Abstract
Objective
Individuals who are at risk of or diagnosed with breast and ovarian cancer often face barriers in seeking help due to embarrassment, complex information needs, and limited access to timely support. Chatbots provide a private, non-judgmental communication interface that may help overcome these barriers. This scoping review systematically examines chatbot applications in breast and ovarian cancer care, focusing on their design, functionalities, evaluation outcomes, and implementation barriers and facilitators.
Methods
A comprehensive search was conducted through PubMed, Medline, Embase, CINAHL, Cochrane Library, and Web of Science databases using keywords related to chatbots, breast cancer, and ovarian cancer. Studies were screened and reviewed following PRISMA-ScR guidelines.
Results
Nineteen studies met the inclusion criteria, covering educational support (n=16), clinical support (n=14), and psychosocial support (n=11). NLP served as the primary technical foundation (n=12), with rule-based and retrieval-based approaches equally represented (n=7 each) and a growing adoption of LLM-driven approaches (n=5). Studies reported consistently high user satisfaction, though RCT evidence on clinical efficacy showed mixed results with benefits varying by patient subgroups. Qualitative studies identified adoption barriers including data privacy concerns and patient expectations exceeding chatbot scope. While breast cancer applications showed promising outcomes, research on diagnosed ovarian cancer patients remains limited.
Conclusions
Chatbots show potential as complementary tools in breast and ovarian cancer care. Future development should focus on standardised evaluation frameworks, specialized ovarian cancer applications, and optimizing the balance between automation and human oversight in clinical settings.
Introduction
Breast and ovarian cancers are major malignancies with profound impacts on global health that are frequently studied together due to their shared risk factors and similar treatment approaches. According to the World Health Organization, 1 breast cancer ranks as the most prevalent cancer among females in 157 out of 185 countries, with an annual incidence of 2.3 million diagnoses, while ovarian cancer accounts for around 313,959 new cases annually. 2 These cancers share key risk factors including advanced age, hormonal factors, family history, more specifically, the hereditary breast and ovarian cancer (HBOC) syndrome characterized by BRCA1 and BRCA2 gene mutations. 3 Despite advances in cancer screening and diagnosis, many patients continue to face substantial challenges in navigating the complexities of treatment, care coordination, and psychosocial support throughout their cancer journey. Studies indicated that approximately 20-50% of breast cancer patients experienced treatment non-adherence due to psychosocial factors, financial barriers, and difficulty managing complex treatment regimens.4–7 Additionally, rising patient volumes have overwhelmed oncology services, leading to prolonged wait times and communication challenges.8,9 Traditional care approaches often fall short in providing personalised care coordination, real-time monitoring, and timely intervention. 10 Many patients face significant information needs regarding diagnostic procedures, treatment options, side effects, and self-management strategies, with limited access to reliable and personalised resources between diagnosis and follow-up visits. 11 These challenges underscore the need for innovative healthcare solutions to support patients with knowledge acquisition and enhance their overall treatment experience through personalised emotional support and resources tailored to individual needs.
