Risk Factors for Breast Cancer at Young Ages in Twins: An International Population-Based Study

Transcription

1 Risk Factors for Breast Cancer at Young Ages in Twins: An International Population-Based Study Anthony J. Swerdlow, Bianca L. De Stavola, Birgitta Floderus, Niels V. Holm, Jaakko Kaprio, Pia K. Verkasalo, Thomas Mack Background: Breast cancer etiology in women may relate to exposures early in life as well as in adulthood, but it has been difficult to gain information on childhood variables, and evidence on their effects is very limited. Comparison of risk factor levels between affected probands and unaffected twins in twin pairs can provide a unique method to investigate risk factors that act in early life. Methods: We conducted four population-based case control studies of breast cancer risk in twins in Denmark, England and Wales, Finland, and Sweden and pooled the data from these studies. The case patients were 400 women with incident breast cancer before the age of 50 years, and the control subjects were their twin sisters who had not had breast cancer by that time. Data on risk factors (e.g., demographic and anthropomorphic variables, reproductive history, and family history) were collected by interview and by a mailed questionnaire and were analyzed by calculating matched odds ratios [ORs]. All statistical tests were two-sided. Results: The risk of breast cancer was increased for women who were less obese (OR = 1.44, 95% confidence interval [CI] = 1.08 to 1.91) or taller (OR = 1.27, 95% CI = 0.95 to 1.70) than their co-twin at age 10 years, for women who developed breasts earlier than their co-twin (OR = 1.53, 95% CI = 1.14 to 2.06), and for women who had a smaller waist-to-hip ratio at age 20 years than their co-twin (OR = 1.79, 95% CI = 1.00 to 3.21). Analysis according to family history of breast cancer showed that the associations of childhood height and weight with risk of breast cancer were only apparent in women without a family history of breast cancer. Conclusions: Childhood growth before puberty may affect the risk of premenopausal breast cancer, at least in women without a family history of breast cancer. The distribution of body fat in young adulthood may also be related to breast cancer risk. [J Natl Cancer Inst 2002;94: ] There is increasing evidence that the etiology of breast cancer may reflect the influences of exposures that occur over many different stages of a woman s life: as well as genetic factors, there is evidence for factors that act prenatally (1), in childhood and puberty (2), in relation to reproduction (2) and, for postmenopausal breast cancer, factors that act during and after menopause (2). Many of these factors are probably hormonal and/or nutritional and are difficult to measure, especially in retrospect for factors that acted during much earlier periods of a woman s life. The variables that have been investigated in most epidemiologic studies are probably very indirect markers of the underlying factors and have often been imprecisely ascertained. This situation particularly applies to exposures occurring in childhood, for which there have been almost no data based on information gathered at the time, and the variables that have been investigated for instance, anthropometric variables and diet are extraordinarily difficult for study subjects to remember with accuracy. The comparison of exposures in affected versus unaffected twins provides a novel approach to address these issues, because ascertainment of risk factors at young ages might be better obtained by retrospective comparison between the two sisters than by comparison of a woman with her contemporaries in the general population. However, twins with breast cancer are not common and, to our knowledge, studies of twins with breast cancer have been conducted on a population basis only in Denmark, England and Wales, Finland, and Sweden. We therefore combined and analyzed data from these four previously unpublished studies to investigate breast cancer etiology. PATIENTS AND METHODS Our analyses are based on data aggregated from four case control studies that were conducted in Denmark, England and Wales, Finland, and Sweden. In each study, twins with breast cancer were identified on a population basis as described below, and data on potential risk factors were then collected during interviews or from written questionnaires obtained from the case patients with breast cancer and from their unaffected co-twins, who were the control subjects. The studies were performed with the approval of the appropriate local ethical review board in each country and with the informed consent of the subjects. A common case control design was used for the studies conducted in the four countries, and the questionnaire used was coordinated between the four studies so that it was similar, with modification for local circumstances. Information collected from the questionnaires included the following risk factors: demographic variables, anthropometric variables in childhood and adulthood, pubertal and reproductive history, exercise history, and history of breast cancer in first-degree relatives. Subjects were asked to respond to questions about these factors as they pertained to the subject themselves and, where different, as they pertained to Affiliations of authors: A. J. Swerdlow, Section of Epidemiology, Institute of Cancer Research, Sutton, U.K.; B. L. De Stavola, Department of Epidemiology and Population Health, London School of Hygiene and Tropical Medicine, U.K.; B. Floderus, Institute of Environmental Medicine, Karolinska Institutet, Stockholm, Sweden; N. V. Holm, Department of Epidemiology, Institute of Public Health, University of Southern Denmark, and Department of Oncology and Radiotherapy, Odense University Hospital, Odense, Denmark; J. Kaprio, Department of Public Health, University of Helsinki, and Department of Mental Health and Alcohol Research, National Public Health Institute, Helsinki, Finland; P. K. Verkasalo, Department of Public Health, University of Helsinki; T. Mack, Department of Preventive Medicine, University of Southern California Keck School of Medicine, Norris Comprehensive Cancer Center, Los Angeles. Correspondence to: Anthony J. Swerdlow, D.M., Section of Epidemiology, Institute of Cancer Research, Cotswold Rd., Sutton, Surrey SM2 5NG, U.K. ( a.swerdlow@icr.ac.uk). See Notes following References. Oxford University Press 1238 ARTICLES Journal of the National Cancer Institute, Vol. 94, No. 16, August 21, 2002

