PS49: Agentic AI and LLMs for Biostatisticians: Pros and Cons of an Emerging Technology in a Transdisciplinary Field

The biostatistical profession is navigating a profound productivity paradigm shift within a rapidly changing digital landscape. As the field transitions from traditional statistical software to reasoning large language models (LLMs) and agentic artificial intelligence (AI), the role of the biostatistician must evolve with it. This session will explore the pros of increased efficiency and innovation alongside the cons of technical reliability and potential social penalties associated with AI adoption. Speakers from industry, academia, regulatory bodies, and the social sciences will provide insights on how to foster impactful transdisciplinary decision-making while preserving the statistical rigor essential for regulatory sciences. We will cover advancements in agentic AI and context engineering frameworks to demonstrate how biostatisticians can utilize self-improving approaches to automate quotidian workflows and improve efficiencies. We will also discuss recently published results of a cross-sectional survey highlighting the experiences of over 50 biostatisticians using LLMs. Related, we will present and discuss the impact of LLM-based applications on various biostatistical tasks from a recently published article, including eight proposed guidelines for biostatisticians working with LLMs. To represent the non-statistical functions that biostatisticians collaborate with, we will incorporate perspectives from the social sciences to discuss recently published results on social evaluation penalties related to the adoption of AI tools. Lastly, we will incorporate FDA speakers to further illuminate how AI-related advancements impacting regulatory submissions can be developed through close interaction with regulatory agencies. Agentic AI and LLMs are already reconstituting core biostatistical workflows, shifting practitioners' time away from routine execution. Session attendees will leave with strategies for integrating agentic systems into their workflows that preserve statistical rigor, account for cross-functional viewpoints, and align with regulatory expectations.