Breast Cancer Risk in Primary Care: Implications for Chemoprevention

Background  Some women may benefit from taking tamoxifen citrate for breast cancer prevention if the probability of benefit outweighs that of adverse events. We determined the proportion of women aged 40 to 69 years attending general internal medicine practices who were potentially eligible for tamoxifen chemoprevention and calculated the maximum proportion of breast cancers that could be prevented.

Methods  Six hundred five women aged 40 to 69 completed self-administered questionnaires in the waiting rooms of 10 general internal medicine practices in North Carolina in 2001.

Results  Among white women, 9.0% (95% confidence interval [CI], 5.1%-15.2%) in their 40s, 24.0% (95% CI, 18.2%-31.0%) in their 50s, and 53.4% (95% CI, 46.1%-61.3%) in their 60s had a 5-year Gail model estimated breast cancer risk of 1.66% or greater. Among black women, 2.9% (95% CI, 0%-15.0%) in their 40s, 7.1% (95% CI, 1.1%-24.4%) in their 50s, and 13.0% (95% CI, 3.1%-34.3%) in their 60s had a similar risk. When adverse events were considered in white women, 10% or fewer in all age groups were potentially eligible for chemoprevention. The maximum proportion of breast cancers prevented in eligible women was 6.0% to 8.3%.

Conclusions  Small numbers of women in primary care practices are eligible for discussions about chemoprevention; the maximum proportion of breast cancers prevented if eligible women take tamoxifen is also small. Challenges lie in targeting discussions to the most appropriate women and in finding new chemoprevention strategies that have less risk of harms.

For the past 30 years, clinicians have used screening with mammography and breast examination as the primary approach for control of breast cancer. Despite the important role of screening, randomized trials document its modest effect on breast cancer mortality.^1,2 New evidence suggests consideration of another approach: chemoprevention of breast cancer with selective estrogen receptor modulators.

Results of the National Surgical Adjuvant Breast and Bowel Project's Breast Cancer Prevention Trial (BCPT) showed a 49% reduction in the incidence of breast cancer (175 cases of invasive breast cancer among 6599 women in the placebo arm vs 89 cases among 6576 women in the tamoxifen citrate arm) among high-risk women 35 years and older who were randomized to tamoxifen, a selective estrogen receptor modulator, compared with placebo.³ The BCPT also found that tamoxifen increased the risk of serious adverse outcomes, such as endometrial cancer and thromboembolic events (eg, pulmonary embolism, deep venous thrombosis, stroke). Although not FDA-approved for breast cancer prevention, raloxifene hydrochloride, another selective estrogen receptor modulator, has been shown to reduce the incidence of breast cancer by a similar amount.⁴ Raloxifene also increases the incidence of deep venous thrombosis and pulmonary embolism, but not endometrial cancer or stroke.⁵

Expert groups suggest that clinicians discuss tamoxifen chemoprevention with women who could realize a net health benefit.^6-8 This includes women with a high potential of benefit and a low potential of harm from taking tamoxifen.

The relative risk reduction in breast cancer incidence in the BCPT was the same for all risk groups; women with a higher risk of breast cancer experienced a greater absolute benefit from taking tamoxifen.^5,9 Because the BCPT also found that tamoxifen increases the incidence of serious adverse events primarily among women older than 50 years and other results show that the baseline risk (ie, in the absence of tamoxifen) for these complications increases with age, older women have a greater probability of harm from tamoxifen.⁹ Therefore, the balance between tamoxifen benefits and harms is most favorable for women at increased risk for breast cancer who are aged 35 to 50 or for women older than 50 who have a low probability of adverse events (ie, have had a hysterectomy and have no history of or risk factors for thromboembolic events).

These women are the ones most eligible for discussions about chemoprevention with tamoxifen. The most commonly used method of assessing breast cancer risk is the Gail model. This is a multivariate logistic regression model developed by Gail et al¹⁰ from data in a large cohort study of breast cancer screening. The model, available on the Internet,¹¹ has proven accurate in 2 population-based studies.^12,13

It is unknown how many women in typical primary care practice meet the criteria of increased breast cancer risk and low risk of adverse events from tamoxifen. It is also not clear how great an effect discussions about tamoxifen could make on the incidence of breast cancer. Many women, including women in their 40s, are worried about their risk of developing breast cancer¹⁴ and may be interested in considering breast cancer chemoprevention with tamoxifen.

