SAFE.AI: Developing and Testing an AI-based Hybrid Chatbot for Financial Empowerment in Rural Cancer Care

N/ANot Yet RecruitingNCT07410689

University of Utah60 enrolled

Overview

This is a randomized, two-arm, parallel-group pilot trial investigating a new chatbot tool designed to support cancer patients and caregivers, particularly those in rural communities. Approximately 60 participants will be randomized 1:1 to interact with either a hybrid chatbot or an AI-enabled chatbot. Participants will use their assigned chatbot to obtain clear and helpful information related to insurance, travel costs, and other financial aspects of cancer care.

Costs of cancer care will approach $246 billion by 2030, making cancer one of the most expensive health conditions for individuals. Cancer-related financial hardships negatively impacts psychological wellbeing, health-related quality of life, medication adherence, and decisions to delay or forgo care. Rural cancer patients and families have a higher prevalence of financial hardships, incur greater travel-related expenses, face unique employment and income stressors, and have lower access to specialized cancer care services and providers-- including those that support financial needs. Few financial toxicity interventions are designed for the needs of rural cancer patients and families. While financial navigation can effectively reduce cancer patients' out-of-pocket costs, cancer programs' financial and rural patient navigation services, including at the Huntsman Cancer Institute (HCI), are overstrained. Most centers respond to financial hardships reactively rather than proactively, and programs are less equipped to assist with non-medical sources of cancer costs, such as travel and employment hardships. To address this gap, Self-Advocacy for Financial Empowerment (SAFE) resource toolkit with a community advisory board consisting of rural cancer patients, caregivers, and healthcare stakeholders. Community-engaged research also identified the need for individualized and accessible information about financial resources and supports, stigma as a barrier to seeking help, and the time and resource-intensive nature of financial navigation that limits the penetration and reach of these essential services among rural communities impacted by cancer. Chatbots, or conversational agents, are a type of artificial intelligence (AI) system that applies machine learning to reproduce realistic human conversations. Scripted chatbots, based on clearly defined information boundaries, offer accurate, reliable, and individualized responses to questions. Conversely, AI-based chatbots that use large language models (LLM) like GPT4, can address ambiguous, open-ended questions while continuing to preserve privacy. Chatbots facilitate individualized, chunked information that enhances complex information communication, promotes users' privacy and support needs, and addresses workforce challenges.\[ GARDE-Chat, an open-source platform, has been established for health system-level risk assessment and genetic testing for hereditary cancer at HCI. GARDE-Chat supports scripted, hybrid, and AI-chatbots. Prior to this pilot test, GARDE-Chat will be used to create a chatbot designed to provide responses for financial toxicity, based on the SAFE toolkit content and verified resources to develop the scripted version of the chatbot. A large language model component of the chatbot will be incorporated for the hybrid version that will enable users to ask more complex and open-ended questions, refined with community stakeholder input. This is a randomized, two-arm, parallel-group pilot trial investigating a new chatbot tool designed to support cancer patients and caregivers, particularly those in rural communities. Approximately 60 participants will be randomized 1:1 to interact with either a hybrid chatbot or an AI-enabled chatbot. Participants will use their assigned chatbot to obtain clear and helpful information related to insurance, travel costs, and other financial aspects of cancer care. Enrolled participants will complete three surveys (pretest, posttest, and 2-week follow-up) and interact with the chatbot prior to the posttest. Participants can interact with the chatbot between the posttest and the 2-week followup as they choose. Participants will share feedback on the usefulness, ease of use, and overall experience using the chatbot and complete pre-and posttest measures to assess preliminary efficacy for secondary outcome measures. All research activities will be done online.

Conditions

Cancer Financial Toxicity

Interventions

Rule-Based ChatbotOTHER

The rule-based (scripted) SAFE.ai chatbot is a guided conversational tool built to provide structured, accurate, and consistent information to rural cancer patients and caregivers experiencing cancer-related financial toxicity. This chatbot is grounded in the Self-Advocacy for Financial Empowerment (SAFE) resource toolkit, which was co-developed with a community advisory board (CAB) composed of rural patients, caregivers, nurses, and financial navigation experts across HCI's five-state catchment area. All scripted responses reflect priorities identified during qualitative needs assessment sessions, ensuring that content is culturally aligned with rural patient experiences and real-world financial challenges. The chatbot follows a rule-based decision tree. Users progress through the conversation by selecting a response from a set of fixed options displayed on-screen. This ensures that all content is clinically vetted, safe, consistent, and aligned with evidence-based practices.

