Journal of the National Cancer Institute Advance Access published online on November 13, 2007
JNCI Journal of the National Cancer Institute, doi:10.1093/jnci/djm224
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
© The Author 2007. Published by Oxford University Press.
ARTICLES |
Predicting Risk of Breast Cancer in Postmenopausal Women by Hormone Receptor Status
Affiliations of authors: Los Angeles Biomedical Research Institute at Harbor-UCLA Medical Center, Torrance, CA (RTC); Fred Hutchinson Cancer Research Institute, Seattle, WA (GLA, AKA); Department of Preventive Medicine, State University of New York, Stony Brook, NY (DSL); Department of Epidemiology and Public Health, Albert Einstein College of Medicine, Bronx, NY (TR); Department of Medicine, University of California Davis, Sacramento, CA (SY); Department of Obstetrics and Gynecology, University of Wisconsin, Madison, WI (GS); Department of Medicine, Evanston Northwestern Healthcare, Evanston, IL (CAR); Department of Medicine, University of California, Irvine, CA (FAH)
Correspondence to: Rowan T. Chlebowski, MD, PhD, Los Angeles Biomedical Research Institute at Harbor-UCLA Medical Center, 1124 West Carson Street, Torrance, CA 90502 (e-mail: rchlebowski{at}gmail.com).
 |
ABSTRACT |
|---|
 Top Abstract Context and CaveatsParticipants and MethodsResultsDiscussionFundingReferencesNotes |
|---|
Background: Strategies for estrogen receptor (ER)–positive breast cancer risk reduction in postmenopausal women require screening of large populations to identify those with potential benefit. We evaluated and attempted to improve the performance of the Breast Cancer Risk Assessment Tool (i.e., the Gail model) for estimating invasive breast cancer risk by receptor status in postmenopausal women.
Methods: In The Women's Health Initiative cohort, breast cancer risk estimates from the Gail model and models incorporating additional or fewer risk factors and 5-year incidence of ER-positive and ER-negative invasive breast cancers were determined and compared by use of receiver operating characteristics and area under the curve (AUC) statistics. All statistical tests were two-sided.
Results: Among 147 916 eligible women, 3236 were diagnosed with invasive breast cancer. The overall AUC for the Gail model was 0.58 (95% confidence interval [CI]=0.56 to 0.60). The Gail model underestimated 5-year invasive breast cancer incidence by approximately 20% (P<.001), mostly among those with a low estimated risk. Discriminatory performance was better for the risk of ER-positive cancer (AUC = 0.60, 95% CI = 0.58 to 0.62) than for the risk of ER-negative cancer (AUC = 0.50, 95% CI = 0.45 to 0.54). Age and age at menopause were statistically significantly associated with ER-positive but not ER-negative cancers (P=.05 and P=.04 for heterogeneity, respectively). For ER-positive cancers, no additional risk factors substantially improved the Gail model prediction. However, a simpler model that included only age, breast cancer in first-degree relatives, and previous breast biopsy examination performed similarly for ER-positive breast cancer prediction (AUC=0.58, 95% CI= 0.56 to 0.60); postmenopausal women who were 55 years or older with either a previous breast biopsy examination or a family history of breast cancer had a 5-year breast cancer risk of 1.8% or higher.
Conclusions: In postmenopausal women, the Gail model identified populations at increased risk for ER-positive but not ER-negative breast cancers. A model with fewer variables appears to provide a simpler approach for screening for breast cancer risk.
Prior knowledge Because many postmenopausal women must be screened to identify a population who will benefit from tamoxifen treatment for breast cancer risk reduction, methods to rapidly identify such a population are needed. Study design Data from the observational study and the clinical trial cohorts of the Women's Health Initiative were used. Four prediction models were investigated, including the Gail model (tested in the clinical trial cohort) and three logistic regression models (trained on the observational study cohort and tested on the clinical trial cohort). Contribution A model with only three risk factors—age, breast cancer in first-degree relatives, and previous breast biopsy examination—performed nearly as well as the Gail model for the prediction of estrogen receptor (ER)–positive breast cancer. Implications The new model with fewer variables than the Gail model may be as effective at identifying women at high risk for ER-positive breast cancer who would benefit from risk reduction interventions. Limitations Information on atypical hyperplasia, reproductive hormone levels, mammogram breast density, and bone mineral density, all risk factors for breast cancer, were not available. The Gail model was the only model evaluated; other models are in clinical use.
|
The Breast Cancer Risk Assessment Tool (i.e., the Gail model) is used to predict risk of invasive breast cancer (including both estrogen receptor [ER]–positive and ER-negative disease) in women 35 years of age or older (1,2). However, strategies to reduce breast cancer risk, including use of tamoxifen, raloxifene, and aromatase inhibitors, influence almost exclusively ER-positive disease, with the use of raloxifene and aromatase inhibitors limited to postmenopausal women (3,4). In addition, although many women could potentially benefit (5), risk–benefit considerations indicate that a large number of postmenopausal women must be screened to identify a population who gain net benefit from tamoxifen use (6,7). As a result, methods to rapidly identify a population of postmenopausal women at increased risk of ER-positive breast cancer are needed.
