Patient and Physician Perspectives on Using Risk Prediction to Support Breast Cancer Surveillance Decision Making.

Introduction
Professional guidelines1–5 recommend that breast cancer survivors undergo annual mammography (‘surveillance imaging’) for early detection of recurrent or new cancers to reduce morbidity and mortality.6–10 Breast cancer survivors have up to a four-fold higher risk of interval second breast cancers relative to the general screening population.11 Interval cancers are those that are diagnosed within 12 months following a negative mammogram, and are associated with worse prognosis than those detected by mammography. While the screening mammography benchmark for interval cancer rates in cancer naïve populations is 0.8 per 1,000 tests, the interval cancer rate among women with a personal history of treated breast cancer is 2.1 – 3.6 per 1,000 surveillance mammograms.11,12 Yet the risk of interval second breast cancer is distributed unevenly across the survivor population: hormonal receptor status, stage at diagnosis, time since diagnosis as well as non-clinical factors (age, breast density) and social determinants of health are associated with interval cancers.13–15
While annual mammography is standard for breast cancer survivors, guidelines also address the potential benefit of adding magnetic resonance imaging (MRI) or ultrasound for women at high risk (herein referred to as supplemental surveillance), particularly in terms of cancer detection rates.16 Given the variation in interval cancer risk and known risk factors for being diagnosed with an interval second breast cancer, there are opportunities to leverage risk prediction tools to identify those at highest risk who might benefit from supplemental screening with the goal of improving surveillance outcomes. While risk prediction models are currently used in clinical practice to select women for supplemental screening in cancer naïve populations at elevated risk,17,18 no models are in current clinical use for guiding selection of women for supplemental surveillance imaging.
Little is known about the acceptability of a risk-based approach to surveillance imaging recommendations, although data suggest that breast cancer survivors desire more support to understand and choose individualized surveillance plans with confidence.19 What we know about using risk prediction and shared decision-making to guide supplemental screening comes from the cancer-naïve population, where tools have remained under-utilized.20,21 This has been in part due to a lack of comfort in risk assessment tool use and interpretation by physicians, little integration with electronic medical records, time constraints impeding their use in clinical encounters, and differential acceptability and uptake across patient populations and clinical specialities.22–24 For risk prediction tools to support shared decision making, they must be accepted and trusted by patients and physicians in addition to being easily accessible and accurate.21,25–28 The use of a risk prediction tool could support shared decision-making in tailoring supplemental surveillance imaging, incorporating cancer history, risk, and personal preferences related to the benefits and harms of various surveillance intervals and modalities.29 But how patients and physicians perceive the integration of risk prediction tools for cancer surveillance and their relationship to shared decision-making is an underdeveloped area of research.
The Breast Cancer Surveillance Consortium (BCSC) has developed a preliminary risk model to predict cumulative absolute 5-year risk of an interval second breast cancer diagnosis, combining risk factors for second breast cancers with parameters influencing cancer detection by mammography.13,15 To facilitate clinical implementation30 of such a model for supporting risk-based surveillance decision making in breast cancer survivors, we sought to identify facilitators and barriers to risk-based surveillance from both patient and physician perspectives.

Methods

Study Design
This qualitative study is part of a larger project by the Breast Cancer Surveillance Consortium (BCSC) aimed at understanding and overcoming barriers to the use of risk-based imaging in breast cancer surveillance. The BCSC is a collaborative network of six breast imaging registries and a statistical coordinating center that seek to assess and improve the equitable delivery and quality of breast cancer screening and surveillance in the United States.31 The study was reviewed and approved by the Institutional Review Boards of Dartmouth College (STUDY00032613) and participating registries and deemed to meet criteria for federal exemption.
To understand acceptability across the varying clinical contexts in which risk-based surveillance tools might be deployed, participants were recruited from five BCSC registries located across the United States and included physicians from four specialties: radiology, medical oncology, surgical oncology and primary care. The participating registries were the New Hampshire Mammography Network, the Vermont Breast Cancer Surveillance System, the Carolina Mammography Registry, the Northwest Screening and Cancer Outcomes Research Enterprise, and the Kaiser Permanente Washington Registry. We conducted focus groups in November 2023 to gather data from patients to maximize the expression of different points of view in a supportive environment.32 We conducted in-depth semi-structured interviews with physicians because it granted confidentiality and the exploration of different approaches to the use of tools across specialties and geographic regions. These telephone or video-based interviews were conducted from January to April 2024.

