SWIFT-Active Screener

SWIFT-Active Screener prioritizes documents using active learning and applies a negative binomial model to estimate remaining relevant records, thereby accelerating and quantifying the screening phase of systematic reviews in bioinformatics.

Key Features:

• Active Learning Integration: Employs active learning to dynamically incorporate reviewer feedback and prioritize documents by likelihood of relevance.
• Collaborative Support: Supports multi-user collaboration for coordinated screening by multiple reviewers.
• Recall Estimation with Negative Binomial Model: Uses a negative binomial statistical model to estimate the number of relevant articles remaining in the unscreened list.
• Efficiency and Time Savings: Simulation studies on 26 systematic review datasets reported an average of 95% recall after screening 40% of references and, for datasets with over 5,000 references, 95% recall after screening 34% of references.
• Decision-Making Aid: Provides explicit recall estimation to inform decisions about when to stop screening a prioritized list.

Scientific Applications:

• Systematic literature synthesis in bioinformatics: Accelerates and quantifies the document screening phase for large-scale systematic reviews in bioinformatics.
• Evidence aggregation for genomic and protein interaction studies: Supports synthesis of literature relevant to genomic data and protein interactions by prioritizing likely-relevant studies.

Methodology:

The approach combines active learning for dynamic prioritization, a negative binomial model for recall estimation, and validation via simulation studies on 26 systematic review datasets.