Breast cancer is the leading cause of cancer mortality in women worldwide [1]. Almost 11,400 women a year died of breast cancer in the United Kingdom between 2015 and 2017 [2]. However, the mortality rate of breast cancer is falling, with a reduction from 60 per 100,000 in 1989 to 33 per 100,000 in 2017 [2,3]. This trend correlates with the introduction of widespread breast cancer screening using x-ray mammography [4]. In the United Kingdom, mammography is offered to women aged 50 to 70 years through the National Health Service (NHS) Breast Cancer Screening Programme [3]. Screening is estimated to reduce the relative risk of breast cancer mortality by 20% [5] and is linked to many lives saved each year [6].

However, mammography is not suitable for all groups who could benefit from breast cancer screening [7]. For example, 10% of invasive breast cancers occur in younger women (aged <45 years) in the United Kingdom at a rate of 235 per 100,000 [8], a group for whom mammography is not recommended because of its considerably decreased sensitivity in dense breast tissue [7]. This is particularly concerning as young women diagnosed with breast cancer are at higher risk of developing aggressive subtypes and have a poorer prognosis [9]. Mammography is also inappropriate for pregnant women as the low-dose radiation used poses a potential risk during lactation [9]. Furthermore, some women may be unable to tolerate mammography because of pain or discomfort [10]. A breast cancer screening modality that extends services to these groups has the potential to save years of life [11]. Identifying a screening alternative has been challenging. The most effective means of diagnosis (eg, triple assessment and magnetic resonance imaging) are often invasive or costly and unfeasible as screening modalities [12].

A promising alternative is clinical breast examination (CBE) [13,14]. A recent randomized controlled trial of CBE breast cancer screening involving >150,000 women in India demonstrated a 15% reduction in breast cancer–related mortality and a 10% relative risk reduction in the diagnosis of stage-III or stage-IV disease [15]. An overview of systematic reviews assessing the effectiveness of CBE screening identified indirect evidence that CBE has the same effect as mammography when performed well [16]. However, there are several challenges to ensuring that CBE is consistently “well performed.” CBE screening effectiveness may be affected by variations in examination proficiency, training of health care professionals (HCPs), and a lack of standard documentation [17-21].

Recent advances in technology may provide a solution to these challenges. Robotics-assisted procedures have expanded rapidly in recent decades [22,23], and existing literature suggests that health users are increasingly more accepting of artificial intelligence (AI) and machine learning algorithms in cancer screening [24-26]. It is theoretically feasible to create a fully automated robotic CBE (R-CBE) platform by combining soft robotic technology and machine learning algorithms trained by breast specialists. This could offer much-needed standardization of CBE. R-CBE also has the potential to extend screening services to currently underserved groups as it is not reliant on radiation or affected considerably by tissue density. As health policy makers are discussing a risk-stratification approach to breast cancer screening, the cheap and low-risk modality of R-CBE may find further use as part of a strategy to classify the personalized risk level of an individual by measuring physiological properties such as mammographic density [27,28].

The Automated Robotic Examination Intelligent System (ARTEMIS), a novel robotic system for automated CBE, is currently being developed by our research team with support from Cancer Research United Kingdom (CRUK). ARTEMIS aims to combine soft robotic technology with a machine learning platform to allow for fully automated CBE and interpretation of results. The platform could be used by women without direct clinical supervision (Figure 1). A prototype is currently in the early stages of development [29-31]. Although such a platform may be capable of effectively performing and interpreting CBE, the voices of potential users are essential in determining how this should be designed and implemented. Creating technology and a service that is acceptable to end users (and meets their needs) will be crucial in determining the uptake of this type of technology.

Very little published literature is available on the acceptability of intelligent systems that interface directly and independently with users. We identified only 1 study assessing the acceptability of autonomous robotic systems that interface directly with users in health care. This study used robotics to perform basic patient assessment tasks (eg, measuring vital signs and inserting intravenous catheters) and concluded that this would be acceptable [32]. We did not identify any publications exploring the acceptability of intelligent robotic services that directly interact with users in cancer screening or diagnostics.

This study investigated whether R-CBE is theoretically acceptable to potential users and explored the attitudes, perspectives, and concerns of women regarding the use of intelligent robotic technology in breast cancer screening. It identified key factors that determine whether (and how) the technology would meet the needs of patients, allowing these to be integrated into the prototype design. We adopted the definition of acceptability proposed by Sekhon et al [33]: “a multi-faceted construct that reflects the extent to which people receiving a healthcare intervention consider it to be appropriate, based on anticipated or experienced cognitive and emotional responses to the intervention.” We conducted a web-based survey of 155 women in the United Kingdom to investigate the following questions: (1) Is there a perceived need for R-CBE? (2) What elements of the R-CBE user interface are most important to women? (3) Is this technology likely to be acceptable to potential users? To the best of our knowledge, this is the first study assessing the acceptability of a fully automated and intelligent patient examination system that interacts directly with users for breast cancer screening.

