The Role of Generative AI for Scalable External Control Arms

Clinical trials are increasingly constrained by rising costs, recruitment challenges, and ethical concerns surrounding placebo use. External control arms (ECAs) — comparator cohorts constructed from real-world data (RWD) — could reduce these burdens, but their adoption remains limited. Each ECA requires substantial effort to curate data, map trial protocols, construct cohorts, and validate causal estimates. At the same time, advances in interoperable health data infrastructure, regulatory pathways, and causal inference are creating the conditions for more credible and scalable use of external evidence. We argue that generative artificial intelligence (AI) could fundamentally change the methodological and operational constraints of ECA development by connecting trial protocols with heterogeneous clinical data through clinical abstraction, eligibility and endpoint interpretation, and cohort generation. However, these advances do not eliminate threats to observational validity, regulatory requirements, or errors that can propagate through automated workflows. We therefore propose an evidence-grade framework for AI-enabled ECA construction comprising interoperable RWD, trial-ready variables, protocol-to-cohort translation, and mandatory causal and statistical validation. The central distinction is that generative AI can make scientifically appropriate ECAs operationally scalable, but it cannot make scientifically inappropriate ECAs valid.