Chatbots, or conversational agents, are defined as computer programs designed to simulate conversation with human users. 12 The emergence of chatbots has revolutionized health information dissemination and access. Initially, chatbots relied on basic keyword-matching techniques, which restricted their abilities to comprehend variations in language and context. 13 With advancements in artificial intelligence (AI), and large language models (LLMs), chatbots evolved to incorporate more sophisticated response generation methods, such as rule-based systems utilizing predefined patterns, retrieval-based systems selecting responses from existing databases, and generative models creating dynamic responses through machine learning (ML) algorithms.14–16 This evolution has significantly enhanced chatbots’ ability to process complex medical information and maintain context-aware conversations, particularly valuable in healthcare settings requiring effective communication and support efficiently. 17
Currently, the integration of chatbots into oncology has significantly enhances patient care in multiple ways. For example, in prostate and lung cancer care, chatbots have helped patients and caregivers understand complex health information and manage treatment expectations.18,19 In colorectal cancer management, chatbots collect family history for risk assessment and educate at-risk individuals about inherited cancer syndromes before clinical visits. 20 Some applications in head and neck cancer have also addressed caregiver needs and optimized communication pathways. 21 Moreover, healthcare professionals reported improved workflow efficiency as chatbots handle routine inquiries and preliminary assessments, allowing them to focus on complex cases and specialized care. 22 More importantly, chatbots’ ability to facilitate anonymous interactions has led to increased disclosure of sensitive health information compared to traditional clinical consultations.23,24 These benefits position chatbots as valuable tools for personalizing cancer care delivery. 25
Despite this emerging evidence, few studies have specifically examined chatbot applications tailored to breast and ovarian cancer alone.26–28 There is one review exploring chatbot applications in breast cancer care, but it was limited to patient education, leaving other potential functions in breast cancer management unexplored. 29 Considering that both breast and ovarian cancers share genetic predispositions (e.g., BRCA1/2 mutations), exhibit similar pathophysiological processes, and have similar treatment modalities (e.g., PARP inhibitors), we review them together to provide a comprehensive understanding of chatbot applications in oncology.
This scoping review aims to address the current gap by systematically mapping chatbot applications across populations at risk of, experiencing, and recovered from breast and ovarian cancer. It aims to synthesize current evidence on chatbot characteristics, functions, outcomes, and implementation considerations, providing insights for future development and integration into clinical workflows. Specifically, the research questions are as follows:
(1) What chatbot applications have been developed for breast and ovarian cancer care; (2) What are the technical approaches and functions of these chatbots; (3) What evaluation outcomes have been reported; (4) What implementation facilitators and barriers have been identified?
Materials and methods
This scoping review was conducted following the five-stage methodological framework proposed by Arksey and O'Malley 30 and enhanced by Levac et al. 31 The study adheres to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for scoping reviews (PRISMA-ScR), 32 with the detailed checklist provided in Supplementary Material (S1 Table). This study was registered in the Open Science Framework (https://doi.org/10.17605/OSF.IO/P26Z3).
Search strategies
An extensive search strategy was formulated in consultation with a research librarian for six academic databases, including PubMed, Medline, Embase, CINAHL, Cochrane Library, Web of Science. For this review, we operationalized chatbots as computer programs that simulate human conversation through text or voice interactions, encompassing various terms such as conversational agents and dialogue systems. Additionally, artificial intelligence was considered as the underlying technology that enables natural language processing (NLP) and ML capabilities in these systems. Based on these operational definitions, the search strategy included three groups of search terms: (1) Breast cancer (e.g., “Breast Neoplasms”, “Breast Cancer Lymphedema”, “Unilateral Breast Neoplasms”); (2) Ovarian cancer (e.g., “Ovarian neoplasm”, “Epithelial Carcinoma”, “Ovarian tumour”); (3) Chatbot (e.g., “Chatbot”, “Artificial intelligence”, “Conversational agent”). The full search syntax for each database, including all Boolean operators and controlled vocabulary is provided in Supplementary Material (S2 Table). The search was limited to English-language peer-reviewed articles published between 01 January 2015 and 01 January 2025, spanning a 10-year period, to incorporate the early evolution of chatbot technology. This timeframe was selected as Agarwal et al. 33 identified post-2015 as the period when chatbots began to be widely utilized for health-related purposes. Additionally, a second-round search covering 01 January 2025 to 26 February 2026 was performed using the same databases and syntax to ensure this review reflects the most recent literature.
Eligibility criteria
Studies were considered eligible for inclusion if they met the following criteria: (1) investigated chatbot applications in populations at risk of, experiencing, or recovered from breast or ovarian cancer; (2) published in peer-reviewed journals; (3) written in English; (4) published between January 1, 2015, and February 26, 2026; and (5) available in full-text. Exclusion criteria included: (1) studies focusing on cancers other than breast or ovarian cancer; (2) review articles, protocols, commentaries, or viewpoint pieces; (3) experimental research without human subject testing.