Participant recruitment and eligibility
We used purposeful sampling to recruit participants.33 Eligible patients were adults who: (1) had a history of breast cancer (American Joint Committee on Cancer stage zero non-invasive breast cancer (DCIS) or stage 1–3 invasive cancer34), (2) completed treatment for breast cancer with the exception of adjuvant endocrine, HER2 therapies, and pembrolizumab, which are long-term regimens to prevent cancer recurrence, (3) had mammography within the last four years at a participating BCSC registry, (4) could participate in a virtual focus group by video or phone, and (5) communicated in English. Participating BCSC registry sites mailed recruitment letters with the study’s objectives, eligibility criteria, and procedures to eligible patients in their registry. Interested participants were directed to complete a screening questionnaire independently online or with assistance from the study team over the phone. The questionnaire included sociodemographic questions where participants self-reported cancer history, race, ethnicity, age, and education. Once screening was complete, eligible participants were scheduled for one of four focus groups. All elements of consent were reviewed with participants, although written documentation of informed consent was waived due to the minimal risk nature of this research.
Physicians were recruited through the five BCSC registries above and the San Francisco Mammography Registry. Eligibility criteria required that they work: (1) directly with women who have had breast cancer, and (2) in any of the following specialties— primary care, medical oncology, surgical oncology, and radiology. The registries emailed physicians using a recruitment letter outlining the study objectives, eligibility criteria, a link to the online screening questionnaire, and a research information sheet. Interested physicians either completed the online screening questions or directly emailed the study team for eligibility verification and scheduling. Recruitment continued at each site until 5–8 physicians representing every specialty was achieved. All patient and physician participants received a $100 gift card as a thank you for their time.

Data collection

Patient focus groups
We conducted four 75-minute, semi-structured focus groups via video conferencing, each with 5–7 patients. The study aimed to recruit patients representing diversity in race, ethnicity, and education status by including an overrepresentation of women meeting diverse demographic characteristics in the initial mailing. We assigned participants to groups to maximize diversity across these characteristics within each focus group. The focus group guide was co-developed by the project team consisting of physicians and researchers with backgrounds in oncology, radiology, primary care, decision-making, and qualitative methods. Questions covered patient feedback on their breast cancer surveillance experiences, barriers and facilitators to their engagement in surveillance decision-making, and their perspectives on a risk prediction tool designed to support decision-making for breast cancer surveillance. Open-ended questions were asked followed by probes to elicit more detailed responses. Because the user interface was not yet fully developed, focus group participants were provided with a description of the proposed risk prediction tool (see Box 1). The focus group guide including the tool description was piloted one-on-one with four breast cancer survivors who informed modifications in the interest of clarity, relevance, and question flow. Two study team members (NB, SS) were present for all focus groups, with one leading the facilitation and the other taking notes and providing technical support. One interviewer has doctoral-level training as a social worker, while the other is doctoral candidate. Both identify as women and have extensive experience in qualitative research methods across a variety of topical areas.

Physician interviews
We conducted semi-structured interviews designed to last between 30 and 45 minutes with physicians over video conference or via phone. The interview guide was co-developed by the project team and covered physicians’ experiences with making recommendations for breast cancer surveillance, and their acceptance of a risk prediction tool for surveillance decision-making. Like the focus groups, the description of the risk prediction tool was read mostly verbatim during the interviews (Box 1). The guide and tool description were piloted with a medical oncologist who has worked with breast cancer patients for over 20 years. The interviews were conducted by (NB, SS). Focus groups and interviews were audio recorded and professionally transcribed verbatim.

Data analysis
All transcripts were entered into a qualitative data analysis program (Dedoose, SocioCultural Research Consultants, LLC; version 9.2.012). The codebook was developed based on the principles of theoretical thematic analysis,35,36 driven by the study team’s interest in the use of risk prediction tools to support shared decision-making. Coding were guided by a priori research questions and objectives, with emergent topics added as needed. A subset of the same transcripts was separately coded by two coders (NB, SS) who also conducted focus groups and interviews. They met with a third study team member (CG), who is a trained health services researcher with expertise in shared decision-making and cancer screening, to review the application of codes and their definitions. The two coders independently coded the subsequent transcripts with a finalized codebook, moving towards more abstract levels of coding and deriving themes from the data using thematic memos.32 We cross-verified the focus group and interview themes through confirming and disconfirming evidence. Although participant member-checking was not performed, all findings were critically evaluated by the larger multi-disciplinary study team that included primary care, oncology, and radiology clinicians and researchers and any discrepancies were resolved through collaborative discussion as to minimize potential bias of any one perspective.