Hybrid ChatbotOTHER

The hybrid SAFE.ai chatbot builds on the existing rule-based system by integrating a large language model (LLM) layer to support more flexible, open-ended, and conversational interactions. While the rule-based chatbot provides structured conversations through predefined content, the hybrid approach allows users to ask complex or personalized questions about financial toxicity. To ensure safety and accuracy, the hybrid chatbot is not allowed to generate responses from the open internet. By combining the consistency of rule-based logic with the adaptability of an LLM, the hybrid chatbot will enable users to ask follow-up questions, describe nuanced financial situations, request clarification in their own words, and receive more tailored guidance while still ensuring adherence to SAFE content.

Eligibility

Sex: ALLMin age: 18 Years

Medical Language ↔ Plain English

Inclusion Criteria: * Adults (18 years and older) * Cancer patients or financially responsible caregivers of cancer patients who: * Reside in the Huntsman Cancer Institute's five-state catchment area (Utah, Idaho, Wyoming, Montana, and Nevada), * Are able to read and write in English, and * Live in a rural area, defined by endorsement of a residential ZIP code classified as non-metropolitan (RUCA codes 4-10) per the USDA Rural-Urban Commuting Area Codes. Exclusion Criteria: * Respondents who do not live within this 5-state region--The SAFE toolkit material was developed for the Huntsman Cancer Institute patient population which serves UT, ID, MT, WY, \& NV * Respondents who only speak Spanish or other exclusively non-English speaking groups--The large language model for the chatbot will be developed in the English language. As non-English language training for the chatbot is not part of the scope of this study, participants who are unable to read and write in English may not be appropriate. As such, we will not be recruiting participants who only speak Spanish or other exclusively non-English speaking groups. Future studies will include adaptation of the chatbot to other languages. * Caregivers who are not primarily responsible for the financial aspects of patients' cancer care--The topic of financial hardship of cancer care may be less relevant for caregivers who are not financially involved in care. * Non-rural dwelling

Locations (1)

Huntsman Cancer Institute/ University of Utah

Salt Lake City, Utah, United States

Outcomes

Primary Outcomes

Single-Item Helpfulness Rating

Helpfulness is assessed using a 5-point Likert-scale item with response options ranging from Very unhelpful (1) to Very helpful (5). Total scores range from a minumum of 1 to a maximum of 5, with lower scores indicating lower perceived helpfulness, and higher scores indicating greater perceived helpfulness.

Time frame: up to 2 weeks

System Usability Scale (SUS)

The SUS is a 10-item, 5 point Likert scale (0 = strongly disagree, 5 = strongly agree). Total scores range from a minimum of 0 to a maximum of 100, with lower values indicating less usability and higher values indicating higher usability.

Time frame: up to 2 weeks

Acceptability of Intervention Measure (AIM)

The AIM is a 4-item, 5-point Likert scale (1 = strongly disagree, 5 = strongly agree). Total scores range from a minimum of 4 a maximum of 20, with lower values indicating lower acceptability and higher values indicating higher acceptability.

Time frame: up to 2 weeks

Chatbot User Satisfaction (CUS) - Satisfaction Subscale

The Satisfaction subscale of the CUS measure is a 5 item Likert-scale ( 1 = strongly disagree, 5 = strongly agree). Total scores range from a minimum of 5 to a maximum of 25, with lower values indicating lower user satisfaction, and higher values indicating higher satisfaction.

Time frame: up to 2 weeks

Trust in Automated Systems Test (TOAST)

The TOAST is a 9-item 7-point Likert scale (1 = strongly disagree, 7 = strongly agree). Total scores range from a minimum of 9 to a maximum of 63, with lower values indicating lower trust, and higher values indicating higher trust.

Time frame: up to 2 weeks

Secondary Outcomes

Data from ClinicalTrials.gov

This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.