2 their co-twins. For some variables, the questions required absolute answers (for instance, the number of children borne). Many questions, however, required relative answers (for instance, which twin was taller at a particular age). In Denmark, a national register of all twins born in the country from 1870 through 1910 was compiled in 1954 from birth registers and was later expanded to include same-sex twins born from 1911 through A questionnaire was mailed to these twins (or to their relatives, if the twin was deceased) to determine zygosity and medical history. Twin pairs in which either member had died before the age of 6 years were excluded, as were pairs in which either twin had emigrated, irrespective of their date of emigration. All other twin pairs were considered eligible for the study. Zygosity was identified for over 92% of eligible twins. Breast cancers among the eligible twins were identified by their responses to questionnaires on their health status that were periodically sent to the twins themselves (the last questionnaire was sent in 1979), by examination of death certificates for twins who had died, and by matching of the twin register against the Danish Cancer Register for cancers incident from 1943 through 1979 (3). Only breast cancers with a verification of diagnosis from hospital records or other medical reports were included. Information on risk factors for breast cancer in twin pairs with at least one twin alive was obtained by interviews that were conducted from 1983 through The Finnish twin cohort was created in 1974 by a computer search of the Central Population Register and consisted of all same-sex twins born before 1958 in Finland who were still alive on January 1, A questionnaire to determine zygosity was mailed to these twins in 1975, and breast cancers in female twin pairs were ascertained by matching the twins to the national cancer register for cancers incident from 1953 through 1983 (4). All breast cancer cases were histologically verified, and the diagnosis was checked by a breast cancer surgeon. The breast cancer case patients and their co-twins were interviewed in person during 1983 and 1984 about risk factors (5). The earliest cohort of the Swedish Twin Registry was established in 1961 and consisted of all same-sex twin pairs who were born in Sweden from 1886 through 1925 in which both members of the pair were alive and had answered a questionnaire in 1961 that included questions on zygosity (6). The registry was expanded in 1971 by ascertaining in a similar way the zygosity of twins who were born from 1926 through 1967; a questionnaire was sent to those in this latter group of twins who were born from 1926 through 1958 (7). The study population for our study contained all female twins from the earliest Swedish cohort who were born from 1900 through 1925 and all female twins from the expanded cohort who were born from 1926 through Information on cancer incidence was obtained by linkage to the national cancer registry. Twins who were diagnosed with breast cancer from 1958 through 1982 were selected as potential case patients for the Swedish case control study on breast cancer risk factors. Twin pairs were eligible for the study if the breast cancer diagnosis could be verified by medical records (which were scrutinized by an oncologist) and if at least one member of the pair was alive on January 1, Breast cancer patients and their co-twins were interviewed during personal visits to their homes from 1983 through 1985 or, in a few instances, asked in a mailed questionnaire for information on risk factors. There is no national twin register in England or Wales; thus, twins with breast cancer were identified in a different way. National cancer registration records for breast cancers incident from 1971 through 1989 in women who were born in 1931 or later were cross-matched against national birth registers and against the National Health Service Central Register, in which twins are virtually always entered sequentially at birth. The breast cancer diagnoses were those accepted by the cancer registry. Information on zygosity and potential risk factors for breast cancer was ascertained from responses to questionnaires mailed to the twins and from telephone interviews conducted with the twins from 1994 through 1996 (8). Statistical Analysis For the combined analyses, the four datasets were checked for omissions and logical errors and then analyzed in parallel, using the same methods and categorization for each. Results for the four datasets analyzed separately were checked for heterogeneity. Because there was no statistically significant heterogeneity in any instance, the results presented are for the overall dataset for the four countries combined. The case control data included breast cancers incident in women at a wide spectrum of ages. Our analysis is restricted to breast cancers incident in women aged less than 50 years, however, because this age group is the surrogate usually taken for premenopausal women and also because dichotomization of data at age 50 years has been found to modify anthropometric associations of breast cancer (9). (Although the questionnaires collected information on menopausal status, that information could not be used to categorize the co-twins for this analysis because the questions about menopausal status addressed to the interviewee about her co-twin referred to menopausal status at the time of interview, not at breast cancer diagnosis). For twin pairs in which both women had been diagnosed with breast cancer, the following criteria were used to decide which woman would be considered the case patient and which would be considered the control subject. If the earlier date of incidence in the pair was before the period for which population-based ascertainment of breast cancer had been conducted in that country (such extra cancers were revealed by the questionnaires), the pair was excluded completely from analysis because the first cancer would not fulfill the study criteria to be a case and the second cancer would lack a control subject (i.e., an unaffected twin) at the time of cancer incidence. If the first breast cancer in the twin pair occurred during the period of incidence included in the study for that country, the twin with the cancer was considered the case patient and her co-twin, who at that time had not had breast