The survey consisted of 5 open-ended questions and 17 closed questions with a separate free-text section for respondents to share additional information. A brief overview of the proposed ARTEMIS concept was provided to respondents (Multimedia Appendix 1). This included Figure 1 and a description of how the user might interact with the palpation platform but had no technical details or any information on the accuracy of the device. Our aim was to allow the respondents to freely think about factors that might affect their use of the hypothetical service without imposing any of our priors.

Key constructs were identified based on a review of the health intervention acceptability and health technology literature, our broader knowledge of health technology, and support from the CRUK patient and public involvement specialist team and the London In Vitro Diagnostics Co-operative. We did not identify a fully validated model suitable for our research questions; instead, 2 frameworks were combined with questions selected to cover essential constructs from both. The first was the Theoretical Framework of Acceptability for health care interventions proposed by Sekhon et al [33]. The second was the Unified Theory of Acceptance and Use of Technology developed by Venkatesh et al [34], which has been widely used in research exploring the acceptance of ITs [35]. The resultant key constructs encompassed affective attitudes, perceived effectiveness, ethicality, self-efficacy, effort expectancy, social influence, and facilitating conditions.

The questions were carefully designed to illuminate implicit assumptions and ensure that all key constructs were considered while maintaining an accessible and nonleading language [36]. After a multistage drafting process, the survey was collated and tested on close contacts and members of the associated research department for appropriateness, readability, and ease of use to produce a final draft (Multimedia Appendix 1).

Closed questions allowed us to quantify the overall level of acceptability and desirability of specific features of the service (eg, interface, timing, and preferred location). Thematic analysis of qualitative data provided insights into the quantitative findings. This added richness to our understanding of potential users’ perspectives and attitudes toward the proposed ARTEMIS R-CBE and allowed us to build a more complete picture of acceptability.

Female respondents aged between 20 and 70 years were recruited through a web-based survey linked to the CRUK patient involvement opportunities web page and newsletter and the People in Health West of England and Imperial Human Behaviour and Experience network mailing lists. This nonprobability, voluntary response sampling strategy was chosen because of its quick recruitment rate and ability to serve the exploratory nature of the study. With no hypothesis to test, the aim of the survey was to develop an initial understanding of the needs of the population, and so the bias introduced by self-selection was considered acceptable.

The 15-minute web-based survey was hosted on Qualtrics (Qualtrics International Inc), and 2 attention-check questions were added to ensure that respondents read each question carefully and also to exclude nonhuman (automated) respondents; this resulted in the expulsion of 1 set of responses because the attention questions were answered incorrectly. A further 15 questionnaires were discarded because they were incomplete, including incomplete attention questions, and 3 were discarded because they did not meet the inclusion criteria. This meant that, of 174 responses initiated, 155 (89.1%) completed the survey over 6 weeks between August 2020 and September 2020. Summing the size of each of the mailing lists gives a response rate of 9.26% (155/1674).

Quantitative analysis was conducted in MATLAB (MathWorks), and differences between groups based on demographics were identified using a 2-sample Kolmogorov-Smirnov test, explored using 1-tailed t tests, and reported where significant (full results available in the data set referenced in the Data Availability section). Pearson correlations were calculated where appropriate to quantify the strength of the associations. CIs were calculated for ranked questions assuming that the preferences were equidistant (1>2>3...).

Qualitative data were analyzed using a method designed around thematic analysis [36]. This allows for detailed exploration of patterns across a data set using a latent approach, with researchers gaining a rich understanding of respondents’ perspectives [36].

Themes were identified after familiarization with the open-text responses. To this end, 2 researchers independently identified a set of key themes within the responses, chosen with relevance to identifying the factors that influenced the respondents’ acceptance of the hypothetical technology. After combining these sets of themes, a single researcher divided each theme into concepts that tightly grouped responses within each theme. Salient ideas from these grouped concepts were then extracted to describe the outcomes of the responses as a whole.

The raw data are available from the source provided in the Data Availability section at the end of this paper.

The study received ethics approval from the Imperial College Research Ethics Committee (20IC6129).

Responses were generally positive for a potential R-CBE service that is at least equivalent to a nonrobotic alternative. The overwhelming majority of respondents reported that they would use R-CBE screening if it were offered. Respondents recognized the potential of R-CBE to address an unmet need in current screening services by providing regular reassurance, reducing interpersonal embarrassment, reducing screening-associated pain, improving appointment availability, and offering rapid results. All these are barriers to screening uptake recognized in the existing literature [37].

The survey showed that a high level of sensitivity and specificity of this technology is an essential factor for acceptability. User acceptance in our survey was dependent on R-CBE being a highly accurate system.

