Extended Results for Enhancing Abstract Screening Classification in Evidence-Based Medicine: Incorporating Domain Knowledge into Pre-trained Models

Evidence-based medicine (EBM) is a foundational element in medical research, playing a crucial role in shaping healthcare policies and clinical decision-making. However, the rigorous processes required for EBM, particularly during the abstract screening phase, pose substantial challenges to researchers. While many have sought to automate this stage using Pre-trained Language Models (PLMs), these efforts often face obstacles due to the specificity of the domain, especially when dealing with EBM studies related to both human and animal subjects. To address this, our initial research presented a state-of-the-art (SOTA) transfer learning approach that enhanced four abstract screening by embedding domain-specific knowledge into PLMs without modifying their base weights utilising the concepts of adapters. Extending the previous work, in this study, we evaluate the same methodology on four animal and human EBM datasets. Our evaluation, conducted on the further four EBM abstract screening datasets, demonstrates that the proposed method significantly improves the screening process and outperforms strong baseline PLMs.