To address these issues, we surveyed women aged 40 to 69 years visiting their general internal medicine physicians in central North Carolina. To identify women with increased breast cancer risk, we asked questions to calculate the estimated 5-year breast cancer risk for each woman using the Gail model.¹⁰ To identify women at low risk for adverse events from tamoxifen, we also asked questions about previous hysterectomy and risk factors for thromboembolic conditions. Finally, we asked questions about the degree to which women were worried about breast cancer. We calculated the maximum potential effect of tamoxifen on the expected number of breast cancers that would be diagnosed during the next 5 years.

This study was a part of the Making Prevention Work study funded by the National Cancer Institute. For this study, we recruited 10 community general internal medicine practices in the 4 cities included in the larger study; all agreed to participate. We excluded solo practices to maximize the efficiency of recruitment at each study site; we excluded academic practices because we were focused on community practices. The Institutional Review Board of the School of Public Health at The University of North Carolina at Chapel Hill approved the study.

Trained research assistants consecutively approached women aged 40 to 69 in the waiting rooms of the participating practices from January 15, 2000, to May 30, 2000, requesting that they complete a 5- to 10-minute anonymous self-administered written questionnaire. Women were recruited on Tuesdays, Wednesdays, and Thursdays in each practice for 2 consecutive weeks. The questions asked about breast cancer risk factors used in the Gail model calculations, about factors that affect the risk for harms with the use of tamoxifen (history of thromboembolic conditions, diabetes mellitus, hypertension, and hysterectomy), and about worry regarding breast cancer. Women with a history of breast cancer or colon cancer were excluded.

The Gail model estimates breast cancer risk based on information about 7 risk factors: age, ethnic group, number of first-degree relatives with breast cancer, age at menarche, age at first live birth, number of breast biopsies, and presence of atypical hyperplasia in the biopsy specimen.¹⁰ The Gail model was developed using data from a group of mostly white women; modifications to the model were later made to allow for race-specific determinations of breast cancer risk.³ The analysis in this article was limited to women who identified themselves as white or black because few women of other ethnic groups were surveyed (<5% of the larger study).

The Gail model also requires information about the presence of atypical hyperplasia on breast biopsy samples. Because most women do not know this information, 2 approximations were used. One involved asking a proxy question, "Have you ever had an abnormal biopsy?" Response options were "yes," "no," and "unknown." Of the 145 women who had had at least 1 biopsy, 13 (9.0%) responded affirmatively. In other studies, 2% to 17% of breast biopsy specimens showed atypical hyperplasia.¹⁵ Only "yes" responses were included in the calculation for atypical hyperplasia. As another approach, all women with biopsies were coded as "unknown" for atypical hyperplasia. Both approaches gave nearly identical results; the first approach is reported herein.

Because age is an important risk factor for breast cancer and for harms from tamoxifen, all analyses were stratified by decade of age (40-49, 50-59, and 60-69 years). The frequencies of demographic information and breast cancer risk factors for the Gail model were examined separately for white and black women. Five-year breast cancer risk estimates were computed for each individual using FORTRAN code for the Gail model obtained from the National Cancer Institute (D. Pee, PhD, written communication, June 4, 2000). Exact binomial confidence intervals were calculated.

To determine potential risk factors for harms from tamoxifen for breast cancer prevention, women were asked questions about their medical history. Women responded yes or no to questions about whether a physician had ever told them that they had high blood pressure, diabetes mellitus, blood clots in the legs, or blood clots in the lungs. Although hypertension and diabetes mellitus are not specific contraindications to the use of tamoxifen, they were included as risk factors because these conditions also increase the risk of stroke. An intact uterus was included as a risk factor for tamoxifen for women older than 50 years.

Women with an estimated 5-year breast cancer risk of at least 1.66% were defined as having increased breast cancer risk. Women were defined as having a low risk for adverse events from tamoxifen if they were aged 40 to 49 without a history of thromboembolic events (deep venous thrombosis or pulmonary embolism), hypertension, or diabetes mellitus. Women aged 50 to 69 were required also to have had a hysterectomy. Discussion with women about tamoxifen was considered to be appropriate if they had an increased risk of breast cancer and a low risk for adverse events from tamoxifen. Because the small number of black women in our study limited the ability to reliably estimate the percentage of black women meeting both of these criteria, this analysis was restricted to white women.

