Leveraging large language models to enhance regional genomic medicine econsults

Care Progress LLC received a $400,618 Project Grant from the National Human Genome Research Institute (NHGRI) under the Human Genome Research program (CFDA 93.172) for research commencing September 20, 2025 and concluding August 31, 2026. The award funds development of an artificial intelligence (AI) assistant tool that leverages large language models (LLMs) to enhance guideline-concordant genetic e-consultations for primary care physicians (PCPs) and clinical genetics professionals. The tool addresses the critical gap between complex clinical genetics practice guidelines and PCPs' limited time and expertise to determine when genetic testing is warranted or which specific tests to order, thereby improving clinical decision-making and patient outcomes.

The project encompasses two primary aims: (1) developing extensible methods to digitize clinical practice genetics guidelines for integration with LLMs to enable semi-automated e-consultation recommendations, including automated assessment of patient eligibility for specific genetic tests and genetics consultations; and (2) demonstrating usability of a prototype web-based chatbot through user-centered design principles, including recruitment of PCPs for usability reviews and feedback collection. This approach offers superior scalability compared to traditional clinical decision support rules, as it automatically adapts to guideline updates without requiring manual reconfiguration, ensuring PCPs maintain current, evidence-based practice standards in genomic medicine delivery.

Abstract

Primary care physicians (PCPS) have limited genetics knowledge, including knowing if and when to order genetic/genomic testing and which tests to order. While genetics and other specialty societies have created robust clinical practice guidelines to define the appropriate genetic evaluation for defined indications, they are typically complex and PCPS lack sufficient time or expertise to assess the need for genetics consultation or testing for a specific patient. Further, given the ever-increasing and evolving medical knowledge base, PCPS lack time to stay current with guideline updates. As a result, PCPS may make clinical decisions that are not fully informed and result in substandard care, with potentially suboptimal clinical outcomes. To address these issues, we propose to leverage large language models (LLM) to create an AI assistant tool to help PCPS and clinical genetics professionals enhance guideline-concordant genetic econsultations. Our approach is superior to creating clinical decision support rules, which have not been robust and must be manually updated whenever guidelines change. Aim 1 focuses on developing methods to combine large language models with clinical practice genetics guidelines for quality, semi-automated econsult recommendations. In doing so, we will develop extensible methods for digitizing guidelines so that llms can access them to determine if a given patient meets criteria for ordering a specific test, whether a genetics consultation is warranted and other tasks. Our second aim is to demonstrate the usability of a prototype among PCPS. Using user-centered design principles, we will build a web-based chatbot prototype of our innovation. We will then recruit PCPS and perform usability reviews, including administering a usability questionnaire. We will also obtain qualitative feedback from PCPS. If we are successful, we will demonstrate the feasibility of our tool to enable PCPS to make informed clinical decisions at the point of care, including ordering genetic testing, which will enhance econsultations with clinical genetics professionals, including reducing the time a patient has to wait for consultation and treatment, and ultimately improving clinical outcomes.