cancer, was considered the control subject. Any subsequent breast cancer in the co-twin was not analyzed as a second case because, even if it occurred during the study incidence period, there was no unaffected twin then available who could be used as a control subject. (The shortest interval between cancer occurrences in twin pairs in which both members were diagnosed with breast cancer was 2 months). The risk of breast cancer in relation to exposure variables was examined by calculating matched odds ratios (ORs) using conditional logistic regression (10), with 95% confidence intervals (CIs) based on a normal approximation. All statistical tests were two-sided. All analyses (except those involving reproductive variables) were adjusted for parity, which is a major known risk factor for breast cancer and a potential confounder. We did not adjust for age at first birth, because information about that factor was missing for a large number of subjects. Linear trends and Journal of the National Cancer Institute, Vol. 94, No. 16, August 21, 2002 ARTICLES 1239

3 heterogeneity of risks between countries and between subgroups by family history and zygosity were assessed by conditional likelihood ratio tests (10). Although for some twin pairs, both sisters had replied to the questionnaires, for other twin pairs, only one sister had replied. We therefore had to decide whether to include these latter pairs in our analysis by using one sister s replies about both members of the pair. To make this decision, we assessed the reliability of cross-reports (reports by one twin about her co-twin) for each risk factor by cross-classifying the replies given by each subject about herself and her co-twin in pairs in which both of the twins had replied. For the risk factors presented in this article, we found that contradictory replies (e.g., each twin reporting that she was taller than the other at a particular age) were uncommon. Therefore, when both members of a pair had responded, we analyzed the data from both the case patient and her co-twin, after excluding the few pairs with contradictory replies; in pairs for which only one sister had responded, we used her responses about both members of the pair. In instances where both sisters had responded and one had reported a difference between the sisters for a comparative variable and the other had reported no difference between the sisters for the same variable, we used the response that reported a difference for the analysis. We used this approach because we believed that it was more likely that a response of the same for a variable such as height would include small differences, because it is unlikely that there would actually be no measurable difference. The analyses were repeated omitting pairs in which one twin reported a difference and the other reported no difference but, because the results were similar to those including such pairs and because the significant results remained so, these subset analyses have not been presented. In addition to examining risks of breast cancer in the dataset overall, we also examined the risks in two subgroupings that we decided on a priori: monozygotic and dizygotic twins and familial and nonfamilial breast cancer cases. For the latter categorization, all case patients who were known to have had a first degree relative with breast cancer (including twin pairs in which both sisters had breast cancer) were classified as familial breast cancer cases, and the remainder were classified as nonfamilial breast cancer cases. We obtained data on cancer incidence in the co-twins from national cancer registries, which at the time of data extraction were considered complete through the end of 1986 for Denmark, through the end of 1993 for England and Wales, through the end of 1996 for Finland, and through the end of 1995 for Sweden. The information used about cancers in first-degree relatives other than co-twins was that collected during the interviews and from responses to the questionnaires. RESULTS This study included 400 cases of breast cancer that were diagnosed in women who were younger than 50 years and were a member of a twin pair: 50 cases were from Denmark, 217 were from England and Wales, 27 were from Finland, and 106 were from Sweden (Table 1). One hundred fifty-two of the cases of breast cancer occurred in women who were younger than 40 years, and the remaining 248 cases occurred in women who were aged years. Responses to questionnaires were available from both sisters in 203 twin pairs, from only the case patient of 35 twin pairs, and from only the control subject of 162 twin pairs. Zygosity was known for 374 (93.5%) of the twin pairs; 149 were monozygotic and 225 were dizygotic. Among the case patients, 72 had a family history of breast cancer (as defined in the Patients and Methods section) and 328 had no such history. In 38 pairs of twins, both sisters had had breast cancer at the time of the first interview. For the twins who originated in England and Wales and in Sweden, data were available to calculate the time from diagnosis to interview: for case patients the average time was 10.9 years (standard deviation 5.7 years), and for control subjects (from diagnosis date for their matched case) it was, on average, 12.5 years (standard deviation 6.1 years). Table 2 shows the relative risks of breast cancer according to anthropometric characteristics of the twins at various ages, after adjustment for parity and for height and/or weight at the same age. (Without adjustment, the results were in the same direction but less marked.) Breast cancer risk was not associated with relative weight at birth (OR for women who were less heavy than their co-twin at birth 1.03, 95% CI 0.82 to 1.29). Breast cancer risk was greater in women who were less heavy than their co-twin in childhood and women who were taller than their co-twin in childhood: the increase was statistically significant for women who weighed less than their co-twin at age 10 years (OR 1.44, 95% CI 1.08 to 1.91) but not statistically Table 1. Breast cancer in twins: description of eligible twin pairs and those included in the study, by country Country of twins origin Denmark England and Wales Finland Sweden No. of potentially eligible twin pairs * 134 No. of pairs included in the study (i.e., questionnaire response for at least one sister) No. of pairs in which both members were dead 9 12 NK* 5 No. of refusals, women too ill, or no trace for both members of pair* * 23 Year of birth, range Year of breast cancer diagnosis for case patients, range Year of questionnaire, range Age at diagnosis for case patients (y), range *Because of the study design in Finland, the number of otherwise eligible pairs in which both twins were dead is not known (NK), but it is probably very small. In addition to exclusions because of refusal, five pairs of otherwise eligible twins in Finland were excluded because one member of the pair had died, and interviews in Finland were only attempted for pairs in which both sisters were alive. At time of approach for interview ARTICLES Journal of the National Cancer Institute, Vol. 94, No. 16, August 21, 2002