Results
Study participants comprised 23 patients who took part in the focus groups and 40 physicians who were interviewed. Despite efforts to recruit a diverse patient sample, the final sample consisted of mostly non-Hispanic White women, of which 78% had at least a 4-year college degree. Patient participants came from Vermont (48%), New Hampshire (30%), Washington (9%), North Carolina (9%), and New York (4%). We did not collect race/ethnicity data for physicians. Each registry recruited 5 to 7 physicians. Most physicians represented academic practices (70%) and the remaining represented community or other practices. The demographic characteristics of patients and physicians are provided in Table 1 and an overview of main themes is displayed in Figure 1.

Theme 1: Perspectives on how a risk prediction tool would enhance and add value to patient-centered care
1a.
Features that promote confidence in the risk prediction tool to support surveillance decision-making.
Patients and physicians described evidence and conditions that would promote confidence that a risk prediction tool was acceptable and appropriate for supporting surveillance recommendations. Some patients expressed the need to be provided with enough background knowledge about the tool’s development to feel confident about the tool’s recommendation and its use in supplemental surveillance decision-making:
“For me, it would be dependent on how much I felt that the tool was based on good research. Because I’d love to do less testing if, in fact, it’s not necessary, but I’d have to be really convinced. It’d take quite a bit to convince me that [testing] was not necessary. But if I was convinced, I’d be happy with that.” (Patient)
Likewise, physicians discussed the need to have enough background information to use the tool to inform surveillance decision-making:
“I think I would seek it out if I had a better understanding of its purpose and when it should be employed… But if I had a better foundational understanding of what the tool does and how it works, how it was built, and what pushes somebody into the category of recommending additional imaging, once I had that deeper baseline understanding, then I would know in which patients to utilize it or not.” (Medical Oncologist)
“Well, I would assume the tool incorporates all sorts of evidence, and I guess I’d just like to see that in advance. Or at least have the people who manage our breast care program look through it in depth and look at all the research, and make sure they agree with the thought process behind it. And then we’d apply it blanket-wide as an organization.” (Radiologist)
While evidence supporting a risk prediction tool was valued by both groups, physicians across specialties emphasized the need for a demonstrable improvement in cancer outcomes as a result of supplemental surveillance generated by tool use:
“I would also need some prospective evidence that doing the supplemental surveillance for the patients who are identified as high risk by the calculator actually leads to earlier detection of cancers than just annual mammography alone. And then there’s the next level of evidence, which would be that you actually improve people’s survival accounting for some kind of lead time bias through this. So, there’s a few tiers of understanding the evidence basis for how the tool was even developed in the first place. The next thing, understanding that you pick up cancers earlier and then the third picking up cancers earlier actually improves patient outcomes.” (Medical Oncologist)
“I’d like to see trials showing not only that people have a higher risk of having recurrent cancers, but that using something other than a mammogram would actually catch those cancers earlier and improve outcomes.” (Primary Care Physician).

1b.
Risk prediction was valued for its ability to personalize and tailor decision-making to individual patients, with both groups expressing more comfort with tools supporting more, but not less care.
Patients and physicians discussed the benefit of having a risk prediction tool that could be tailored to the patient, integrating multiple family history, cancer type, and medical history factors. Most physicians felt this would provide an advantage over existing guidelines and “blunt tools” being used currently, suggesting a risk calculator could help them make surveillance recommendations based on objective evidence specific to each patient. Others agreed that it could support a more nuanced, patient-centric approach:
“There are differences in guidelines and then there are differences in patient risk. And the guidelines don’t take into account the personalized approach. You need to personalize your surveillance based on your patient and their risk factors and their breast density and their history. So, I mean taking the guidelines and applying all those personalized measures to it sounds like the dream.” (Medical Oncologist)
“I think it just makes it very objective. I think most patients, they want to know your advice based off your experience and based off objective data. So, to be able to provide that to them with a tool that’s been validated, and it can be just very helpful and make them more comfortable about their decision-making or shared decision-making with their provider.” (Radiologist)
Some patients discussed the potential for the risk prediction tool to add value to their care by enhancing the physician’s assessment of their surveillance plan. Patients perceived that the tool may lead to better decision-making, viewing the tool to include the latest research and objective evidence in the process:
“It sounds all positive to me. I mean, it helps counter any kind of bias a doctor might have. I think we all rely a lot on our own experience when we make decisions and that could prevent a doctor from really seeing the big picture as well as they could if they [use] very objective evidence and not just what they’ve seen in their practice. I don’t know what downside it could have.” (Patient)
“As it turns out, doctors are human too. You know what you know, and you have good days and bad days or whatever. Anytime you’re able to make it easier for people to use the latest research in a helpful way, to me, that makes a lot of sense.” (Patient)
“I’m thinking of doctors using it as a checklist... So, in a way, it is a way to standardize the way that analysis gets done from one physician to another. And if it’s been found that some physicians miss certain steps and maybe outcomes could be better, if they hadn’t missed those steps, then that’s certainly a positive.” (Patient)
While physicians and patients expressed that they valued personalized care arising from the use of risk prediction, they voiced more comfort with using a tool that would support additional imaging, but were more cautious when it came to a tool that would recommend less surveillance imaging.
“In my case, my oncologist and surgeon recommended twice a year for several years after. I would be nervous with this decision-making tool that they might not allow that.” (Patient)

