ELISA: Easy-to-use, Live, Interactive Support Avatar

Patients starting new cancer treatment must absorb complex information (treatment schedule, medication, side effects, safety instructions), often while coping with a new diagnosis or progression. Recall averages around 60%, with lower rates after bad-news consultations, so information often needs repeating before treatment starts, adding to nurse workload.

Dutch cancer incidence is projected to rise from 118,000 (2019) to ~156,000 diagnoses/year by 2032, with ~1.4 million cancer survivors by then, compounding existing staff shortages. Existing digital tools (websites, videos) offer only static, non-personalized information, whereas LLM-based systems allow natural-language, personalized interaction.

ELISA pairs AI-generated educational videos with an interactive avatar restricted to physician-approved content, referring patients to their care team otherwise. This trial tests whether ELISA plus shortened nurse-led education yields non-inferior recall versus standard education, alongside patient experience measures.

Objective(s) The primary objective of this study is to determine whether AI-supported education consisting of AI-generated videos and an interactive, life-like avatar, in addition to shortened nurse-led education, is non-inferior to standard nurse-led education alone with respect to patients' recall of treatment information when starting a new line of anti-cancer treatment. The secondary objective is to assess patients' experience with and usage of the AI-supported patient education.

Study type A prospective, single-center, randomized controlled non-inferiority trial Study population Adult patients, 18 years or older, who have an indication to start a new line of anti-cancer therapy at Erasmus MC Medical Oncology Methods Eligible patients, after giving eConsent, are randomized 1:1 to the control or study group. The study group gets access to ELISA. Both groups will receive an educational session with an oncology nurse. Two days after the educational session both groups receive a digital version of the Netherlands Patient Information Recall Questionnaire (NPIRQ) to test information retention. The study group will receive an additional online Chatbot Usability Questionnaire (CUQ) one month after getting access to ELISA.

Burden and risks The additional burden for participants in participating in this study is minimal and consists of watching an informational video of less than 10 minutes and completing brief questionnaires, which will take no more than 40 minutes in total. The AI-generated educational videos are static and its content is based on the existing informational brochures for oncologic treatments, which are thoroughly reviewed by oncologists and nurses. The interactive, life-like avatar is based on a large language model (LLM). As with all LLM's, there is a small chance it may occasionally hallucinate, giving inaccurate information. To minimize this risks, information to answer participants' questions is limited to information from the existing informational brochures for oncologic treatments. The interactive, life-like avatar cannot independently generate or retrieve new information beyond the scope of this information. Furthermore, the interactive life-like avatar is explicitly prompted to not answer questions beyond the information provided, to never give medical advice and to not ask for personal information.

Recruitment and consent Eligible patients are identified and informed by their oncologists during routine clinic visits. Eligible patients, after giving eConsent are randomized 1:1 to the control or study group.