AMERICAN JOURNAL OP EPIDEMIOLOGY
Vol. 131, No. 6
Copyright © 1990 by The Johns Hopluni University School of Hygiene and Public Health All rights reserved
Printed m U S A.
Original Contributions AGE AT ONSET AS AN INDICATOR OF FAMILIAL RISK OF BREAST CANCER ELIZABETH B. CLAUS,1 NEIL J. RISCH,1J AND W. DOUGLAS THOMPSON1 Claus, E. B. (Dept of Epidemiology and Public Health, Yale U., New Haven, CT 06510), N. J. Risen, and W. D. Thompson. Age at onset as an indicator of familial risk of breast cancer. Am J Epidemiol 1990; 131:961-72. The familial risk of breast cancer was investigated in a large population-based, case-control study conducted by the Centers for Disease Control. The data set included 4,730 histologically confirmed breast cancer cases aged 20-54 years and 4,688 controls who were frequency matched to cases by geographic region and 5-year categories of age. Family history of breast cancer among first-degree female relatives of cases and controls was utilized. To identify factors associated with familial risk of breast cancer, a Cox proportional hazards model was used, modeling time to onset of breast cancer among mothers and sisters. Case relatives were at greater risk than control relatives. Among relatives of cases, a significant increase in the risk of breast cancer was associated with decreasing age at onset of the case and with having an additional relative affected with breast cancer. The hazard ratio for the mother of a case with breast cancer diagnosed at 50 years of age was 1.7 (95% confidence interval (Cl) 1.4-2.0), compared with 2.7 (95% Cl 2.2-3.2) and 4.3 (95% Cl 3.3-5.6) for the mother of a case whose diagnosis occurred at 40 and 30 years of age, respectively. The hazard ratio for the sister of a case with an unaffected mother and at least one affected sister in addition to the case was 3.6 (95% Cl 2.1-6.1) when the case was diagnosed at age 50, compared with 5.8 (95% Cl 3.4-10.0) and 9.4 (95% Cl 5.3-16.7) when the case was diagnosed at 40 and 30 years of age, respectively. The hazard ratio for the sister of a case with an affected mother and no additional affected sisters was 5.9 (95% Cl 3.9-8.9) when the case was diagnosed at age 50, compared with 9.4 (95% Cl 6.2-14.4) and 15.1 (95% Cl 9.4-24.3) when the case was diagnosed at 40 and 30 years of age, respectively. The hazard ratio for the sister of a case with both an affected mother and at least one affected sister aside from the case was 17.1 (95% Cl 9.4-31.3) when the case was diagnosed at age 50, compared with 27.5 (95% Cl 15.0-50.3) and 44.2 (95% Cl 23.5-83.2) when the case was diagnosed at 40 and 30 years of age, respectively. No effect of case's menopausal status and biiaterality was found, indicating that in addition to a positive family history, age at onset is the strongest indicator of a possible genetic subtype of breast cancer in these data. breast neoplasms; genetics; risk
Editor's note: For a discussion of this It is well documented that a positive fampaper and that by Mettlin et aL immediately ily history is associated with an increased following, see page 984. risk of breast cancer (1-9). Within this Received for publication June 13,1989, and in final form November 3, 1989.
Abbreviations: Cl, confidence interval; SEER, Surveillance, Epidemiology, and End Results. 961
962
CLAUS ET AL.
are more likely to be genetic from those which are not. These factors include parity, age at menarche, menopausal status, age at menopause, history of benign breast disease, laterality, age at first pregnancy, number of livebirths, and number of stillbirths, as well as the number and type of relatives affected with breast cancer. Few other studies have looked systematically at how factors other than age and laterality may help to differentiate between genetic and nongenetic subgroups. The analytic approach of this 3tudy is to model the age-specific incidence of breast cancer in relatives of the cases and controls using risk factor information measured on the cases and controls themselves. The critical issue being addressed in this study is the possibility of discerning potential genetic het1 Department of Epidemiology and Public Health, erogeneity. If women with different risk Yale University School of Medicine, New Haven, CT. * Department of Human Genetics, Yale University factor patterns also show differences in the amount of risk associated with their relaSchool of Medicine, New Haven, CT. Reprintrequeststo Dr. E. B. Claus, Department of tives, then it may be possible to use these Epidemiology and Public Health, Yale University School of Medicine, 60 College Street, New Haven, clinically defined variables to identify women and their families who are espeCT 06510. The authors wish to acknowledge the contributors cially genetically predisposed. group of women with a positive family history, bilaterality of a relative's breast cancer as well as the age at which the relative is affected have been shown to modify the familial risk. Bilateral breast cancer and early onset have been associated with the highest risk to relatives (1, 2, 5, 6, 9). These differences in risk indicate that age at onset and bilaterality may be useful variables for identifying subgroups of breast cancer patients in whom genetic factors play a relatively important etiologic role. In addition to bilaterality and age at onset of breast cancer, this study investigates the possibility that other risk factors associated with breast cancer might be useful in distinguishing breast cancer cases which
to the Cancer and Steroid Hormone Study Study Design and Coordination at the Division of Reproductive Health, Center for Chronic Disease Prevention and Health Promotion, Centers for Disease Control: Principal Investigator, George L. Rubin; Project Director, Phyllis A. Wingo; Project Associates, Nancy C. Lee, Michele G. Mandel, Herbert B. Peterson. Data Collection Centers Principal Investigators: AtlantaRaymond Greenberg; Connecticut—J. Wistar Meigs, W. Douglas Thompson; Detroit—G. Marie Swanson; Iowa—Elaine Smith; New Mexico—Charles Key, Dorothy Pathak; San Francisco—Donald Austin; Seattle—David Thomas; Utah—Joseph Lyon, Dee West. Pathology Review Principal Investigators: Fred Gorstein, Robert McDivitt, Stanley J Robboy. Project Consultants: Lonnie Burnett, Robert Hoover, Peter M. Layde, Howard W. Ory, James J. Schlesselman, David Schottenfeld, Bruce Stadel, Linda Webster, Colin White. Pathology Consultants: Walter Bauer, William Christopherson, Deborah Gersell, Robert Kunnan, Allen Paris, Frank Vellios. The Cancer and Steroid Hormone Study was supported by interagency agreement no. 3-Y01-HD-81037 between the Centers for Disease Control and the National Institute of Child Health and Human Development, with additional support from the National Cancer Institute. E. B. Clau* was supported by USPHS training grant no. 2-T32-ES07085. N. Risch and E. B. Claus were supported by NIH grant no. GM39812 from NIGMS. W. D. Thompson «nd data collection in Connecticut were supported by contract no. 200-800561 from the Centers for Disease Control.
