Presentation (CTI)

OBJECTIVES: To compare the automated deduplication performance of EasySLR™ and EndNote™ 20 using datasets from multi-database systematic literature reviews (SLRs).
METHODS: Deduplication performance was assessed using five completed SLRs. Literature searches included PubMed and at least one additional database (EuropePMC, Embase, or Cochrane Library). Tool performance was benchmarked against a curated reference dataset in Microsoft Excel^®. Within Excel, record titles were standardised using TRIM, CLEAN, and SUBSTITUTE functions, after which MATCH function was used to identify duplicates. These records were subsequently manually verified using additional bibliographic details. For both platforms, duplicate detection was performed using identical matching criteria based on publication title and year. Other parameters were excluded because of inconsistent availability and formatting across databases. Records were classified as true positives, false positives, false negatives, or true negatives, and performance was evaluated using sensitivity, specificity, precision, and accuracy.
RESULTS: Both tools demonstrated comparable deduplication performance across the five datasets. Accuracy ranged from 99.1% to 100% for Endnote™ and from 99.2% to 100% for EasySLR™, while specificity remained consistently high across all projects, ranging from 99.4% to 100% for both platforms. Sensitivity exceeded 99% in four datasets but was lower in one dataset (93.2% for EndNote™ and 94.5% for EasySLR™), where records were retrieved from Embase, PubMed, and the Cochrane Library. These findings suggest that cross-database formatting inconsistencies remain a challenge for automated duplicate detection regardless of the platform used. A hierarchical matching strategy that applies strict matching criteria across multiple bibliographic fields (e.g., DOI, title, year, journal, volume, and issue) to enable bulk deduplication, followed by title-only matching for case-by-case review of unresolved records, may further improve duplicate detection.
CONCLUSIONS: Automated deduplication tools demonstrated robust and comparable performance in multi-database SLRs when consistent bibliographic metadata were available. Integrating deduplication within end-to-end review platforms has the potential to enhance workflow efficiency and methodological consistency.