4 Table 2. Relative risks of breast cancer in relation to anthropometric measures at various ages and to age at pubertal milestones* Measures or milestones case patients control subjects same reponse for case patients and control subjects No. of instances in which data are not known or twin pairs disagreed OR (95% CI) for factor-positive Height Taller at age 7 y (0.91 to 1.61) Taller at age 10 y (0.95 to 1.70) Taller at age 20 y (0.94 to 1.58) Weight Less heavy at birth (0.82 to 1.29) Less heavy at age 7 y (0.97 to 1.69) Less heavy at age 10 y (1.08 to 1.91) Less heavy at age 20 y (0.98 to 1.63) Body shape at age 20 y Larger waist (0.66 to 1.10) Larger hips (0.77 to 1.28) Larger breasts (0.88 to 1.47) Puberty First period earlier (0.88 to 1.42) First regular period earlier (0.94 to 1.64) Breast development earlier (1.14 to 2.06) *Twenty-two pairs of twins were omitted from these analyses because data on the adjustment variables were missing. OR odds ratio; CI confidence interval. In no instance did more than eight (2%) pairs disagree. All ORs were adjusted for parity. ORs for height were also adjusted for weight at the same age and vice versa (because of this adjustment, the analyses for weight are concerned more with obesity than with weight per se and are discussed in relation to obesity in this paper). ORs for pubertal variables were adjusted for height and weight at age 10 years. Referent group is factor-negative in case patients (i.e., control subject positive). 10 years in Nordic countries; 11 years in England and Wales. 20 years in Nordic countries; 18 years in England and Wales. P.01, Wald test. P.005, Wald test. significant for women who were taller than their co-twin at age 10 years (OR 1.27, 95% CI 0.95 to 1.70). After further adjustment for age at menarche, the risk of breast cancer for women who weighed less than their co-twin at age 10 years was essentially unchanged (OR 1.42, 95% CI 1.07 to 1.89; P.02; data not shown), and adjustment for weight at age 20 years left the OR unchanged but widened the CI. A previous study (11) found that women who were overweight in both childhood and young adulthood had a lower risk of breast cancer than those who were overweight on only one of, or neither of, these occasions. We therefore examined the risk of breast cancer for women who were heavier than their co-twin at both 10 and 20 years of age and found that these women had virtually the same risk of breast cancer as women who were heavier than their co-twin only at age 10 years (data not shown). This result was not surprising, given that most of the women who were heavier than their twin at age 10 years were also heavier at age 20 years. Because increased height at age 10 years may indicate an early growth spurt, we examined the breast cancer risk for women who were taller than their co-twin at age 10 years but not at age 20 years and found that those women had a lower risk of breast cancer than did women who were the same height as their co-twin at both ages (OR 0.80; 95% CI 0.35 to 1.83) (data not shown). The results for weight adjusted for height were effectively analyses of obesity. We also analyzed risk of breast cancer in relation to obesity by a second method, that is, by analyzing risk in subcategories cross-classified by height and weight for instance, taking as more obese than their co-twin women who were both heavier and shorter than their sister. The results from this analysis (data not shown) were similar to those of weight adjusted for height. The risk of breast cancer was not statistically significantly associated with the relative size of the waist, hips, or breasts of the twins at age 20 years (Table 2). Women who had had a smaller waist-to-hip ratio than their co-twin (i.e., a smaller waist and same-sized or larger hips, or a same-sized waist and larger hips) had an increased risk of breast cancer compared with women who had a larger waist-to-hip ratio than their co-twin (OR 1.79, 95% CI 1.00 to 3.21 [data not shown]), but this result was based on only 55 twin pairs who reported one of these two options for waist-to-hip ratio and was of borderline statistical significance. Women who reached menarche before their twin and those who menstruated regularly before their twin had statistically nonsignificantly increased risks of breast cancer. Women who developed breasts before their twin had a statistically significantly increased risk of breast cancer (P.005; OR 1.53, 95% CI 1.14 to 2.06). These analyses were adjusted for parity and for height and weight at the same age (Table 2); the results were in the same direction, albeit less strongly, in the crude, unadjusted analyses. After further adjustment for age at menarche, the risk of breast cancer for women who developed breasts before their twin remained statistically significant (OR 1.59, 95% CI 1.17 to 2.16) (data not shown). When we restricted the analyses in Table 2 to twin pairs in which both sisters replied to the questionnaires, the effect of weight adjusted for height at age 10 years was little changed by the restriction and remained statistically significant (OR 1.60, 95% CI 1.09 to 2.35), the effect of height adjusted for weight at age 10 years became statistically significant (OR 1.49, 95% CI 1.01 to 2.19), and the effect of earlier breast development ceased to be statistically significant (OR 1.13, 95% CI 0.77 to 1.65). Journal of the National Cancer Institute, Vol. 94, No. 16, August 21, 2002 ARTICLES 1241