Previous studies have investigated the distribution of various risk factors by casecontrol status (3, 7). In this study, these risk factors, measured in cases and controls, are used to define subgroups of the cases. These subgroups are then compared for level of risk among the relatives. In addition, each relative is also classified according to type of relationship (mother vs. sister), and if a sister, whether the mother is also affected, and whether at least one additional sister is affected. Our ability to detect subgroups of cases whose relatives display distinct frequencies of breast cancer depends on the relation between genetic predisposition and the measured risk factors in producing breast cancer risk. For example, if a risk factor multiplies the risk in genetically predisposed and nonpredisposed individuals by the same amount, then separating cases by risk factor profile (e.g., nulliparous vs. parous) will not yield distinct groups in terms of genetic predisposition. On the other hand, if the
AGE AT ONSET AND FAMILIAL RISK OF BREAST CANCER
risk factor modifies the risk in genetically predisposed individuals to a different degree than in non-predisposed individuals, then cases with different risk factor profiles will be genetically distinct. For example, suppose that nulliparity increases risk only among genetically non-predisposed individuals. Then, among breast cancer cases, those without the risk factor (e.g., paxous women) will have a higher genetic predisposition than those with the risk factor (e.g., nulliparous women), and hence the former group will display a higher level of risk among relatives. Similarly, characteristics of a case's disease, such as bilaterality, may be useful in separating cases into genetically distinct subgroups. Therefore, if associations between risk factors measured in cases and breast cancer risk in relatives are found, it may be possible to use such information to divide breast cancer patients into subgroups that are genetically more homogeneous and to then look at patterns of transmission within these subgroups. The identification of subgroups may also help to identify families that are more likely to be informative for genetic linkage studies. MATERIALS AND METHODS
Study population Data are taken from the Cancer and Steroid Hormone Study, a multicenter, population-based, case-control study conducted by the Centers for Disease Control. The data consist of 4,730 histologically confirmed breast cancer cases 20 to 54 years of age and 4,688 controls. The cases were registered between December 1, 1980 and December 31, 1982 at one of eight Surveillance, Epidemiology, and End Results (SEER) Centers of the National Cancer Institute. The eight centers include the cities and metropolitan areas of Atlanta, Detroit, San Francisco, and Seattle, the four urban counties of Utah, and the states of Connecticut, Iowa, and New Mexico. All cases were interviewed within 6 months of the diagnosis of their first primary breast
963
cancer; age at onset and current age are therefore essentially equivalent for cases. There were 72 bilateral breast cancer cases, all of whom were diagnosed with bilateral breast cancer at time of interview. Controls were frequency matched to cases according to geographic region and 5-year categories of age. Cases and controls with a previous history of breast cancer or a breast biopsy of unknown outcome were excluded from the study. Information on family history was collected on all first-degree female relatives as well as maternal and paternal grandmothers for cases and controls. Data describing the number of affected maternal and paternal aunts were also collected. An extensive amount of additional information was collected for the cases and controls including complete pregnancy and menstrual history, history of benign breast disease, age at onset of disease, alcohol and cigarette usage, breast surgery, sociodemographic variables, and use of oral contraceptives. A detailed description of the study may be found elsewhere (10). Analysis Age-specific Kaplan-Meier estimates of cumulative risk of breast cancer to firstand second-degree female relatives of the cases and controls, stratified by race, were computed using BMDPlL (11). Cumulative risks of breast cancer for relatives of cases were also compared according to the age at onset of breast cancer in the case and according to whether her disease was unilateral or bilateral. Cumulative risks of breast cancer for relatives of controls were compared according to the current age of the .control. Hazard rates were computed for first- and second-degree relatives of the cases and controls using a Cox proportional hazards regression model and BMDP2L. Daughters of cases and controls were excluded from all analyses as only two daughters of cases as well as two daughters of controls were reported to have had breast cancer. Relatives with unknown current age or age at death were eliminated from
964
CLAUS ET AL.
