Grant Number:	1R21CA294008-01A1 Interpret this number
Primary Investigator:	Banerjee, Imon
Organization:	Mayo Clinic Arizona
Project Title:	A Clinical Data-Driven Digital Medical Assistant for Personalized Cancer Patient Support
Fiscal Year:	2026

PROJECT SUMMARY/ABSTRACT Prostate cancer remains the second-leading cause of cancer-related death in the United states. Patients report a significant decline in mental health immediately after the diagnosis. The situation is made worse by the stigma around discussion of prostate cancer symptoms especially sexual symptoms with healthcare providers. Under the stress and anxiety caused by the diagnosis, patients often turn to web search engines for information. Studies have shown information about prostate cancer online can be incorrect and misleading. Automated conversational agents pose a unique solution to this problem where patients can access information in a private setting, reducing the stigma around discussion about disease. However, only a limited number of such agents are designed for oncological care. Most oncology-focused patient-facing conversational agents provide a one-size-fits-all solution and have limited capacity to adapt to individual patient characteristics and clinical contexts, instead operating primarily under rule-based frameworks.They also have no access to patients’ clinical records, instead relying on patients or healthcare providers to provide relevant information. Advancements made in the field of natural language processing in terms of development of large language models (LLM) have given rise to general purpose conversational agents like ChatGPT, however, their use for dissemination of information about sensitive topics like medical care is highly inadvisable. A recent study on the use of LLMs specifically for cancer treatment information found about 35% LLM-generated responses to include treatment recommendation discordant with NCCN guidelines. We propose PCa-PDA, a clinical data driven personalized digital medical assistant designed specifically for prostate cancer patients which will have access to patients’ medical records stored in electronic health records (EHR). We have developed a domain specific LLM for prostate cancer (PCaLLM) using reliable and relevant clinical data and domain sensitive training strategies. Under the proposed project, we will build on PCaLLM by aligning it with instructional tasks and injecting knowledge of treatment guidelines for prostate cancer set by societies such as American Society of Clinical Oncology (ASCO) and National Comprehensive Cancer Network (NCCN). PCa-PDA will be built on top of this knowledge-injected PCaLLM by training it to learn from retrospective patient data to optimize its answers for best possible clinical outcome. PCa-PDA will personalize its answers to each patient based on clinical records of that patient. With the help of Mayo Clinic Center for Digital Health infrastructure support team, PCa-PDA will be designed as a FHIR-compatible tool and will be integrated with Mayo Clinic patient portal for a pilot study to assess its utility in a real-world scenario. This will also ensure possibility for wide-spread adoption as an assistive tool integrated with secure patient portals with FHIR store to provide reliable and clinically accurate conversational support to patients while observing industry standards of security and privacy. PCaLLM and PCa- PDA will be released under open-source license to support wide-ranging academic research.





None