Large Language Models in Clinical Trial Recruitment: Sociotechnical and Economic Framework Development Study

Large language models (LLMs) have shown substantial promise in patient-trial matching, but most published studies still evaluate the performance under controlled technical conditions rather than within real recruitment workflows. Less is known about how LLM-enabled clinical artificial intelligence (AI) systems should be embedded into organizational settings where privacy constraints, human oversight, patient-facing concerns, and governance costs shape deployment outcomes.

This study develops a theory-grounded conceptual framework for analyzing how LLM-enabled clinical AI can be integrated into clinical trial recruitment workflows and how such integration affects operational, governance, and economic outcomes.

Using structured conceptual analysis and targeted evidence synthesis, the study draws on recent literature on LLM-based patient matching, human-AI collaboration in clinical settings, and clinical AI governance. Sociotechnical systems theory and transaction cost economics are integrated to build the LLM-Embedded Clinical Recruitment Architecture (LECRA) and to derive 6 testable propositions.

LECRA conceptualizes recruitment as a closed-loop sociotechnical and economic system, spanning data complexity, model configuration and processing, human-AI collaboration, and economic and governance consequences. The revised framework identifies privacy constraints, hallucination risk, bias, patient trust, oversight intensity, and regulatory validation costs as key moderators of performance. It also reframes recruitment performance as a multidimensional construct and outlines an empirical roadmap for future testing.

LECRA offers a more deployment-sensitive account of when LLM-enabled recruitment is likely to create value and when the benefits may be offset by coordination, compliance, or trust-related frictions. This framework is intended to support future empirical studies and more realistic implementation decisions rather than to claim validated superiority of LLM-assisted recruitment.