Research Screener: a machine learning tool to semi-automate abstract screening for systematic reviews

Area of Science:

• Evidence Synthesis
• Health Informatics
• Machine Learning in Research

Background:

• Systematic reviews and meta-analyses are crucial for evidence-based policy and practice.
• Abstract screening is the most time-intensive phase of systematic reviews, averaging 33 days.
• Current automated tools have limitations and face barriers to adoption.

Purpose of the Study:

• To introduce Research Screener, a semi-automated machine learning tool.
• To provide initial evidence on the utility of Research Screener for abstract screening.

Main Methods:

• Conducted three sets of analyses: simulation, interactive, and sensitivity.
• Evaluated the tool's utility using simulated and real-world examples.

Main Results:

• Research Screener achieved workload savings of 60-96% across 11 reviews.
• Real-world analysis showed a time saving of 12.53 days and a financial saving of USD 2444.
• Scanning 50% of articles is highly likely to identify 100% of eligible papers.

Conclusions:

• Research Screener effectively reduces the burden of systematic reviews.
• The tool maintains scientific rigor while improving efficiency.
• Facilitates comprehensive systematic reviews with reduced manual effort.