Comparing Artificial Intelligence versus Human Screening in Systematic Reviews

Abstract Introduction Systematic reviews are essential for informing health policy and practice. Artificial intelligence (AI) automates the article screening process and produces time savings, although the performance of AI screening compared to traditional human screening remains uncertain. We undertook this study to compare the performance of two agentic AI tools, namely Loon Lens TM and Catchii, to one another and to humans at the title and abstract screening level. We also compared Loon Lens to humans at the full-text screening level. Methods We developed a de novo research question on the association between any of three ambient air pollutants – carbon monoxide, ozone, nitrogen dioxide – and the onset or worsening of Parkinson’s disease. A health sciences librarian developed the literature search strategy and we proceded with human screening guided by PRISMA. We uploaded the retrieved citations and the eligibility criteria to both AI tools and compared screening results using sensitivity, specificity, positive predictive value, negative predictive value, concordance, kappa, and F1 score. We compared the calculated performance statistics to those obtained by naïve guessing and regressed concordance (agree or disagree with the human reference standard) onto confidence scores provided by Loon Lens, which assigned a confidence level (‘Very High’, High’, Medium’, or ‘Low’) to each of its screening decisions. Human screening was the reference standard against both AI tools; Catchii was the reference standard against Loon Lens. Results At title and abstract screening, Loon Lens outperformed Catchii when humans were the reference standard. At full-text screening, most disagreements centered around articles Loon Lens included and humans excluded. At both screening levels, higher confidence scores were associated with lower odds of disagreement between Loon Lens and human screeners. Discussion Given the panoply of available AI screening tools and their differential performance, plus the rapidly evolving nature of AI technology, researchers should pilot test their chosen tool at the start of each review. Sensitivity, kappa, and F1 are the optimal performance statistics to employ, especially at title and abstract screening, where the imbalance between proportions of included and excluded citations can inflate concordance and negative predictive value. What is new? Key findings Strongest screening agreement seen among excluded articles Abstract wording affects AI tools’ performance Model instructions and wording of eligibility criteria also affect performance What this adds to what is known? Up-to-date evidence comparing the performance of two AI tools to one another and humans at title/abstract and full-text screening What is the implication? AI tools should be pilot tested at the outset of any review Plain Language Summary Systematic reviews help health practitioners and policy makers decide on courses of action. However, these reviews can take a long time to finish. We wanted to know whether artificial intelligence (AI) tools can replace humans when reviewing hundreds or thousands of research papers for inclusion in systematic reviews (a practice called ‘screening’). We compared two AI tools, Loon Lens™ and Catchii, to human screeners in the area of exposure to three air pollutants (carbon monoxide, ozone, or nitrogen dioxide) and occurrence of Parkinson’s disease. At title-and-abstract screening, the two AI tools had the strongest agreement with each other and with human screeners for papers that were not relevant. Loon Lens generally outperformed Catchii when both AI tools were compared to humans. At full-text screening, disagreements mainly happened when Loon Lens included a paper that humans decided to exclude. Loon Lens assigned each screening decision a ‘confidence’ level (Very High, High, Medium, Low): the higher the confidence level, the less often Loon Lens disagreed with humans. Overall, AI can serve alongside human screeners, though it is not ready to replace them yet. Researchers should test their chosen AI tool when starting a review. This will help them get the best performance out of their tool.