Optimizing Clinical Trial Screening with LLMs: A New Era in Healthcare Technology

Clinical trials are foundational to medical research, serving as the cornerstone of innovation and a pathway to improve patient outcomes. In this regard, integrating large language models and related AI systems into this process may prove transformative in addressing many challenges. Therefore, the focus of this study was to develop a clinical trial patient screening system to identify patients for COVID-19 clinical studies while bypassing traditional, time-consuming methods. To achieve this, we introduced automated extraction of eligibility criteria, mapping of medical entities to standardized codes (e.g., SNOMED, RxNorm, and LOINC), and an LLM-assisted query engine to improve patient recommendations. The present study reported a mean accuracy of 88.8% for entity extraction. Subsequently, in terms of standard code mapping, the embedding-based approach demonstrated reliable performance, achieving 97% and 90.25% retrieval accuracies for concepts and abbreviations, respectively. In the meantime, a strong funneling of patients with a percentage-wise match was created using LLM and rule-based query engines. Overall, this study demonstrates the potential of an end-to-end automated approach that leverages state-of-the-art AI to enhance the precision, scalability, and efficiency of identifying eligible patients; this research represents a substantial advance in clinical trial recruitment.