all analyses. For affected relatives with known current age but unknown age at onset, their age at onset was estimated by using the average age at onset for affected individuals whose current age matched that of the relative. To predict the age-specific risk of breast cancer in mothers and sisters of cases and controls, information on breast cancer risk factors, measured on the cases and controls, was used in a Cox proportional hazards regression model. These risk factors included: bilaterality (for cases only), history of benign breast disease, parity, number of livebirths and stillbirths, age at first pregnancy, age at onset of breast cancer, age at menarche, menopausal status, and age at last menstrual period. The dependent variable was defined as years to onset of breast cancer, measured on the mothers and sisters of the cases and controls. To allow for heterogeneity of risk factors across casecontrol status, separate terms for cases and controls for all of the risk factors were included. To more specifically examine familial risk relations, eight additional variables were created. These eight variables (four for the relatives of cases and four for the relatives of controls, respectively) identify each individual by relationship to the case or control either as a mother, as a sister with an affected mother and at least one affected sister aside from the case, as a sister with an affected mother and no affected sisters aside from the case, as a sister with an unaffected mother but at least one affected sister aside from the case, or as a sister with an unaffected mother and no affected sisters aside from the case. Four second-order terms were also included to allow the effect of age at onset of the case to be heterogeneous across different types of relatives; for example, if the effect of the case's age at onset is more pronounced for the sister of a case with an affected mother compared with a sister of a case with an unaffected mother. In addition to the above analyses, breast cancer cases were divided by type of breast cancer family history (i.e., number and type
of relatives affected with breast cancer). The distributions of the remaining risk factors (i.e., percentage of cases with a positive history of benign breast disease, average age at first menstrual period, etc.) were then observed across the varying family history patterns. RESULTS
Kaplan-Meier estimates of cumulative risk of breast cancer to mothers and sisters of both cases and controls, stratified by race, are presented in tables 1 and 2. By age 89, the risk of developing breast cancer was 15.17 percent for mothers of white cases versus 8.97 percent for mothers of white controls. By age 69 (the last age at which the mother of a black case was reported as having breast cancer), the risk of breast cancer for the mother of a black case was 6.26 percent versus 2.91 percent for mothers of black controls. The pattern is the same for sisters of cases and controls. Cumulative risk of breast cancer by age 64 for the sister of a white case was 7.60 percent versus 4.60 percent for the sister of a white control. Among sisters of black women, cumulative risk of breast cancer by age 54 (the last age at which the sister of a black case was reported as having breast cancer) was 3.76 percent for sisters of black cases versus 2.33 percent for sisters of black controls. At all age periods, mothers and sisters of cases, regardless of race, showed a risk of at least one and one-half times that of mothers and sisters of controls, and in general, the risk of developing breast cancer for first-degree relatives of cases was at least two times that of first-degree relatives of controls. The hazard rate of breast qancer for the mother of a white case was 1.98 times the rate for the mother of a white control, while for sisters of white cases, the rate was 2.28 times the rate for sisters of white controls. The hazard rate of breast cancer for the mother of a black case was 2.02 times the rate for the mother of a black control, while for sisters of black cases, the rate was 1.65 times the rate for sisters of black controls.
965
AGE AT ONSET AND FAMILIAL RISK OF BREAST CANCER TABLE 1
Cumulative probability (percent) ± standard error of breast cancer m mothers of cases and controls, by race, Cancer and Steroid Hormone Study. 1980-1982 Mothers of Age (in yean) of mother
0-24 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64 65-69 70-74 75-79 80-84 85-89 90-94 No. at risk No. with breast cancer
Black
White Controls
Case* 0.07 0.20 0.50 1.18 2.33 3.58 4.87 6.14 7.45 9.19 11.16 13.41 14.43 15.17
0.00 0.02 0.23 0.45 0.76 1.57 2.04 2.93 3.69 4.58 5.91 7.25 8.97 8.97 11.43
± 0.04 ± 0.07 ± 0.11 ±0.17 ± 0.24 ± 0.30 ± 0 35 ± 0.39 ± 0.44 ± 0.50 ± 0.59 ± 0.77 ± 0.92 ± 1.18*
4,019 379
± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
0.00 0.02 0.06 0.11 0.14 0.20 0.23 0.28 0.32 0.37 0.47 0.62 0.94 0.94 2.59
Case*
Controls 0.00 0.00 0.22 0.65 0.88 1.11 1.88 1.88 2.31 2.91 3.76
0.00 ± 0.00 0.00 ± 0.00 0.41 ± 0.36 0.62 ± 0.36 1.05 ± 0.47 1.95 ± 0.64 2.92 ± 0.80 4.06 ± 0.97 5.82 ± 1.24 6.26 ± 1.30
4,026 189
± 0.00 ± 0.00 ± 0.22 ± 0.38 ± 0.44 ± 0.49 ± 0.66 ± 0.66 ± 0.79 ± 0.98 ± 1.30
481 11
489 23
* Cumulative risks are presented up until the last time interval in which a breast cancer occurred. TABLE 2
Cumulative probability (percent) ± standard error of breast cancer in sisters of cases and controls, by race. Cancer and Steroid Hormone Study, 1980-1982 Sisters of' Age (in years) of sister 0-24 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64 65-69 No. at risk No. with breast cancer
Black
White Cases 0.01 0.10 0.50 1.17 2.42 3.73 4.96 6.47 7.60 12.15
± ± ± ± ± ± ± ± ± ±
0.02 0.04 0.10 0.16 0.25 0.34 0.43 0.60 0.98 2.05*
5,612 173
Controls 0.06 0.08 0.18 0.45 0.84 1.79 2.15 2.85 4.60
± ± ± ± ± ± ± ± ±
0.03 0.04 0.06 0.10 0.15 0.25 0.29 0.41 0.85
5,772 74
Case* 0.00 0.30 0.53 0.94 1.26 2.53 3.76
± 0.00 ± 0.17 ± 0.24 ± 0 33 ± 0 40 ± 0.70 ± 0.98
1,138 18
Controls 0.10 0.22 0.22 0.36 0.36 1.07 2.33 2.33 7.27
± ± ± ± ± ± ± ± ±
0.10 0.15 0.15 0.21 0.21 0.46 0.86 0.86 3.53
1,037 11
* Cumulative risks are presented up until the last time interval in which a breast cancer occurred.
