Pharmaceut Med. 2026 Sep 6. doi: 10.1007/s40290-026-00635-9. Online ahead of print.
ABSTRACT
Clinical trial recruitment continues to fail at scale, with eligible patients often invisible to routine screening despite being present in electronic health records. Disease stage, treatment response, biomarker results, and clinical reasoning are frequently documented in narrative form rather than structured fields, placing eligibility-relevant information beyond the reach of conventional recruitment workflows. This narrative review examines applications of artificial intelligence (AI), and particularly large language models, across the recruitment pipeline. The evidence is organized around two ceilings. The discovery ceiling reflects the limits of identifying eligible candidates in clinical data, and current evidence suggests AI can raise it, improving discovery and screening efficiency in defined workflows. The enrollment ceiling reflects the human, logistical, ethical, and institutional barriers that persist after a patient has been identified. AI can make invisible patients visible, but visibility is not enrollment, and AI alone cannot remove these barriers. Upstream applications in trial design, predictive enrichment, site selection, and patient-facing engagement are also reviewed, together with governance requirements for equity, privacy, regulation, and workflow integration. Current evidence remains concentrated in retrospective evaluations, with limited prospective, multicenter validation and few demonstrated gains in enrollment, representativeness, or cost effectiveness. Recruitment AI is therefore best treated as a governed, human-supervised support tool that is locally validated and evaluated against downstream outcomes such as enrollment, diversity, burden, and cost.
PMID:42701994 | DOI:10.1007/s40290-026-00635-9