RENata: Development of a GPT-based Methodological Assistant for Structuring Renal Carcinoma PDTA in Romagna

The integration of AI in healthcare is transforming clinical workflows, particularly through the application of LLMs. This dissertation presents the development of RENata, a custom GPT-based methodological assistant designed to support the drafting of a Percorso Diagnostico-Terapeutico Assistenziale (PDTA) for renal carcinoma in Romagna. The project, conducted in collaboration with the Data Unit team of IRCCS IRST, aimed to create an AI-driven tool that ensures adherence to national and regional guidelines while facilitating multidisciplinary collaboration. The study follows a systematic methodology involving literature review, clinical guideline analysis, AI model training, and prompt engineering techniques. The GPT system was fine-tuned using prompt engineering techniques to enhance accuracy, reduce hallucinations, and ensure structured outputs. The development process included iterative feedback from clinicians, refining RENata’s responses and improving domain specificity. Internal testing demonstrated RENata’s readiness to be employed in a working setting. However, limitations such as the model’s reliance on a static knowledge base and the absence of real-time medical data integration were identified. Future work will focus on dynamic updates, API integrations with medical databases, and further evaluation in working clinical settings. The project highlights the potential of AI in streamlining PDTA drafting, optimizing healthcare workflows, and supporting decision-making, while reinforcing the need for human oversight in AI-assisted medical documentation.