We therefore examined models of breast cancer risk in participants of the Women's Health Initiative (WHI). Our objective was to facilitate the identification of postmenopausal women at increased risk for ER-positive invasive breast cancer as candidates for potential risk reduction interventions.
|
Participants and Methods |
|---|
|
TopAbstractContext and CaveatsParticipants and MethodsResultsDiscussionFundingReferencesNotes |
|---|
Study Population
The WHI is a large multicomponent study designed to test three chronic disease risk reduction strategies and to examine risk factors for these conditions in postmenopausal women. Details of the implementation of both the observational study with 93676 participants and the four randomized clinical trials with 68132 participants that evaluated menopausal hormone therapy, a low-fat dietary intervention, and calcium–vitamin D supplementation have been published (8). Briefly, women were recruited at 40 clinical centers in the United States largely through direct mailings (9). Postmenopausal women, who were aged 50–79 years and unlikely to move or die within 3 years, were eligible. Each randomized trial had additional eligibility requirements related to safety and the intervention under test. Potential participants who were not eligible or interested in the randomized trials entered the observational study. All clinical trials excluded women with a history of breast cancer and required that the baseline mammogram and clinical breast examination not be suspicious for breast cancer. Breast screening was not required for enrollment in the observational study enrollment. For these analyses, women with previous invasive breast cancer, previous noninvasive breast cancer, previous mastectomy, or less than 5 years follow-up were excluded, leaving 147916 women who met all eligibility criteria.
All participants provided written informed consent. Human subjects committee approval at each participating institution was provided.
Data Collection
Participants provided data on demographics; medical, reproductive, and family medical histories; and lifestyle factors, such as smoking and alcohol use, and physical activity (10). Menopausal hormone therapy use (i.e., use of estrogen alone or combined estrogen plus progestin) was ascertained through an interviewer-administered questionnaire.
The Gail model variables include age, ethnicity, age at menarche, age of the mother at the birth of her first live child, number of first-degree relatives with breast cancer (0, 1, or >1), number of previous breast biopsy examinations (0, 1, or >1), and the presence or absence of atypical hyperplasia in the biopsy specimen (http://brca.nci.nih.gov/brc/questions.htm). Calculations of 5-year risk estimates from the modified Gail model for women in this report were made by the National Surgical Adjuvant Breast and Bowel Project statistical center by following their usual coding procedures on data from individual WHI participants, courtesy of Dr Joseph Costantino. Because historical information on atypical hyperplasia was not collected in the WHI, all women with previous breast biopsy examinations are coded as "unknown" for this last variable.
Follow-up and Breast Cancer Ascertainment
Breast cancer incidence and mammography use were updated annually (in the observational study) or semiannually (in the clinical trial) by mail or telephone questionnaires. Self-reported breast cancers were verified by centrally trained WHI physician adjudicators who reviewed pathology reports (11). Final adjudication and coding were performed at the WHI Clinical Coordinating Center by use of the Surveillance, Epidemiology, and End Results Program (12). Only the 3263 invasive breast cancers diagnosed within 5 years of enrollment and confirmed by central review were included as events, 713 in situ breast cancers were excluded from analyses, and 144680 women with no invasive breast cancer were coded as control subjects. The 363 case patients with missing or borderline information on ER status were excluded from subgroup analyses.
Statistical Analyses
Risk of invasive breast cancer was initially assessed with the Gail model and coded as a four-level nominal variable with categories of no invasive breast cancer or invasive breast cancer that was considered as ER-positive and progesterone receptor (PR)–positive tumors, ER-positive and PR-negative tumors, or ER-negative tumors. There were too few ER-negative invasive cancers to subdivide this group by PR status. After initial receiver operating characteristic (ROC) analyses, the two ER-positive invasive breast cancer categories were combined resulting in three final nominal variables of no invasive breast cancer, ER-positive invasive breast cancer, and ER-negative invasive breast cancer.
To explore whether other models could predict ER-positive and ER-negative breast cancer with similar or improved accuracy, the data were divided into the distinct observational study and clinical trial cohorts. The prediction models that we investigated were the Gail model, which was tested in the clinical trial cohort, and three logistic regression models, which were trained on the observational study cohort and tested on the clinical trial cohort. The first logistic regression model included the Gail model risk factors; the second model added parity, breastfeeding, smoking, alcohol, body mass index, physical activity, duration of previous estrogen-alone use, and duration of previous estrogen plus progestin use. The third, simpler model used only a subset of modified Gail model risk factors (age, number of first-degree relatives with breast cancer [coded as 0 or 
1], and number of previous breast biopsy examinations [coded as 0, 1, or >1]).
The discriminatory accuracy of each model was assessed and compared by use of the ROC curve (R version 2.3 and R library ROCR, R Development Core Team, http://www.R-project.org) and the corresponding area under the curve (AUC). For models developed in the observational study cohort, the ROC and AUC were obtained by applying the models to the independent clinical trial cohort. ROC curves plot the true-positive rate (sensitivity) versus the false-positive rate (1 – specificity) at a continuum of thresholds; a participant is predicted to have breast cancer if her estimated probability of breast cancer exceeds a particular threshold. An ROC curve that corresponds to a fair-coin toss classifier (i.e., a nonpredictive model) is a straight line connecting the coordinates (0,0) to (1,1) and has an AUC of 0.50. An ROC curve that corresponds to a perfect classifier is a pair of vertical and horizontal lines connecting the coordinates (0,0) to (0,1) to (1,1) and has an AUC of 1.00.
To aid in our understanding of the ROC analysis, a single multinomial logistic regression model (in SAS PROC LOGISTIC version 9.1; SAS Institute, Cary, NC) was used in the pooled clinical trial and observational study cohorts to examine whether risk factors were associated with ER-positive and ER-negative invasive breast cancer, separately and combined, and whether the associations differed by receptor status. Specifically, the estimated coefficients of the risk factors were allowed to vary by ER status. This model included the Gail model risk factors and the additional risk factors described above; to increase power, family history of breast cancer was recoded as 0 or 1 or more. All risk factors were included regardless of statistical significance. Odds ratios (ORs), confidence intervals (CIs), and P values for tests of main effects and for tests of heterogeneity between tumor types were based on Wald statistics. All statistical tests were two-sided.
