OBJECTIVES: With the rise of artificial intelligence (AI) to support the labor-intensive screening phase of literature reviews, much of the focus has been around the accuracy of AI tools. However, the impact of replacing human reviewers for title and abstract screening with AI remains largely undefined. This study continues previous assessments of AI (Cichewicz et al. 2022) to determine the potential time and cost savings when implementing AI tools in the conduct of literature reviews.

METHODS: Five previously completed systematic literature reviews (SLRs) on epidemiology, treatment patterns, health utilities, treatment guidelines, and SLRs and/or meta-analyses were replicated using DistillerSR AI reviewer. The proportion of references screened by the AI reviewer was used to estimate the number of hours and associated cost differential between literature reviews conducted entirely by humans compared with reviews employing AI as an independent or second screener.

RESULTS: After 10% of references were used to train the AI, the number of references to review across the replicated SLRs ranged from 472 to 1695 (approximately 17 to 62 hours with dual screening). With a pre-defined prediction threshold of 0-0.2 for excludes and 0.8-1 for includes, 73% to 91% of these references were screened by the AI reviewer. Based on this volume, AI could save approximately 6 to 27 hours of a human reviewer’s time. This amounts to roughly 47% to 91% of time and budget savings in single-screened reviews and 23% to 45% in dual-screened reviews.

CONCLUSIONS: While this does not account for the other burdensome aspects of conducting literature reviews, the use of AI has the potential to substantially reduce the volume of resources and timeframe in which reviews can be conducted and minimize labor costs. These savings may be more impactful when employing AI in literature reviews with higher volumes of references.