Preliminary Screening for Hereditary Breast and Ovarian Cancer Using an AI Chatbot as a Genetic Counselor: Clinical Study

Introduction

The rapid adoption of digital health tools has significantly enhanced health care efficiency and quality [1]. Chatbots enable communication with patients. Their use includes delivering remote health services (eg, patient support, education, and health promotion) and providing administrative assistance (eg, managing routine tasks and supporting research). One key advantage of chatbots is resource optimization. By handling routine tasks, chatbots let health care professionals (HCPs) focus on clinical duties. Chatbots standardize health care and expand access, increasing patient opportunities for appropriate services, even in regions with limited medical staff [2]. In genetic medicine, which requires specialized expertise, chatbots can collect family medical history and assist HCPs in communicating with at-risk relatives [3,4]. However, practical implementation in genetic medicine remains limited, particularly in Japan. We developed a chatbot for preliminary screening of hereditary breast and ovarian cancer (HBOC) [5] and conducted a clinical study to validate its feasibility and potential usefulness in a clinical setting with real patients.

Methods

Overview

The chatbot system [5] was planned to be tested on 10 patients to estimate its medical accuracy and ability to perform preliminary screening, such as obtaining correct medical and family histories and assessing the risk of HBOC according to the National Comprehensive Cancer Network guidelines. The system uses Watson, IBM’s cognitive computing service. Preoperation questionnaires and postoperation interviews validated data on the system’s usability and acceptability. Additional details are provided in Multimedia Appendix 1.

Ethical Considerations

This study was approved by the institutional review board of Kanagawa Cancer Center (2020 - epidemiology 78). Written informed consent was obtained from all participants. An ID was assigned for each participant, and personal information was excluded. The correspondence table between individuals and IDs is stored in a secure database that is not connected to the internet, and the database terminal is stored under double-locked security. We provided each participant with JP ¥2000 (~US $15) as an honorarium.

Results

A total of 11 participants enrolled between October and December 2020 (Table 1). All participants completed the chatbot operation (Multimedia Appendix 2). Preliminary screening results are shown in Table 1. Qualitative data from postoperation interviews are shown in Table 2, with key comments in Multimedia Appendix 3.

Discussion

This study demonstrated that implementing the chatbot for preliminary HBOC screening is feasible in a clinical setting. Two factors support this conclusion. First, the chatbot achieved a 100% completion rate, higher than previous studies [4]. Most participants reported excellent usability in the qualitative interviews, suggesting that the system’s interface positively influenced completion. Despite limited experience, participants successfully completed the operation, indicating its accessibility regardless of technological literacy. It could be adapted for both patients visiting hospitals and healthy individuals eligible for cancer screening. Previous studies have confirmed that chatbots can effectively enhance cancer screening by automating tasks and increasing participation rates [6,7]. Participant interviews indicate that there is potential to provide opportunities for genetic medicine to patients living in remote areas or those who experience anxiety regarding in-person consultations. If the system can serve a broader patient population, it can not only act as a substitute for expert-led preliminary screening and quality standardization but also enhance accessibility to genetic medicine. Second, the chatbot accurately determined participants’ eligibility for BRCA1/2 testing. Given the medical accuracy maintained in the 2 phases of our study, the findings support clinical application, although further research with diverse medical and family histories is needed for validation.

Based on the interviews, we identified areas for improvement. Discrepancies between the medical and family histories reported by the CGCs and those recorded using the chatbot suggest inadequacies in the chatbot’s interview, which could lead to missing individuals at high risk for HBOC. These discrepancies arose from incomplete questions from the chatbot, participants’ omissions, or communication errors. Thus, the chatbot’s questions should be extended to include all information gathered during the CGC interviews. Participants also reported difficulty viewing the family tree image, indicating that larger images and text or voice confirmations could improve accuracy. Furthermore, inappropriate wording in the chatbot’s responses may have led to communication errors, highlighting the need to review the chatbot’s language. We are currently improving the system.

We could not evaluate the chatbot’s impact on patient visit behavior because this study excluded physician feedback. Future research should evaluate whether the chatbot increases genetic counseling and testing rates.

This study suggests that chatbots can effectively gather diverse information. Previous research has shown that chatbots enhance communication [3,8-10]. Therefore, chatbots could be used for routine medical tasks beyond preliminary screenings, potentially improving HCPs’ effectiveness and streamlining workflow.