overweight and obesity affect prostate cancer risk, nor did weight gain or height; but there was an indication of an increased risk of prostate cancer in lean subjects, especially for BMI at age 40. However, the number of lean subjects was small in most of the analyses, which made risk estimates unstable in this category.
In the co-twin control analyses, a 50% increase in the risk of breast cancer was found in obese subjects, and a similar pattern was evident for prostate cancer. However, these risk estimates were very unstable due to the small sample size in the co-twin control analyses, which gives wide 95% CIs.
Our study has several advantages. The nationwide Swedish Cancer Registry has been documented to be 95% complete for prostate cancer and 98–99% complete for cancers of the breast and female genitalia.22 Approximately 97% of cases are morphologically verified.23 Information was available on a large number of potentially important confounding factors, though we found that control of confounding did not change the results. All information about height, weight and confounding factors was collected prospectively, which means that differential exposure misclassification cannot affect the results. Another strength is the co-twin control analyses, where the unaffected twin is treated as a matched control to the twin with the disease. This method allows for control of genetic influences and early childhood environment. Due to power considerations, both monozygotic and dizygotic twin pairs were included; therefore, only partial control of genetic factors could be made. For perfect control of genetic factors, ideally, only monozygotic twin pairs should be used. For breast and prostate cancers, we found somewhat different results in the co-twin control analyses. This could be due either to random variation because of the small sample size or to some genetic or early environmental factor that could mask an association between BMI and these diseases in the cohort analyses.
Our study has some limitations as well. Information about height and weight was collected before the start of follow-up, and weight changes during the study period of up to 29 years could lead to nondifferential misclassification of BMI. Furthermore, the anthropometric information was self-reported, which could also lead to nondifferential misclassification of BMI. This would cause a dilution of the effect estimates. Furthermore, the number of cases of ovary cancer was too small to detect modest effects. The median age at baseline was 56 years; therefore, evaluation of the effect of BMI on premenopausal tumor risk was not possible.
Increased risk of endometrial cancer associated with high BMI has consistently been described in many previous studies.4, 5, 6, 10, 11, 24, 25 Similar to the present findings, the majority of studies have also reported a greater excess risk at advanced ages. However, the effect of overweight and obesity at younger ages on the risk of developing endometrial cancer later in life has not been fully explored. One limitation has been that the numbers of subjects with overweight and obesity at a young age have been low, and as several studies have used quartiles or tertiles to categorize BMI, the highest BMI category has encompassed subjects who would not be classified as overweight according to the WHO criteria.5, 11, 24 Our results suggest that although BMI close to diagnosis has the largest impact on endometrial cancer risk, overweight and obesity at ages 25 and 40 may further increase the risk. Effects of long-term overweight and obesity have also been shown in some previous studies.10, 25 Furthermore, Terry et al.26 analyzed the association between weight and endometrial cancer risk in the Swedish Twin Registry and found a similar pattern, though their results were based on a shorter follow-up and quartiles were used to categorize weight. Our finding of increased risk related to weight gain supports the results of previous studies,10, 11, 24 though a few studies did not report similar findings.5, 13
One proposed mechanism for the association between obesity and cancer is through hormonal effects. Levels of circulating estrogens increase with increasing body mass through different mechanisms. The androgen precursor androstenedione is transformed to estrone in adipose tissue, which is the primary source of estrogen in postmenopausal women.27 Furthermore, obesity is inversely associated with levels of sex hormone–binding globulin (SHBG), which is a protein that reduces the availability of estrogen in the body. In men, levels of testosterone and SHBG have been inversely associated with BMI.28 Obesity is also related to insulin resistance, compensatory hyperinsulinemia and increased levels of free IGF-I,29 which may be related to carcinogenesis.30
Results for other cancer forms are not as consistent. The majority of studies suggest that obesity is associated with a decreased risk of premenopausal breast cancer and a modestly increased risk of postmenopausal breast cancer.4, 6, 7 In the present study, we did not evaluate premenopausal breast cancer, but we found a protective effect of overweight and obesity at age 25 in relation to later development of breast cancer, while overweight and obesity later in life were associated with increased risk. Weight gain has generally been associated with increased breast cancer risk,31, 32 which was also confirmed in the present study. Our finding of increased risk of breast cancer in tall women has been reported earlier.7, 31, 32, 33 It has been suggested that early obesity may protect against premenopausal breast cancer by causing more frequent anovulatory ovarian cycles.9 The effect of early obesity on postmenopausal breast cancer is less clear.34 It has been hypothesized that continued obesity after the teenage years would lead to an increased risk of postmenopausal breast cancer,9 but we found evidence of a protective effect of overweight also at young adulthood.
Studies of ovarian cancer have provided inconsistent results, with some showing an increased risk associated with obesity, others no association and some even a protective effect.4, 6, 35, 36, 37 Increased risk of ovarian cancer related to height has been reported previously but only for the histologic subtype of borderline serous tumors.38 Other studies found no association39 or even an inverse relationship.40
The majority of studies of prostate cancer related to BMI have found no association, though some have found positive associations with total or with certain subtypes of prostate cancer.6, 8, 12, 13, 41, 42 However, no consistent patterns in the results have emerged. Two studies report increased risk of developing prostate cancer at younger ages6, 43 and no association43 or even a protective effect6 at older ages. However, most other studies have not presented age-specific results. The lack of significant findings in our data suggests that the effect of BMI on prostate cancer risk is at the most modest.
In conclusion, our results confirm that obesity increases the risk of some hormone-dependent cancers in women. We also demonstrate that age is an important effect modifier of the cancer risk associated with obesity. The risk increased with age for cancer of the corpus uteri, while for breast cancer risk was increased only among women over 70 years at diagnosis. The results also indicate that obesity in young adult life may influence cancer risk later in life, increasing the risk of cancer of the corpus uteri and decreasing the risk of breast cancer. Differences in the age distribution between studies as well as differences in the age at which information about obesity was collected might explain inconsistencies in the results between different studies of, e.g., breast cancer and prostate cancer. Other explanations might be differences in the definition and measurement of anthropometric factors as well as in the categorization of BMI.
- 2WHO. Obesity: preventing and managing the global epidemic. WHO Technical Report Series 894. Geneva: WHO, 2000.
- 17WHO. Physical status: the use and interpretation of anthropometry. WHO Technical Report Series 854. Geneva: WHO, 1995. 312–344, 420, 423, 452.
- 23National Board of Health and Welfare. Cancer incidence in Sweden 1995. Stockholm: Norstedts Tryckeri, 1998.