Strategies to Identify and Recruit Women at High Risk for Breast Cancer to a Randomized Controlled Trial of Web-based Decision Support Tools

Chemoprevention with antiestrogens, including selective estrogen receptor modulators (SERM) and aromatase inhibitors (AI), has been shown in randomized controlled trials to reduce the incidence of invasive breast cancer by up to 50% to 65% among women at high risk for developing breast cancer (1–6). Multiple national medical organizations, including the U.S. Preventive Services Task Force and the American Society of Clinical Oncology, recommend that physicians discuss antiestrogen chemoprevention with high-risk women (7, 8). However, uptake of breast cancer chemoprevention among high-risk women is estimated to be less than 15% (9–11). Use of chemoprevention is particularly low among racial/ethnic minorities, which can lead to increased health disparities in these populations (12). Barriers to chemoprevention uptake include lack of routine breast cancer risk assessment, insufficient clinician and patient knowledge about chemoprevention, and concerns about side effects (10, 13, 14). There is therefore a clear need to identify strategies to improve chemoprevention uptake among women at high risk for breast cancer.

Integration of breast cancer risk assessment into the clinical workflow and the development of innovative approaches to communicate breast cancer risk could improve chemoprevention uptake among high-risk women. In particular, decision support tools for patients and providers could facilitate breast cancer risk assessment and shared decision-making regarding breast cancer prevention. We have developed web-based decision support tools, RealRisks and Breast cancer risk NAVigation (BNAV) for patients and healthcare providers, respectively, that include interactive educational modules on breast cancer risk assessment, genetic testing, screening, chemoprevention, and other prevention strategies (15, 16). We evaluated these tools in a randomized controlled trial (RCT) among a racially/ethnically diverse group of women, with the hypothesis that RealRisks and BNAV would improve accuracy of breast cancer risk perception, facilitate referrals to specialty clinics, and increase chemoprevention uptake among high-risk women (17).

In addition to evaluating these outcomes, our RCT evaluating these decision support tools provided a unique opportunity to examine recruitment strategies that can identify a large, racially/ethnically diverse population of women at high risk for breast cancer. Such strategies could also be utilized for other clinical studies targeting high-risk women to improve recruitment yield, including among racial/ethnic minorities. We evaluated five strategies to identify women at high risk for breast cancer for recruitment to our trial, and compared recruitment yield and patient demographics of each strategy.

We enrolled a total of 300 high-risk women and 50 healthcare providers to an RCT of chemoprevention decision support from November 2016 to March 2020 at Columbia University Irving Medical Center (CUIMC) in New York, NY (17). Women were eligible for enrollment if: (i) age 35 to 75 years; (ii) at high risk for breast cancer either because they had a history of lobular carcinoma in situ (LCIS) or had an estimated 5-year invasive breast cancer risk ≥1.67% according to the Gail (18) or Breast Cancer Surveillance Consortium (BCSC; ref. 18) models and/or a lifetime risk ≥20% according to the Gail model; (iii) no prior history of breast cancer or ductal carcinoma in situ (DCIS); (iv) no prior use of a SERM or AI for breast cancer chemoprevention; (v) no prior bilateral mastectomies; (vi) English or Spanish speaking; (vii) had a healthcare provider at CUIMC; (viii) had access to the internet; (ix) had access to text messaging and/or e-mail; and (x) able to provide informed consent. Potential participants were first screened to determine eligibility, and then underwent informed consent in person, over the phone, or online in either English or Spanish. Providers were identified and approached for recruitment after one of their patients was identified as being potentially eligible for the study. Enrolled healthcare providers were primary care providers (internists, family medicine providers, gynecologists), as well as specialists (breast surgeons, medical oncologists, nurse practitioners) who practiced at CUIMC. The study was approved by the Institutional Review Board (IRB) at CUIMC. The study team obtained written informed consent from patient prior to enrollment, and the studies were conducted in accordance with recognized ethical guidelines (i.e., the Declaration of Helsinki).

Patients completed a baseline survey including questions on perceived breast cancer risk, chemoprevention knowledge, and breast cancer worry, and then were randomized 1:1 to either receive access to the web-based decision support tool RealRisks in addition to standard educational materials related to breast cancer risk and chemoprevention (intervention arm), or receive access to standard educational materials alone (control arm). If an enrolled provider's patient was randomized to the intervention arm, he or she was given access to the BNAV tool. Patients were administered surveys at 1 month, 6 months, and after the next clinical encounter with their healthcare providers at CUIMC.

