Accelerating Clinical Trial Recruitment Through Automated Eligibility Screening with Multi-Modal Deep Learning

Abstract

Clinical trial recruitment remains a critical bottleneck in medical research, with approximately 80% of trials experiencing significant delays due to inadequate patient enrollment. Traditional manual screening approaches require substantial time and resources while yielding suboptimal accuracy in matching patients to appropriate trials. This paper presents a novel multi-modal deep learning framework that integrates structured electronic health record data with unstructured clinical narratives to automate eligibility screening processes. The proposed architecture employs transformer-based encoders for clinical text processing, coupled with specialized neural networks for structured data analysis, unified through an attention-based fusion mechanism. Experimental validation demonstrates substantial improvements over existing methods, achieving 92.3% accuracy in eligibility prediction while reducing screening time by 73%. The framework successfully processes heterogeneous medical data sources, including diagnosis codes, laboratory results, medication histories, and physician notes, enabling rapid identification of suitable trial candidates. Performance analysis across multiple clinical domains confirms the generalizability and robustness of the approach.