Using Natural Language Processing to Explore Patient Perspectives on AI Avatars in Support Materials for Patients...

Area of Science:

• Health Informatics
• Artificial Intelligence in Healthcare
• Patient Education

Background:

• Informed patients improve satisfaction, quality of life, and health outcomes.
• Traditional patient information methods (booklets, leaflets) are often ineffective.
• Generative AI offers a potential solution for efficient educational video production.

Purpose of the Study:

• To utilize natural language processing (NLP) for analyzing patient feedback on AI-generated educational videos.
• To assess patient responses to an AI-generated video about the breast cancer follow-up program.

Main Methods:

• Survey distributed to 400 breast cancer patients; 98 free-text responses analyzed.
• Applied NLP techniques: sentiment analysis, topic modeling (BERTopic), summarization, and TF-IDF word clouds.
• Evaluated effectiveness of NLP models in extracting insights from unstructured patient feedback.

Main Results:

• 81% of responses were positive or neutral; negative feedback focused on the AI avatar.
• Topic modeling identified key themes: treatment pathway, video content, AI avatar, and minimal content responses.
• Positive sentiment for treatment pathway and video clarity; negative sentiment for AI avatar impersonality.

Conclusions:

• NLP techniques successfully generated insights from patient feedback on generative AI educational content.
• Generative AI videos are a viable, cost-effective alternative to traditional patient education, though AI avatar design requires refinement.
• AI-generated content should supplement, not replace, human interaction in healthcare.