Introduction

Artificial intelligence (AI) is a general term used to describe a computer system modelling intelligent behaviour with minimal human intervention [1]. Prior work has investigated using subfields of AI in medicine to harness various types of data [2]. Applications include generative AI, which can generate new language or other data, and predictive AI, which can predict specific outcomes. AI approaches have been investigated to assist human-model interaction in many areas of medicine including surgical consultation, medical education, disease diagnosis, and pharmaceutical research and treatment [3]. The application of AI in oncology has grown substantially in recent years [4, 5]. Cancer is the second leading cause of death worldwide, and disproportionately affects individuals with restricted healthcare access, especially in rural areas and low- and middle-income countries [6–8]. AI has the potential to improve access to specialized care and address these health inequities, particularly by enhancing early detection and facilitating remote diagnostics [4].

Within cancer care, researchers have used AI to predict survival, for predictive supportive care needs, and to improve detection and diagnosis of cancer [9–12]. Specifically, AI algorithms have demonstrated high accuracy in identifying malignant lung nodules, assisting radiologists in improving early detection and treatment decisions [13]. In colorectal cancer, AI has been employed to detect polyps and other early-stage cancer indicators via colonoscopy and histopathology analysis [14]. In mammography, AI enhances breast cancer detection by improving radiologists’ performance and accuracy particularly for early-stage cancers [15].

To ensure these advances deliver patient-centered care, we must consider patients’ perspectives on how AI should be used in medical care [16]. Patients with cancer can experience a distinct psychosocial toll, given the extended duration, severity, and chronic nature of their illness [17, 18]. This experience may influence their attitudes on the use of AI in their cancer care that may be different from other medical care they receive. To date, a mixed methods systematic review explored patients from many fields of medicine and public attitudes toward AI [19]. The review found that patients and the public generally accepted the use of AI in their medical care but expressed concerns and preferred human supervision of such models. This review found some studies involving patients with cancer, including those finding that patients had a positive attitude toward AI in melanoma diagnostics in dermatology when AI preserves the human physician-patient relationship [20, 21]. Another study found that Chinese patients with cancer generally accepted the use of AI but preferred the oncologist over the AI when their opinions diverged [22]. This review provided an overview of patients and the public’s attitudes on AI in their care and briefly discussed this from the perspective of a patient with cancer, but the attitudes of patients with cancer were not the focus of the review or explored in depth. We also identified a narrative review examining patients’ perceptions of AI used in breast cancer diagnosis [23]. The review outlined positive patient sentiments toward AI improving diagnostic accuracy, but patients had concerns regarding the trustworthiness of an AI diagnosis and preferred that AI to complement the radiologist rather than replace them. These reviews only examined a few studies of AI in cancer care, leading to a need for a review to more comprehensively assess patients’ attitudes toward AI in cancer care beyond the fields of skin and breast cancer screening.

To address this gap, this protocol outlines a systematic scoping review that will collate prior work to understand the attitudes of patients with cancer toward AI in their care. Specifically, this scoping review will update and narrow the focus of Young and colleagues by including only patients with cancer [19]. Also, this review will expand the focus of Pesapane and colleagues by including patients with all cancer types and explore papers that address patients’ perspectives beyond only the use of AI in imaging for breast cancer screening [23].

We utilised a proposed conceptual framework for understanding how patients evaluate AI in healthcare to establish our search criteria [24]. According to this framework, we included the search terms: “experiences”, “beliefs”, and “attitudes” and added additional relevant search terms to ensure a comprehensive search. Our protocol seeks to establish a systematic scoping review that will be inclusive but rigorous by following widely adopted practices for a systematic scoping review.

Methods

The objective of this review is to map existing literature, identify knowledge gaps, and outline opportunities for future research. To achieve this, we will conduct a scoping review, which is particularly suited for mapping the breadth of available research and identifying gaps where evidence is limited [25]. Scoping reviews are also useful for clarifying definitions of key concepts in emerging fields, providing an overview of the research landscape, and highlighting areas where further study is needed [26, 27]. Additionally, they can facilitate stakeholder engagement by summarizing existing knowledge and guiding future research directions based on ongoing developments [27, 28]. We registered this protocol with the Open Science Framework (https://osf.io/8zph9).

The scoping review will follow a six-stage methodological framework and will report using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guideline (PRISMA-ScR) [26, 29].