To estimate absolute risks and prevent downward bias, missing receptor status was dealt with by a multiple imputation model. Because the main goal of this particular analysis is estimation, data from the WHI observational study and clinical trials were combined. Calibration was verified by use of the Hosmer and Lemeshow goodness of fit test (13).
|
Results |
|---|
|
TopAbstractContext and CaveatsParticipants and MethodsResultsDiscussionFundingReferencesNotes |
|---|
Women in both the clinical trial and observational study cohorts were ethnically diverse and had a mean age of 63 years (range=50–79 years), with 21% being 70 years or older (Table 1). Serial mammography was common, with an average of 0.7 mammogram per year in both cohorts. Given the large sample size, tests for statistical significance between cohorts were highly statistically significant for nearly all characteristics. There appeared to be substantive differences between the cohorts in age, ethnicity or race, body mass index, family history, previous benign breast biopsy examination, and use of hormone therapy. Among 147916 women eligible for these analyses, 3236 developed invasive breast cancer within 5 years. ER status was borderline or missing for 363 breast cancer patients, and 2412 ER-positive and 461 ER-negative invasive breast cancers were diagnosed.
|
The ROC curves for predicting invasive breast cancer by use of the Gail model 5-year risk probabilities for the 64568 women in the WHI clinical trial produced an AUC of 0.58 (95% CI=0.56 to 0.60). The AUC provided reasonable prediction of ER-positive and PR-positive tumors and ER-positive and PR-negative tumors (AUC=0.60 for both). Given the similar ability to predict ER-positive and PR-positive tumors and ER-positive and PR-negative tumors, these categories were combined in subsequent modeling. For all ER-positive tumors, the usual Gail model threshold of 1.67%, used for defining increased risk, corresponds to approximately 50% sensitivity and 65% specificity. For ER-negative tumors, the Gail model was comparable to a random process (AUC=0.50, 95% CI=0.45 to 0.54) (Fig. 1).
|
The accuracy of the Gail model estimated probabilities (calibration) for estimating the incidence of invasive breast cancer was assessed by comparing observed cases of breast cancer in the cohort with the number predicted by the Gail model (as the sums of the individual estimated Gail model probabilities of breast cancer). The Gail model 5-year risk estimate statistically significantly underestimated the number of breast cancers diagnosed within 5 years by approximately 20% (3236 observed versus 2562 expected, P<.001) with the disparity concentrated among those with lower Gail model risk estimates (
1.37% expected 5-year risk). The total Gail model estimate was more closely aligned with the number of ER-positive breast cancers observed (Table 2).
|
To better understand this difference in predictive performance between estimated breast cancer incidence from the Gail model and the observed incidence in the WHI cohort, we examined breast cancer risk factors by hormone receptor status (ER-positive tumors, ER-negative tumors, and cancer-free for 5 years) (Table 3). In multinomial logistic regression analyses, the 5-year probability of ER-positive breast cancer in the entire cohort was statistically significantly associated with age, ethnicity, family history of breast cancer, number of previous breast biopsy examinations, age at menopause, parity, age at first birth, smoking status, alcohol use, and body mass index. African American women were at statistically significantly lower risk of ER-positive disease than white women (OR=0.68, 95% CI=0.54 to 0.85).
|
Prior breast biopsy examination and body mass index were the only risk factors that were statistically significantly associated with ER-negative disease. The smaller number of ER-negative tumors may have precluded detection of some associations; however, the data indicate a different pattern of association for most factors with ER-positive disease than with ER-negative disease. For ER-negative cancers, odds ratios for age, family history, age at menopause, greater parity, smoking, and alcohol use were all close to 1.0. The association between breast cancer risk and chronologic age (P=.05), race or ethnicity (P=.01), and age at menopause (P=.04) differed statistically significantly across ER disease subtypes (Table 1). Age at menarche was not statistically significantly associated with either ER-positive or ER-negative breast cancer.
To determine whether other models could improve Gail model performance for prediction of ER-positive breast cancer, the effects of the Gail model risk factors were reestimated in a logistic regression model that used the WHI observational study cohort as a training set, and the resulting model was applied to WHI clinical trial cohort as a test set. This approach differed from the original approach in that it did not explicitly account for competing risks. However, only 5.3% of women in the observational study and 4.7% of women in the clinical trial died or were lost to follow-up within 5 years of enrollment. Expanding this model to add parity, breastfeeding, smoking, alcohol, body mass index, physical activity, duration of previous estrogen-alone use, and duration of previous estrogen plus progestin use produced ROC curves in a similar pattern with only slight improvements in the AUC statistics (Fig. 2). By focusing on only ER-positive tumors, we simplified the model to include only age, first-degree relatives with breast cancer (coded as 0 or 1), and number of previous breast biopsy examinations (coded as 0, 1, or >1) and we obtained an AUC of 0.58 (95% CI=.56 to 0.60). The difference between the Gail model and the simplified model at the estimated 1.8% probability threshold was small and not statistically significant (difference=0.02, 95% CI=–0.04 to 0.08) (Fig. 3). The absolute risk prediction results for ER-positive invasive breast cancer from the simplified model by age increments are shown in Table 4. To prevent the downward bias of the predicted 5-year risk of ER-positive invasive breast cancers, missing receptor status was dealt with by multiple imputations. The imputation model included age, ethnicity, and age at menopause. These variables were chosen from Table 3 because these variables suggested differing risks by receptor status.