1c.
Using the risk prediction tool to supplement, but not replace, shared decision-making.
Physicians and patients valued the relational aspects of shared decision-making and as such, they expected that any risk prediction tool would supplement the patient-physician relationship and enhance current standards of care. Patients viewed the tool as being used “in conjunction with the actual interaction with your doctor” and not becoming “the decision-making guide.” Several patients described a potential downside of using the tool if it took away from their personal relationship with or expertise of their clinical team:
“I just am too nervous about a tool becoming too important in the decision-making process and the relationship with the doctor and the patient, some of the emotional pieces we’ve talked about, some of the intuitive pieces we’ve talked about, some of the lifestyle pieces that go into decision-making. I would be too nervous that it would replace some of those other pieces.” (Patient)
“But I have to be honest, my initial reaction was a little skeptical... As long as it’s supplementing, but I guess I just don’t want to move away from the physician taking in the whole picture of me sitting there as well. I don’t want it to replace that. As long as it’s supplemental and not they just check off the boxes and say, ‘Okay, this is the way to go.’” (Patient)
One focus group was concerned with the potential for unintended consequences, including financial implications, as these two participants described:
“There’s always the whole issue of unintended consequences... I guess it does make me a little nervous that somebody’s going to say, ‘I’m going to spin all your data in here and come up with the answer, and I don’t need to know the answer anymore because this machine or this tool’s going to tell me.’ It does, it makes you a little nervous.” (Patient)
“I think that’s a very good point to make sure there’s not some unintended consequence so that it’s used to reduce care. Or if the insurance companies get hold of it, the way our medical system is, then it’s like, ‘Oh, good. Then we don’t really need the doctor in this thing, so we’re not going to pay for that.’” (Patient)
While trust as a barrier to the use of a risk prediction tool was primarily shared amongst patients, a few physicians corroborated this position. Physicians expressed that proper administration of the tool could reduce patient anxiety when implemented by physicians who have existing relationships with patients:
“I think it generates some healthy dialogue, probably, especially if a difference has been recommended. And in some of the patients who have anxiety, meaning that they want more scans, if the risk predictor came out low, that would be helpful to maybe calming them down a little bit and saying, ‘We’ve even done this extra step and look to see if you’re at high risk and you’re not.’ And so that might be helpful in relieving their anxiety a little bit, about wanting something more.” (Medical Oncologist)

Theme 2: Risk prediction tools can be a means to improve communication about risk.
2a.
Sharing the results of the risk-prediction tool with patients was an important part of the shared decision-making process.
Physicians discussed the value of receiving the results in such a way that would support discussions with patients, providing an opportunity to explain the risk score, and talk through recommendations. Like physicians, many patients noted the importance of having the results of a risk prediction tool shared by a provider to improve their understanding and inclusion in decision-making:
“It would be really helpful particularly if it was something that I could pull up and kind of go through with the patient in the clinic room. I have a lot of patients who I think would be really interested in seeing that information and would help them with the decision-making as well.” (Medical Oncologist)
“I would think that it could be a great tool if it was something that you were going over with your doctor. It’s that you’re educating both the patient…with this kind of tool, it could say, ‘Well, these are the things that increase your risk or make it wiser to do these other tests,’ or whatever. But I think that it could be a really useful tool to help doctors organize information and explain to people their risk.” (Patient)
“Personally, I like the one-on-one face-to-face. If you’re reading it at home and it’s medically complicated in terminology or whatever, your mind can run circles. It’s better if it comes to you, you can hand them a copy of the report, but you can do it one-on-one with the doctor when you’re sitting there with them.” (Patient)
As above, risk prediction tools were viewed positively when used as part of the patient-physician interaction, supporting in-person discussions about surveillance.

