Large Language Models Across the Clinical Trial Lifecycle: A Scoping Review of Applications, Methodologies, and Validation Gaps

Abstract Large language models (LLMs) are increasingly being incorporated into clinical trial research, yet evidence of their utility is fragmented across isolated tasks, trial stages, model architectures, and evaluation strategies. This fragmentation makes it difficult to determine where LLMs provide meaningful value, where evidence remains preliminary, and what barriers limit their translation into routine clinical research. We therefore systematically characterized generative LLM applications across the clinical trial lifecycle, defined here as the interconnected stages of trial design, participant recruitment, trial conduct, and data analysis. We reviewed 71 studies published between January 2023 and August 2025 and examined their applications, technical approaches, and validation strategies. LLM applications were concentrated in Analysis (40.8%) and Recruitment (31.0%), followed by Design (18.3%) and Conduct (5.6%). Among all studies, 34 (47.9%) reported better performance than traditional machine-learning or previous language-model comparators, whereas 32 (45.1%) did not include such comparisons. Human-based evaluation was reported in 25 studies (35.2%), but prospective and operational validation remained limited. By mapping applications and evidentiary maturity across the full clinical trial lifecycle, this review identifies a substantial gap between rapidly expanding technical capabilities and evidence supporting real-world implementation. These findings provide a structured evidence base for prioritizing rigorous validation, transparent governance, and clinically meaningful evaluation as LLMs transition from experimental tools toward operational components of clinical research.