Home // HEALTHINFO 2025, The Tenth International Conference on Informatics and Assistive Technologies for Health-Care, Medical Support and Wellbeing // View article
From Abstracts to Full Texts: The Impact of Context Positioning in LLM-Based Screening Automation
Authors:
Elias Sandner
Marko Zeba
Igor Jakovljevic
Alice Simniceanu
Luca Fontana
Andre Henriques
Andreas Wagner
Christian Gütl
Keywords: systematic review; screening automation; full-text screening; LLM.
Abstract:
Screening for relevant research is among the most time-intensive phases of a Systematic Review (SR), significantly impacting its timeliness and resource requirements. Automation through Large Language Model (LLM) promises substantial efficiency gains, potentially reducing the human screening workload and mitigating the risk of reviews becoming outdated prior to publication. Existing research has primarily explored LLM applications in Title & Abstract (TiAb) screening, achieving promising sensitivity but limited investigation into Full-Text (FT) screening. This study extends the 5-tier prompting approach, originally developed for TiAb screening, to FT screening. An experimental evaluation was conducted using the LLaMA 3.1 8B model on five real-world SR datasets. Two FT prompting strategies were tested: one that directly adapted the 5-tier TiAb approach to FT screening, and another addressing the known ’lost-in-themiddle’ phenomenon by positioning eligibility criteria before and after the full text. Findings indicate that providing FT context improves workload reduction considerably, nearly doubling it in some cases, though sensitivity slightly decreased compared to TiAb screening. Notably, positioning eligibility criteria both before and after FT significantly improved performance, highlighting the importance of the prompt structure. These results demonstrate that careful prompt engineering enhances LLM effectiveness in FT screening, balancing the critical trade-off between sensitivity and workload reduction. Overall, this research underscores the potential of LLM-based FT screening, providing valuable insights into prompt optimization for systematic review automation.
Pages: 57 to 62
Copyright: Copyright (c) IARIA, 2025
Publication date: October 26, 2025
Published in: conference
ISSN: 2519-8491
ISBN: 978-1-68558-312-5
Location: Barcelona, Spain
Dates: from October 26, 2025 to October 30, 2025