P303: Artificial Intelligence (AI) Use Cases for Scientific Communications & Medical Information
Poster Presenter
Katie Moy
Associate Director, Medical Information & Review
Acadia Pharmaceuticals United States
Objectives
To explore the use of a generative AI chatbot in generating scientific communications and medical information content and enhancing processes.
Method
This case study was conducted using a GPT-4 series, generative AI chatbot to analyze data and generate text. Medical Information and Scientific Communications use cases were tested, including FAQ generation, medical review reference QC, and custom and standard response content development.
Results
The generative AI chatbot demonstrated several capabilities and limitations in our use cases. It was effective in identifying text information from large documents, summarizing data, paraphrasing text, and creating simple tables. These strengths are useful in finding and creating targeted summaries from lengthy reports and publications. However, the AI chatbot faced challenges in finding data from complex tables, compiling data from multiple sources, following existing document templates, creating figures, creating complex tables, and integrating into existing workflows. These challenges resulted in incomplete generated content and required additional follow-up and accuracy checks. The AI chatbot also demonstrated limitations in accuracy, reliability, and reproducibility, sometimes generating inconsistent results.
Conclusion
Generative AI has the potential to increase productivity and significantly enhance scientific communications and medical information processes. The AI can streamline creation of scientific communications and medical information content, potentially saving time and resources. However, there are challenges related to accuracy (e.g., hallucinations), integration in existing workflows, and reproducibility that need to be addressed. Ensuring accuracy and consistency of the final content is crucial, and a key limitation of generative AI. Trained industry professional effort cannot be replaced completely – human oversight is still required to review, edit, and manage the overall workflow and create usable content. Additional exploration in medical affairs and advancements in AI capabilities could improve its usability and effectiveness.