Women at high risk for breast cancer were identified and recruited using five recruitment strategies, as described below (Fig. 1). Strategies 1 to 3 aimed to identify patients using existing databases and the electronic health record (EHR), while strategy 4 identified patients through clinical encounters with enrolled healthcare providers, and strategy 5 relied on patient self-referral. Women who were identified through multiple strategies were characterized using a hierarchy: Referrals by enrolled healthcare providers (strategy 4) > self-referrals (strategy 5) > recruitment from the Know Your Risk: Assessment at Screening (KYRAS) study (strategy 1) > diagnosis of atypical hyperplasia/LCIS (strategy 3) > breast cancer risk assessment from the EHR (strategy 2). Women were therefore only considered as having been identified using one category during analyses.

We previously enrolled women to the KYRAS study, an observational study at CUIMC that evaluated self-reported breast cancer risk factors among women at the time of screening mammography (19). Women were approached for enrollment to the KYRAS study by a member of the study team during their routine screening mammography at the Avon Breast Imaging Center at CUIMC, and participants completed a one-time comprehensive survey on demographic characteristics and breast cancer risk factors that was used to calculate 5-year and lifetime breast cancer risk according to the Gail model. A subset of patients also agreed to be recontacted for future studies. Among the 2,019 women enrolled in the KYRAS study between November 2014 and December 2015, 76.8% identified as Hispanic and 17.7% met high-risk criteria based upon an estimated 5-year invasive breast cancer risk ≥1.67% according to the Gail model (19). For the chemoprevention decision support RCT, the study team specifically assessed eligibility among high-risk women enrolled in the KYRAS study who had agreed to be recontacted, and then contacted them via e-mail, mailed letter, or telephone.

Information related to established breast cancer risk factors, including age, race/ethnicity, first-degree family history of breast cancer, benign breast biopsies, and mammographic density, was extracted from the EHR for women undergoing screening mammography at CUIMC, as previously described (20). Estimated 5-year risks of invasive breast cancer were calculated using the BCSC model. Among 9,514 women, 1,443 (15.2%) met high-risk criteria based upon a 5-year invasive breast cancer risk ≥1.67% according to the BCSC model. Of note, this automated EHR-based strategy might have underestimated risk in some patients through inaccurate or missing information. Approximately 88% of women had unknown family history of breast cancer and 95% did not have breast pathology reports in the CUIMC EHR system, which was classified as none or unknown for breast cancer risk calculation. After contact was made with potential participants, we clarified and collected any missing breast cancer risk information by self-report to confirm eligibility for the trial.

Women with a diagnosis of atypical hyperplasia or LCIS between 2007 and 2015 were identified from the International Classification of Diseases (ICD)-9/10 codes in the EHR (610.9/N60.99 and 233.0/D05/0.90, respectively), including outpatient medical records at CUIMC and the New York Presbyterian Hospital (New York, NY; NYPH) tumor registry (21). Manual review of the EHR was done in patients with LCIS to confirm the diagnosis, as LCIS and DCIS share the same ICD-9/10 codes. For women with more than one diagnosis, the most advanced lesion was used, with LCIS considered more advanced than atypical hyperplasia. Breast cancer risk was calculated according to the BCSC model using patient information collected from aggregate data extraction from the EHR. Similar to strategy 2, once the study team made contact, we collected additional breast cancer risk information by self-report.

Healthcare providers enrolled in the RCT consented to allow research staff to approach their high-risk patients for potential enrollment. Any high-risk women identified through the three recruitment strategies above were merged with clinic schedule data for enrolled healthcare providers, including primary care providers as well as specialists providing high-risk consultations (breast surgeons and medical oncologists). An e-mail was sent to enrolled providers by the study team informing them that a high-risk patient was scheduled to see them in clinic the following week and requesting permission to contact the patient. Women were contacted to determine their eligibility by the study team prior to their clinic visit. Providers then had an opportunity to discuss enrollment in the study during the clinic visit. Enrolled healthcare providers could also refer patients who met high-risk criteria for breast cancer directly to the study team, without prior identification by the study team. Tying recruitment efforts to the clinical encounter therefore allowed providers to encourage participation of their high-risk patients.

Recruitment flyers providing information about the study and contact information for the study team were distributed throughout the medical campus community along with online distribution via the RecruitMe platform (https://recruit.cumc.columbia.edu/), an online recruitment tool at CUIMC that provides information about current clinical trials. Flyers contained information including eligibility criteria, study intervention, evaluations, and compensation and were posted throughout the CUIMC campus before recruitment started. Women who responded were contacted by phone and prescreened to determine their eligibility, with calculation of breast cancer risk according to the Gail model using patient-reported risk factor data collected during screening. Women without a medical provider at CUIMC were excluded from the study.

