Barriers to minority participation in breast carcinoma prevention trials
Breast carcinoma prevention trials must recruit large cohorts of women who have an above-average risk of developing breast carcinoma. Recruitment for the Study of Tamoxifen and Raloxifene (STAR) trial required volunteers to complete a risk assessment questionnaire form (RAF). Women whose estimated risk of developing breast carcinoma in the next 5 years was ≥ 1.67% based on the Gail model were invited to participate in STAR. Less than 4% of participants in the previously conducted P1 (tamoxifen vs. placebo) trial were minority women. We, therefore, studied barriers to minority participation in STAR among black, white, and Hispanic women who completed an RAF.
The authors analyzed the association of Gail model risk factors, education, and insurance with race/ethnicity using chi-square tests and two-sided P values. They developed logistic regression models of trial eligibility, controlling for the Gail model risk factors, education, and insurance status.
Among 823 women who completed an RAF, white women were 10 times as likely as Hispanic women and 45 times as likely as black women to be eligible for STAR. Age at first birth (P = 0.04), having an affected first-degree relative (P < 0.0001), having had a biopsy (P < 0.0001), education (P < 0.0001), and insurance status (P < 0.0001) varied by race/ethnicity. All variables except insurance status were associated with eligibility when race was excluded from the model. In a model that included race/ethnicity, the same factors remained statistically significant.
These findings suggested that both the race/ethnicity adjustment and socioeconomic factors were barriers to eligibility for and contribute to low minority participation in breast cancer prevention trials. Cancer 2005. © 2005 American Cancer Society.
Breast carcinoma chemoprevention trials have begun to show great promise. However, their findings have not translated into widespread community use in part because, for valid research reasons, only women at above-average risk were eligible to participate in these trials. In addition, because breast carcinoma incidence rates are lower among racial/ethnic minorities than among white women, relatively few minority women were eligible to participate. These limitations are unfortunate because most breast carcinomas occur in women without special risk factors, and without minority participation, differences in the effects of interventions by race/ethnicity cannot be studied.1–5
In 1998, Breast Cancer Prevention Trial (BCPT) investigators reported that, among women at higher than average risk, breast carcinoma incidence was 50% lower in the group assigned to take tamoxifen.6 The BCPT was the first to show that taking a pill could reduce breast carcinoma risk, and with 13,388 subjects across the United States and Canada, it was one of the largest breast carcinoma prevention studies ever completed.6–10 In a recent analysis of pooled data from 13 National Surgical Adjuvant Breast and Bowel Project (NSABP) trials, black and white women with breast carcinoma were found to derive similar benefits from using tamoxifen to prevent contralateral breast carcinoma. The hazard ratios for tamoxifen users versus nonusers were 0.74 for black women and 0.76 for white women. However, black patients with breast carcinoma were less likely than white patients to receive tamoxifen treatment.11 Similar comparative analyses could not be conducted in the BCPT because < 4% of its participants were minorities.
In 1999, the NSABP undertook the Study of Tamoxifen and Raloxifene (STAR), to compare the effectiveness and side effects of tamoxifen with those of raloxifene, an estrogen receptor modifier.12 Recently closed to enrollment, the STAR trial has randomly assigned 19,747 postmenopausal women, at > 500 participating institutions in the United States, Puerto Rico, and Canada, to take tamoxifen or raloxifene for 5 years.13
Although the NSABP has devoted a great deal of effort to recruiting minority women,6, 7 < 7% of STAR participants are members of racial/ethnic minority groups. Only 2.5% are of African descent (black women), and 2% were born in or are descended from those born in the Spanish-speaking countries of the Western Hemisphere (Hispanic women).13 Yet, black and Hispanic women together account for > 20% of the U.S. population and 12% of female patients with breast carcinoma, and breast carcinoma mortality rates are higher among black women than among women of European descent (white women).4, 14, 15
When we began recruiting subjects for the STAR trial, we were aware that minority recruitment for the BCPT had been disappointing. The purpose of the current study, initiated in July 2001, was to analyze the association between race/ethnicity and STAR enrollment at our site.
MATERIALS AND METHODS
The Herbert Irving Comprehensive Cancer Center at Columbia University (New York, NY) is part of a large urban medical center, The New York Presbyterian Health Care System. The Cancer Center's catchment area, the Washington Heights neighborhood, is home to 350,000 people, of whom 85% are Hispanic or black.
Our recruitment efforts for the STAR trial followed a protocol used nationwide and approved by the institutional review boards of the participating institutions. An outreach coordinator or recruiter made contact with healthy women seen in faculty practice offices, clinics, small health care facilities, free mammography screening sites, churches, community-based organizations, and breast carcinoma support groups (educating survivors of breast carcinoma so that they could educate their first-degree female relatives [FDR]). In one-on-one or small group meetings, study staff provided education on breast health, breast carcinoma, and clinical trials. Women with no personal history of breast carcinoma were asked to complete a risk assessment form (RAF) and some additional questions about sociodemographic factors. Recruiters administered the study questionnaires in English or Spanish depending on the respondent's preference. Women were eligible for the study if they had a Gail model risk > 1.67%, were > 35 years, were postmenopausal, and had no history of cancer.