|
|
|
All women older than 55 years with either a previous biopsy examination or a first-degree relative with breast cancer had a 5-year risk of invasive breast cancer that was greater than 1.8%. In an exploratory analysis, the same process was applied to African American participants in the clinical trial and observational study cohorts. For these African American women, women aged 60 years or older with a previous biopsy examination and a positive family history of breast cancer had a 5-year risk greater than 1.8% (Table 4).
|
Discussion |
|---|
|
TopAbstractContext and CaveatsParticipants and MethodsResultsDiscussionFundingReferencesNotes |
|---|
In a large cohort of postmenopausal women (50–79 years of age at entry), the discriminatory performance of the Gail model was similar to that observed in previous studies (14,15), but it underestimated the observed 5-year invasive breast cancer incidence by approximately 20%. Discriminatory performance and model calibration were somewhat better for estimating population-based risk of ER-positive breast cancers, but the performance of the Gail model for predicting ER-negative breast cancers was equivalent to chance alone. Incorporation of several additional risk factors provided only a small improvement in prediction. However, a simpler model that incorporated fewer variables was nearly as accurate as the Gail model in predicting ER-positive breast cancer risk and would be more accessible for routine and rapid prescreening in the prevention or routine care setting, in which breast cancer risk is only one measure among the many requiring assessment.
In our analyses, the association of putative risk factors with breast cancer prediction varied by ER status. For ER-positive breast cancers among postmenopausal women, the Gail model components of age, ethnicity, age of the mother at birth of her first live child, number of first-degree relatives with breast cancer, and number of prior breast biopsy examinations were statistically significantly associated with breast cancer risk. Age at menarche was not associated with risk. Also associated with the risk of an ER-positive tumor were age at menopause, parity, and body mass index. In contrast, only prior breast biopsy examination and body mass index were statistically significantly associated with the risk of ER-negative breast cancer, although the smaller number of patients with ER-negative breast cancer may have limited our ability to detect some associations, particularly for race or ethnicity.
Although the Gail model has been validated for predicting total breast cancer risk in several settings (14,15), including both pre- and postmenopausal women, the utility of the Gail model for predicting ER-positive compared with ER-negative breast cancers in postmenopausal women has not been previously recognized to our knowledge. In these analyses, a woman's age, a Gail model component, was associated only with the risk of ER-positive breast cancer but not with the risk of ER-negative breast cancer. As a result, the Gail model appears to have the ability to differentially predict breast cancer risk by receptor subgroup in postmenopausal women.
Despite its well-documented predictive performance (14,15), the Gail model statistically significantly underestimated breast cancer incidence in the WHI cohort of postmenopausal women. The initial four studies that validated the original Gail model included not only postmenopausal women but also some much younger premenopausal women (1,16–18), and two of these studies (17,18) entered women no older than 54 and 61 years at entry, respectively. In these studies, mammography was not prespecified and was rarely used before the mid-1980s. Because tumors detected by mammography are more likely than those detected by other means to be ER-positive tumors than ER-negative tumors (19) and older women are substantially more likely than younger women to develop ER-positive tumors (20), it is likely that these older studies did not optimally detect ER-positive disease. In contrast, the WHI cohort included only postmenopausal women who were 50 years or older, with a substantial number older than 70 years, and had comprehensively used mammography. In addition, a secular change in breast biopsy procedures had occurred, beginning in the early 1990s, away from open surgical biopsy examination to the common use of image-guided percutaneous core biopsy examinations, which are associated with less morbidity and a lower threshold for use (21). As a result of these changes, the breast biopsy rate per 100000 Medicare beneficiaries increased by 43% between 1999 and 2004 (22). These factors likely contributed to a more comprehensive ascertainment of all patients with breast cancer in the WHI cohort, especially for those with ER-positive disease, and likely enhanced our ability to discriminate between the influences on ER-positive and ER-negative disease. Future development of breast cancer risk models, consequently, should consider premenopausal and postmenopausal women separately and segregate risk by hormone receptor status.
Other groups have examined the risk factors associated with breast cancer in hormone receptor subgroups (20,23–25). In a meta-analysis of observational studies, parity and age at first child's birth were associated with ER-positive and PR-positive tumors but not with ER-negative and PR-negative breast cancers (22). In another report (19), statistically significant heterogeneity among the four ER and PR categories was observed for some risk factors but not for prior breast biopsy examinations, family history of breast cancer, alcohol use, and height. Direct comparison with our analyses was precluded by differences in study populations, breast cancer receptor categories, and the risk factors examined.
Despite evaluation of multiple additional risk factors, we did not identify a model that would clearly improve the accuracy of risk prediction for ER-positive breast cancers over that of the Gail model. However, we did identify a simpler model with only three variables—age, family history of breast cancer in first-degree relatives, and previous breast biopsy examination—which had predictive accuracy approaching that of the Gail model.
Although many women could potentially benefit from tamoxifen use, there has been reluctance to incorporate Gail model breast cancer risk assessment in routine clinical practice (6,26–28). In one survey (6), only 11% of California primary care physicians had used the Gail model for risk assessment in the past year. In a recent national survey of primary care providers (29), only 16% agreed that "it is easy to determine" who is eligible for breast cancer risk reduction strategies and only 25% had prescribed tamoxifen for risk reduction in the past year. As a result, development of new, simpler risk models is an identified research priority (30). The simplified model described above provides a straightforward approach to initial screening for risk of ER-positive breast cancer in postmenopausal women and does not require computer use. Women who were 55 years or older with either a first-degree relative with breast cancer or a previous breast biopsy have 5-year risk of 1.8% or higher (which is higher than the Gail model threshold of 1.67%).