2b.
Participants felt it was critical that the results be presented in non-medical jargon, with attention to literacy and language.
Both physicians and patients cautioned the use of jargon or complicated results, especially if patients receive the results outside of a clinical encounter. Most physicians felt that patients should have access to their results, but some were uncertain whether all patients could enter their data accurately or interpret results independently. Patients discussed preferring to receive results from their doctor without seeing “all these technical terms being used” and in a way that patients “can clearly understand what is going on.” Many noted the need to consider literacy levels, use patient-centric language to improve comprehension, and include visuals with the risk rating:
“There’s been a fair amount of research shown that statistics aren’t great with patients, or at least not all patients. And numbers get lost... So [if] the scientists behind it can evaluate that and then talk in more layman’s terms to patients so that they make sure that they understand. As opposed to just presenting them the numbers themselves directly and letting them try to pick through it and analyze it.” (Radiology)
“Eighth grade language probably and maybe clear text, but also some infographic kind of presentation that sort of gives somebody a visual as well as a text-based interface because people sometimes do better to understand things when they’re prepared or presented visually.” (Primary Care)
“Cultural and also literacy. I think they’re both so incredibly important. Most Americans in fact read at the fifth-grade level and if it’s a tool that they’d be involved in, that really needs to be taken into consideration.” (Patient)
It was also articulated that having simple and straightforward results requiring minimal interpretation would be beneficial for physicians, especially those without specialized knowledge about interval breast cancer or recurrence risks:
“I don’t think the average clinician is great at risk estimating. I like pictographs, but not all patients can understand them. They take time to understand, so it depends on in what setting.” (Primary Care)
“Yeah, I mean, I think if it’s exclusively clinician-facing, then the clinician has to develop their own skillset and separate set of tools about then how to communicate that back to a patient. And depending on how engaged patients want to be in decision-making around their health, they may press the clinician in ways that then they’re not really prepared to deal with if the tool doesn’t really serve that purpose.” (Medical Oncology)
In sum, risk prediction tools were perceived positively when positioned as a supplement to the patient-physician relationship and supported by strong, transparent evidence. Results that are easy to interpret and that enhance patient involvement in surveillance decision making were desired by both groups.