5 When we repeated the analyses in Table 2 separately for monozygotic and dizygotic twins (shown for selected variables in Table 3), the effects of height, weight, and puberty on breast cancer risk were almost always greater in the dizygotic twins than in the monozygotic twins. These findings were based on small numbers, however, and the only factors with a statistically significant effect on breast cancer risk in dizygotic twins were weight at age 10 years and earlier breast development (Table 3). The anthropometric factors were also examined separately for familial and nonfamilial breast cancers (Table 3), as defined in the Patients and Methods section. The effects of childhood height and weight on risk of breast cancer were apparent for the nonfamilial cases. They were not apparent, albeit with wide confidence intervals, for the familial cases. For earlier age at breast development, each group showed an appreciably increased risk, although this increase was statistically significant only within the nonfamilial group (Table 3). For earlier age at menarche, however, both groups showed similar effects (OR in familial cases 1.12, 95% CI 0.65 to 1.93; OR in nonfamilial cases 1.12, 95% CI 0.85 to 1.47). For each of the measures listed in Table 3, the OR for monozygotic twins was not statistically significantly different from the OR for dizygotic twins, and the OR for familial cases was not statistically significantly different from the OR for nonfamilial cases. The ORs of breast cancer associated with participation in sports at various ages were less than 1 but were not statistically significant (Table 4), and there was no consistent pattern of the effect of sports participation in data that were analyzed separately for monozygotic and dizygotic twins (data not shown). The effect on risk of participation in sports 5 years before breast cancer diagnosis was statistically significant for the women whose breast cancer was familial (OR 0.23, 95% CI 0.06 to 0.90) but not for those whose breast cancer was nonfamilial (OR 0.81, 95% CI 0.55 to 1.18) (data not shown). Because Bernstein et al. (12) found a differential effect of exercise on breast cancer risk according to whether women were parous or nulliparous, we examined separately the effect of adult exercise on breast cancer risk among parous and nulliparous women. The ORs in these groups were 0.71 (95% CI 0.49 to 1.03) and 3.00 (95% CI 0.27 to 33.72), respectively (data not shown). Being parous (before the year of cancer diagnosis) was statistically significantly associated with a reduced risk of breast cancer (P<.001) (Table 5). There was less marked evidence of decreasing risk of breast cancer with greater parity (data not shown). Earlier age at first birth was also associated with a reduced risk of breast cancer, although the trend was not statistically significant within the parous group (P.33). (When the nulliparous women were included in the analysis, the trend was statistically significant [P.004]). Separate analyses among monozygotic and dizygotic twins showed that being parous was statistically significantly associated with a decreased risk of breast cancer within each group (Table 6), but age at first birth was not associated with risk within either group (among parous women, P trend among monozygotic twins.99, and P trend among dizygotic twins.16; data not shown). In separate analyses among familial and non- Table 3. Relative risks of breast cancer in relation to selected factors for monozygotic and dizygotic case patients and for familial and nonfamilial case patients* Measures or milestones case patients control subjects same response for case patients and control subjects data are not known or twin pairs disagreed OR (95% CI) for factor-positive Zygosity Taller at age 10 y Monozygotic twins (0.56 to 1.90) Dizygotic twins (0.94 to 1.90) Less heavy at age 10 y Monozygotic twins (0.58 to 1.88) Dizygotic twins (1.10 to 2.19) Breast development earlier Monozygotic twins (0.60 to 1.93) Dizygotic twins (1.17 to 2.40) Familial breast cancer Taller at age 10 y Familial cases (0.41 to 1.81) Nonfamilial cases (1.00 to 1.89) Less heavy at age 10 y Familial cases (0.44 to 1.75) Nonfamilial cases (1.16 to 2.18) Breast development earlier Familial cases (0.73 to 2.67) Nonfamilial cases (1.14 to 2.22) *Familial case patients were women who were diagnosed with breast cancer and had a first-degree female relative with breast cancer, and nonfamilial case patients were women who were diagnosed with breast cancer and did not have a first-degree female relative with breast cancer. OR odds ratio; CI confidence interval. Twenty-two pairs of twins were omitted from all analyses in this table because data on the adjustment variables were missing. An additional 22 pairs of twins (i.e., 44 pairs in all) were omitted from the zygosity analyses because data on zygosity were missing. All ORs were adjusted for parity. ORs for height were also adjusted for weight at same age, and vice versa (because of this adjustment, the analyses for weight are more concerned with obesity than with weight per se and are discussed in relation to obesity in this paper). ORs for breast development were adjusted for height and weight at age 10 y. The referent group is factor-negative in case patients (i.e., control subject positive). Age 10 y in Nordic countries; age 11 y in England and Wales. P.01, Wald test. P.005, Wald test. P.004, Wald test. P.006, Wald test ARTICLES Journal of the National Cancer Institute, Vol. 94, No. 16, August 21, 2002

6 Table 4. Relative risks of breast cancer in relation to sporting activity* case patients control subjects same response for case patients and control subjects data are not known or twin pairs disagreed OR (95% CI) for factor-positive More time in sporting activities at age 10 y (0.70 to 1.33) at age 15 y (0.66 to 1.27) 5 y before diagnosis (0.56 to 1.09) *Twenty-two pairs of twins were omitted from all analyses in this table because data on the adjustment variables were missing. An additional 49 pairs of twins (i.e., 71 pairs in all) were omitted from the analyses of sport 5 y before diagnosis because breast cancer occurred in the case patient in England and Wales at an age under 35 y (see footnote ). OR odds ratio; CI confidence interval. In no instance did more than 14 pairs disagree. Adjusted for parity, and for height and weight at age 10 y, except for sport 5 y before diagnosis, which is adjusted for parity at diagnosis and for height and weight at age 20 y. The referent group is factor-negative in case patients (i.e., control subject positive). Age 10 y in Nordic countries; up to age 11 y in England and Wales. Age 15 y in Nordic countries; age y in