In previous analyses of the WHI cohort (31), African American women were identified as being at substantially lower risk for ER-positive breast cancer but at substantially higher risk for ER-negative, high-grade breast cancers with poor prognosis. These analyses suggest that African American women who were 60 years or older with a first-degree relative with breast cancer and a previous breast biopsy examination may have a 5-year breast cancer risk of 1.8% or higher. Given the sample size in this subgroup, additional studies are needed to confirm this finding. Because evaluation of the Gail model in multiethnic populations that include African American women have been preliminary (32–33) or disappointing (34), further model development in minority populations of African American and Hispanic women is a research priority.
The demonstration that the Gail model and our proposed simpler model can predict the risk of ER-positive breast cancer has several clinical implications. Although the selective estrogen receptor modulator (SERM) tamoxifen is approved by the Federal Drug Administration (FDA) for breast cancer risk reduction (35), a large number of postmenopausal women must be screened to identify potential candidates. By one calculation, only one of 142 screened postmenopausal women who were 60–79 years has a favorable balance for tamoxifen use (6). The SERM raloxifene has recently been approved by the FDA for breast cancer risk reduction in postmenopausal women as well and has a somewhat more favorable side effect profile (36). The current Gail model or our simpler model could identify populations of postmenopausal women appropriate for further consideration for breast cancer risk reduction interventions. For interactions with individual women in the clinic, the limitations of any risk assessment estimate should be acknowledged and the risk or benefit of any intervention should be carefully considered (3,26).
Benign breast disease, a recognized breast cancer risk factor, may be able to integrate hormone exposures and breast tissue response that lead to mammographic breast alterations and indications for a biopsy examination. In the future, histologic subclassification of benign breast disease (nonproliferative versus proliferative and the magnitude of lobular involution), which further differentiates breast cancer risk (37–39), may lead to more reliable risk estimation.
The strengths of this study include the prospective design, a large racially representative population that is well characterized for breast cancer risk factors and has serial assessment of mammography use, central adjudication of breast cancer pathology reports, and information on breast cancer hormone receptor status. In addition, the risk assessment models evaluated in this report were developed in one WHI population (the observational study cohort) and independently tested in a separate population (the clinical trial cohort), in which many of the risk assessment procedures and breast cancer outcomes were determined similarly.
Our study has several limitations. Information on atypical hyperplasia, a Gail model component, was not available in our cohort. However, the Gail model does allow for missing histologic subclassification and that was how we coded the WHI data. We evaluated only one model and recognize other risk prediction models are in clinical use (29). We selected the Gail model because it is the most commonly cited model, is the most frequently used by care providers in the United States, and is the basis for the FDA approved indication for use of tamoxifen for prevention.
Another study limitation is that reproductive hormone levels, mammogram breast density, and bone mineral density, which are strongly associated with breast cancer risk (40–43), were not available for these analyses. However, analyses incorporating these risk factors have provided mixed results. Although estrogen levels have been commonly related to breast cancer (40,43,44), in a randomized trial involving women at relatively high risk of breast cancer, estrogen and testosterone levels were not associated with either subsequent breast cancer risk or risk reduction by tamoxifen (45). The addition of mammographic breast density has not improved (46) or only modestly improved (47,48) breast cancer prediction over the Gail model. To our knowledge, the influence of bone mineral density addition to Gail model prediction has not been reported. At present, therefore, the role of these three factors on the assessment of breast cancer risk in routine clinical practice remains to be determined.
In summary, we found that among postmenopausal women who receive regular mammographic examinations, the Gail model predicts ER-positive breast cancer risk at the population level but does not predict the risk of ER-negative tumors. A model incorporating only age, family history of breast cancer, and previous breast biopsy examinations provides a simpler approach for initial identification of populations of postmenopausal women at elevated risk for ER-positive breast cancer who may benefit by further evaluation for risk reduction interventions. Future attempts to improve breast cancer risk models should consider premenopausal and postmenopausal women separately and segregate risk by hormone receptor status.
|
Funding |
|---|
|
TopAbstractContext and CaveatsParticipants and MethodsResultsDiscussionFundingReferencesNotes |
|---|
The Women's Health Initiative program is funded by the National Heart, Lung, and Blood Institute, National Institutes of Health, Department of Health and Human Services.
|
NOTES |
|---|
|
TopAbstractContext and CaveatsParticipants and MethodsResultsDiscussionFundingReferencesNotes |
|---|
A list of key investigators involved in this research follows. A full listing of WHI investigators can be found at the following Web site: http://www.whi.org.
Program Office: Elizabeth Nabel, Jacques Rossouw, Shari Ludlam, Linda Pottern, Joan McGowan, Leslie Ford, and Nancy Geller (National Heart, Lung, and Blood Institute, Bethesda, MD).
Clinical Coordinating Center: Ross Prentice, Garnet Anderson, Andrea LaCroix, Charles L. Kooperberg, Ruth E. Patterson, Anne McTiernan (Fred Hutchinson Cancer Research Center, Seattle, WA); Sally Shumaker (Wake Forest University School of Medicine, Winston-Salem, NC); Evan Stein (Medical Research Labs, Highland Heights, KY); Steven Cummings (University of California at San Francisco, San Francisco, CA).
