1185 - Rad Onc Smart Learn: Empowering Radiation Oncology Trainees through a Personalized AI-Driven Educational Copilot

05:40pm - 05:45pm ET

Room 254

Presenter(s)

Abass Conteh, MD, PhD - Stanford Healthcare, Palo Alto, CA

A. M. Conteh¹, J. S. Chiang¹, C. Parikh², A. Bhandarkar³, and M. K. Buyyounouski⁴; ¹Department of Radiation Oncology, Stanford University, Stanford, CA, ²Department of Medicine, Greenwich Hospital, Greenwich, CT, ³Department of Radiation Oncology, Mayo Clinic, Rochester, MN, ⁴Stanford University, Stanford, CA

Purpose/Objective(s): The ABR Radiation Oncology Certifying Oral Exam is historically difficult, first-time pass rates have recently declined, and preparation is limited by variable access to structured practice. Rad Onc Smart Learn (RSL) is a web-based platform featuring AI-enabled case-based mock oral examinations with adaptive follow-up questioning, optional speech-to-text responses, and rubric-aligned feedback. We conducted a pilot study to describe feasibility, usability, and self-efficacy responses among pilot users.

Materials/Methods: We conducted a prospective, descriptive feasibility pilot of RSL. In-app surveys were administered before and after mock oral sessions linked to participant pseudonyms. Eligible participants were radiation oncology resident pilot testers (n=7). Measures included self-reported preparedness, confidence in clinical knowledge and oral communication under time pressure, knowledge gap clarity, usability (Single Ease Question [SEQ] and UMUX-Lite), perceived performance change, and likelihood of future use and recommendation. Participants without a matched post-session survey were included in feasibility counts but excluded from pre-post comparisons. Responses were summarized descriptively; free-text responses were thematically grouped.

Results: Seven participants completed 11 pre-session and 7 post-session surveys across 6 disease sites. Baseline barriers most commonly included time constraints, lack of feedback, and uncertainty about what to study. Among matched pre-post pairs (n=6), median preparedness increased from 1 to 3 (scale 1–5) and median oral communication confidence from 2 to 3. Post-session, respondents rated the platform easy to use (SEQ median 7 [IQR 6–7]) and 86% agreed the system fit their workflow. All post-session respondents (100%) perceived performance as better or much better versus before RSL; none reported no change or worsening. Median likelihood of reuse and recommendation to a co-resident were both 5/5. Commonly requested improvements included greater case and question diversity within disease sites.

Conclusion: In this small pilot, RSL demonstrated feasibility within resident workflows with favorable preliminary signals in preparedness, confidence, and satisfaction. Floor level baseline scores and absence of a control arm limit causal interpretation. These findings support RSL as a promising, scalable adjunct to board preparation that may broaden access to structured, feedback-rich oral practice. Larger evaluation with objective performance metrics and a comparator arm is warranted.