Accelerating Systematic Literature Reviews with an AI-Powered Screening Tool
At ISPOR Europe 2025, Paul Loustalot from Quinten Health presented a new AI-powered tool designed to accelerate and enhance systematic literature reviews (SLRs) by semi-automating title and abstract screening. The tool enhances efficiency while preserving the scientific rigor required in evidence-based clinical research.
Why Systematic Reviews Need Innovation
Systematic literature reviews remain essential for evidence synthesis in clinical research, guidelines, and health technology assessment. However, manually screening hundreds or thousands of publications is time-consuming and labor-intensive.
Traditional tools such as keyword search engines or BM25 often fail to achieve high recall without substantial manual effort. The emergence of Natural Language Processing (NLP)and Large Language Models (LLMs) offers new opportunities to automate a demanding step: title/abstract screening.
An AI-Powered Tool for Screening
After the initial PubMed keyword search phase, the AI tool evaluates publications against reviewer-defined criteria. The goal is a hybrid workflow where AI accelerates screening while human reviewers ensure scientific validity.
The prompting framework relies on:
1. Uncertainty-aware prompting to avoid hallucinated decisions : the LLM model is given the possibility to say it is unsure about a criteria if the information is missing or unclear from the title and abstract.
2. Few-shot learning : the LLM uses expert-generated examples to guide the model.
Mistral-Large 2024 was used for all inference experiments.
Experimental Validation
Five published systematic reviews were replicated to test the approach. Each PubMed query was re-run, and AI-generated shortlists were compared with expert selections.
Results show that the AI-assisted workflow:
• reduced screening effort by 46–91% (2× to 10× faster),
• maintained recall between 96% and 100% (missed fewer than 5% of expert-selected studies)
A Powerful Hybrid Approach
This hybrid human–AI approach supports faster, more reproducible, and more transparent evidence generation. It enables efficient SLR processes for regulatory assessment, HTA submissions, and clinical development.
Conclusion
The tool demonstrates that AI can substantially accelerate systematic literature reviews without compromising rigor. Hybrid AI-assisted screening offers a scalable and reliable method for producing high-quality evidence in clinical and health economics research.
Official ISPOR Poster
You can view the official ISPOR Europe 2025 poster here: Development of an AI-Powered Tool to Accelerate and Enhance Systematic Literature Reviews .