Clinical Centers: Sylvia Wassertheil-Smoller (Albert Einstein College of Medicine, Bronx, NY); Jennifer Hays (Baylor College of Medicine, Houston, TX); JoAnn Manson (Brigham and Women's Hospital, Harvard Medical School, Boston, MA); Annlouise R. Assaf (Brown University, Providence, RI); Lawrence Phillips (Emory University, Atlanta, GA); Shirley Beresford (Fred Hutchinson Cancer Research Center, Seattle, WA); Judith Hsia (George Washington University Medical Center, Washington, DC); Rowan Chlebowski (Los Angeles Biomedical Research Institute at Harbor-UCLA Medical Center, Torrance, CA); Evelyn Whitlock (Kaiser Permanente Center for Health Research, Portland, OR); Bette Caan (Kaiser Permanente Division of Research, Oakland, CA); Jane Morley Kotchen (Medical College of Wisconsin, Milwaukee, WI); Barbara V. Howard (MedStar Research Institute/Howard University, Washington, DC); Linda Van Horn (Northwestern University, Chicago/Evanston, IL); Henry Black (Rush Medical Center, Chicago, IL); Marcia L. Stefanick (Stanford Prevention Research Center, Stanford, CA); Dorothy Lane (State University of New York at Stony Brook, Stony Brook, NY); Rebecca Jackson (The Ohio State University, Columbus, OH); Cora E. Lewis (University of Alabama at Birmingham, Birmingham, AL); Tamsen Bassford (University of Arizona, Tucson/Phoenix, AZ); Jean Wactawski-Wende (University at Buffalo, Buffalo, NY); John Robbins (University of California at Davis, Sacramento, CA); F. Allan Hubbell (University of California at Irvine, CA); Howard Judd (University of California at Los Angeles, Los Angeles, CA); Robert D. Langer (University of California at San Diego, LaJolla/Chula Vista, CA); Margery Gass (University of Cincinnati, Cincinnati, OH); Marian Limacher (University of Florida, Gainesville/Jacksonville, FL); David Curb (University of Hawaii, Honolulu, HI); Robert Wallace (University of Iowa, Iowa City/Davenport, IA); Judith Ockene (University of Massachusetts/Fallon Clinic, Worcester, MA); Norman Lasser (University of Medicine and Dentistry of New Jersey, Newark, NJ); Mary Jo O'Sullivan (University of Miami, Miami, FL); Karen Margolis (University of Minnesota, Minneapolis, MN); Robert Brunner (University of Nevada, Reno, NV); Gerardo Heiss (University of North Carolina, Chapel Hill, NC); Lewis Kuller (University of Pittsburgh, Pittsburgh, PA); Karen C. Johnson (University of Tennessee, Memphis, TN); Robert Brzyski (University of Texas Health Science Center, San Antonio, TX); Gloria E. Sarto (University of Wisconsin, Madison, WI); Denise Bonds (Wake Forest University School of Medicine, Winston-Salem, NC); Susan Hendrix (Wayne State University School of Medicine/Hutzel Hospital, Detroit, MI).
Dr R. T. Chlebowski is a consultant for Astra-Zeneca, Novartis, Lilly, and Organon, and he has conducted research funded by Lilly.
The authors thank the WHI investigators and staff for their outstanding dedication and commitment. The National Institutes of Health had input into the design and conduct of the study.
|
REFERENCES |
|---|
|
TopAbstractContext and CaveatsParticipants and MethodsResultsDiscussionFundingReferencesNotes |
|---|
(1) Costantino JP, Gail MH, Pee D, Anderson S, Redmond CK, Benichou J, et al. Validation studies for models projecting the risk of invasive and total breast cancer incidence. J Natl Cancer Inst (1999) 91:1541–8.
(2) Freedman AN, Seminara D, Gail MH, Hartge P, Colditz GA, Ballard-Barbash R, et al. Cancer risk prediction models: a workshop on development, evaluation, and application. J Natl Cancer Inst (2005) 97:715–23.
(3) Chlebowski RT, Col N, Winer EP, Collyar DE, Cummings SR, Vogel VG, et al. American Society of Clinical Oncology technology assessment of pharmacologic interventions for breast cancer risk reduction including tamoxifen, raloxifene and aromatase inhibition. J Clin Oncol (2002) 20:3328–43.
(4) Kinsinger LS, Harris R, Woolf SH, Sox HC, Lohr KN. Chemoprevention of breast cancer: a summary of the evidence for the US preventive services task force. Ann Intern Med (2002) 137:59–69.
(5) Freedman AN, Graubard BI, Rao SR, McCaskill-Stevens W, Ballard-Barbash R, Gail MH. Estimates of the number of US women who could benefit from tamoxifen for breast cancer chemoprevention. J Natl Cancer Inst (2003) 95:526–32.
(6) Brewster AM, Christo DK, Lai H, Helzlsouer K. Breast carcinoma chemoprevention in the community setting. Estimating risks and benefits. Cancer (2005) 103:1147–53.[Medline]
(7) Lewis CL, Kinsinger LS, Harris RP, Schwartz RJ. Breast cancer risk in primary care: implications for chemoprevention. Arch Intern Med (2004) 164:1897–903.
(8) Anderson GL, Manson J, Wallace R, Lund B, Hall D, Davis S, et al. Implementation of the Women's Health Initiative study design. Ann Epidemiol (2003) 13:S5–17.[CrossRef][Web of Science][Medline]
(9) Hays J, Hunt JR, Hubbell FA, Anderson GL, Limmacher M, Allen C, et al. The Women's Health Initiative recruitment methods and results. Ann Epidemiol (2003) 13(Suppl):S18–77.[CrossRef][Web of Science][Medline]
(10) Langer RD, White E, Lewis CE, Kotchen JM, Hendrix SL, Trevisan M, et al. The Women's Health Initiative observational study: baseline characteristics of participants and reliability of baseline measures. Ann Epidemiol (2003) 13(Suppl):S107–21.[CrossRef][Web of Science][Medline]
(11) Curb JD, McTiernan A, Heckbert SR, Kooperberg C, Stanford J, Nevitt M, et al. WHI Morbidity and Mortality Committee. Outcomes ascertainment and adjudication methods in the Women's Health Initiative. Ann Epidemiol (2003) 13(Suppl):S122–8.[CrossRef][Web of Science][Medline]
(12) National Cancer Institute. About SEER. Available at http://www.seer.cancer.gov/. [Last accessed: July 29 2007.].