Data extraction and review process
The selection process was performed using EndNote 21, through separate screening of titles and abstracts for relevance. Data were extracted independently by two authors using a standardised data extraction form developed in Microsoft Excel. The extracted data were categorized into four main domains: (1) study characteristics (including year of publication, study design, country of origin, age group, population, and sample size); (2) chatbot characteristics (including name of chatbots, AI techniques, Dialogue system methods, launch methods, and functions); (3) chatbot evaluation outcomes (including user satisfaction, usability, and information quality); and (4) reported limitations (both chatbot-related and research methodology limitations). The detailed data extraction framework is presented in Supplementary Material (S3 Table). Subsequently, full-text documents were retrieved and subjected to comprehensive qualitative review, guided by the above-mentioned inclusion criteria. Prior to screening, a calibration exercise was conducted where two independent reviewers each evaluated a subset of 100 articles independently. A 72% inter-rater agreement was obtained on the initial set. Discrepancies were discussed to refine the eligibility criteria and improve screening consistency before moving to the formal screening. In the formal screening process, two independent reviewers assessed the titles and abstracts of the articles and manually annotated them as “included,” “excluded” or “uncertain” based on the definition and eligibility criteria defined. Articles marked as “uncertain” or with conflicting decisions were flagged for further review. Any disagreements were resolved through discussion between the two reviewers, and a third reviewer was consulted when consensus could not be reached. In accordance with scoping review methodology, no quality appraisal was conducted, 34 as the primary aim was to map the breadth of chatbot applications in breast and ovarian cancer care.
Results
The search was conducted in two rounds across six academic databases. The initial search yielded 5761 articles, while the second-round update search identified an additional 1118 articles, resulting in a combined total of 6879 records. After eliminating 2539 duplicate records, 4340 articles remained for titles and abstracts screening. Of these, 4082 records were excluded based on the inclusion criteria, leaving 258 full-text articles retrieved for further review. Following full-text screening, 239 studies were excluded due to other cancer types, non-English language, non-human studies, and review or commentary articles. 19 studies were ultimately included in this review (Figure 1). PRISMA Flow diagram of the study selection process.
Characteristics of the included studies
This review examined 19 studies published between 2019 and 2026, with more than 60% published in 2023 or later. The geographical breakdown shows a predominance of research from Western countries, with 26.3% (n=5) from the United States and 47.4% (n=9) from European nations, including Germany, Norway and France. Contributions from Asia accounted for 21.1% (n=4), with studies from Taiwan, Japan, and South Korea, while African contributions accounted for 5.3% (n=1).
Characteristics of included studies.
Note. RCT: randomized controlled trial; NR: not reported; HBOC: hereditary breast and ovarian cancer; MDT: multidisciplinary team; BRCA: breast cancer gene; RAG: retrieval-augmented generation; APAIS: Amsterdam Preoperative Anxiety and Information Scale.
Technical and functional characteristics of chatbots applications
Technical characteristics
Technical characteristics of chatbots applications.
Notes. RAG: retrieval-augmented generation; NR: not reported; LLMs: large language models; ML: machine learning; NLP: natural language processing; CWS: Chinese word segmentation.
Functional characteristics
Three functional domains were consistently represented, including clinical support, educational support, and psychosocial support. Most studies incorporated more than one domain. Educational support was the most prevalent (n=16), closely followed by clinical support (n=14), with psychosocial support appearing less frequently (n=11). Figure 2 illustrates the subcategories within each domain and the number of studies addressing each. Detailed chatbot functionalities for each study are provided in Supplementary Material (S4 Table). Distribution of chatbot functionalities used in breast and ovarian cancer care.