The risk assessment form
The data entered on the RAF are values for the variables used in a formula called the Gail model to assess individual risk for breast carcinoma. The Gail model is based on data collected in the 1970s from 280,780 white women participating in a trial of screening mammography.16–18 In this large sample, Gail et al.19 analyzed factors believed to be associated with breast carcinoma risk and developed a formula intended for use in estimating individual 5-year and lifetime risk of developing breast carcinoma. The variables used in the formula are current age, age at menarche, number of previous breast biopsies, abnormal biopsy results, age at first live birth, and number of FDRs affected with breast carcinoma. To be eligible for the STAR trial, a woman must have a ≥ 1.67% risk of developing breast carcinoma over the next 5 years according to the model.
The first report on the Gail model noted that the data reflected the experience of white women only. The study also included a caveat that the formula should be used only among women who received annual mammograms.19 Because no minority women were included in the sample from which the Gail model was developed, and because minority women have lower breast carcinoma incidence rates than white women, an adjustment factor for race was added to the model used for risk assessment in chemoprevention trials.6
An additional questionnaire, developed and used only at our site, collected data on educational attainment and health insurance coverage.
Women who identified their race/ethnicity as other than black, white, or Hispanic were eliminated from the current analysis because they were too few in number to be analyzed separately. To determine whether study participants differed by race/ethnicity in Gail model risk for breast carcinoma, we conducted analyses of variance of the age variables in the model by race/ethnicity. We also compared the proportions with an affected (with breast carcinoma) FDR, a history of breast biopsy, a history of abnormal biopsy results, and health insurance coverage, and the distributions in different categories of educational attainment by race/ethnicity using chi-square tests with two-sided P values. We then conducted logistic regression analyses of eligibility for prevention trial participation as determined by the NSABP (based on a 5-year Gail model risk ≥ 1.67%), using as independent variables the Gail model factors as well as educational attainment and insurance status.
From July 1, 2001 to October 31, 2004, when the study was closed to further accrual, 847 women volunteered to learn about and be screened for the STAR trial. Of these, 24 were excluded from analysis because they identified their race/ethnicity as other than black, white, or Hispanic. Of the remaining 823, 164 were found eligible for the STAR trial.
Table 1 shows the distribution of Gail model and socioeconomic factors by race/ethnicity among 823 black, white, or Hispanic women who completed an RAF. Race/ethnicity was statistically, significantly associated with almost all the variables analyzed. White women were younger at menarche, older at first live birth, and more likely to be nulliparous, to have an affected FDR, and to have had a breast biopsy than black or Hispanic women. White women also were more likely to have attended graduate school and to be eligible for the STAR trial. Among women who had received a biopsy, white women were more likely than black or Hispanic women to report a history of lobular carcinoma in situ or atypical hyperplasia, and less likely to report that they did not know their biopsy results.
Table 1. Gail Model and Socioeconomic Factors among Subjects Completing a Risk Assessment Form for the STAR Trial, by Race/ethnicitya
|Current age||57.5 (8.9)||55.1 (7.8)||57.1 (8.7)||55.9 (8.2)||0.14|
|Age at menarche||12.8 (1.5)||13.2 (1.8)||13.1 (1.9)||13.1 (1.8)||0.03|
|Age at first live birth||24.6 (7.7)||21.9 (6.0)||21.1 (6.2)||22.1 (6.4)||< 0.0001|
|Have an affected FDR||35.1||13.7||20.4||18.4||< 0.0001|
|Ever had a biopsy||55.6||20.1||24.5||26.7||< 0.0001|
|Of those who had a biopsy|| || || || || |
| Atypical hyperplasia||25.3||13.9||2.8||16.0||0.07|
| Lobular carcinoma in situ||18.7||3.7||2.8||8.7||0.0007|
|Educational attainment|| || || || ||< 0.0001|
| High school or less||9.4||48.0||11.8||35.6|| |
| College||40.2||42.2||73.6||47.6|| |
| Graduate education||50.4||9.9||14.6||16.7|| |
|Any health insurance coverage||73.9||32.1||60.5||44.0||< 0.0001|
|Eligible for STAR trial||67.2||10.9||10.2||19.9||< 0.0001|
Table 2 shows odds ratios (ORs) with 95% confidence intervals generated by 2 logistic regression models analyzing factors associated with eligibility for the STAR trial. Model 1 included the age-related Gail model variables, having an FDR, having had a biopsy, educational attainment, and insurance status. Model 2 also included race/ethnicity. All the Gail variables were statistically significant predictors of eligibility in both models. By far the strongest predictor was a family history of breast carcinoma. Women with an affected FDR were 64 times as likely in Model 1, and 114 times as likely in Model 2, as women without such a history to be eligible for the trial. In Model 2, the second strongest predictor was race/ethnicity. Controlling for the other factors, white women were > 45 times as likely as black women and Hispanic women were approximately 4 times as likely as black women to be eligible. The third strongest predictor in Model 2, and the second strongest predictor in Model 1, was biopsy history. For example, women who had received a biopsy were > 20 times as likely as other women to be eligible. Educational attainment also was associated strongly with eligibility—women who had attended graduate school were more than three times as likely to be eligible for the trial as other women. No statistically significant association of eligibility with having health insurance was observed in either model, but the ORs changed from 1.7 to 0.9 when race was taken into account.