When the Kaplan-Meier curves for mothers of black and white cases were compared using the generalized Wilcoxon (Breslow) log rank test, the null hypothesis of no difference in time to onset of breast cancer was rejected (x2 = 4.10, df = 1, p =
0.04). However, when the same curves for the sisters of black and white cases were compared, the difference in time to onset of breast cancer due to race was not significant (x2 = 0.79, df = 1, p = 0.37). By age 89, the risk of developing breast cancer for
966
CLAUS ET AL.
mothers of white controls was 8.97 percent, essentially the reported lifetime risk for white females in the United States. However, the risk by age 74 for mothers of black controls was 3.76 percent, a rate lower than the estimated population risk of approximately 7 percent to age 74 among black women (12-14). As a result of this discrepancy and the relatively small number of nonwhite cases and controls (approximately 10 percent of the data set), the remaining analyses included only the white cases and controls. The cumulative risk of breast cancer by age 89 for a maternal grandmother of a white case was 7.34 percent versus 5.42 percent for the maternal grandmother of a white control. The cumulative risk of breast cancer among paternal grandmothers was generally lower than those reported for maternal grandmothers regardless of the age investigated. For example, the cumulative risk of breast cancer by age 89 for the paternal grandmother of a white case was 5.39 percent versus 3.70 percent for the paternal grandmother of a white controlThe hazard rate of breast cancer for the maternal grandmother of a white case was 1.23 times the rate for the maternal grandmother of a white control; for paternal grandmothers of white cases, the rate was 1.46 times the rate for paternal grandmothers of white controls. The difference in risk between maternal and paternal grandmothers is likely to be due to a higher rate of underreporting among paternal relatives than among maternal relatives. It is notable, however, that the hazard ratios comparing relatives of cases to relatives of controls were similar for the two types of grandmothers. Due to overall underreporting, second-degree relatives were also left out of subsequent analyses. Table 3 lists the cumulative probability of breast cancer in sisters of white cases given the mother's breast cancer status. The cumulative risk by age 39 for a sister with an affected mother was 4.95 percent versus 0.84 percent for a sister without an affected mother. By age 59, the risk was
TABLE 3
Cumulatwe probability (percent) ± standard error of breast cancer in sisters of white cases given mother's breast cancer status, Cancer and Steroid Hormone Study, 1930-1982 . Sisters with: Age (in years) of sister
Mother affected with breast
0-24 25-29 30-34 36-39 40-44 45-49 50-54 55-59 60-64 65-69
0.00 ± 0.00 0.71 ± 0 41 2.27 ± 0.75 4.95 ± 1.14 7.94 ± 1.52 12.20 ± 2 07 15.92 ± 2.57 15.92 ± 2.57 25.26 ± 9.10*
0.04 ± 0.03 0.04 ± 0.03 0.34 ± 0.09 0.84 ±0.14 1.94 ± 0.24 3.01 ± 0.32 4.03 ± 0.41 5.67 ± 0.61 7.94 ± 1.03 11.17 ±2.12
No. at risk No. with breast cancer
472
5,140
40
133
cancer
Mother not affected with breast cancer
* Cumulative risks are presented up until the last time interval in which a breast cancer occurred.
15.92 percent for a sister with an affected mother versus 5.67 percent for a sister without an affected mother. The hazard rate of breast cancer for a sister of a case with an affected mother was 8.48 times the rate for a sister of a case without an affected mother. The risk of breast cancer among relatives of white cases was investigated according to whether the case was diagnosed with unilateral or bilateral breast cancer. The hazard rate of breast cancer among sisters of white bilateral cases was 0.95 times that of sisters of white unilateral cases. The hazard rate of breast cancer among mothers of white bilateral cases was 0.66 times that pf mothers of white unilateral cases. Cumulative risk of breast cancer for a mother or a sister was dependent on the age at which the case was affected with breast cancer (figure 1). The lifetime risk, as well as the risk at any given age period, for a first-degree relative appeared to decrease as the age at onset of the case increased. For example, the risk of breast cancer by age 59 for the mother or sister of
967
AGE AT ONSET AND FAiMJLIAL RISK OF BREAST CANCER
0.30 -I Age at onset of case 20-29 30-39 40-49 50-54
£ 0.20 -
B 0.10 -
I 0.00
85 FIGURE 1. Observed cumulative risk of breast cancer for mothers and sisters of white cases by age at onset of the case, Cancer and Steroid Hormone Study, 1980-1982.
a white case diagnosed with breast cancer subgroups by current age of the controls by age 29 was 15.60 percent versus 10.27 (figure 2). A log rank test ( x s = 5.12, df = percent when the case was diagnosed be- 3, p = 0.15) indicated that there was no tween 30 and 39 years of age, 6.31 percent significant difference among the Kaplanwhen the case was diagnosed between 40 Meier curves of the relatives based on the and 49 years of age, and 4.10 percent when control's current age. the case was diagnosed between 50 and 54 ' The results of the Cox proportional hazyears of age. The hypothesis of no differ- ards regression models are presented in taence in time to onset of breast cancer ble 4. Using relatives of cases and controls, among the relatives in the four case onset two models were examined. In the first groupings was rejected (x2 = 62.29, df = 3, model, pregnancy variables were included p
Vol. 131, No. 6
Copyright © 1990 by The Johns Hopluni University School of Hygiene and Public Health All rights reserved
Printed m U S A.
