Presentation (CTI)

OBJECTIVES: Systematic literature reviews (SLRs) are integral for evidence generation, but pharmaceutical companies face challenges in leveraging SLR insights due to lengthy processes and high resource demands. Additionally, SLRs quickly become outdated with rapid knowledge growth. We developed ActiveSLR, an artificial intelligence (AI)/large language model (LLM) based software, saving time and resources while enabling regular insights generation. This study assessed ActiveSLR’s efficiency compared to Microsoft Excel based traditional SLR processes. It also assessed the performance of proprietary deduplication algorithm for removing duplicates when searching multiple databases.
METHODS: We created four projects with 330-2,000 citations from MEDLINE and EMBASE for testing ActiveSLR’s AI based deduplication algorithm. Two projects out of them, with 500 and 1,000 citations, were screened with LLM based screening assistance and data were extracted from the selected studies using an integrated data extraction (DE) module.
RESULTS: ActiveSLR identified 100% duplicates in all four projects (12-15 minutes) compared to 74% for EndNote (5-10 minutes with false duplicate in one project), and manual review (100% in 3-6 hours). Reviewers took an additional 2-3 hours to achieve 100% accuracy if EndNote deduplication was done first. Thus, ActiveSLR offered an 11-12X efficiency gain in deduplication compared to EndNote. The AI-powered Population-Intervention-Comparator-Outcome (AiPICO) identifier improved efficiency by 50-109% for title-abstract screening and 60-100% for full-text screening compared to the traditional SLR process. ActiveSLR identified most free publications, saving 49% of time compared to manual searching. Data extraction by ActiveSLR also saved an average of 37.5% time compared to manual extraction in Microsoft Excel.
CONCLUSIONS: ActiveSLR cumulatively saved 42% (39-45%) of time compared to Microsoft Excel based traditional approaches for the end-to-end SLR process. We are the first to present efficiency gain of an automated SLR tool compared to traditional SLR process. Independent evaluation by academic researchers is warranted for comparing performance of such automated SLR tools.