For women identified as potentially eligible for the trial, initial contact was made by e-mail (if available) or mailed letter that included brief information about the RCT and contact information for the study team. Women with available e-mail addresses were prioritized for contact, given that we hoped to enrich for younger women at high risk for breast cancer with access to the internet. This initial contact attempt was followed by up to two follow-up phone calls using phone numbers either provided by patients (if enrolled in KYRAS) or listed in the EHR. Of note, our study team included two coordinators bilingual in English and Spanish, allowing for discussion and consent in a patient's preferred language. Patients who enrolled in the study and completed the study evaluations received up to $95 as an incentive for participation.

We conducted descriptive statistics to compare sociodemographic characteristics across the five recruitment sources, limiting our sample to women who had received at least one contact from the study team in the form of e-mail, letter, and/or phone call. The frequency distribution of age, race/ethnicity, and mean 5-year invasive breast cancer risk were compared across the five recruitment sources using χ² tests and two-sample t tests for categorical and continuous variables, respectively. We also performed descriptive analyses to compare the distribution of age, race/ethnicity, and mean breast cancer risk scores among contacted high-risk women who enrolled in the RCT versus those who did not enroll. Statistical analysis was conducted using SAS version 9.4 (SAS Institute).

The data generated in this study are available upon request from the corresponding author.

We identified a total of 6,229 high-risk women, of whom 3,663 were contacted via e-mail, mailed letter, and/or telephone for participation (Fig. 2). Of those contacted, 204 women did not meet eligibility criteria on preliminary screening: 127 were not between 35 and 75 years old, 38 did not have a medical provider at CUIMC, 25 had previously undergone bilateral mastectomy or breast augmentation (which did not allow breast cancer risk calculation according to the BCSC model), and 14 had a history of breast cancer. Of the 3,459 high-risk women (94.4%) who were contacted and eligible for enrollment, 16.3% were identified through prior enrollment in the KYRAS study, 37.8% through breast cancer risk data from the EHR, 15.0% through a diagnosis of atypical hyperplasia/LCIS identified in the EHR, 29.3% through clinical visits with enrolled healthcare providers, and 1.6% through the recruitment flyers distributed in the community or through online recruitment. All of these 3,459 were contacted by e-mail or letter, while approximately 19% had one phone contact attempt and 43% had two phone contact attempts from the study team. Ultimately, 300 women (8.7% of eligible women contacted) were enrolled in the study and underwent randomization.

The baseline characteristics of eligible high-risk women who were contacted (n = 3,459), stratified by recruitment source, are shown in Table 1. The mean age of eligible women was 58.7 years (SD, 9.7), and 47% of women were non-Hispanic White, 15% non-Hispanic Black, 29% Hispanic, 5% Asian, and 20% other/unknown. Age was significantly different among women based upon recruitment source (P < 0.001); for example, the mean age of women recruited through recruitment flyers and online recruitment was 47.4 years (SD, 8.8) compared with 63.9 years (SD, 7.6) for women recruited from the KYRAS study. Race/ethnicity also varied significantly by recruitment source (P < 0.001); for example, the majority (54.9%) of women recruited from the KYRAS study were Hispanic, while the majority of women identified through breast cancer risk information in the EHR or diagnoses of atypical hyperplasia/LCIS were non-Hispanic White. There were also significant differences in the mean 5-year risk of invasive breast cancer among women based upon recruitment source, with highest mean 5-year risk among those women with AH/LCIS (3.51%; SD, 2.75%).

Of the 300 women enrolled in the study, the greatest proportion of participants (44.7%) was identified through clinical encounters with enrolled healthcare providers, while 27.2% were recruited via the KYRAS study, 11.7% through breast cancer risk information in the EHR, 10.3% through diagnoses of atypical hyperplasia/LCIS, and 6.0% through flyers and online recruitment. However, flyers and online recruitment had the highest enrollment yield, with 33.3% of women contacted through that method enrolling in the study. Recruitment yields were 14.5% for recruitment from the KYRAS study, 13.2% for clinical encounters with enrolled healthcare providers, 6.0% for diagnoses of atypical hyperplasia/LCIS, and 2.7% for women identified through breast cancer risk information from the EHR.

We also compared characteristics among eligible high-risk women who were enrolled and not enrolled (Table 2). Enrolled patients were younger than those not enrolled, with a mean age of 57.3 years (SD, 10.1) versus 58.9 years (SD, 9.7), respectively. Enrolled patients were also more likely to be non-Hispanic White than not enrolled patients (58.5% vs. 45.4%, respectively) and had a higher mean 5-year invasive breast cancer risk (3.02% vs. 2.31%, respectively). Overall, we were able to enroll a racially/ethnically diverse high-risk population, including 13.7% non-Hispanic Black, 22.4% Hispanic, and 4.7% Asian.

