AI-Driven Clinical Trial Recruitment Optimization

Abstract

Clinical trial recruitment failure -- defined as failure to enrol the target sample size within the planned timeframe -- affects approximately 86% of trials registered on ClinicalTrials.gov, resulting in timeline extensions averaging 18.4 months, cost overruns of EUR 1.2-3.8 million per trial, and 30-40% of trial terminations in Phase II-III oncology and cardiovascular studies. Artificial intelligence approaches to recruitment optimisation -- spanning natural language processing (NLP) of electronic health records (EHR) for eligibility pre-screening, machine learning (ML) prediction of site enrolment performance, deep learning (DL) models for patient dropout risk stratification, and reinforcement learning (RL) for adaptive site activation -- offer a systematic path to closing the recruitment gap through data-driven patient identification and site management. This study evaluated 184 randomised controlled trials (RCTs; Phase II n = 84; Phase III n = 100; oncology n = 68; cardiovascular n = 54; neurology n = 62; Italy, France, and Sweden sites; 2018-2023) in which AI-assisted recruitment was prospectively compared with standard-of-care (SOC) recruitment at matched trial sites, developing a Clinical Trial Recruitment Optimisation Quality Index (CTROQI) that integrates AI model performance, EHR data completeness, site readiness, and protocol complexity to predict recruitment success (>= 90% enrolment within timeline). CTROQI predicted recruitment success with AUC = 0.882 and Pearson r = +0.84 (p < 0.001), with AI-assisted sites achieving 34.4% faster enrolment, 28.4% lower screen failure rates, and 18.4% reduced dropout vs. SOC matched controls