To determine women's worry about breast cancer, we asked "How worried are you about getting breast cancer?" Responses offered were "very worried," "somewhat worried," and "not worried." Women were classified as worried if they responded that they were very or somewhat worried.

Efficient approaches to identifying women with high breast cancer risk were also examined. Age, family history, and history of previous biopsies are weighted most heavily in estimating breast cancer risk in the Gail model. Positive responses to family history and previous biopsy within age categories were compared with full Gail model risk calculations. In accord with the BCPT, increased 5-year breast cancer risk was defined as a 1.66% or higher 5-year probability of developing breast cancer.³

Using the distribution of individual Gail model risk estimates for women in the sample, we calculated the expected total number of breast cancers for 10 000 similar white women in each age group (40-49, 50-59, and 60-69 years) during the next 5 years (10 000 × [the sum of the probabilities of breast cancer/the total number in the sample] = the number of cancers expected). In a similar manner, we calculated the expected number of breast cancers for the subgroup of women with increased breast cancer risk and low risk of harms from tamoxifen. Assuming that complete adherence to 5 years of tamoxifen use would reduce these cancers by 49%,³ the maximum percentage of breast cancers in each group that would be prevented by tamoxifen was estimated. These last 2 analyses were limited to white women because the number of black women in the sample was too small for stable calculations.

To confirm the estimates for breast cancer risk levels in women aged 40 to 49, information collected from similar surveys of women in the same 10 internal medicine practices from July 15, 1997, to February 28, 1998, (n = 1452, 85.0% response rate) was also analyzed. The questions for the Gail model risk factors and data analysis procedures were identical to those already described.

Six hundred five women (84.9% of women we approached) agreed to complete the questionnaire. The most common reasons for nonparticipation were short waiting times before the office visit and feeling too ill. Nonrespondents and respondents had the same mean age. Other details about nonrespondents were not available for comparison.

Participants were predominately white (79%-89% of each group by decade). Almost all had finished high school and had health insurance (Table 1). The proportion of women who reported that they had ever had a breast biopsy ranged from 8.6% in younger black women to 32.1% in older white women; the proportion of white women reporting a biopsy was higher than that for black women in each age group. The proportion of women reporting a family history of breast cancer in a first-degree relative ranged from 2.9% to 21.4% by decade and racial group.

The proportion of women with an estimated 5-year risk of breast cancer of 1.66% or higher as calculated by the Gail model increased with age. Estimated risks for white women were higher in each age group compared with black women in the same age group (Table 2). Among women in their 40s, 9.0% (95% confidence interval [CI], 5.1%-15.2%) of white women and 2.9% (95% CI, 0%-15.0%) of black women had an estimated 5-year breast cancer risk of 1.66% or greater. This proportion increased to 24.0% (95% CI, 18.2%-31.0%) for white women and 7.1% (95% CI, 1.1%-24.4%) for black women in their 50s and 53.4% (95% CI, 46.1%-61.3%) for white women and 13.0% (95% CI, 3.1%-34.3%) for black women in their 60s.

To confirm the findings in a larger sample, we analyzed data collected earlier in the same study from the same practices on a larger number of women in their 40s. In this sample of 1452 women, 8.0% (95% CI, 6.8%-9.8%) of white women and 4.0% (95% CI, 2.7%-5.2%) of black women had estimated breast cancer risks of 1.66% or higher (R.P.H., oral communication, August 9, 2000), similar to the proportions already reported. Earlier data were not collected from women aged 50 to 69.

Three (25.0%) of the women aged 40 to 49 with increased breast cancer risk had 1 or more conditions that could potentially increase adverse events from tamoxifen. Among women with increased breast cancer risk in their 50s and 60s, 30 (65.2%) and 87 (85.3%), respectively, had 1 or more such conditions.

Nearly 7% (6.7%) (95% CI, 3.1%-11.8%; n = 9) of white women in their 40s had an increased risk of breast cancer and no conditions that would increase their likelihood of adverse events from tamoxifen (Table 3). Six percent (95% CI, 2.9%-10.7%; n = 12) of women in their 50s and 9.8% (95% CI, 6.2%-14.9%; n = 19) of women in their 60s met these requirements for discussions about tamoxifen.

