We invited adult women diagnosed with a first invasive, stage I-III breast cancer during 2013–2017 to participate in a one-time interviewer-administered survey. We aimed to include a diverse population of White, Black, and Hispanic participants. Participants had to understand and speak English or Spanish and had to receive some or all of their cancer care (at least 3 visits) at a participating center (Dana-Farber Cancer Institute [DFCI] or Boston Medical Center [BMC] in Boston, MA or Columbia University Irving Medical Center [CUIMC] in New York, NY). The Institutional Review Board of each participating center approved the study (protocol number 17–612) and the study conforms to the standards set forth in the Declaration of Helsinki.

The survey (Data Supplement, File 1) was designed to be completed in 25–40 min. It included eight sections: (1) general health[28]; (2) breast cancer knowledge [17], perception of recurrence risk[29], genetic testing utilization[30], decision-making styles[31, 32]; (3) treatments recommended/received, reasons for initiating/not initiating treatments[33,34,35,36], adherence[36, 37], knowledge about treatment rationales[35]; (4) social support[38], religiosity[39, 40]; (5) information-seeking preferences[41,42,43], trust [42], health care experiences[35]; (6) discrimination [44, 45], health literacy[46], numeracy[47]; (7) demographics (age, race, ethnicity, language, country of origin, highest educational level, marital status, finances [48, 49], primary health insurance, employment status, impact on work/finances)[48, 49]; and (8) mood[50] and social activity [51]. Whenever possible, we used or adapted previously validated survey items. Surveys underwent cognitive testing with 10 patients to ensure comprehension and clear mapping of answers to response options. The results presented herein focus on data collected in Sects. (2), (5), and (6).

Figure 1 displays the study schema. For patients treated at Boston centers (Fig. 1A), we utilized electronic medical record reports to identify patients meeting eligibility criteria who were Black, Hispanic, or White, oversampling for Black and Hispanic patients. Study invitations and postage-paid opt-out cards were mailed to eligible patients in English and Spanish. At CUIMC (Fig. 1B), study staff identified and contacted eligible women who previously agreed to contact for research.

Schema for patient approach at Boston centers (Dana-Farber Cancer Institute and Boston Medical Center) (Fig. 1A) and Columbia University Irving Medical Center (Fig. 1B)

Interviews were conducted (via telephone or in-person) by trained staff. Patients provided verbal, informed consent at the start of the interview, including permission to review medical records. The survey was conducted in English or Spanish using computer-assisted interview software. Data from the survey were entered directly into the survey software. All participants were provided a $20 gift card upon interview completion. We abstracted participants’ medical records to confirm tumor characteristics and treatment, when available.

Interviews were conducted between June 1, 2018, and June 30, 2020; 96% of participants were diagnosed with breast cancer > 2 years before their interview; all were within 5 years. Accrual slowed during the COVID-19 pandemic, where in-person contact/consenting was not permitted; we halted study accrual once 326 women enrolled (187 from DFCI, 100 from BMC, 39 from CUIMC; 5 were interviewed in Spanish), fewer than the planned enrollment of 500. Because analyses focused on breast cancer knowledge (i.e., agreement of medical record and patient report for tumor characteristics), 13 women were excluded (2 Black, 1 Native American, 1 White, 9 missing race) for incomplete medical record information.

Given the different contact procedures at each site, we calculated response, cooperation, refusal, and contact rates for each center separately using American Association for Public Opinion Research (AAPOR) definitions[52] (Supplemental Fig. 1A-1C). Response rates varied by site: DFCI = 30.3% (cooperation rate 58.4%, refusal rate 21.6%, contact rate 51.9%); BMC = 38.5% (cooperation rate 63.3%, refusal rate 22.3%, contact rate 60.8%). CUIMC response rates were not easily measurable given the uncertain denominator of patients receiving/reading broadcast emails (CUIMC cooperation rate = 76.5%, refusal rate = 2.3%; Fig. 2C).

Proportion of participants answering tumor variable questions correctly by race and ethnicity

Findings were significant for each comparison (stage, p = 0.003; grade, p = 0.004; hormone receptor [HR] status, p = 0.003; human epidermal growth factor receptor 2 [HER2] status, p = 0.005)

Dependent variables of interest were derived from the questions in Supplemental Table 1.[17, 35] The survey (English version) is provided in the Data Supplement [17, 35]. Patients were asked about knowledge of their breast cancer tumor characteristics, including stage, grade, hormone receptor (HR) status, and human epidermal growth factor receptor 2 (HER2) status; hereafter called “breast cancer knowledge.” Participants were considered knowledgeable about each characteristic if their response matched the medical record documentation. Two participants with unconfirmed stage in medical records were categorized as ‘correct’ for stage, regardless of their response. We examined each item and the sum of the number of correct answers (range 0–4).

Patients were asked about understanding of treatment rationale(s) for each recommended therapy in the neo/adjuvant setting by adapting prior surveys [35]. Each question on treatment rationale and preferences was examined and reported individually. Participants were only asked treatment rationale questions about recommended treatments, defined by participant report of whether treatments were recommended. Patients also reported whether any recommended treatments were initiated and completed.

We additionally inquired about information-seeking preferences, whether understanding treatment rationales was important in decision-making, and the degree to which participants felt information was provided to them by their clinicians.

Patients self-reported their race and ethnicity, categorized as Non-Hispanic (NH) White, NH Black (or African American), Hispanic, or Other/Unknown (race and ethnicity data were missing for 17 patients who were excluded from analyses). We additionally documented age, general health, country of birth, education, treatments recommended/received, preferred decision-making role, and insurance, categorized as in Table 1. Health literacy and numeracy were each categorized based on responses to three questions, and we summarized the proportion of participants by race and ethnicity reporting the most health literacy[46] and numeracy for each item [47].

We examined participant characteristics and individual breast cancer knowledge and the responses to questions on treatment rationales for recommended therapies by race and ethnicity. We examined associations of breast cancer knowledge and understanding treatment rationales with initiating recommended treatment(s) using contingency tables and Chi-square tests among those recommended for each treatment modality. Because missing/non-responses were infrequent for each knowledge question (< 8% for each item), we excluded patients who did not provide a response for that particular question only. Finally, we examined associations of knowledge with each of the three health literacy[46] and numeracy questions[47] using contingency tables and chi-square tests. Sample sizes for each analysis are described below. Since treatment recommendation and receipt were self-reported, we examined agreement between self-report and medical records among participants with medical record treatment information; agreement was 95–97% across treatments. Two-sided P values < 0.05 were considered statistically significant.