We were able to identify and recruit a large cohort of women at high risk for breast cancer using five strategies, which included recruitment of patients previously enrolled in a clinical study evaluating breast cancer risk, identification of high-risk women using clinical information extracted from the EHR, identification of women with atypical hyperplasia or LCIS through diagnostic codes, clinical encounters with enrolled healthcare providers, and patient self-referral through recruitment flyers. Through these strategies, we enrolled a racially/ethnically diverse cohort of women to our study evaluating decision support tools for women at high risk for breast cancer, with over 40% of enrolled women from underrepresented minority groups. While most eligible patients were identified through data in the EHR, the majority of enrolled patients were recruited through clinical encounters with enrolled providers or from a cohort of high-risk women previously enrolled in a clinical study. Although the highest recruitment yield came from flyers and online recruitment, this strategy only accounted for a small fraction of the women screened and enrolled. Recruited patients had higher mean 5-year breast cancer risk by the BCSC model than patients who were not enrolled, and patients with atypical hyperplasia or LCIS had the highest breast cancer risk, consistent with the known at least four-fold increased risk of breast cancer associated with these breast lesions (22) and the inclusion of atypical hyperplasia and LCIS in the BCSC model (18).

Overall, we contacted over 3,000 high-risk women in order to enroll 300 participants to our randomized controlled trial of web-based decision support tools. Therefore, we needed to screen a large high-risk population to meet our accrual goal. While we were able to meet our recruitment goals, the overall recruitment processes varied with level of efficiency, and could present a challenge to trials with limited resources and time for accrual. A yield of 10% study participation is consistent with low accrual to prevention trials when compared with accrual to trials for patients with active disease, such as diabetes (23), and highlights the ongoing need to identify efficient and effective recruitment methods for trials among healthy, high-risk populations.

Our finding that the majority of enrolled patients were recruited through direct discussion with and referrals from their healthcare providers is in agreement with previous studies demonstrating that provider recommendations increased the likelihoods both of chemoprevention uptake and participation in chemoprevention studies (11, 24, 25). For example, in one survey of women at high risk for breast cancer, those who were advised to enroll in a chemoprevention trial by their primary physicians were 13 times more likely to participate (24). In a systematic review of 63 studies (with 1,681 adult patients) that evaluated barriers and facilitators to clinical trial participation for diseases including cancer and chronic medical conditions, major themes of patient-reported barriers to clinical trial participation included skepticism, fear, and mistrust as well as a lack of awareness of opportunities (26). However, major facilitators described by patients included trust in medical staff and information from their providers about clinical trials, as well as a sense of altruism and sense of belonging and connectedness. While automated methods can identify a large number of high-risk women, strategies involving direct communication with patients either by their providers or study team remained important in our study to achieve enrollment targets and should be utilized in future studies among chemoprevention and other studies to enhance recruitment yield. In addition, the high yield among patients who self-referred through flyers and online recruitment might reflect patient motivation to participate. While this study did not capture patient or provider perspectives on enrollment, this is a potential area of study in future studies in order to better evaluate barriers and facilitators to enrollment among a population of high-risk women.

We were able to recruit a racially/ethnically diverse cohort of high-risk women, which is representative of the predominantly Hispanic population of women undergoing screening mammography at our medical center (19) but also underscores the need to utilize a variety of recruitment strategies for more inclusive clinical trial enrollment. The National Surgical Adjuvant Breast and Bowel Project (NSABP) Study of Tamoxifen and Raloxifene (STAR) trial, in which high-risk women were randomized to tamoxifen or raloxifene, similarly utilized multiple strategies to improve minority accrual, including community outreach programs, presentations at national medical organization meetings, and workshops addressing the importance of minority accrual (27). These efforts resulted in enrollment of a more diverse cohort of women than previous breast cancer chemoprevention trials, with 21% of participants identifying as racial/ethnic minorities compared with less than 4% in the NSABP Breast Cancer Prevention Trial (28). For our study, having a diverse research team, including research staff who were bilingual Spanish/English speakers was useful in addressing language barriers, building trust, and recruiting minority women. Previous studies have demonstrated that a shared facilitator of clinical trial participation among racial/ethnic minorities is the presence of culturally-matched research personnel, study materials in patients’ languages, as well as discussion with providers they knew and trusted (29). Increasing representation of racial/ethnic minorities in clinical trials investigating breast cancer prevention strategies is critical to evaluate their efficacy in these populations with the goal of addressing health disparities. This underscores the importance of avoiding a “one size fits all” approach to patient recruitment, with the goal of facilitating inclusive clinical trial enrollment.

