Garbage In, Garbage Out: Can AI Reduce the Impact of Human Error in eCOA Localized Text Migration?
Author(s)
Jonathan R. Norman, PGDip.
Director, Localization & Scale Management, YPrime, Malvern, PA, USA.
Director, Localization & Scale Management, YPrime, Malvern, PA, USA.
OBJECTIVES: eCOA localization requires that existing translations (such as those licensed from copyright holders or taken from the public domain) are migrated from their paper formats (Word or Excel) into an eCOA system. The quality of this step directly impacts the duration of the eCOA proofreading process, as errors necessitate additional rounds of screen report generation and linguist proofreading. This is a known pain point and rate-limiting factor for sponsors attempting to accelerate study startup. The objective of YPrime’s research was to determine whether a localization methodology replacing human-led migration with AI-powered migration would prevent errors, leading to higher quality outputs and reduced timelines for sponsors.
METHODS: YPrime reviewed a convenience sample of projects (n=15) to identify errors introduced through human-led migration, which resulted in further rounds of screen report generation and linguist review. YPrime then repeated the migration process of the same materials, keeping all process steps the same except for replacing the human migration activities with AI. The outputs were then compared to the linguist-approved outputs from the sample projects to identify any errors introduced by the AI-powered migration process and to determine which methodology achieved the desired outcome with the fewest rounds of review.
RESULTS: Of the previously identified errors, only a fraction reoccurred. Further, no new errors were introduced by the AI-powered process. Therefore, YPrime concluded that the majority of the reviewed projects would have been approved at least one round earlier as a result of an AI-powered migration process, with an average timeline saving of around 67% across the projects reviewed.
CONCLUSIONS: YPrime identified that error in human-led migration is a driving factor in the number of rounds of review required before eCOA screen report approval and AI-powered migration is able to largely eliminate these errors, resulting in significant timeline savings for sponsors.
METHODS: YPrime reviewed a convenience sample of projects (n=15) to identify errors introduced through human-led migration, which resulted in further rounds of screen report generation and linguist review. YPrime then repeated the migration process of the same materials, keeping all process steps the same except for replacing the human migration activities with AI. The outputs were then compared to the linguist-approved outputs from the sample projects to identify any errors introduced by the AI-powered migration process and to determine which methodology achieved the desired outcome with the fewest rounds of review.
RESULTS: Of the previously identified errors, only a fraction reoccurred. Further, no new errors were introduced by the AI-powered process. Therefore, YPrime concluded that the majority of the reviewed projects would have been approved at least one round earlier as a result of an AI-powered migration process, with an average timeline saving of around 67% across the projects reviewed.
CONCLUSIONS: YPrime identified that error in human-led migration is a driving factor in the number of rounds of review required before eCOA screen report approval and AI-powered migration is able to largely eliminate these errors, resulting in significant timeline savings for sponsors.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
CO125
Topic
Clinical Outcomes, Methodological & Statistical Research
Topic Subcategory
Clinical Outcomes Assessment
Disease
No Additional Disease & Conditions/Specialized Treatment Areas