ScreenGPT – Developing and Evaluating a Custom Large Language Model to Support Cervical Cancer Prevention
Angyal Viola Zsuzsanna
Health Sciences
Dr. Nagy Zoltán Zsolt
SE Semmelweis Szalon
2026-05-20 13:00:00
Public health
Dr. Ács Nándor
Dr. Dinya Elek
Dr. Tóth Zoltán
Dr. Molnár László
Dr. Maurovich Horvat Pál
Dr. Girasek Edmond
Dr. Kaposi Ambrus
As a summary, this study presents the successful development of a custom GPT-based model designed to support public health efforts in cervical cancer screening. The model was built by processing established medical guidelines and applying prompt engineering strategies using the OpenAI GPT-4 and GPT-4o within the Streamlit framework, utilizing Python for implementation. The system was deployed through the Streamlit platform and underwent a three-week beta testing phase involving 115 participants. Of these, 112 completed a user experience questionnaire, offering valuable insights into various aspects of system performance.
Furthermore, two independent medical practitioners assessed the system’s responses to the 30 most frequently asked questions for clinical accuracy. The findings underscore the potential of AI-driven tools to enhance cervical cancer screening efforts and address healthcare workforce limitations, especially in resource-constrained environments. By providing tailored, interactive health information, such systems can improve public health literacy and foster greater engagement in preventive care. A key strength of the approach lies in its capacity to distill complex clinical information into accessible, actionable guidance for diverse user populations. Moreover, the model’s ability to deliver empathetic and motivational dialogue, emulating interactions with a knowledgeable, supportive interlocutor, significantly contributes to user engagement and trust.
In line with existing literature, the results suggest that while AI is not a substitute for healthcare professionals, it serves as a valuable complement. These systems can optimize healthcare delivery by streamlining routine tasks and expediting access to reliable information, thereby enabling clinicians to dedicate more time to individualized, high-priority care.
Sustained user engagement is vital for the long-term success of digital health interventions. Future research should explore strategies to improve personalization, user-centered design, and the integration of interactive educational components aimed at encouraging continued usage and positive user referrals. Subsequent development phases can also aim to expand the model’s conversational scope to encompass other critical preventive screenings, including those for breast and colorectal cancers.