We also were able to identify a large group of women at high-risk for breast cancer using automated extraction of EHR data for risk calculation. While recruitment yield was low for this source, automated breast cancer risk calculation could help to identify women at high risk for breast cancer by overcoming barriers to routine breast cancer risk calculation including time constraints during clinical encounters and insufficient knowledge of breast cancer risk among primary care providers (14). We previously demonstrated that there is moderate agreement between breast cancer risk calculated using EHR-derived information and patient self-report, and also that EHR might identify more women at high risk for breast cancer because of pathologic risk factors including atypical hyperplasia/LCIS compared with patient self-report (20). However, given the potential for missing or incomplete information in the EHR for risk factors including race/ethnicity, family history of breast cancer and prior breast biopsy, breast cancer risk estimates using EHR-derived data could serve as an initial screening tool to identify high-risk patients that can guide more comprehensive breast cancer risk assessment during clinical encounters. We are currently evaluating the use of automatically-populated breast cancer risk data from the EHR in our RealRisks decision aid (30), including among high-risk women with atypical hyperplasia or LCIS enrolled in an ongoing multicenter cluster RCT, “Making Informed Choices on Incorporating Chemoprevention into Care” [MiCHOICE; Southwest Oncology Group (SWOG) 1904; NCT04496739].

Strengths of our study include our large cohort of eligible women, racially/ethnically diverse patient population, evaluation of multiple recruitment strategies, and use of EHR data to efficiently identify eligible high-risk women for enrollment. Limitations include that it was a single-center study at an academic medical center, which might limit the generalizability of our findings. We had incomplete information on reasons for refusal to enroll in our study, which could have informed future efforts to reduce barriers to enrollment. We also did not collect additional patient information for all patients identified through our recruitment strategies, including information such as health insurance, employment status/income, and distance from healthcare facilities, limiting our ability to evaluate patient-level factors beyond age, race/ethnicity, and breast cancer risk that predicted enrollment. In addition, comparison of breast cancer risk of women across the five recruitment strategies might be limited by incomplete and missing information for breast cancer risk factors including family history of breast cancer and prior breast biopsy, particularly for EHR-based identification strategies. This missing data could have led to underestimation of breast cancer risk in potentially eligible patients, and therefore our strategies relying on automated EHR-based methods might have failed to identify potential participants, including racial/ethnic minorities. This could limit the application of these strategies as the primary source of potential participants in larger scale trials, without the use of additional strategies relying on patient-reported breast cancer risk factors.

In conclusion, we were able to identify and successfully recruit a large cohort of racially/ethnically diverse high-risk women from multiple recruitment sources to a randomized controlled trial evaluating the web-based decision support tools, RealRisks and BNAV. We are currently conducting a larger multicenter cluster randomized controlled trial, MiCHOICE (SWOG S1904) evaluating the effect of RealRisks and BNAV on chemoprevention informed choice among women with high-risk breast lesions including atypical hyperplasia or LCIS (NCT04496739). The recruitment strategies described here could be useful for future breast cancer chemoprevention trials to allow for more efficient identification of high-risk women and improve enrollment yields.

No discloures were reported.

J.E. McGuinness: Methodology, writing–original draft, writing–review and editing. G. Bhatkhande: Data curation, formal analysis, writing–original draft, writing–review and editing. J. Amenta: Data curation, investigation, methodology, project administration, writing–review and editing. T. Silverman: Resources, data curation, investigation, methodology, project administration, writing–review and editing. J. Mata: Resources, data curation, methodology, project administration, writing–review and editing. A. Guzman: Resources, data curation, investigation, project administration, writing–review and editing. T. He: Resources, data curation, methodology, writing–review and editing. J. Dimond: Resources, software, investigation, methodology, writing–review and editing. T. Jones: Investigation, methodology, writing–review and editing. R. Kukafka: Conceptualization, resources, supervision, funding acquisition, investigation, methodology, writing–review and editing. K.D. Crew: Conceptualization, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, writing–review and editing.

The study was supported by the NIH, NCI NIH R01 CA226060–01A1 (to K.D. Crew and R. Kukafka), R01 CA177995–04S1 (to K.D. Crew and R. Kukafka), R38 CA231577 (to K.D. Crew), P30 CA013696 P30 (to K.D. Crew), UL1TR001873 (to K.D. Crew). The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.