(13) Hosmer DW, Hosmer T, Le Cessie S, Lemeshow SA. comparison of goodness-of-fit tests for the logistic regression model. Stat Med (1997) 16:965–80.[CrossRef][Web of Science][Medline]
(14) Rockhill B, Spiegelman D, Byrne C, Hunter DJ, Colditz GA. Validation of the Gail et al. model of breast cancer risk prediction and implications for chemoprevention. J Natl Cancer Inst (2001) 93:358–66.
(15) Bondy ML, Lustbader ED, Halabi S, Ross E, Vogel VG. Validation of a breast cancer risk assessment model in women with a positive family history. J Natl Cancer Inst (1994) 86:620–5.
(16) Gail MH, Brinton LA, Byar DP, Corle DK, Green SB, Schairer C, et al. Projecting individualized probabilities of developing breast cancer for white females who are being examined annually. J Natl Cancer Inst (1989) 81:1879–86.
(17) Wingo PA, Ory HW, Layde PM, Lee NC. The evaluation of the data collection process for a multicenter, population-based case-control design. Am J Epidemiol (1988) 128:206–17.
(18) Spiegelman D, Colditz GA, Hunter D, Hertzmark E. Validation of the Gail et al. model for predicting individual breast cancer risk. J Natl Cancer Inst (1994) 86:600–7.
(19) Porter PL, El-Bastawissi AY, Mandelson MT, Lin MG, Khalid N, Watney EA, et al. Breast tumor characteristics as predictors of mammographic detection: comparison of interval- and screen-detected cancers. J Natl Cancer Inst (1999) 91:2020–8.
(20) Colditz GA, Rosner BA, Chen WY, Holmes MD, Hankinson SE. Risk factors for breast cancer according to estrogen and progesterone receptor status. J Natl Cancer Inst (2004) 96:218–28.
(21) Levin DC, Rao VM, Frangos AJ, Parker L, Sunshine JH. Current practice patterns and recent trends in breast biopsy among radiologists and surgeons. J Am Coll Radiol (2006) 3:707–9.[CrossRef][Medline]
(22) Ghosh K, Melton J III, Suman VJ, Grant CS, Sterioff S, Brandt KR, et al. Breast biopsy utilization. Arch Intern Med (2005) 165:1593–8.
(23) Ma H, Bernstein L, Pike MC, Ursin G. Reproductive factors and breast cancer risk according to joint estrogen and progesterone receptor status: a meta-analysis of epidemiological studies. Breast Cancer Res (2006) 8. R43.
(24) Rosenberg LU, Einarsdottir K, Friman EI, Wedren S, Dickman PW, Hall P, et al. Risk factors for hormone receptor-defined breast cancer in postmenopausal women. Cancer Epidemiol Biomarkers Prev (2006) 15:2482–8.
(25) Chen WY, Colditz GA. Risk factors and hormone-receptor status: epidemiology, risk-prediction models and treatment implications for breast cancer. Nat Clin Pract Oncol (2007) 4:415–23.[CrossRef][Web of Science][Medline]
(26) Kaplan CP, Haas JS, Pérez-Stable EJ, Gregorich SE, Somkin C, Des Jarlais G, et al. Breast cancer risk reduction options: awareness, discussion, and use among women from four ethnic groups. Cancer Epidemiol Biomarkers Prev (2006) 15:162–6.
(27) Fosket J. Constructing "high risk women": the development and standardization of a breast cancer risk assessment tool. Sci Technol Human Values (2004) 29:291–313.[Abstract]
(28) Savage L. Researchers wonder why high-risk women are not taking chemoprevention drugs [news]. J Natl Cancer Inst (2007) 99:913–4.
(29) Armstrong K, Quistberg DA, Micco E, Domchek S, Guerra C. Prescription of tamoxifen for breast cancer prevention by primary care physicians. Arch Intern Med (2006) 166:2260–5.
(30) Freedman AN, Seinara D, Gail MH, Hartge P, Colditz GA, Ballard-Barbash R, et al. Cancer risk prediction models: a workshop on development, evaluation, and application. J Natl Cancer Inst (2005) 97:715–23.
(31) Chlebowski RT, Chen Z, Anderson GL, Rohan T, Aragaki A, Lane D, et al. Ethnicity and breast cancer: factors influencing differences in incidence and outcome. J Natl Cancer Inst (2005) 97:1619–20.
(32) Bondy ML, Newman LA. Breast cancer risk assessment models: applicability to African American women. Cancer (2003) 97:230–5.[CrossRef][Medline]
(33) Newman LA, Gail MH, Selvan M, Bondy M, Rockhill B, Colditz GA, et al. Proposed revision of the Gail breast cancer risk assessment model for African American women (2003) American Society of Clinical Oncology.
(34) Adams-Campbell LL, Makambi KH, Palmer JR, Rosenberg L. Diagnostic accuracy of the Gail model in the Black Women's Health Study. Breast J (2007) 13:332–6.[CrossRef][Web of Science][Medline]
(35) Fisher B, Costantino JP, Wickerham DL, Redmond CK, Kavanah M, Cronin WM, et al. Tamoxifen for prevention of breast cancer: report of the National Surgical Adjuvant Breast and Bowel Project P-1 Study. J Natl Cancer Inst (1998) 90:1371–88.