Stage 2: Identification of relevant studies

Our eligibility criteria are outlined in Table 1 using the Population-Concept-Context framework. The following electronic databases will be searched: MEDLINE (OVID), EMBASE (OVID), PsycINFO (EBSCOHost), and CINAHL (EBSCOHost).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Eligibility criteria according to Population-Concept-Context framework.

https://doi.org/10.1371/journal.pone.0317276.t001

Stage 3: Search strategy and study selection

Following Joanna Briggs Institute’s (JBI) recommended three step search strategy, we conducted an initial limited search of MEDLINE (OVID) and EMBASE (OVID) [23]. Based on this initial limited search, relevant articles were selected and keywords in the title and abstract, along with relevant index terms were identified. These keywords and index terms were then used to develop a final search strategy and to conduct a literature search across the identified databases for this review. Furthermore, the reference lists of literature included in this scoping review will be searched for additional relevant studies. The final search strategy found in Table 2 was developed with support from subject librarians at the University of British Columbia.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. List of search terms.

https://doi.org/10.1371/journal.pone.0317276.t002

Study Selection: Using Covidence, at least two independent reviewers will screen the titles and abstracts of the articles identified by the literature search. Afterwards, the full text review of the relevant articles will be reviewed by the same reviewers and the reviewers will determine if the article meets our criteria. Our inclusion and exclusion criteria were determined using the Population-Concept-Context (PCC) framework [30]. Disagreements between reviewers will be resolved by consensus or by a third independent reviewer on the team.

Stage 4: Charting the data

A tabular chart will be developed by two independent reviewers to collect relevant information from the selected articles. Table 3 outlines a preliminary list of PCC [30].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Preliminary table of charting elements and associated questions for data.

https://doi.org/10.1371/journal.pone.0317276.t003

Discussion

AI is being investigated for a diversity of applications in cancer care. To ensure we use these tools in a patient-centered manner, it is essential to understand patients’ perspectives and attitudes toward AI in their treatment, especially as AI becomes a topic more readily discussed in the public sphere. Initial studies have found that patients both accept the use of AI in medical care but have concerns about it operating without physician supervision [36]. Given the distinct complexity of cancer care, including its impacts on psychosocial health and its combination of chronicity and severity, we believe a focus of this question on cancer care is warranted [37]. For example, patient perspectives may differ if AI tools are being used to influence treatment decisions that can directly impact their chances of surviving cancer, compared to being used to guide treatment decisions for a low-acuity condition. To our knowledge, this will be the first scoping review to map the existing literature on the attitudes of patients with cancer regarding AI in their medical care. We anticipate that this scoping review will reveal gaps in the existing literature and provide direction to better understand patient’s needs to guide clinical deployment of AI in cancer care. Based on prior research, we anticipate that cancer patients would generally express optimism about the use of AI in their care. However, we also expect heightened concerns regarding its unsupervised use, potential privacy issues, and specific themes related to end-of-life care, particularly whether AI might be employed to discontinue curative-intent treatments.

Our intention with this review is to include a broad set of studies in adult cancer care. We believe that deferring cancer care for children and adolescents is justified due to the unique characteristics of this population, including the distinct types of cancers they face, expected differences in technological familiarity and comfort, as well as variations in autonomy and their capacity to provide informed consent [38, 39].

Our search criteria are based on a framework proposed for examining patients’ attitudes regarding artificial intelligence in healthcare [24]. According to this framework, we included the search terms: “experiences”, “beliefs”, and “attitudes”. To ensure a comprehensive search, we included additional relevant search terms identified after an initial review of terms used in the literature that are similar.

A potential limitation of the novelty of this scoping review may be finding results that are similar to reviews conducted for general medical care. However, given the specific nature of cancer, we believe investigating whether themes are similar or different is warranted. Other potential limitations of this scoping review include excluding papers that are not in English, secondary research, and grey literature. We recognize that findings from secondary research could be relevant, but we chose to exclude these papers to avoid duplication. The exclusion of grey literature was necessary to keep the review practical. Despite these exclusions, we believe our decisions will make the review focused and manageable, contributing to clear and specific conclusions that can better inform future research, policy-making, and clinical practice.

Supporting information

Acknowledgments

The authors are grateful to the University of British Columbia librarian, Jane Jun, for her contributions to the search strategy.