Cancer patients' comfort with artificial intelligence in oncology care and first-round "AI chemo-teach" quality improvement initiatives.

e13652 Background: Artificial intelligence (AI)-powered tools are rapidly advancing oncology care, but patients’ acceptance of AI is crucial for patient-centered implementation. This study aims to analyze key factors that predict oncology patients’ comfort with various AI applications, including the effect of clinician oversight, and to apply these findings to an “AI chemo-teach" quality improvement (QI) initiative using AI-generated chemotherapy educational materials in adult patients with new gastrointestinal malignancies. Methods: This ongoing QI project initially used convenience surveying among adult cancer patients being seen at UCLA Health oncology clinics from October to December 2025. Data was collected regarding demographics, health literacy scores via BRIEF (a validated screening tool), familiarity with and attitudes towards AI, and comfort with AI-assistance in five theoretical use cases: administrative tasks, treatment decision support, and patient messaging as well as AI-generation of chemotherapy education materials (with and without clinician review of outputs). Comfort was assessed on a 10-point Likert scale. Statistical analyses included Kruskal-Wallis tests, Mann-Whitney U tests, and linear mixed-effects modeling. Results: Among 47 patients surveyed, most were male (51.1%), White (46.8%) or Asian (21.3%), and had a bachelor’s degree or higher (55.3%). Median age was 62 years, and 7 patients (14.9%) screened with low health literacy. Patient comfort was highest for AI-assistance with administrative tasks (6.53 ± 3.25), followed by treatment decision support (5.85 ± 3.22) and patient messaging (5.11 ± 3.07) (p = 0.067). Adding clinician review to AI generation of chemotherapy education materials significantly improved patient comfort by 3.19 points (from 4.19 ± 3.19 without clinician review to 7.38 ± 3.39 with clinician review; p < 0.001). Higher familiarity with AI (β = 0.40, p < 0.001) and positive attitude towards AI (β = 0.58, p < 0.001) were the strongest predictors of high comfort, and low health literacy significantly predicted poor comfort (p < 0.05). Age, sex, race/ethnicity, and education level were not significant predictors. Conclusions: Oncology patients’ comfort with AI in clinical care trended higher for lower-risk administrative tasks. Clinician review of AI-generated chemotherapy educational materials significantly improves comfort scores, and a similarly positive effect may be seen in other use cases. Familiarity and positivity with AI were dominant drivers of comfort, supporting a role for patient education about AI itself. Ultimately, AI integration in oncology care for low-risk clinical tasks for which providers remain “in-the-loop" are likely to be accepted by patients. Findings are informing the "AI chemo-teach" QI initiative for which the first plan-do-study-act cycle launched in December 2025.