(36) Vogel V, Costantino JP, Wickerham DL, Cronin WM, Cecchini RS, Atkins JN, et al. Effects of tamoxifen versus raloxifene on the risk of developing invasive breast cancer and other disease outcomes: the NSABP Study of Tamoxifen and Raloxifene (STAR) P-2 Trial. JAMA (2006) 295:2727–41.
(37) Hartmann LC, Sellers TA, Frost MH, Lingle WL, Degnim AC, Ghosh K, et al. Benign breast disease and the risk of breast cancer. N Engl J Med (2005) 353:229–37.
(38) Santen RJ, Mansel R. Benign breast disorders. N Engl J Med (2005) 353:275–85.
(39) Arpino G, Laucirica R, Elledge RM. Premalignant and in situ breast disease: biology and clinical implications. Ann Intern Med (2005) 143:446–57.
(40) The Endogenous Hormones and Breast Cancer Collaborative Group. Endogenous sex hormones and breast cancer in postmenopausal women: reanalysis of nine prospective studies. J Natl Cancer Inst (2002) 94:606–16.
(41) Boyd NF, Guo H, Martin LJ, Sun L, Stone J, Fishell E, et al. Mammographic density and the risk and detection of breast cancer. N Engl J Med (2007) 356:227–36.
(42) Cauley JA, Lucas FL, Kuller LH, Vogt MT, Browner WS, Cummings SR. Bone mineral density and risk of breast cancer in older women:the study of osteoporotic fractures. Study of Osteoporotic Fractures Research Group. JAMA (1996) 276:1404–8.
(43) Tamimi RM, Byrne C, Colditz GA, Hankinson SE. Endogenous hormone levels, mammographic density, and subsequent risk of breast cancer in postmenopausal women. J Natl Cancer Inst (2007) 99:1178–87.
(44) Cummings SR, Duong T, Kenyon T, Cauley JA, Whitehead M, Krueger KA. Multiple Outcomes of Raloxifene Evalution (MORE) Trial. Serum estradiol level and risk of breast cancer during treatment with raloxifene. JAMA (2002) 287:216–20.
(45) Beattie MS, Costantino JP, Cummings SR, Wickerham DL, Vogel VG, Dowsett M, et al. Endogenous sex hormones, breast cancer risk, and tamoxifen response: an ancillary study in the NSABP Breast Cancer Prevention Trial (P-1). J Natl Cancer Inst (2006) 98:110–5.
(46) Tice JA, Cummings SR, Ziv E, Kerlikowske K. Mammographic breast density and the Gail model for breast cancer risk prediction in a screening population. Breast Cancer Res Treat (2005) 94:115–22.[CrossRef][Web of Science][Medline]
(47) Barlow WE, White E, Ballard-Barbash R, Vacek PM, Titus-Ernstoff L, Carney PA, et al. Prospective breast cancer risk prediction model for women undergoing screening mammography. J Natl Cancer Inst (2006) 98:1204–14.
(48) Chen J, Pee D, Ayyagari R, Graubard B, Schairer C, Bryne C, et al. Projecting absolute invasive breast cancer risk in white women with a model that includes mammographic density. J Natl Cancer Inst (2006) 98:1215–26.
Manuscript received May 23, 2007; revised September 5, 2007; accepted October 9, 2007.
Editorial about this Article
Related Article in JNCI
CiteULike 
Connotea 
Del.icio.us What's this?
J Natl Cancer Inst 2007 99: 1657-1659.
J Natl Cancer Inst 2007 99: 1653.
This article has been cited by other articles:
|
|
 |
|
  C. J. Crandall, A. K. Aragaki, R. T. Chlebowski, A. McTiernan, G. Anderson, S. L. Hendrix, B. B. Cochrane, L. H. Kuller, and J. A. Cauley New-Onset Breast Tenderness After Initiation of Estrogen Plus Progestin Therapy and Breast Cancer Risk Arch Intern Med, October 12, 2009; 169(18): 1684 - 1691. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 K. Visvanathan, R. T. Chlebowski, P. Hurley, N. F. Col, M. Ropka, D. Collyar, M. Morrow, C. Runowicz, K. I. Pritchard, K. Hagerty, et al. American Society of Clinical Oncology Clinical Practice Guideline Update on the Use of Pharmacologic Interventions Including Tamoxifen, Raloxifene, and Aromatase Inhibition for Breast Cancer Risk Reduction J. Clin. Oncol., July 1, 2009; 27(19): 3235 - 3258. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. R. Cummings, J. A. Tice, S. Bauer, W. S. Browner, J. Cuzick, E. Ziv, V. Vogel, J. Shepherd, C. Vachon, R. Smith-Bindman, et al. Prevention of Breast Cancer in Postmenopausal Women: Approaches to Estimating and Reducing Risk J Natl Cancer Inst, March 18, 2009; 101(6): 384 - 398. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 T. M. Peters, A. Schatzkin, G. L. Gierach, S. C. Moore, J. V. Lacey Jr., N. J. Wareham, U. Ekelund, A. R. Hollenbeck, and M. F. Leitzmann Physical Activity and Postmenopausal Breast Cancer Risk in the NIH-AARP Diet and Health Study Cancer Epidemiol. Biomarkers Prev., January 1, 2009; 18(1): 289 - 296. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
L. R. Howe and S. M. Lippman Modulation of Breast Cancer Risk by Nonsteroidal Anti-inflammatory Drugs J Natl Cancer Inst, October 15, 2008; 100(20): 1420 - 1423. [Full Text] [PDF]
|
|
|
|
|
|
M. H. Gail, W. F. Anderson, M. Garcia-Closas, and M. E. Sherman Absolute Risk Models for Subtypes of Breast Cancer J Natl Cancer Inst, November 21, 2007; 99(22): 1657 - 1659. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||