Evaluation of AI-guided questioning versus standard information delivery in a simulated newly diagnosed breast cancer population.

Abstract

1623 Background: Patients newly diagnosed with breast cancer frequently experience information overload and anxiety, which can impair understanding and participation in shared decision-making. Artificial intelligence (AI)–guided educational approaches may support individualized information delivery; however, their impact on patient understanding and emotional response remains incompletely characterized. Methods: We conducted a prospective, randomized vignette-based survey study among adults (≥18 years) asked to imagine receiving a new diagnosis of early-stage breast cancer. Participants were randomized 1:1 to receive either standard written information or AI-guided questioning that adapted presentation of identical content based on stated preferences and level of detail. Outcomes included a composite breast cancer knowledge score (range 0-20) (primary endpoint), perceived information overload, anxiety, decision confidence, and clarity (all Likert-scale measures, range 1-5). Group comparisons were performed using Welch's two-sample t-test. Exploratory subgroup analyses included educational level and health-related background. Results: 102 participants completed the survey (51 per group). Mean knowledge scores were higher in the AI-guided group compared with standard information (17.6 vs 13.2, p=0.00022). Participants receiving AI-guided questioning reported lower information overload (2.31 vs 3.35, p=0.00012), lower anxiety (3.67 vs 4.16, p=0.031), greater decision confidence (4.06 vs 2.84, p=1.2×10⁻⁶), and higher perceived clarity (4.20 vs 3.18, p=2.9×10⁻⁵). Exploratory analyses showed greater benefits among participants without graduate education or health-related background (both p<0.001), with smaller, non-significant differences in those with higher education or healthcare experience. Conclusions: In a simulated newly diagnosed breast cancer population, AI-guided questioning was associated with improved understanding, reduced information overload and anxiety, and greater decision confidence compared with standard information delivery. Exploratory analyses suggested that benefits were more pronounced among participants without graduate-level education or medical background, supporting a potential role for human-centered AI in equitable patient education. Given exploratory nature and unequal subgroup distributions, confirmation in prospectively stratified studies is needed. Outcomes by information delivery approach. Outcome (range) Standard information (n=51) AI-guided questioning (n=51) p value Knowledge score (0-20) 13.2 17.6 0.00022 Information overload (1-5)† 3.35 2.31 0.00012 Anxiety (1-5)† 4.16 3.67 0.031 Decision confidence (1-5)‡ 2.84 4.06 1.2×10⁻⁶ Clarity (1-5)‡ 3.18 4.20 2.9×10⁻⁵ †Lower scores indicate better outcome. ‡Higher scores indicate better outcome.