Artificial intelligence and the future of inflammatory bowel disease trial recruitment: from bottleneck to breakthrough

Recruitment for clinical trials has become one of the most expensive, time-consuming, and failure-prone steps in modern drug development. Across multiple fields, slow enrollment is a leading cause of trial delays, premature termination, and escalating costs, with downstream consequences for patients awaiting access to new therapies and for sponsors investing substantial resources in research and development.1,2 This is particularly true in inflammatory bowel disease (IBD), for which fewer than 1% of patients participate in clinical trials. Furthermore, growing competition among concurrent studies has intensified pressure on already limited recruitment pipelines.3,4 As trial eligibility criteria become more complex and clinician workload continues to rise, reliance on physician-driven identification of eligible participants is increasingly unsustainable. Against this backdrop, Gupta and colleagues have reported a prospective evaluation of a large language model-based (LLM) artificial intelligence (AI) system designed to improve screening efficiency for IBD clinical trials.5 LLMs are deep learning systems trained on vast text-based datasets to interpret unstructured clinical information and generate probabilistic outputs at scale.6 Gupta et al. applied an LLM-based AI system to analyze electronic medical record (EMR) data and screen for eligibility in 2 clinical trials.5 The system reviewed EMR data without prescreening or manual filtering and assigned probabilistic scores reflecting the likelihood that each patient met trial-specific inclusion and exclusion criteria. In the prospective phase, the system screened more than 7400 patients scheduled for upcoming clinic visits and identified 22 high-probability candidates for a phase 2a ulcerative colitis trial. During the same 4-week period, no patients were referred through traditional physician-driven pathways. Similarly, for a phase 3 Crohns disease trial, the AI system flagged 18 potential candidates compared with only 1 physician referral, which was also identified by the AI tool. Together, these findings illustrate the substantial gains in screening yield that can be achieved by shifting from clinician-dependent referral to systematic, AI-enabled population screening.