AI-Assisted Screening

Covidence and ASReview are estimated to reduce screening workload by 20 to 54 percent while minimising bias. Robot Screener acts as the second reviewer for remaining title and abstract screening, with conflicts resolved by a second human reviewer. Two human reviewers independently screen 5% to 10% of records to train the Robot Screener before it is used on remaining records. Nested Knowledge uses AI classifiers that learn from project-specific screening decisions and generate inclusion probabilities for records. ASReview performs semi-automated title and abstract screening using active researcher-in-the-loop machine learning. In annual updates, an experienced reviewer independently screens 10% of all records to reduce the risk of false exclusions, with additional 10% checks if false negatives are found. Covidence manages references, supports blinded independent screening, ranks studies by likely inclusion, and allows custom exclusion labels.