ACCURACY AND EFFICIENCIES ASSOCIATED WITH AI-BASED LINKING OF CLINICAL TRIAL PUBLICATIONS IN SYSTEMATIC REVIEWS
Author(s)
Md Sohail Aman, M.Pharm1, Abhra R. Choudhury, BDS, MDS1, Kopal Dixit, MSc1, Anup Shaw, Sr., BTech1, Iain Fotheringham, BSc2;
1PharmaQuant Insights Pvt. Ltd., Kolkata, India, 2PharmaQuant International Limited, Winchester, United Kingdom
1PharmaQuant Insights Pvt. Ltd., Kolkata, India, 2PharmaQuant International Limited, Winchester, United Kingdom
OBJECTIVES: Systematic literature reviews (SLR)s often include multiple citations from the same clinical study. Failure to identify and link these can lead to duplicate data extraction, increased reviewer burden and potential double-counting in subsequent synthesis. Manual identification is resource-intensive and error-prone, particularly in large evidence bases. This study evaluates the accuracy and efficiency of ActiveSLR®’s novel AI-supported linking.
METHODS: Two SLRs (1,497 and 2,700 citations) were conducted using search results from MEDLINE and Embase. Following population-intervention-comparator-outcomes (PICO) driven title/abstract and full-text screening in ActiveSLR®, included studies were exported to Excel. One reviewer used ActiveSLR®’s AI-supported trial identification feature to link citations, while a second reviewer used Excel to independently link them. Time required for linking was recorded for each, and efficiency was assessed based on the time difference between ActiveSLR® and Excel-based manual linking. AI-supported versus manual linking accuracy was measured, including whether AI introduced incorrect publication links.
RESULTS: Across the two SLRs, a total of 179 citations were included. ActiveSLR® correctly identified trial names for 164 (~92%) citations. Poster booklets where multiple posters were present in a single PDF, required correction. Reviewer validation and editing time both averaged under 10 seconds/citation; On average linking time across both projects (including validation and edits) was ~24.6 seconds/citation. Manual linking in Excel required 3.82 minutes/citation on average, corresponding to an efficiency gain of ~89.3% in ActiveSLR®. Subsequent assessment of the time difference between ActiveSLR® and Excel to select studies eligible for data extraction, is underway.
CONCLUSIONS: ActiveSLR® demonstrated efficiency gains over Excel for linking multiple publications from the same clinical study, with the potential for efficiency to increase further with review size. AI-supported study linking remains underexplored in evidence synthesis but may markedly reduce reviewer burden in what can be a time-consuming task.
METHODS: Two SLRs (1,497 and 2,700 citations) were conducted using search results from MEDLINE and Embase. Following population-intervention-comparator-outcomes (PICO) driven title/abstract and full-text screening in ActiveSLR®, included studies were exported to Excel. One reviewer used ActiveSLR®’s AI-supported trial identification feature to link citations, while a second reviewer used Excel to independently link them. Time required for linking was recorded for each, and efficiency was assessed based on the time difference between ActiveSLR® and Excel-based manual linking. AI-supported versus manual linking accuracy was measured, including whether AI introduced incorrect publication links.
RESULTS: Across the two SLRs, a total of 179 citations were included. ActiveSLR® correctly identified trial names for 164 (~92%) citations. Poster booklets where multiple posters were present in a single PDF, required correction. Reviewer validation and editing time both averaged under 10 seconds/citation; On average linking time across both projects (including validation and edits) was ~24.6 seconds/citation. Manual linking in Excel required 3.82 minutes/citation on average, corresponding to an efficiency gain of ~89.3% in ActiveSLR®. Subsequent assessment of the time difference between ActiveSLR® and Excel to select studies eligible for data extraction, is underway.
CONCLUSIONS: ActiveSLR® demonstrated efficiency gains over Excel for linking multiple publications from the same clinical study, with the potential for efficiency to increase further with review size. AI-supported study linking remains underexplored in evidence synthesis but may markedly reduce reviewer burden in what can be a time-consuming task.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
MSR82
Topic
Methodological & Statistical Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
No Additional Disease & Conditions/Specialized Treatment Areas