Original Contributions AGE AT ONSET AS AN INDICATOR OF FAMILIAL RISK OF BREAST CANCER ELIZABETH B. CLAUS,1 NEIL J. RISCH,1J AND W. DOUGLAS THOMPSON1 Claus, E. B. (Dept of Epidemiology and Public Health, Yale U., New Haven, CT 06510), N. J. Risen, and W. D. Thompson. Age at onset as an indicator of familial risk of breast cancer. Am J Epidemiol 1990; 131:961-72. The familial risk of breast cancer was investigated in a large population-based, case-control study conducted by the Centers for Disease Control. The data set included 4,730 histologically confirmed breast cancer cases aged 20-54 years and 4,688 controls who were frequency matched to cases by geographic region and 5-year categories of age. Family history of breast cancer among first-degree female relatives of cases and controls was utilized. To identify factors associated with familial risk of breast cancer, a Cox proportional hazards model was used, modeling time to onset of breast cancer among mothers and sisters. Case relatives were at greater risk than control relatives. Among relatives of cases, a significant increase in the risk of breast cancer was associated with decreasing age at onset of the case and with having an additional relative affected with breast cancer. The hazard ratio for the mother of a case with breast cancer diagnosed at 50 years of age was 1.7 (95% confidence interval (Cl) 1.4-2.0), compared with 2.7 (95% Cl 2.2-3.2) and 4.3 (95% Cl 3.3-5.6) for the mother of a case whose diagnosis occurred at 40 and 30 years of age, respectively. The hazard ratio for the sister of a case with an unaffected mother and at least one affected sister in addition to the case was 3.6 (95% Cl 2.1-6.1) when the case was diagnosed at age 50, compared with 5.8 (95% Cl 3.4-10.0) and 9.4 (95% Cl 5.3-16.7) when the case was diagnosed at 40 and 30 years of age, respectively. The hazard ratio for the sister of a case with an affected mother and no additional affected sisters was 5.9 (95% Cl 3.9-8.9) when the case was diagnosed at age 50, compared with 9.4 (95% Cl 6.2-14.4) and 15.1 (95% Cl 9.4-24.3) when the case was diagnosed at 40 and 30 years of age, respectively. The hazard ratio for the sister of a case with both an affected mother and at least one affected sister aside from the case was 17.1 (95% Cl 9.4-31.3) when the case was diagnosed at age 50, compared with 27.5 (95% Cl 15.0-50.3) and 44.2 (95% Cl 23.5-83.2) when the case was diagnosed at 40 and 30 years of age, respectively. No effect of case's menopausal status and biiaterality was found, indicating that in addition to a positive family history, age at onset is the strongest indicator of a possible genetic subtype of breast cancer in these data. breast neoplasms; genetics; risk
Editor's note: For a discussion of this It is well documented that a positive fampaper and that by Mettlin et aL immediately ily history is associated with an increased following, see page 984. risk of breast cancer (1-9). Within this Received for publication June 13,1989, and in final form November 3, 1989.
Abbreviations: Cl, confidence interval; SEER, Surveillance, Epidemiology, and End Results. 961
962
CLAUS ET AL.
are more likely to be genetic from those which are not. These factors include parity, age at menarche, menopausal status, age at menopause, history of benign breast disease, laterality, age at first pregnancy, number of livebirths, and number of stillbirths, as well as the number and type of relatives affected with breast cancer. Few other studies have looked systematically at how factors other than age and laterality may help to differentiate between genetic and nongenetic subgroups. The analytic approach of this 3tudy is to model the age-specific incidence of breast cancer in relatives of the cases and controls using risk factor information measured on the cases and controls themselves. The critical issue being addressed in this study is the possibility of discerning potential genetic het1 Department of Epidemiology and Public Health, erogeneity. If women with different risk Yale University School of Medicine, New Haven, CT. * Department of Human Genetics, Yale University factor patterns also show differences in the amount of risk associated with their relaSchool of Medicine, New Haven, CT. Reprintrequeststo Dr. E. B. Claus, Department of tives, then it may be possible to use these Epidemiology and Public Health, Yale University School of Medicine, 60 College Street, New Haven, clinically defined variables to identify women and their families who are espeCT 06510. The authors wish to acknowledge the contributors cially genetically predisposed. group of women with a positive family history, bilaterality of a relative's breast cancer as well as the age at which the relative is affected have been shown to modify the familial risk. Bilateral breast cancer and early onset have been associated with the highest risk to relatives (1, 2, 5, 6, 9). These differences in risk indicate that age at onset and bilaterality may be useful variables for identifying subgroups of breast cancer patients in whom genetic factors play a relatively important etiologic role. In addition to bilaterality and age at onset of breast cancer, this study investigates the possibility that other risk factors associated with breast cancer might be useful in distinguishing breast cancer cases which
to the Cancer and Steroid Hormone Study Study Design and Coordination at the Division of Reproductive Health, Center for Chronic Disease Prevention and Health Promotion, Centers for Disease Control: Principal Investigator, George L. Rubin; Project Director, Phyllis A. Wingo; Project Associates, Nancy C. Lee, Michele G. Mandel, Herbert B. Peterson. Data Collection Centers Principal Investigators: AtlantaRaymond Greenberg; Connecticut—J. Wistar Meigs, W. Douglas Thompson; Detroit—G. Marie Swanson; Iowa—Elaine Smith; New Mexico—Charles Key, Dorothy Pathak; San Francisco—Donald Austin; Seattle—David Thomas; Utah—Joseph Lyon, Dee West. Pathology Review Principal Investigators: Fred Gorstein, Robert McDivitt, Stanley J Robboy. Project Consultants: Lonnie Burnett, Robert Hoover, Peter M. Layde, Howard W. Ory, James J. Schlesselman, David Schottenfeld, Bruce Stadel, Linda Webster, Colin White. Pathology Consultants: Walter Bauer, William Christopherson, Deborah Gersell, Robert Kunnan, Allen Paris, Frank Vellios. The Cancer and Steroid Hormone Study was supported by interagency agreement no. 3-Y01-HD-81037 between the Centers for Disease Control and the National Institute of Child Health and Human Development, with additional support from the National Cancer Institute. E. B. Clau* was supported by USPHS training grant no. 2-T32-ES07085. N. Risch and E. B. Claus were supported by NIH grant no. GM39812 from NIGMS. W. D. Thompson «nd data collection in Connecticut were supported by contract no. 200-800561 from the Centers for Disease Control.