England and Wales. Five years before diagnosis in Nordic countries; ages y in England and Wales. Analyses for England and Wales were restricted to the sets in which breast cancer occurred in the case after age 34 y. Table 5. Relative risks of breast cancer in relation to reproductive factors familial breast cancer cases (Table 6), being parous was associated with a statistically nonsignificant decreased risk of breast cancer for women with a family history of breast cancer (OR 0.44, 95% CI 0.14 to 1.44) and a statistically significant decreased risk for women without such a history (OR 0.34, 95% CI 0.19 to 0.61). The effect of age at first birth was apparent in the nonfamilial cases, but risk estimates were unstable (based on small numbers) in the familial cases. There was no statistically significant heterogeneity between the results for monozygotic and dizygotic twins or between the results for women with familial breast cancer and those with nonfamilial breast cancer shown in Table 6. DISCUSSION No. of case patients No. of control subjects OR (95% CI)* Parity Nulliparous (referent) Parous (0.21 to 0.61) Age at first birth, y < (referent) (0.66 to 2.23) (0.79 to 3.07) (0.66 to 3.72) Nulliparous (1.39 to 7.72) Test for linear trend, all categories: P.004; test for linear trend, among parous: P.33 *OR odds ratio; CI confidence interval. Parity not known for 22 pairs (six case patients and 16 control subjects). There was disagreement on parity in 4% of twin pairs. P<.001, Wald test. Age at first birth not known for 136 pairs (105 case patients and 120 control subjects), leading to exclusion of these pairs. P.007, Wald test. The previously reported associations between risk of breast cancer and age at menarche and adult height (2), both factors that appear to be dependent on growth during the prepubertal years, have suggested that part of the etiologic process for breast cancer in women may occur before puberty. This possibility is Table 6. Relative risks of breast cancer in relation to parity for monozygotic and dizygotic case patients and for familial and nonfamilial case patients* No. of case patients No. of control subjects OR (95% CI) Monozygotic case patients Nulliparous (referent) Parous (0.13 to 0.84) Dizygotic case patients Nulliparous (referent) Parous (0.23 to 0.86) Familial case patients Nulliparous (referent) Parous (0.14 to 1.44) Nonfamilial case patients Nulliparous (referent) Parous (0.19 to 0.61) *Familial case patients were women who were diagnosed with breast cancer and had a first-degree female relative with breast cancer, and nonfamilial case patients were women who were diagnosed with breast cancer and did not have a first-degree female relative with breast cancer. Twenty-two pairs of twins were omitted from all analyses in this table because data on parity were missing, and another 22 pairs were omitted from the zygosity analyses (i.e., 44 pairs in all) because data on zygosity were missing. OR odds ratio; CI confidence interval. P.02, Wald test. P.02, Wald test. P<.001, Wald test. strengthened by findings that migration from a country in which risk of breast cancer is low to one in which risk is high has a greater effect in increasing breast cancer risk if the migration is at a young age (13). It is also supported by findings in Norway that correlations between adult height and risk of breast cancer were restricted to women whose peripubertal years occurred during the nutritional deprivations of World War II (14) and that breast cancer incidence on a cohort basis was lower than expected in women who experienced puberty during the War (15). It has so far proven extremely difficult, however, to test more directly the effects of childhood anthropometric variables on breast cancer risk. In several studies of premenopausal breast cancer, adult subjects have been asked about their height and weight (9,16 23) and diet (17,18) in childhood or in adolescence. However, adults have difficulty recalling these variables, Journal of the National Cancer Institute, Vol. 94, No. 16, August 21, 2002 ARTICLES 1243

7 and the potential for misclassification is considerable, especially if a subject is asked to recall absolute values of the variables at a specific age. To our knowledge, only one study (11) has published results of analyses that used anthropometric data recorded during childhood or adolescence. The use of twins, as in our study, provides unique advantages, as well as disadvantages, for addressing the relationship between breast cancer risk and childhood anthropometric variables. On the one hand, a twins study has strengths over a study that compares the subject with her peers in general (9,17,20,22), which is likely to be highly dependent on the subject s retrospective self-image, and over a study in which the subject is asked to recall absolute values for height and weight (16,18,19), which is likely to be highly inaccurate; instead, the use of comparative measures between the twins allows a comparison that is at least reasonably objective for the time to which the comparison applies and is internally validated (by questions to the cotwin). On the other hand, it is extraordinarily difficult to identify and interview a sufficient number of eligible subjects for a twins study; as a consequence, each of the four studies pooled in our article was itself of modest size and had unstable results on its own. For this reason, except for some data on parity for the twins from Denmark (24) and on mammographic patterns for the twins from Finland (5), results of the interviews from the four individual studies have not previously been published. A study that compares twins also has the disadvantage that it may overmatch for certain variables. Consequently, the apparent magnitude of an effect may be reduced, and the absence (or small size) of an effect for a factor in our study does not necessarily indicate that the factor has little or no relation to risk. Although the datasets from the four countries in our study differed in the years of cancer incidence, years of birth, and proportions of eligible subjects who were interviewed, we do not believe that these differences would have biased the results, because all analyses were twin pair-matched and, hence, matched on country and year of birth. Although there was no statistically significant heterogeneity between the country-specific effects, the power of the heterogeneity tests was low because of