Clinical functionalities comprised risk assessment, genetic counseling, symptom monitoring, risk triage and referral, patient medical history collection, and clinical decision-making assistance. Risk assessment (n=5) and genetic counseling (n=5) played a central role. These applications were designed to support hereditary cancer risk evaluation, with one system successfully identifying 27% of participants as high-risk HBOC candidates. 41 Genetic counseling chatbots primarily served as digital pre-consultation tools, guiding patients through pre-test guidance and preliminary interpretation of results before directing them to in-person genetic counseling. 47 Clinical decision-making assistance specifically focused on tumour board decision-making, with one chatbot achieving 70% congruence with multidisciplinary tumour board recommendations. 42 Symptom monitoring capabilities were also implemented, with Chen et al. 22 developing a system that demonstrated high correlation accuracy in breast cancer assessment scoring compared to specialist evaluations.
Educational functions included disease knowledge, treatment procedures, genetic testing information, medication side effects, rehabilitation and lifestyle, and financial and legal information. Disease knowledge was a foundational component across multiple applications, with chatbots providing information about breast cancer epidemiology, symptoms, and prognosis.37,40,53 Treatment procedures were addressed in chatbots explaining radiotherapy protocols 39 and breast biopsy processes, where 87% of patients reported improved understanding after use. 48 Genetic testing information constituted a major focus, with several systems educating patients about BRCA1/2 mutations, hereditary cancer syndromes, and testing implications.36,41,45,50 Rehabilitation and lifestyle guidance was integrated into chatbots offering advice on post-treatment recovery, exercise, diet, and daily activities during radiotherapy.37,39 Financial and legal information, including reimbursement processes and patients’ rights, was incorporated into comprehensive support chatbots like Vik.40,41
Psychosocial support functions, while less commonly emphasized, were nonetheless integral to user experience in several applications. These included anxiety management that helped users cope with procedure-related distress,39,50 fertility-related emotional support addressing concerns about reproductive capacity and family planning following cancer treatment.37,40 Family communication support guiding patients in discussing genetic risks with relatives and children,37,40,43,47 coping strategies providing practical approaches to managing the psychological impact of diagnosis and treatment,38,45 and body image and sexuality support helping patients navigate changes in self-perception and intimate relationships post-treatment.37,51 These support functions showed preliminary positive signals in user-reported psychological wellbeing, with users developing emotional engagement with chatbots,45,50 reducing chemotherapy-related symptoms, 38 and improving medication adherence from 51% in week 1 to 76% in week 540. An emerging trend toward personalised support features was identified, with 5 chatbots incorporating individualized guidance based on patient-specific factors and clinical contexts. Users particularly valued the 24/7 accessibility of these systems, considering them reliable complementary sources of information rather than replacements for healthcare provider interactions. 45
Evaluation outcomes of chatbot applications
Satisfaction, usability, feasibility, and efficacy
Chatbots in breast and ovarian cancer care have been evaluated across satisfaction, usability, feasibility, and efficacy. 5 studies quantitatively assessed satisfaction. In the pilot and RCT studies, Al-Hilli et al. 36 compared Gia with face-to-face genetic counseling among 37 newly diagnosed breast cancer patients, finding no significant difference in median satisfaction scores between groups (30 vs. 30, p=0.19). Chou et al. 51 recorded a mean satisfaction score of 5.43/7 regarding the ovarian cancer-related information provided by the chatbot in a study with 31 ovarian cancer patients. High satisfaction levels were similarly observed in larger cohort studies. Chaix et al. 40 reported a satisfaction rate of 93.95% among 958 breast cancer patients, with 88% acknowledging effective support from the chatbot. Nazareth et al. 41 reported a mean satisfaction score of 4.6/5 across 61,070 users undergoing hereditary cancer risk assessment. Bibault et al. 37 also reported that 83.1% of 142 breast cancer patients found the chatbot’s responses helpful when compared to a multidisciplinary board comprising surgical, medical, and radiation oncologists. Regarding usability, 4 studies reported that patients found the chatbot interface intuitive and easy to use, regardless of their technological literacy.38,43,45,48 Even those with minimal exposure to digital tools reported being able to navigate the systems successfully, with many completing their engagements without issues. For instance, in the study by Chetlen et al., 48 74.5% of users indicated a better understanding of medical procedures through chatbot interaction. Regarding feasibility, Sato et al. 50 conducted a pilot study to evaluate the feasibility of using chatbots to collect family history information for HBOC, finding a high 100% completion rate and an average task time of 18 minutes.