Table 2. ORs for the Association of STAR Trial Eligibility with Gail Model and Socioeconomic Variables
|Age at menarche||0.84||0.71–0.99||0.83||0.69–1.01|
|Age at first live birth||1.08||1.03–1.13||1.07||1.02–1.12|
|Breast carcinoma in a FDR|| || || || |
|Biopsy history|| || || || |
|Race/ethnicity|| || || || |
| Black|| || ||1.00||Referent|
| White|| || ||45.50||13.54–152.91|
| Hispanic|| || ||4.10||1.44–11.69|
|Educational attainment|| || || || |
| Graduate education||1.00||Referent||1.00||Referent|
| High school or less||0.18||0.07–0.42||0.31||0.11–0.89|
|Insurance status|| || || || |
A logistic regression model stratified by race (data not shown) indicated that the race adjustment factor used in the Gail model served, as intended, to dampen the effects of the other Gail model variables on eligibility among black and Hispanic women.
Among the 823 black, white, and Hispanic women in our sample, race/ethnicity was statistically significantly associated with all the Gail model variables and was a much stronger determinant of eligibility for the STAR trial than we had anticipated.
As others have observed, minority women were more likely to be parous and were younger than white women at first birth.20 The lower prevalence of family history among minority women may reflect, in part, their relatively short life expectancy and that of their FDRs.14 Poor women, especially immigrants, may also be less likely than others to be aware of relatives' cancer diagnoses.
The association of race/ethnicity with biopsy history may be related to health care access. Although differences by race/ethnicity in mammographic screening history appear to be diminishing,21 several studies have found that white women with breast carcinoma are more likely than minorities to be diagnosed at an early stage15 and that white women are more likely to have timely follow-up of abnormalities found on mammograms.22, 23
In a logistic regression model that included race/ethnicity (Table 2), white women were 45 times as likely, and Hispanic women 4 times as likely, as black women to be eligible for the STAR trial. Even in the postmenopausal age group targeted for the trial, differences by race/ethnicity in breast carcinoma incidence rates cannot fully account for the strength of these associations.15 They certainly cannot account for the differences in eligibility between Hispanic and black women. In Surveillance, Epidemiology and End Results program data, breast carcinoma incidence rates are lower among Hispanic women than among black women.15
The difference between black and white women in age group-specific incidence is only approximately 25%, yet white women were 45 times as likely as black women to be eligible for the STAR trial. Race/ethnicity was a barrier, in part, because it was incorporated into the model specifically for that purpose. The investigators who added that variable to the model must have assumed that the other Gail model variables would not account for the difference in breast carcinoma incidence between white and minority women. However, our data tend to refute that assumption.
Historically, the health status of the black population has been significantly poorer than that of the general population.24, 25 Lack of insurance coverage and education helps to explain part of that disparity.26, 27
Although the racial/ethnic composition of our study population is different from that of the United States as a whole, we believe that the associations we observed between race/ethnicity and other factors in that population can be generalized. The white women in our sample may have been somewhat more affluent and educated than those recruited at other sites. However, among the 500 STAR sites, our site was 1 of the top 10 in minority recruitment, and we have no reason to believe that our minority study participants were different from those recruited elsewhere.
Our data do not include women who refused to complete an RAF or who were not approached. Such women might have differed from our subjects with respect to the factors we studied, but we have no reason to believe so or to believe that such women would be more likely to volunteer for future breast carcinoma prevention trials than they were for the STAR trial.
Our analysis supports the misgivings of the investigators who developed the Gail model regarding its use in a population whose screening history is unknown or spotty. Our findings suggest that the model may have limited applicability as a screening tool for underserved women in chemoprevention trials. Unfortunately, our data do not suggest a better way to assess their risk, but the relatively high breast carcinoma mortality/incidence ratios observed among poor and minority women argue for a systematic effort to obtain such data. Certainly our data suggest that the Gail model adjustment for race/ethnicity is unnecessary.
Even without adjustment for race, the Gail model would probably underestimate risk in minority women because they differ from white women with respect to the other Gail model variables.28 A large cohort study of minority women is needed to generate data on the basis of which a new risk model can be developed. Such a study will be difficult to conduct, but the effort is necessary if we wish to develop prevention strategies that help to reduce racial disparities in breast carcinoma outcomes. Until the results of such a study are available, in addition to eliminating adjustment for race/ethnicity from the Gail model, investigators could consider the use of relative rather than absolute risk in determining eligibility. At least until we understand the factors that affect breast carcinoma risk in underserved populations better than we do now, we need to lower the barriers to study participation associated with race/ethnicity and socioeconomic status.4
The authors gratefully acknowledge Rossy Sandoval, whose dedicated efforts in recruiting minority women to complete risk assessment forms for the Study of Tamoxifen and Raloxifene trial played an important role in making the current analysis possible.