Previous studies have investigated the distribution of various risk factors by casecontrol status (3, 7). In this study, these risk factors, measured in cases and controls, are used to define subgroups of the cases. These subgroups are then compared for level of risk among the relatives. In addition, each relative is also classified according to type of relationship (mother vs. sister), and if a sister, whether the mother is also affected, and whether at least one additional sister is affected. Our ability to detect subgroups of cases whose relatives display distinct frequencies of breast cancer depends on the relation between genetic predisposition and the measured risk factors in producing breast cancer risk. For example, if a risk factor multiplies the risk in genetically predisposed and nonpredisposed individuals by the same amount, then separating cases by risk factor profile (e.g., nulliparous vs. parous) will not yield distinct groups in terms of genetic predisposition. On the other hand, if the
AGE AT ONSET AND FAMILIAL RISK OF BREAST CANCER
risk factor modifies the risk in genetically predisposed individuals to a different degree than in non-predisposed individuals, then cases with different risk factor profiles will be genetically distinct. For example, suppose that nulliparity increases risk only among genetically non-predisposed individuals. Then, among breast cancer cases, those without the risk factor (e.g., paxous women) will have a higher genetic predisposition than those with the risk factor (e.g., nulliparous women), and hence the former group will display a higher level of risk among relatives. Similarly, characteristics of a case's disease, such as bilaterality, may be useful in separating cases into genetically distinct subgroups. Therefore, if associations between risk factors measured in cases and breast cancer risk in relatives are found, it may be possible to use such information to divide breast cancer patients into subgroups that are genetically more homogeneous and to then look at patterns of transmission within these subgroups. The identification of subgroups may also help to identify families that are more likely to be informative for genetic linkage studies. MATERIALS AND METHODS
Study population Data are taken from the Cancer and Steroid Hormone Study, a multicenter, population-based, case-control study conducted by the Centers for Disease Control. The data consist of 4,730 histologically confirmed breast cancer cases 20 to 54 years of age and 4,688 controls. The cases were registered between December 1, 1980 and December 31, 1982 at one of eight Surveillance, Epidemiology, and End Results (SEER) Centers of the National Cancer Institute. The eight centers include the cities and metropolitan areas of Atlanta, Detroit, San Francisco, and Seattle, the four urban counties of Utah, and the states of Connecticut, Iowa, and New Mexico. All cases were interviewed within 6 months of the diagnosis of their first primary breast
963
cancer; age at onset and current age are therefore essentially equivalent for cases. There were 72 bilateral breast cancer cases, all of whom were diagnosed with bilateral breast cancer at time of interview. Controls were frequency matched to cases according to geographic region and 5-year categories of age. Cases and controls with a previous history of breast cancer or a breast biopsy of unknown outcome were excluded from the study. Information on family history was collected on all first-degree female relatives as well as maternal and paternal grandmothers for cases and controls. Data describing the number of affected maternal and paternal aunts were also collected. An extensive amount of additional information was collected for the cases and controls including complete pregnancy and menstrual history, history of benign breast disease, age at onset of disease, alcohol and cigarette usage, breast surgery, sociodemographic variables, and use of oral contraceptives. A detailed description of the study may be found elsewhere (10). Analysis Age-specific Kaplan-Meier estimates of cumulative risk of breast cancer to firstand second-degree female relatives of the cases and controls, stratified by race, were computed using BMDPlL (11). Cumulative risks of breast cancer for relatives of cases were also compared according to the age at onset of breast cancer in the case and according to whether her disease was unilateral or bilateral. Cumulative risks of breast cancer for relatives of controls were compared according to the current age of the .control. Hazard rates were computed for first- and second-degree relatives of the cases and controls using a Cox proportional hazards regression model and BMDP2L. Daughters of cases and controls were excluded from all analyses as only two daughters of cases as well as two daughters of controls were reported to have had breast cancer. Relatives with unknown current age or age at death were eliminated from
964
CLAUS ET AL.