Efficacy was evaluated across four RCTs, with two focusing on clinical outcomes and two employing non-inferiority trials. Tawfik et al. 38 assessed the chatbot’s impact on chemotherapy-related side effects among breast cancer patients, finding significant reductions in the frequency, severity, and distress associated with physical and psychological symptoms, alongside notably higher self-care efficacy compared to the control group. In contrast, Lee et al. 39 enrolled breast cancer patients undergoing radiotherapy and found no significant improvement in overall anxiety levels. However, their subgroup analysis revealed younger patients (aged ≤50) showed significantly reduced anxiety, suggesting age-related differential effects. The remaining two studies employed non-inferiority designs to evaluate chatbots against conventional clinical consultation. Al-Hilli et al. 36 found patients receiving genetic counseling via chatbot demonstrated no significant differences in knowledge comprehension compared to those counseled in person by a genetic counselor, with a genetic testing uptake rate of 100% in both groups. Similarly, Bibault et al. 37 demonstrated that chatbot was non-inferior to a multidisciplinary oncology board in delivering breast cancer-related information, with comparable success rates (69% vs. 64%, p<0.001).
Performance and information quality
Information quality was formally assessed in 8 studies. High concordance was demonstrated in genetic counseling applications36,50 and screening assessments, 22 while treatment planning applications showed variable results with congruence rates ranging from 70% 42 to 16.05% 49 when compared with professional team recommendations. The evaluation methods for information quality varied considerably across studies, with few reporting validated measurement tools. Only a small number of studies employed standardised assessment instruments, such as 5-point satisfaction rating scales 37 and knowledge assessment questionnaires. 36 Most studies relied on custom evaluation frameworks, including specialist comparison protocols, 22 tumour board decision concordance metrics,42,49 genetic counselor validation protocols, 50 and technical performance indicators such as fallback rates.43,45
Challenges in chatbot implementation
Technical challenges were identified in 9 studies, primarily concerning chatbot comprehension abilities and language accessibility. Two studies specifically noted that similar wordings could lead to misinterpretation of essential information.22,43 Language accessibility emerged as another significant barrier, particularly evident in one U.S. study where English-only support limited effectiveness in Spanish-speaking populations. 41 Two studies reported insufficient database support in recognizing specific medical terms,37,49 while three studies noted limitations in accurate diagnosis and screening due to incomplete patient information collection through conversations.42,49,50 Additionally, clinical integration challenges focused on human-chatbot interaction limitations. Studies identified insufficient capabilities in establishing emotional connections with patients 45 and addressing complex, personalised issues. 38 Multiple studies emphasized that current chatbot systems should complement rather than replace healthcare professionals.37,38,50 Methodological constraints were also noted, including small sample sizes and the need for standardised evaluation frameworks,36,37 suggesting opportunities for more robust research designs in future studies.
