Integrating Health Behaviour and AI/ML Theories: A Case for Pre-Screening Prediction in Industry-Sponsored Clinical Trials

Recruitment inefficiencies remain a critical barrier in clinical research, often leading to delayed timelines, underpowered studies, and reduced generalisability. While algorithmic solutions have improved enrolment logistics, many models overlook behavioural and time-sensitive factors that influence patient progression through the recruitment process. This study addresses these limitations by integrating behavioural theory, artificial intelligence (AI), machine learning (ML), and survival analysis into a unified recruitment optimisation framework. Patient pre-screening interactions (screener question and response pair) were mapped to Health Belief Model (HBM) constructs perceived susceptibility, severity, benefits, and barriers using Named Entity Recognition (NER) within large language models (LLMs). These behavioural features were incorporated into an XGBoost classifier to predict phone screening progression, yielding strong performance (AUC = 0.7398; accuracy = 67.04%). SHAP analysis revealed perceived barriers and composite social vulnerability indicators as dominant predictors, while severity and benefits contributed marginally, highlighting the value of embedding behavioural variables early in screening design. Kaplan-Meier survival analysis showed that the site contact probability declined sharply after referral, falling below 50% by day 11. This underscores the temporal fragility of patient engagement and the inadequacy of static models. Incorporating survival-derived probabilities into dynamic scoring enables timely prioritisation of referrals during peak responsiveness. This behaviourally informed, temporally adaptive screening framework enhances recruitment precision while supporting site-level efficiency. It advances methodological practice by aligning AI/ML capabilities with behavioural theory and offers practical, ethically grounded implications for improving patient recruitment in clinical trials. The findings contribute to both scholarly understanding and the evolving architecture of AI/ML within clinical trial recruitment.