The UCSF Health Problem

Patients with chronic liver diseases, including those with cirrhosis and those awaiting transplantation, have unique and often stringent nutritional requirements.  Malnutrition and sarcopenia are prevalent among this patient population, directly impacting clinical outcomes, transplant candidacy, and overall quality of life.  While clinical nutrition guidelines and registered dietitian consults are invaluable, standard recommendations frequently overlook the patient’s ethnic and cultural food preferences.  This lack of cultural tailoring can result in suboptimal adherence, missed opportunities to optimize nutritional intake, and diminished patient satisfaction.  Therefore, there is an unmet need to provide treating clinicians with culturally-appropriate, evidence-based, and patient-specific nutritional recommendations that resonates with that person’s dietary norms and preferences.

How Might AI Help?

We propose to leverage a Retrieval-Augmented Generation (RAG) approach powered by a Large Language Model (LLM) that references established nutrition and liver disease guidelines.  RAG is an approach in which LLMs are enhanced by integrating them with vetted external data sources, allowing them to generate more accurate and contextually relevant responses.  In addition to standard clinical and lab data that would be transferred via Fast Healthcare Interoperability Resources (FHIR) application progressing interface (API) calls, we will also leverage patient-reported cultural and ethnic background information to tailor nutritional recommendations.  Specifically, the generative AI approach would:

• Offer culturally appropriate recommendations

• By integrating a repository of traditional food items, cooking methods, and culturally specific dietary patterns, the LLM can suggest meal plans or dietary adjustments that align with both the patient’s clinical needs and cultural norms.

• Provide personalized nutritional guidance

• Patient clinical factors, such as liver disease severity, comorbidities (e.g., renal impairment), and body anthropomorphic measurements, will be gathered via FHIR calls and fed into the model to generate customized nutrition strategies. 
• Patient cultural/dietary preferences, such as cultural identification and dietary limitations, would be gathered through semi-structured forms entered by the treating clinician

• Conduct adaptive learning and continuous updates

• The LLM can continuously learn from new guidelines, emerging research, and aggregated nutritional data.

This solution has the potential to improve adherence to nutritional recommendations and enhance the overall patient experience working with the hepatology and liver transplantation teams.

How Would an End-User Find and Use It?

This RAG-LLM tool would be embedded as a specialized tab within the ambulatory encounters in the APeX EHR.  The tab will provide the end-user, defined as a clinical provider or dietitian, with a display of relevant clinical data along with standardized and free-text selections for the patient’s cultural and preferred nutritional preferences.  There would also be a “Generate” button, which will take the EHR information and user-inputted preferences, to produce personalized dietary recommendations.  This information could then be populated into the progress note and the patient instructions.

• FHIR-Based Data Retrieval and Processing

• The system will automatically retrieve relevant patient data (e.g., labs, medications, severity of liver disease, comorbidities) via FHIR calls.
• The system will also populate the patient’s demographics along with semi-structured input regarding the patient’s stated cultural background.

• Interaction with the RAG-LLM

• There will be a semi-structured box with selections for common dietary restrictions (e.g. “low-sodium,” “renal,” “nut allergy,” “lactose intolerant”) and free-text to pose specific nutritional questions (e.g., “What are culturally-appropriate protein sources for this patient who follows a South Asian diet and requires a low-sodium meal plan?”).

• AI-Generated Guidance

• The LLM synthesizes the patient’s clinical profile, cultural data, and evidence-based nutritional guidelines, returning a concise set of meal plans or tips that align with recognized best practices in liver disease nutrition.  This information could then be pushed into the progress note and patient instructions.

Example of AI Output

What Are the Risks of AI Errors?

Implementing an LLM for culturally-aware nutritional guidance introduces several potential risks: 1. Incomplete or Inaccurate Recommendations, 2. Overreliance on AI, 3. Bias or Cultural Misalignment, and 4. Privacy and Security Concerns (due to integrating demographic and cultural information into the RAG-LLM API pipelines).  To mitigate these risks, we will conduct rigorous validation, ensure continuous updates to the guideline library, and maintain a clear disclaimer that final decisions rest with licensed professionals.

How Will We Measure Success?

We will evaluate this solution using a mix of clinical, operational, and patient-centered metrics:

• Clinical Outcomes:  Caloric intake reported by the patient, nutrition labs (prealbumin, albumin), measurements of liver disease severity (MELD) before and after RAG-LLM deployment.
• Implementation Outcomes:  Feedback through patient surveys on cultural relevance, dietary adherence, and overall satisfaction with their nutritional care plan.  Tool utilization and time saved in created personalized meal plans.

Describe Your Qualifications and Commitment

• This project is led by Dr. Jin Ge, MD, MBA.  Dr. Ge is a transplant hepatologist and a data science and AI researcher within the Division of Gastroenterology and DoC-IT.  He serves as the Director of Clinical AI for the Division of Gastroenterology.  He has extensive experiences in building and deploying specialized LLMs using retrieval augmented generation (RAG), such as the LiVersa for liver diseases.  He has previously worked closely with both APeX-Enabled Research and the AI Tiger Team on various digital health projects.  If selected, he will commit at least 10% effort for 1 year towards this project to ensure its success.
• Jennifer C. Lai, MD, MBA is a Professor of Medicine, In Residence, in the Division of Gastroenterology.  She serves as the Director of the UCSF Health Advancing Research in Clinical Hepatology (ARCH), the research arm of the Division of Gastroenterology.  In addition, she is a board-certified Physician Nutrition Specialist (PNS).  She will be contributing her unique expertise and experiences at the intersection of nutritional sciences and transplant hepatology.
• Kathy Pariani, RD is a Registered Dietitian Nutritionist and Certified Diabetes Education Care and Specialist.  Kathy provides nutrition assessments, education and counseling to patients with chronic diseases, specifically those with liver disease, diabetes, obesity and cardiovascular risk factors.  Kathy previously practiced as an acupuncturist and herbalist before earning her Master’s degree in Nutrition and Dietetics from Bastyr University.  Kathy is also a current Fellow in the Evidence Based Fellowship Program at UCSF, researching effective nutrition interventions in patients with Metabolic Associated Steatotic Liver Disease (MASLD).