Additionally, two qualitative studies provided insight into patient perspectives on chatbot adoption. Wollney et al. 46 interviewed breast cancer patients and found that clinician endorsement of the chatbot’s relevance during genetic counseling sessions served as a key facilitator. Similarly, Siglen et al. 47 conducted qualitative interviews with 7 breast and ovarian cancer patients and identified perceived accessibility as the primary driver of willingness to use the chatbot. However, both studies revealed several barriers. Patients expressed concerns regarding the privacy and security of genetic information, particularly the risk of data misuse or unauthorized disclosure. 46 A lack of personal risk perception also emerged as a significant barrier, as individuals who did not consider themselves at high risk found it difficult to recognize the tool’s relevance to their own circumstances. 46 Information clarity represented another prominent concern, with patients worried both about misinterpreting chatbot-generated information and about the chatbot’s limited capacity to comprehend the complexity of their individual queries. 47 Notably, patients frequently posed questions beyond the chatbot’s intended scope, such as those related to treatment and prognosis, resulting in lower response accuracy. This finding underscores the importance of clearly communicating the chatbot’s functional boundaries to patients prior to implementation, so that expectations are appropriately managed from the outset.
Discussion
Principal findings
This scoping review identified 19 studies that investigated the use of chatbots in breast and ovarian cancer care. The majority of chatbot applications targeted breast cancer and addressed three functional domains: clinical support, educational support, and psychosocial support. NLP remained the primary technical foundation, with a growing adoption of LLM-driven and RAG-based approaches. Our analysis revealed consistently high user satisfaction and usability across studies, while RCT evidence on clinical efficacy showed mixed results. The studies also highlighted important limitations in chatbot capabilities, adoption barriers identified through qualitative research, and gaps in implementation evidence.
This review found that current chatbots incorporate clinical, educational, and psychosocial support functions. Early healthcare chatbots were predominantly single-function tools focused on basic patient education, symptom checking, and health information delivery.29,54 Current systems showcase comprehensive integration of multiple care domains. This integration particularly excelled in high-stakes clinical domains, such as genetic counseling and risk assessment, where chatbots showed potential in identifying high-risk HBOC populations. 41 Furthermore, the reviewed cancer-specific chatbots exhibited advanced personalization capabilities, incorporating context-aware support features that adapt to individual patient circumstances and clinical requirements. This evolution towards complex, multi-functional systems represents a notable trend in chatbot applications, though it also highlights the need for more robust evaluation frameworks to assess the feasibility and efficacy of these integrated functions in real-world clinical settings.
Importantly, while breast cancer chatbots show diverse applications, there is a notable absence of chatbots specifically designed for ovarian cancer care, revealing a significant gap in the field. This finding is consistent with Cheung et al., 55 who also found that AI and ML technologies for ovarian cancer care significantly lag behind other oncology areas. This gap may be explained by two key challenges. On one hand, the majority of patients are diagnosed at an advanced stage, with approximately 75% presenting at stage III or IV. 56 At this point, patients face urgent surgical decisions, complex chemotherapy regimens, and palliative care needs that go far beyond what a general health education chatbot can offer. On the other hand, ovarian cancer’s high molecular and histological heterogeneity, both across subtypes and within individual tumours, makes it exceptionally difficult to build a unified information framework or a consistent conversational pathway for patients. 57 This complexity also severely limits the availability of early-stage training data for AI development. Despite these challenges, the unmet need for digital support among ovarian cancer patients should not be overlooked. Compared to other cancer types, ovarian cancer patients report a higher proportion of unmet psychological and systemic support needs, 58 and face unique demands such as managing treatment-induced symptoms, making fertility preservation decisions, and navigating complex surgical options. Future chatbot research could therefore focus on guiding patients through post-treatment recovery, supporting health literacy and follow-up understanding, and raising awareness of early symptoms such as persistent bloating and pelvic pain, which are frequently overlooked and contribute to diagnostic delay. Developing specialized ovarian cancer chatbots represents a critical opportunity to address these specific care gaps and enhance patient outcomes.