small numbers. We found a statistically significantly decreased risk of breast cancer in relation to relative obesity at age 10 years, with an indication that the effect of obesity was greater at this age than at other ages. Our results on obesity are similar to those from the only study (11) that analyzed information on childhood height and weight that was recorded in childhood: in that study, in Hawaii, body mass index at ages years, but not at younger or older ages under 25 years, was statistically significantly inversely related to risk of breast cancer occurring in women younger than 50 years. Breast cancer risk in the Hawaiian study was lowest for subjects who were also overweight in young adulthood, but this was not so in our data. In some (9,16,17,19,23) but not all (18,20,22) non-twin recall-based studies, reported adolescent weight or Quetelet s index (i.e., body mass index) has been found to be inversely associated with risk of premenopausal breast cancer. Because childhood obesity tends to be associated with earlier age at menarche, finding an inverse association between childhood obesity and breast cancer risk is not what would be expected from the well-established association between breast cancer risk and age at menarche; this finding implies that the effect of childhood obesity is not via any contribution to acceleration of puberty. The results of our analyses that adjusted for obesity later in life, suggest that the relationship between childhood obesity and risk was independent of, rather than a correlate of, the fairly well established inverse association (25) of adult obesity to premenopausal breast cancer risk. Our data, and those from the Hawaiian study based on childhood height as reported by parents at the time (11), indicated that taller height at age 10 years was associated with breast cancer risk and that the effect may be greater in relation to height at this age than at other ages. In two non-twin, recall-based case control studies that analyzed childhood height and breast cancer risk, an association between height and risk of subsequent breast cancer was found in one study (9) but not clearly in the other (22). Height might be a risk factor because it might be associated with mammary gland mass or because it is a marker of childhood nutrition. Our analysis of risk in relation to height at ages 10 and 20 years did not support the idea that an early growth spurt might be a risk factor for breast cancer (21). Height (but not body mass index or occurrence of menarche) in prepubertal girls is positively associated with serum insulin-like growth factor (IGF)-I concentration (26). A high serum level of IGF-I a few years before breast cancer diagnosis has been found to be associated with increased risk of premenopausal breast cancer (27), but no data exist on risk of breast cancer in relation to childhood IGF-I levels. Another possible mechanism by which childhood growth might influence risk of breast cancer is by affecting mammary gland mass (28). Breast size, which is likely to be related to, although not an exact measure of, mammary gland mass, has been found to be associated with premenopausal breast cancer risk in some (29 31) but not all (20,30 32) studies. In our study, the results were in the direction of greater risk with greater breast size, but the risk increase was not statistically significant. Young age at menarche has long been recognized as a risk factor for breast cancer, and the age at establishment of regular menstrual cycles has been found to be a risk factor independent of age at menarche, perhaps because regular menstruation might be associated with increased cumulative estrogen exposure (33). In our data, earlier age at the first menstrual period was slightly associated with risk, and age at first regular period more strongly associated with risk, but not statistically significantly in either case. There was, however, a statistically significant association that was independent of age at menarche between breast cancer risk and another pubertal variable age at breast development that has been little investigated. To our knowledge, the only published study (9) to analyze the relationship between breast cancer risk and age at breast development was based on recall comparisons with contemporaries, and it showed results, after adjustment for age at menarche, that were in the same direction as ours. Most stages of breast development occur before menarche. Although reporting bias might have occurred for this variable, it is also possible that age at breast development may be an independent risk factor, because the growth and differentiation of the breast tissues that occur at pubertal development (34) might permanently alter the risk status of the breast; thus, at a given age, the number of years accumulated in this susceptible state would be greater for women whose breasts had developed earlier than for women whose breasts developed later. We found that risk of breast cancer was increased for women with a low waist-to-hip ratio, compared with their co-twin, at age 20 years. Our data on this variable were indirect, however, and the comparison analyzed was between extreme categories of this 1244 ARTICLES Journal of the National Cancer Institute, Vol. 94, No. 16, August 21, 2002

8 variable. The only previously reported data on the relationship between waist-to-hip ratio and the risk of premenopausal breast cancer came from studies that reported waist-to-hip ratios measured at the time of the study interviews; those results were in the same direction as ours in one study (20) but not in two other studies (29,35). A low waist-to-hip ratio could be relevant to breast cancer risk because in premenopausal women such a ratio is associated with increased estrogenic, compared with androgenic, activity (36). Studies have reported that physical exercise after menarche is associated with a reduced risk of premenopausal breast cancer, perhaps by reducing a woman s cumulative exposure to cyclic estrogens and progesterone (12,37). We found ORs in relation to sporting activity that were in this direction but were not statistically significant. The OR in our data for breast cancer in relation to adult exercise was more greatly reduced in women with familial breast cancer than in those with nonfamilial breast cancer, a comparison that, to our knowledge, has not previously been examined in premenopausal women; the OR in our data for