Feasibility and Acceptability of Using Large Language Models for Reporting Health-Related Quality of Life Among Cancer Survivors
Thomas M. Atkinson · California Digital Library · 2023
Not cached yet. This record was fetched live for this request and is stored after the lookup. Reopening the page should serve the cached copy.
California Digital Library · California Digital Library · other · 2023
Abstract
There are currently over 18 million cancer survivors in the United States. Many of these individuals have comorbidities or other factors (e.g., mild neurocognitive impairment) that may be associated with poor health-related quality of life (HRQoL). Because cancer survivors frequently seek information through online resources, some individuals may use large language models (LLMs) such as OpenAI’s Generative Pre-Trained Transformer-3 (GPT-3) based ChatGPT to obtain information about cancer survivorship and potentially report their HRQoL.
LLMs are advanced artificial intelligence (AI) models that are designed to generate human-like text responses by predicting the most probable sequence of words based on input prompts. However, these models may generate and amplify medical misinformation, due in part to a known limitation in language models to sometimes “hallucinate” and generate plausible-sounding yet incorrect/nonsensical answers. Additionally, as the use of LLMs for medical question-answering is likely to grow exponentially, this increased use is associated with AI-based technology concerns about misinformation and privacy, particularly in protected health information (PHI).
There has been a rapid growth in the potential of LLMs in patient-facing applications. However, to date, most published articles in this area focus on the bioethics and potential pitfalls of LLM use. There is a need to create methodologic guidance that incorporates stakeholder views of appropriate LLM use and their essential qualities.
Before leveraging this rapidly evolving technology for self-reporting of HRQoL, it is crucial that we first conduct a methodologically rigorous study with cancer survivors to understand feasibility and acceptability of using LLMs for this purpose, concerns that they may have about using this as part of their current or future healthcare management, as well as features that they would like built into LLMs to improve the user experience.
We propose to enroll 50 adult, English-speaking cancer survivors, and randomize the order of their interaction with two LLMs in separate sessions; one created locally at Memorial Sloan Kettering Cancer Center (MSK), the other is the Azure OpenAI Service licensed to MSK (i.e., ChatGPT with the security, private networking, responsible AI content filtering, and enterprise support of Microsoft Azure). Semi-structured interviews will be conducted collect feedback regarding survivorship experiences.
Additionally, we will complete focus groups with a subset of 30 of these cancer survivors to better understand their concerns related to LLMs and views on desirable characteristics for future use of this technology in the healthcare setting. If we are successful, we will establish a framework to evaluate whether LLMs are also acceptable for broader uses, including patients currently undergoing cancer treatment. Moving beyond conventional psychometric assessment by incorporating unfiltered statements directly from patients and cancer survivors will allow for a better understanding of their experiences and better inform clinical decision making.
Open the resolved locator · 10.48321/d13633
Metadata licence: OpenAlex metadata (CC0)