The dialogue system and AI techniques used in chatbots varied across different care tasks, with rule-based and retrieval-based chatbots predominating in clinical tasks and generative chatbots in patient educational and psychosocial support. Fujihara and Sone 59 observed similar patterns in diabetes care, where deterministic algorithms and rule-based systems were preferred in medication decisions and risk assessment, while ML and neural network approaches were more commonly applied in patient self-management tools and continuous monitoring. Gupta et al. 60 further noted this differentiation, finding that while NLP-enhanced systems excel in processing unstructured clinical data and facilitating patient-provider communication, rule-based systems with clear decision trees are often favored for critical clinical decisions due to their predictability. Rule-based systems excel in providing the transparency and reliability needed for high-stakes clinical decisions, particularly in areas like genetic counseling and risk assessment, where clear and predictable pathways are essential. In contrast, more sophisticated AI technologies are increasingly utilized in patient psychosocial support, where the ability to handle natural language and adapt to contextual nuances enhances the patient experience. For instance, generative AI could play a transformative role in personalised health education or emotional support during critical phases of cancer care, including treatment decisions, perioperative periods, and remission. More importantly, the latest development of digital twins and multimodal chatbot driven by predictive analytics could revolutionize patient care by developing real-time virtual models of patients/caregivers. Ultimately, these advancements could further provide 24/7 continuous information or emotional support across the entire oncology patient journey, bridging the gaps in patient care delivery and improving long-term survivorship outcomes.
The RCT evidence in this review revealed mixed efficacy outcomes. While some trials demonstrated significant reductions in symptom burden 40 and non-inferiority to conventional counseling,36,37 others found no overall improvement in anxiety levels, with benefits limited to specific subgroups such as younger patients. 39 These inconsistent findings suggest that chatbot effectiveness may be moderated by patient characteristics, and highlight the need for larger, well-powered trials with stratified analyses.
Beyond technological considerations, the implementation of chatbots in cancer care presents distinct ethical challenges. Privacy protection in genetic counseling applications and the handling of sensitive cancer-related data require particular attention. The limited geographical diversity in current research, with predominant representation from Western countries, raises concerns about potential algorithmic bias when these systems are deployed across different populations with diverse cultural backgrounds. This underrepresentation is particularly concerning as cultural factors significantly influence both healthcare communication preferences and cancer care approaches. Additionally, the emergence of LLM-driven chatbots introduces specific risks related to misinformation and hallucination, where the system may generate plausible but clinically inaccurate responses. 61 In oncology settings, such errors could directly influence patient decision-making regarding treatment or genetic testing, potentially leading to delayed diagnoses or the pursuit of clinically inappropriate interventions. While RAG-based approaches have been proposed to mitigate hallucination by grounding responses in verified sources,44,53 their reliability in high-stakes clinical contexts remains insufficiently validated. RAG systems in oncology also risk exposing sensitive patient data through external knowledge retrieval unless privacy-preserving mechanisms such as differential privacy or on-premise architectures are adopted. 62 Furthermore, integrating large external knowledge bases introduces latency that can impede real-time clinical use, while retrieval quality across heterogeneous data sources remains inconsistent, undermining accuracy and governance in oncology applications. 63 Moreover, health equity concerns also warrant further attention, as current chatbot implementations predominantly serve English-speaking populations in high-income countries, potentially widening existing disparities in cancer care access for underserved communities. The digital divide and limited AI literacy act as compounding determinants of this inequity, as restricted broadband access, low device ownership, and unfamiliarity with AI-driven tools can systematically exclude vulnerable patients from the benefits of these technologies. 64 Moreover, although high satisfaction levels were reported in both small-scale pilot studies and large-scale implementations, it is important to note that these results may not be directly comparable. Small-scale studies often have limited generalisability, and future research involving broader populations or larger sample sizes is needed to confirm the consistency of these findings. Furthermore, chatbots’ limitations in handling complex emotional support highlight the critical need to maintain appropriate boundaries between automated and human-delivered care. These ethical considerations underscore the importance of developing clear guidelines for chatbot implementation in clinical oncology care, particularly regarding data privacy, algorithmic fairness, and the scope of automation in personalizing patient support.