breast cancer in relation to adult exercise was also lower among parous women than among nulliparous women, adding to previous evidence of the same effect (12). Very high levels of exercise can delay menarche, and hence potentially affect breast cancer risk, but our data did not suggest that exercise at premenarcheal ages was important to risk in a study group that was not selected for its high exercise levels. The increased risk of breast cancer in nulliparous women is well established, but it is less clear whether the effects of nulliparity apply to women with a family history of breast cancer as well as to women without such a history. In our data, parity was associated with a decreased risk of breast cancer of similar magnitude in both of these subgroups as well as within both monozygotic and dizygotic twin pairs. The use of twins in our study design has the strength that, unlike conventional recall-based case control studies, differential completeness of reporting of family history between cases and controls (38) will not have potential to bias the results, because the same family history is used for each member of the twin pair. Reports by breast cancer patients of breast cancer in their first-degree relatives have been found to be 99% accurate (39). The only other study of breast cancer risk in relation to family history and parity in women who were premenopausal or at premenopausal ages found, as we did, that family history did not modify the effect of parity (40). In studies that analyzed a mixture of pre- and postmenopausal breast cancers, results have varied from no modification of the effect of parity by family history (41) to an absent effect of nulliparity or even a reduced risk with nulliparity in women with a family history of breast cancer (42 44). In young women who carry mutations in the BRCA1 or BRCA2 genes, nulliparity has been found to be associated with a decreased risk of breast cancer (45); however, the great majority of familial cases of breast cancer that occur at premenopausal ages do not have these mutations (46). In our study, the effects of childhood height, weight, and breast development on breast cancer risk appeared to apply within dizygotic but not within monozygotic twin pairs; there are no previous data on this relationship. Monozygotic twins, because they are genetically identical, would be expected to show no effect for factors whose variability is of genetic origin. Monozygotic twins might also show less effect than dizygotic twins, however, for prenatal factors [because monozygotic twins usually have a shared chorion, with communication between the fetal circulations (47), whereas dizygotic twins have a dichorionic placenta] and for childhood factors, because monozygotic twins may behave and be treated more similarly in childhood than dizygotic twins. The effects of anthropometric variables at age 10 years on risk were apparent in nonfamilial but not in familial cases of breast cancer. The only other comparable analysis we could find was from a small (50 cases) recall-based case control study of premenopausal bilateral breast cancer, which also indicated that weight and obesity at menarche were associated with risk only in nonfamilial cases (19). The relationship between age at breast development and risk observed in our data was seen in both familial and nonfamilial cases; this familial/nonfamilial subdivision does not appear to have been examined previously. Overall, our results provide a unique type of evidence that suggests that childhood growth before puberty may be an important contributor to risk of premenopausal breast cancer. REFERENCES (1) Trichopoulos D. Hypothesis: does breast cancer originate in utero? Lancet 1990;335: (2) de Waard F, Trichopoulos D. A unifying concept of the aetiology of breast cancer. Int J Cancer 1988;41: (3) Holm NV, Hauge M, Jensen OM. Studies of cancer aetiology in a complete twin population: breast cancer, colorectal cancer and leukaemia. Cancer Surv 1982;1: (4) Kaprio J, Teppo L, Koskenvuo M, Pukkala E. Cancer in adult same-sexed twins: a historical cohort study. Prog Clin Biol Res 1981;69(Pt C): (5) Kaprio J, Alanko A, Kivisaari L, Standertskjold-Nordenstam CG. Mammographic patterns in twin pairs discordant for breast cancer. Br J Radiol 1987;60: (6) Cederlof R, Friberg L, Lundman T. The interactions of smoking, environment and heredity and their implications for disease etiology. A report of epidemiological studies on the Swedish twin registries. Acta Med Scand Suppl 1977;612: (7) Medlund P, Cederlof R, Floderus-Myrhed B, Friberg L, Sorensen S. A new Swedish Twin Registry: containing environmental and medical base line data from about 14,000 same-sexed pairs born Acta Med Scand Suppl 1976;600: (8) Swerdlow AJ, De Stavola BL, Swanwick MA, Maconochie NE. Risks of breast and testicular cancers in young adult twins in England and Wales: evidence on prenatal and genetic aetiology. Lancet 1997;350: (9) Brinton LA, Swanson CA. Height and weight at various ages and risk of breast cancer. Ann Epidemiol 1992;2: (10) Clayton D, Hills M. Statistical models in epidemiology. Oxford (U.K.): Oxford University Press; p (11) Le Marchand L, Kolonel LN, Earle ME, Mi MP. Body size at different periods of life and breast cancer risk. Am J Epidemiol 1988;128: (12) Bernstein L, Henderson BE, Hanisch R, Sullivan-Halley J, Ross RK. Physical exercise and reduced risk of breast cancer in young women. J Natl Cancer Inst 1994;86: (13) Shimizu H, Ross RK, Bernstein L, Yatani R, Henderson BE, Mack TM. Cancers of the prostate and breast among Japanese and white immigrants in Los Angeles County. Br J Cancer 1991;63: (14) Vatten LJ, Kvinnsland S. Body height and risk of breast cancer. A prospective study of 23,831 Norwegian women. Br J Cancer 1990;61: (15) Tretli S, Gaard M. Lifestyle changes during adolescence and risk of breast cancer: an ecologic study of the effect of World War II in Norway. Cancer Causes Control 1996;7: (16) Choi NW, Howe GR, Miller AB, Matthews V, Morgan RW, Munan L, et al. An epidemiologic study of breast cancer. Am J Epidemiol 1978;107: (17) Hislop TG, Coldman AJ, Elwood JM, Brauer G, Kan L. Childhood and recent eating patterns and risk of breast cancer. Cancer Detect Prev 1986; 9: (18) Pryor M, Slattery ML, Robison LM, Egger M. Adolescent diet and breast cancer in Utah. Cancer Res 1989;49: Journal of the National Cancer Institute, Vol. 94, No. 16, August 21, 2002 ARTICLES 1245