We then considered how worried the women eligible for discussion about tamoxifen were about breast cancer. Most (8/9) of the women in their 40s, many (8/12) women in their 50s, and slightly more than half (10/19) of the women in their 60s with increased breast cancer risk and low risk for adverse events reported being very or somewhat worried about breast cancer (Table 3).

To determine if clinicians could accurately estimate the breast cancer risk level of women without applying the full Gail model calculation, we examined the predictive value of the absence of 2 risk factors—a family history of breast cancer and previous breast biopsy—in identifying women who were not at increased risk of breast cancer. No women in their 40s (negative predictive value, 100%) and only 3 women in their 50s (negative predictive value, 97%) who had an estimated 5-year risk of breast cancer less than 1.66% reported a family history of breast cancer or history of breast biopsy (Table 4). Most women in their 40s (73.1% [95% CI, 64.8%-80.1%]) and 50s (64.1% [95% CI, 57.2%-70.6%]) would have needed no further risk assessment, as they did not have either risk factor. In a strategy of calculating full Gail model risk estimates only for women who had at least 1 of these risk factors, about a quarter of women in their 40s and slightly more than one third of women in their 50s would require further risk assessment.

We calculated the potential effect of identifying women eligible for discussions about tamoxifen from a hypothetical cohort of 10 000 white women similar to the women in this study and offering all of them treatment. We did not calculate the effect for black women because of the small number of black women in our sample.

In a hypothetical cohort of 10 000 white women aged 40 to 49 whose 5-year risk estimates correspond to the levels of risk of white women in this study, 109 women at all risk levels would be expected to be diagnosed with invasive breast cancer for 5 years, based on Gail model calculations (Figure 1). For 10 000 women in their 50s, 149 cancers would be expected in the coming 5 years; 217 cancers during 5 years would be expected for 10 000 women in their 60s.

If clinicians discussed chemoprevention with all eligible women and all of these women took tamoxifen for 5 years (assuming a relative risk reduction of 49% as in the BCPT), 6.0% to 8.3% (9-15 cases per 10 000 women) of all breast cancers in women aged 40 to 69 would be prevented. These percentages represent the maximum possible effect of a strategy of discussing chemoprevention with appropriate women.

This study found that from 9.0% (ages 40-49) to 53.4% (ages 60-69) of white women and 2.9% (ages 40-49) to 13.0% (ages 60-69) of black women visiting their primary care physicians had an estimated 5-year breast cancer risk of 1.66% or higher. An efficient way to identify women in their 40s or 50s (but not 60s) with this degree of risk is to first ask about family history of breast cancer and previous breast biopsy, calculating a full Gail model estimate only on those women who answered positively to at least 1 of these questions.

The study further found that 10% or fewer of white women had an increased breast cancer risk and a low risk for tamoxifen adverse events and thus met criteria for discussions about tamoxifen. If all of these women adhered to tamoxifen use for 5 years, the maximum effect in these age groups would be to reduce the number of invasive breast cancers during the next 5 years by 6.0% to 8.3%. If chemoprevention with a drug such as raloxifene could be safely offered also to women in their 50s and 60s who had not had a hysterectomy, the percentage of breast cancers potentially preventable would increase to 11.0% to 13.2%. In actual practice, probably not all such women would take tamoxifen for 5 years; therefore, the actual number of cancers prevented would be smaller. In the 4 trials of breast cancer chemoprevention with tamoxifen, 24% to 40% of the women in the treatment group discontinued the drug.⁵ The inclusion of hypertension and diabetes mellitus as risk factors for adverse events is a cautious approach and leads to a more conservative estimate of the number of women who could potentially benefit from tamoxifen use. Practitioners may be concerned about prescribing tamoxifen for women whose risk of stroke is increased because of such comorbid conditions.

To our knowledge, this study is the first to assess breast cancer risk and the effect of tamoxifen for breast cancer prevention in women from primary care practices. In a study of BlueCross BlueShield of North Carolina enrollees, about 3% of women aged 40 to 45 and 12% aged 50 to 55 had risks of 1.66% or greater.¹⁶ This study's corresponding estimates were 6.3% for women aged 40 to 45 and 18.0% for women aged 50 to 55. Analyses of the Nurses' Health Study found that 3.3% of all breast cancers (in women of all ages) occurred in women who were aged 45 to 74 and were appropriate candidates for tamoxifen (based only on the absence of a uterus).¹³ Theoretically, tamoxifen could prevent about half of these, or 1.6%, assuming that all eligible women took tamoxifen for 5 years. Our study estimated instead the proportion of potentially preventable cancers within each age group, excluding older women at higher risk of adverse events from tamoxifen. The maximum benefit found ranged from 6.0% to 8.3% of breast cancers among women aged 40 to 69.

