A LLM-based Multi-Agent Collaborative Approach for Screening Prioritization towards Automated Systematic Reviews

Abstract Objective Systematic reviews (SRs) are essential for evidence-based practice but remain labor-intensive, especially during title and abstract screening. This study evaluates whether multiple large language model (multi-LLM) collaboration can improve screening prioritization while reducing costs. Methods Abstract screening was framed as a question-answering (QA) task using cost-effective LLMs. Three multi-LLM collaboration strategies were proposed and evaluated, including majority voting by averaging opinions of peers, multi-agent debate (MAD) for answer refinement, and LLM-based adjudication against answers of individual QA baselines. These strategies were evaluated on the CLEF eHealth 2019 Technology-Assisted Review benchmark using standard performance metrics in the domain, including Mean Average Precision (MAP), Recall@k%, and Work Saved over Sampling (WSS). Results Multi-LLM collaboration strategies significantly outperformed QA baselines. Majority voting was the most outstanding, achieving the highest MAP 0.4621 and 0.3409 on the subsets of SRs about clinical intervention and diagnostic technology assessment, respectively, with WSS@95% 0.6064 and 0.6798, enabling in theory up to 68% workload reduction at 95% recall of all included studies. MAD improved weaker models the most. The adjudicator-as-a-ranker method surpassed adjudicator-as-a-judge and was the second strongest approach, but with a significantly higher cost than majority voting and debating. Conclusion Multi-LLM collaboration can substantially improve abstract screening efficiency, and the success lies in model diversity. Making the the best use of diversity, majority voting stands out in terms of both fantastic performance and low cost compared to adjudication.Despite context-dependent gains and diminishing model diversity, MAD is still a cost-effective strategy and a potential direction of further research.