Patient trust in oncologists versus artificial intelligence for cancer-related questions.

Abstract

11105 Background: Artificial intelligence tools, including the recent introduction of ChatGPT Health, are increasingly accessible to patients seeking medical information. In oncology, where discussions surrounding prognosis and treatment are highly sensitive, trust in the source of information is critical. However, patient perspectives on AI tools compared with oncologists for cancer-related questions remain poorly characterized. Methods: We conducted a cross-sectional survey of 75 adult cancer patients at an academic safety-net oncology clinic. The survey assessed patients’ perspectives on technology use and their trust in AI tools like ChatGPT compared with oncologists for cancer-related questions. The primary outcome was patients’ preferred source of trust when asked who they would rely on most to answer a serious question about their cancer. Secondary outcomes assessed attitudes toward physician versus AI trust using Likert-scale items and comfort engaging with AI for cancer-related questions. Binomial proportions are reported with 95% confidence intervals. Spearman correlation and Fisher’s exact tests were used for secondary exploratory analyses. Results: Among 75 respondents, the median age group was 55–64 years; 42.7% were male and 57.3% female. English was the primary language for 53.4% and Spanish for 43.8%. The most common cancer types were gastrointestinal (24.0%), breast (20.0%), and hematologic malignancies (17.3%). For the primary outcome, when asked who they would trust most to answer a serious question about their cancer's future, an overwhelming majority selected their oncologist (97.3%; 73/75; 95% CI 90.7–99.7). No respondents selected ChatGPT/AI (0%; 95% CI 0–4.8), while 2.7% were unsure (95% CI 0.3–9.3). In secondary analyses, greater self-reported comfort with technology was associated with greater comfort engaging with AI tools for cancer-related questions (Spearman ρ = 0.40, p < 0.001). In exploratory subgroup analyses, Spanish speakers showed a trend toward less agreement that they would trust a doctor’s answer more than AI (71.9% vs 87.2%). This may reflect increased trust when medical information is delivered in a patient’s primary language. In addition, a similar directional signal was observed among younger patients (73.3% vs 84.4%). Conclusions: Cancer patients currently demonstrate a strong preference for oncologists over AI tools when addressing serious cancer-related questions, underscoring the central role of physician trust in oncology care. However, comfort engaging with AI varies by technological familiarity, suggesting that this preference may evolve over time as younger, more digitally fluent generations enter care and AI becomes increasingly integrated into healthcare. This study highlights the potential of AI tools to support oncologists in providing more accurate and individualized prognostic estimates for patients with advanced malignancies.