Freedman and colleagues¹⁷ used data from the 2000 National Health Interview Survey to estimate the total number of US women, aged 35 to 79 years, who were eligible for tamoxifen chemoprevention based on Food and Drug Administration eligibility criteria. They estimated that, overall, 15.5% of these women would be eligible, with wide variations by race (18.7% of white women and 5.7% of black women). When they considered potential harms, only 4.9% of white women and 0.6% of black women had a positive benefit-risk index.

The potential effect of chemoprevention on breast cancer incidence may be smaller than might seem likely from the large relative risk reduction in the BCPT. The reasons for the small effect are different for women in different age groups. For women in their 40s, a small proportion of breast cancers occurs in the group of women with increased risk. Therefore, despite the fact that these younger women have a lower likelihood of adverse effects from tamoxifen, the overall effect of chemoprevention is small. For women in their 50s and 60s, the potentially larger number of women with increased breast cancer risk is reduced by a larger number of these women who have a substantial likelihood of adverse effects. Again, the magnitude of effect on the total societal burden of breast cancer is small. Whether women with higher levels of breast cancer worry would be more likely to consider use of tamoxifen for chemoprevention is unknown.

This study has several limitations. First, the number of women in each age category was small, and black women, women with lower education, and women with no health insurance were not well represented in the sample. Similar studies involving larger numbers of such women would be helpful. Second, although the study achieved high response rates, the respondents were from only a small number of primary care practices in one state. Therefore, the generalizability of this study is uncertain; further study using a larger number of women in more primary care practices is needed to confirm these results. Third, there was no way of determining how many of the women who reported having an abnormal biopsy finding actually had atypical hyperplasia. Two different approaches to estimate the overall risk distributions were used, each approach providing a similar result.

The 5-year breast cancer risk distributions in this study demonstrate that the Gail model estimates for black women are significantly lower than those for white women with similar risk factors. For example, a 55-year-old woman with "average" Gail model variables would have a 5-year breast cancer risk of 1.1% if she were white, compared with 0.6% if she were black.

The Gail model is one way to identify women at high risk for breast cancer. Other models have been developed¹⁸ and may be more accurate for some subgroups of patients. All of these models have deficiencies, however. Improved approaches to assessing breast cancer risk could help better target chemoprevention.¹⁹

Whether chemoprevention will find an important place in our armamentarium against breast cancer is still uncertain. Many questions remain, including tamoxifen's effects on breast cancer mortality and other noncancer conditions. Much also depends on whether US women will embrace chemoprevention as many have embraced screening mammography. If future chemopreventive agents have fewer adverse effects (including such quality-of-life effects as hot flashes) and if more older women at higher breast cancer risk can be added to those who might be appropriate for these drugs, a greater proportion of breast cancers could be prevented.

Correspondence: Carmen L. Lewis, MD, MPH, Division of General Medicine and Clinical Epidemiology, Department of Medicine, The University of North Carolina at Chapel Hill, Campus Box 7110, 5039 Old Clinic Building, Chapel Hill, NC 27599-7110 (Carmen_Lewis@med.unc.edu).

Accepted for publication November 28, 2003.

This study was supported by grant 5 R01 CA068378 from the National Cancer Institute, Bethesda, Md, and The University of North Carolina at Chapel Hill Lineberger Comprehensive Cancer Center; and by grant 290-97-0011 from the Agency for Healthcare Research and Quality, Rockville, Md. Dr Lewis is the recipient of Cancer Control Career Development Award for Primary Care Physicians 00-180-01 from the American Cancer Society, Atlanta, Ga.

We appreciate statistical assistance from Bahjat Qaqish, PhD; the editorial assistance of Kathleen N. Lohr, PhD; the assistance of Mitchell Gail, PhD, and David Pee, PhD, for providing the program for the Gail model; and the secretarial assistance of Audrina Bunton.