all analyses. For affected relatives with known current age but unknown age at onset, their age at onset was estimated by using the average age at onset for affected individuals whose current age matched that of the relative. To predict the age-specific risk of breast cancer in mothers and sisters of cases and controls, information on breast cancer risk factors, measured on the cases and controls, was used in a Cox proportional hazards regression model. These risk factors included: bilaterality (for cases only), history of benign breast disease, parity, number of livebirths and stillbirths, age at first pregnancy, age at onset of breast cancer, age at menarche, menopausal status, and age at last menstrual period. The dependent variable was defined as years to onset of breast cancer, measured on the mothers and sisters of the cases and controls. To allow for heterogeneity of risk factors across casecontrol status, separate terms for cases and controls for all of the risk factors were included. To more specifically examine familial risk relations, eight additional variables were created. These eight variables (four for the relatives of cases and four for the relatives of controls, respectively) identify each individual by relationship to the case or control either as a mother, as a sister with an affected mother and at least one affected sister aside from the case, as a sister with an affected mother and no affected sisters aside from the case, as a sister with an unaffected mother but at least one affected sister aside from the case, or as a sister with an unaffected mother and no affected sisters aside from the case. Four second-order terms were also included to allow the effect of age at onset of the case to be heterogeneous across different types of relatives; for example, if the effect of the case's age at onset is more pronounced for the sister of a case with an affected mother compared with a sister of a case with an unaffected mother. In addition to the above analyses, breast cancer cases were divided by type of breast cancer family history (i.e., number and type
of relatives affected with breast cancer). The distributions of the remaining risk factors (i.e., percentage of cases with a positive history of benign breast disease, average age at first menstrual period, etc.) were then observed across the varying family history patterns. RESULTS
Kaplan-Meier estimates of cumulative risk of breast cancer to mothers and sisters of both cases and controls, stratified by race, are presented in tables 1 and 2. By age 89, the risk of developing breast cancer was 15.17 percent for mothers of white cases versus 8.97 percent for mothers of white controls. By age 69 (the last age at which the mother of a black case was reported as having breast cancer), the risk of breast cancer for the mother of a black case was 6.26 percent versus 2.91 percent for mothers of black controls. The pattern is the same for sisters of cases and controls. Cumulative risk of breast cancer by age 64 for the sister of a white case was 7.60 percent versus 4.60 percent for the sister of a white control. Among sisters of black women, cumulative risk of breast cancer by age 54 (the last age at which the sister of a black case was reported as having breast cancer) was 3.76 percent for sisters of black cases versus 2.33 percent for sisters of black controls. At all age periods, mothers and sisters of cases, regardless of race, showed a risk of at least one and one-half times that of mothers and sisters of controls, and in general, the risk of developing breast cancer for first-degree relatives of cases was at least two times that of first-degree relatives of controls. The hazard rate of breast qancer for the mother of a white case was 1.98 times the rate for the mother of a white control, while for sisters of white cases, the rate was 2.28 times the rate for sisters of white controls. The hazard rate of breast cancer for the mother of a black case was 2.02 times the rate for the mother of a black control, while for sisters of black cases, the rate was 1.65 times the rate for sisters of black controls.
965
AGE AT ONSET AND FAMILIAL RISK OF BREAST CANCER TABLE 1
Cumulative probability (percent) ± standard error of breast cancer m mothers of cases and controls, by race, Cancer and Steroid Hormone Study. 1980-1982 Mothers of Age (in yean) of mother
0-24 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64 65-69 70-74 75-79 80-84 85-89 90-94 No. at risk No. with breast cancer
Black
White Controls
Case* 0.07 0.20 0.50 1.18 2.33 3.58 4.87 6.14 7.45 9.19 11.16 13.41 14.43 15.17
0.00 0.02 0.23 0.45 0.76 1.57 2.04 2.93 3.69 4.58 5.91 7.25 8.97 8.97 11.43
± 0.04 ± 0.07 ± 0.11 ±0.17 ± 0.24 ± 0.30 ± 0 35 ± 0.39 ± 0.44 ± 0.50 ± 0.59 ± 0.77 ± 0.92 ± 1.18*
4,019 379
± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
0.00 0.02 0.06 0.11 0.14 0.20 0.23 0.28 0.32 0.37 0.47 0.62 0.94 0.94 2.59
Case*
Controls 0.00 0.00 0.22 0.65 0.88 1.11 1.88 1.88 2.31 2.91 3.76
0.00 ± 0.00 0.00 ± 0.00 0.41 ± 0.36 0.62 ± 0.36 1.05 ± 0.47 1.95 ± 0.64 2.92 ± 0.80 4.06 ± 0.97 5.82 ± 1.24 6.26 ± 1.30
4,026 189
± 0.00 ± 0.00 ± 0.22 ± 0.38 ± 0.44 ± 0.49 ± 0.66 ± 0.66 ± 0.79 ± 0.98 ± 1.30
481 11
489 23
* Cumulative risks are presented up until the last time interval in which a breast cancer occurred. TABLE 2
Cumulative probability (percent) ± standard error of breast cancer in sisters of cases and controls, by race. Cancer and Steroid Hormone Study, 1980-1982 Sisters of' Age (in years) of sister 0-24 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64 65-69 No. at risk No. with breast cancer
Black
White Cases 0.01 0.10 0.50 1.17 2.42 3.73 4.96 6.47 7.60 12.15
± ± ± ± ± ± ± ± ± ±
0.02 0.04 0.10 0.16 0.25 0.34 0.43 0.60 0.98 2.05*
5,612 173
Controls 0.06 0.08 0.18 0.45 0.84 1.79 2.15 2.85 4.60
± ± ± ± ± ± ± ± ±
0.03 0.04 0.06 0.10 0.15 0.25 0.29 0.41 0.85
5,772 74
Case* 0.00 0.30 0.53 0.94 1.26 2.53 3.76
± 0.00 ± 0.17 ± 0.24 ± 0 33 ± 0 40 ± 0.70 ± 0.98
1,138 18
Controls 0.10 0.22 0.22 0.36 0.36 1.07 2.33 2.33 7.27
± ± ± ± ± ± ± ± ±
0.10 0.15 0.15 0.21 0.21 0.46 0.86 0.86 3.53
1,037 11
* Cumulative risks are presented up until the last time interval in which a breast cancer occurred.