Limitations
This review has several limitations worth noting. First, although we conducted a comprehensive search across multiple databases, we only included English-language publications, which may have led to the exclusion of relevant studies published in other languages. Second, the geographical distribution of included studies showed a predominance from Western countries, with limited representation from Asian and African regions. This geographical disparity may stem partially from our English-language inclusion criterion. Third, the heterogeneity in outcome measures and evaluation methods across studies made it challenging to directly compare effectiveness across different chatbot implementations. Fourth, in line with scoping review methodology, no formal quality appraisal was performed. This may limit the ability to evaluate the reliability and validity of the reported findings. Future systematic reviews and meta-analyses are warranted to further validate the effectiveness of these findings.
Future directions
Based on our review findings, several key areas warrant further investigation. Future research should prioritize the development of more sophisticated chatbot designs that can better handle complex clinical scenarios and emotional support needs, particularly in genetic counseling and treatment decision-making processes. 65 Standardised evaluation frameworks need to be established to enable meaningful comparisons across different chatbot implementations and to assess their long-term impact on patient outcomes. More rigorous study designs, particularly RCT with larger sample sizes and longer follow-up periods, are needed to establish the clinical effectiveness of chatbot interventions. 66 Future studies should also focus on implementing chatbots across larger and more diverse geographical samples to enhance the applicability and generalisability of research outcomes. 41 Additionally, there is a need to develop and validate multilingual chatbot systems that can serve diverse patient populations and address cultural-specific needs in cancer care. Importantly, future research should explore the optimal balance between chatbot automation and human oversight to ensure safe and effective deployment in clinical settings.
Conclusion
This scoping review mapped the landscape of chatbot applications in breast and ovarian cancer care, revealing integration of clinical, educational, and support functions with generally favorable user satisfaction. The implementation pattern shows rule-based systems dominating critical decision-making areas while advanced AI approaches serve patient support roles. Despite promising breast cancer applications, a notable absence exists in ovarian cancer-specific chatbots. Key future priorities should include developing specialized applications for ovarian cancer care, establishing standardised evaluation frameworks, enhancing geographical research diversity, and optimizing the balance between chatbot automation and human oversight in clinical settings. Ultimately, although chatbots show promise as supportive tools, they should be viewed as augmenting rather than replacing human healthcare providers, and integrated with appropriate human oversight to ensure patient safety and care quality.
Supplemental material
Supplemental material - Leveraging AI chatbots in supporting breast and ovarian cancer patients: A scoping review
Supplemental material for Leveraging AI chatbots in supporting breast and ovarian cancer patients: A scoping review by Vivian Hui, Lidan Tian, Xinyu Feng, Xiaoling Yuan, Janelle Yorke and Young Ji Lee in Digital Health.
Footnotes
Acknowledgements
The authors acknowledge the support of the Hong Kong Jockey Club STEM Lab for Digital Oncology Care Enhancement (DOCE).
Ethical considerations
Ethical approval was not required for this scoping review as it analyzed only publicly available published literature and did not involve human participants.
Consent to participate
This scoping review did not involve human participants.
Consent for publication
This scoping review does not contain any individual person’s data, images, or videos.
Author contributions
VH: Conceptualization, Methodology, Formal analysis, Writing – original draft, Writing – review & editing, Supervision, Project administration. LT: Conceptualization, Data curation, Formal analysis, Investigation, Writing – original draft, Writing – review & editing. XF: Data curation, Investigation, Validation, Writing – review & editing. XY: Methodology, Data curation, Writing – review & editing. JY: Supervision, Writing – review & editing, Funding acquisition. YJL: Methodology, Supervision, Writing – review & editing.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Hong Kong Jockey Club STEM Lab for Digital Oncology Care Enhancement (DOCE).
Declaration of conflicting interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Data Availability Statement
All data generated or analyzed during this study are included in this published article and its supplementary information files.
Supplemental material
Supplemental material for this article is available online.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
