February 27, 2024 | Mullai Murugan MS, Bo Yuan PhD, Eric Venner PhD, Christie M. Ballantyne, MD, Katherine M. Robinson PharmD, James C. Coons PharmD, Liwen Wang PhD, Philip E. Empey, PharmD, PhD, Richard A. Gibbs PhD
This study evaluates the effectiveness of an AI assistant developed using OpenAI's GPT-4 for interpreting pharmacogenomic (PGx) testing results. The AI assistant employs Retrieval Augmented Generation (RAG), combining retrieval and generative techniques, to provide contextually relevant and accurate responses. The Knowledge Base (KB) includes Clinical Pharmacogenetics Implementation Consortium (CPIC) data, and GPT-4 generates tailored responses through prompt engineering and guardrails. The AI assistant demonstrated high efficacy in addressing user queries, particularly in provider-specific queries requiring specialized data and citations, outperforming OpenAI's ChatGPT 3.5. Key areas for improvement include enhancing accuracy, relevancy, and language in responses. The study highlights the potential of generative AI in transforming healthcare provider support and patient accessibility to complex pharmacogenomic information, emphasizing the need for careful implementation and addressing ethical, regulatory, and safety concerns.This study evaluates the effectiveness of an AI assistant developed using OpenAI's GPT-4 for interpreting pharmacogenomic (PGx) testing results. The AI assistant employs Retrieval Augmented Generation (RAG), combining retrieval and generative techniques, to provide contextually relevant and accurate responses. The Knowledge Base (KB) includes Clinical Pharmacogenetics Implementation Consortium (CPIC) data, and GPT-4 generates tailored responses through prompt engineering and guardrails. The AI assistant demonstrated high efficacy in addressing user queries, particularly in provider-specific queries requiring specialized data and citations, outperforming OpenAI's ChatGPT 3.5. Key areas for improvement include enhancing accuracy, relevancy, and language in responses. The study highlights the potential of generative AI in transforming healthcare provider support and patient accessibility to complex pharmacogenomic information, emphasizing the need for careful implementation and addressing ethical, regulatory, and safety concerns.