When the Kaplan-Meier curves for mothers of black and white cases were compared using the generalized Wilcoxon (Breslow) log rank test, the null hypothesis of no difference in time to onset of breast cancer was rejected (x2 = 4.10, df = 1, p =
0.04). However, when the same curves for the sisters of black and white cases were compared, the difference in time to onset of breast cancer due to race was not significant (x2 = 0.79, df = 1, p = 0.37). By age 89, the risk of developing breast cancer for
966
CLAUS ET AL.
mothers of white controls was 8.97 percent, essentially the reported lifetime risk for white females in the United States. However, the risk by age 74 for mothers of black controls was 3.76 percent, a rate lower than the estimated population risk of approximately 7 percent to age 74 among black women (12-14). As a result of this discrepancy and the relatively small number of nonwhite cases and controls (approximately 10 percent of the data set), the remaining analyses included only the white cases and controls. The cumulative risk of breast cancer by age 89 for a maternal grandmother of a white case was 7.34 percent versus 5.42 percent for the maternal grandmother of a white control. The cumulative risk of breast cancer among paternal grandmothers was generally lower than those reported for maternal grandmothers regardless of the age investigated. For example, the cumulative risk of breast cancer by age 89 for the paternal grandmother of a white case was 5.39 percent versus 3.70 percent for the paternal grandmother of a white controlThe hazard rate of breast cancer for the maternal grandmother of a white case was 1.23 times the rate for the maternal grandmother of a white control; for paternal grandmothers of white cases, the rate was 1.46 times the rate for paternal grandmothers of white controls. The difference in risk between maternal and paternal grandmothers is likely to be due to a higher rate of underreporting among paternal relatives than among maternal relatives. It is notable, however, that the hazard ratios comparing relatives of cases to relatives of controls were similar for the two types of grandmothers. Due to overall underreporting, second-degree relatives were also left out of subsequent analyses. Table 3 lists the cumulative probability of breast cancer in sisters of white cases given the mother's breast cancer status. The cumulative risk by age 39 for a sister with an affected mother was 4.95 percent versus 0.84 percent for a sister without an affected mother. By age 59, the risk was
TABLE 3
Cumulatwe probability (percent) ± standard error of breast cancer in sisters of white cases given mother's breast cancer status, Cancer and Steroid Hormone Study, 1930-1982 . Sisters with: Age (in years) of sister
Mother affected with breast
0-24 25-29 30-34 36-39 40-44 45-49 50-54 55-59 60-64 65-69
0.00 ± 0.00 0.71 ± 0 41 2.27 ± 0.75 4.95 ± 1.14 7.94 ± 1.52 12.20 ± 2 07 15.92 ± 2.57 15.92 ± 2.57 25.26 ± 9.10*
0.04 ± 0.03 0.04 ± 0.03 0.34 ± 0.09 0.84 ±0.14 1.94 ± 0.24 3.01 ± 0.32 4.03 ± 0.41 5.67 ± 0.61 7.94 ± 1.03 11.17 ±2.12
No. at risk No. with breast cancer
472
5,140
40
133
cancer
Mother not affected with breast cancer
* Cumulative risks are presented up until the last time interval in which a breast cancer occurred.
15.92 percent for a sister with an affected mother versus 5.67 percent for a sister without an affected mother. The hazard rate of breast cancer for a sister of a case with an affected mother was 8.48 times the rate for a sister of a case without an affected mother. The risk of breast cancer among relatives of white cases was investigated according to whether the case was diagnosed with unilateral or bilateral breast cancer. The hazard rate of breast cancer among sisters of white bilateral cases was 0.95 times that of sisters of white unilateral cases. The hazard rate of breast cancer among mothers of white bilateral cases was 0.66 times that pf mothers of white unilateral cases. Cumulative risk of breast cancer for a mother or a sister was dependent on the age at which the case was affected with breast cancer (figure 1). The lifetime risk, as well as the risk at any given age period, for a first-degree relative appeared to decrease as the age at onset of the case increased. For example, the risk of breast cancer by age 59 for the mother or sister of
967
AGE AT ONSET AND FAiMJLIAL RISK OF BREAST CANCER
0.30 -I Age at onset of case 20-29 30-39 40-49 50-54
£ 0.20 -
B 0.10 -
I 0.00
85 FIGURE 1. Observed cumulative risk of breast cancer for mothers and sisters of white cases by age at onset of the case, Cancer and Steroid Hormone Study, 1980-1982.
a white case diagnosed with breast cancer subgroups by current age of the controls by age 29 was 15.60 percent versus 10.27 (figure 2). A log rank test ( x s = 5.12, df = percent when the case was diagnosed be- 3, p = 0.15) indicated that there was no tween 30 and 39 years of age, 6.31 percent significant difference among the Kaplanwhen the case was diagnosed between 40 Meier curves of the relatives based on the and 49 years of age, and 4.10 percent when control's current age. the case was diagnosed between 50 and 54 ' The results of the Cox proportional hazyears of age. The hypothesis of no differ- ards regression models are presented in taence in time to onset of breast cancer ble 4. Using relatives of cases and controls, among the relatives in the four case onset two models were examined. In the first groupings was rejected (x2 = 62.29, df = 3, model, pregnancy variables were included p
