Machine-learning assisted screening for evidence synthesis: Methodological case study of the ASReview tool

Area of Science:

• Medical Informatics
• Systematic Review Methodology
• Cancer Research

Background:

• Systematic literature reviews are crucial for evidence-based medicine but are labor-intensive.
• Automating or semi-automating literature screening can significantly reduce workload and time.
• Cancer immunotherapy research requires efficient methods for synthesizing trial data.

Purpose of the Study:

• To evaluate the feasibility, advantages, and limitations of ASReview software for literature screening.
• To populate a database of cancer immunotherapy trials using ASReview.
• To assess ASReview's potential to reduce workload and save resources in systematic reviews.

Main Methods:

• ASReview software was used to rank retrieved records for literature screening.
• The tool's usability, efficiency, and effectiveness in identifying relevant records were assessed.
• A database of cancer immunotherapy trials was populated using the software.

Main Results:

• ASReview demonstrated ease of use and efficiency in identifying relevant records.
• The software has the potential to save resources compared to traditional two-human-reviewer systematic reviews.
• Predefined procedures are essential for maintaining transparency and reproducibility.

Conclusions:

• ASReview is a feasible tool for reducing literature screening workload in systematic reviews.
• The software offers advantages in efficiency but requires careful implementation for transparency and reproducibility.
• Limitations include difficulties in adding references to existing projects and the algorithm's learning behavior.