Discussion
This study qualitatively explored patient and physician perceptions of using a risk prediction tool to support surveillance imaging decisions in breast cancer survivorship and identified facilitators and barriers to their clinical implementation. Broadly, such an approach was perceived to have advantages in improving the objectivity and tailoring of risk assessment, supporting the process of shared decision-making, and potentially improving patient outcomes. Both groups desired transparency about risk model inputs, validation of model performance, and recommendations that could be easily understood by both clinical and patient audiences. Finally, using a risk prediction tool was valued when augmenting clinical expertise, but concerns arose about overreliance on the results, unintended consequences such as insurance implications, potential limitations of patient literacy, and a strong desire to maintain the relational aspects of the patient-physician relationship.
Patients and physicians both affirmed the potential improvement that tailoring surveillance imaging decisions based on risk could have. However, physicians emphasized the desire for a risk prediction tool to enhance outcomes via integrating more nuanced patient data relative to ‘blunt’ guidelines. As the use of risk prediction and Artificial Intelligence (AI) proliferates in medicine, physicians are more accustomed to models that support risk-tailored recommendations. As found in other contexts, they are generally accepting of such tools to support decision-making given they have utility, credibility and are usable.37,38 Of note, physicians in our sample called for risk prediction tools that promoted engagement in tests with demonstrated benefits on morbidity and mortality. This study does not directly address the evidence for supplemental surveillance,16 although findings make it clear that risk prediction tools should be coupled with evidence-based recommendations with demonstrable benefits.
Meanwhile, patients expressed that a risk prediction tool would be most advantageous in supporting physicians or other clinicians with less breast cancer-specific knowledge or experience and/or to reduce physician bias. These patient concerns are not unfounded: studies have found frequent use of non-recommended surveillance tests among breast cancer survivors39,40 and differential patterns of surveillance by race and ethnicity.41,42 Future research is needed to identify model inputs and evidence that promote trust and credibility across physician and patient audiences, as we saw some areas of divergence in terms of what elements of risk prediction are valued by these different groups. Despite the slightly different rationales for its use, both physicians and patients perceived value in tailoring surveillance imaging by risk.
With regard to acceptance of risk model use, patients and physicians supported a tool that would justify more imaging, but patients especially expressed wariness about the consequences of using a tool that suggested providing less imaging. Physicians appeared to assume that the risk prediction tool would be focused on identifying high risk patients for whom additional imaging was warranted and could also provide reassurance for low-risk patients. In contrast, patients expressed concern about low-risk results being used to scale back supplemental surveillance imaging. Patient apprehensions about risk prediction withdrawing care are not new: studies in screening have demonstrated that patients are concerned that algorithms will be used to reduce health care costs, and these perceptions are linked to lower acceptance of such tools.43 Our study found that while concerns about a risk tool suggesting less surveillance were present, participants felt this could be mitigated by ensuring the tool did not supersede physician autonomy, a finding found across several studies of clinical decision support tools.37 To mitigate these concerns, studies have found that it is important to avoid ‘prescriptive’ and overly ‘definitive’ recommendations, and to include supportive references to expert or scientific evidence.44 Indeed, the decision sciences literature has appropriately emphasized the presentation of options in decision-agnostic, balanced, and unbiased ways, with tools to support implementation of approaches that support patient and clinician autonomy.45,46
Aligned with patients’ desire for physician involvement in discussion of the tool’s results, participants cited the need to provide clear interpretation of results to physician and patient audiences as a critical component of using a risk prediction tool. Both groups noted the need for lay summaries, written at recommended grade levels, in line with standards, including the Mammography Quality Standards Act set forth by federal organizations.47–49 These strategies designed to accommodate health literacy are known to support patient involvement and shared decision-making.48,50 Yet producing interpretations that support individual decision-making remains an area of improvement for the field at large.21 End-user involvement in the development of risk prediction tools from the outset may ensure tools developed are relevant and useful to intended audiences (patients, physicians).
Our findings, especially those from patients, suggest that the patient-physician relationship, and a physician’s knowledge of an individual patient is an important component of survivorship care and surveillance decisions. The idea that the physician provides additional value beyond a risk prediction tool is mirrored in the breast cancer risk-based screening literature, where the physician recommendation and emotional support are often cited as important.51 Similar to our findings here, a recent review found that when using AI or prediction algorithms in breast screening, women expect some level of physician control and care, excellent model performance, transparency and governance, and sound reasons for its implementation (e.g., improved outcomes).52 Indeed, clinical frameworks for the use of risk prediction tools likewise recommend a role for shared decision-making, using the results of a tool to augment discussion.53
This study has limitations. While we sought to recruit a diverse sample of patients by race, ethnicity, education, and geography, our focus group sample was predominantly white and college educated. Acceptability and use of risk tools should be further explored in more diverse populations and clinical settings (e.g. community hospitals) to confirm the robustness of these findings. In particular, findings about the need for appropriate and understandable risk outputs will be critical to test with populations that are less educated than the current sample and across different clinical provider populations (e.g. non-physicians).
We were also in part limited by describing the risk prediction tool in general terms, rather than being able to user-test a specific tool. Thus, our findings are somewhat bounded by the description provided to participants and did not permit the identification of desired evidence for model performance with great specificity. Two recent systematic reviews including both hypothetical and validated risk prediction tools to support cancer screening decisions found similar results: Studies reflected our findings that patients and care providers are concerned about model inputs and accuracy, and display hesitancy to reduce screening when results indicate low risk.54,55 Similar concerns have been raised with when testing usability of specific risk prediction tools across a variety of clinical conditions.56–58 Using a broad description of the risk prediction tool allowed us to guide development of this specific risk tool in its early translational phase and also to provide insight on the general approach of using risk prediction tools for surveillance. As the tool is validated and moves toward implementation, we will continue to assess aspects of acceptability and usability to ensure its translational potential.
In conclusion, this study provides new data on patient and physician perceptions of using a risk prediction tool to support surveillance imaging decisions among breast cancer survivors. While several findings mirror what has been documented in the screening (cancer-naïve) population, there are distinctions. The array of physicians (oncologists, primary care, radiologists) involved in survivorship care highlight the wide range of perspectives to consider and include when promoting evidence-based care across settings. Future research should identify care delivery settings and features that promote adoption and use in ways that support improved shared decision making and patient outcomes.