The results of this study also complement the existing literature on AI diagnostics, which suggests that the public has a high level of trust in computerized decision-making in health care and that AI in cancer screening is increasingly accepted [24-26].

The acceptance of R-CBE was qualified. Our results complement the existing literature [38] by identifying high levels of trust as an essential property for the uptake of robotic and automated systems. Our data identified factors that are necessary for an R-CBE service to be considered trustworthy. Key among these are accuracy, usability, and communication. Respondents’ concerns regarding the lack of human connection, data privacy, and regulation of new health technology echoed similar concerns identified in a recent study exploring the public perception of AI mammography reading [26].

Our results indicate that most users are likely to accept autonomous screening if there is a well-established, efficient process for follow-up with a clinician if needed. This agrees with studies to date indicating that people are more accepting of intelligent systems working symbiotically with physicians or HCPs [24,39] but remain ambivalent about those that function independently [25].

This study provides important information to guide decision-making on R-CBE development, determine its viability as an investment, and inform our understanding of public attitudes toward intelligent health technology in cancer screening. Crucially, our results indicated a significant concordance between what is technically feasible and what is acceptable to users. For example, most respondents (128/155, 82.6%) were willing to be examined for up to 15 minutes and were also willing to receive results directly from ARTEMIS (in some format) rather than from an HCP. Research suggests that it is feasible to create an automated R-CBE service based on these acceptability characteristics [29-31].

From these results, we believe that R-CBE may offer a more patient-focused option that has the potential to increase screening uptake provided it can perform examinations with sufficient sensitivity and specificity.

To develop technologies seeking to provide the service of R-CBE or similar, these results provide appropriate targets to be met when evaluating their expected acceptability. For example, several respondents supported R-CBE because it would reduce the burden on the NHS and free up time for HCPs. Although early detection and intervention could reduce progression to advanced disease (and, therefore, reduce the treatment burden on the NHS), this assumption is only valid if R-CBE detects early disease and allows for early intervention without overdiagnosis or excessive referrals to primary or specialist services.

As an investigative survey, the sample size was comparatively small, and the skewed distribution of demographic groups within the sample means that it was insufficiently powered to detect nuanced differences between them. A larger sample size with a demographic distribution representative of the wider population would be needed to identify whether the subtle differences in preference between demographic groups in this study are statistically significant and externally valid in the general population.

The demographics of the respondents were also not representative of the UK population. First, Black and minority ethnic groups were underrepresented. The data may not accurately capture the needs, thoughts, attitudes, and perspectives of these demographics. This is of particular concern as these groups are at an elevated risk of breast cancer and face the greatest barriers to screening [40,41]. Reaching these groups in future research is essential as they may benefit substantially from widening screening. Achieving this is likely to require targeted methods.

In addition, over three-quarters of our sample (119/155, 76.8%) had a degree-level education. Jonmarker et al [25] found a significant association between level of education and level of trust in technology. This is reflected in the very high levels of trust in technology reported in our sample. This reduces the generalizability of our results, with survey respondents being more likely to find R-CBE acceptable than the general population. The non–probability sampling used in this study may also introduce selection bias—it is possible that women who had a history of engaging with existing breast cancer screening programs were more likely to answer the survey, which might have contributed to overestimation of the acceptability of R-CBE screening. The particular method of electronic survey requires respondents to have ready access to a compatible device connected to the internet and be literate at using it, inherently excluding those who do not fulfill both criteria.

The ARTEMIS R-CBE is currently in development (part of this system is described in the study by Jenkinson et al [31]); the responses relate to a theoretical service. Further research will be needed to establish the acceptability of the specific service among users as development continues, as well as an assessment of its cost and accuracy. Future research would benefit from a larger and more diverse sample size that better represents the population. Our team is currently undertaking further qualitative research via focus groups to better understand the requirements of trustworthy and acceptable R-CBE and automated breast cancer screening more generally. Despite the limitations outlined previously, the survey data allowed us to identify key priorities among potential users and provide valuable information for the research team. These findings may provide insights for others working in automated health technology development, particularly for cancer screening.

R-CBE holds promise as a new modality of breast cancer screening. It could address limitations in current screening services, increase screening uptake, and provide a more patient-focused service. This investigative survey demonstrated that there is potential for high levels of acceptability of R-CBE among its target user group and a high concordance between user expectations and technological feasibility. However, the acceptability of R-CBE is conditional on users being confident that it is accurate, easy to use, able to communicate results sensitively, and well integrated with health services. These findings will contribute directly to prototype development and will be of interest to other researchers developing automated cancer screening and related health technologies. This study highlights the fact that the development of new technologies raises ethical and practical issues. The importance of public and patient involvement in health technology development to address these issues should not be underestimated. Patient and public involvement at each stage of development will be key to ensure that any future service meets the needs of the public.