OBJECTIVES: Comparative Review is a key quality assurance step in the Clinical Outcome Assessment (COA) localization process (Linguistic Validation or LV). In order to improve efficiency of this process, our goal is to automate steps where possible to reduce time and spend getting products to market. The purpose of this research is to determine if Comparative Review can effectively be automated using GenAI while maintaining the quality and thoroughness of COA and LV methodologies.

METHODS: A prompt was developed for input into a GenAI engine with the goal of identifying discrepancies between original assessment and correlated back translation content. The prompt requests that GenAI define that discrepancy in a way that would help the linguists involved to update and refine the existing forward translation to best reflect the original concepts. The outputs are compared to determine how often discrepancies are found by the GenAI engine versus the human resources and assessed for clarity of the explanations provided.

RESULTS: The initial results are promising, with clear, concise descriptions of original assessment and back translation discrepancies at an overall preliminary accuracy rate of 97%, matching 84% of feedback when compared to human output, and in 12% of segments, finding discrepancies not noted in the human output. In 3%, human output found potential discrepancies that AI noted as equivalent due to the usage of synonyms or localization changes.

CONCLUSIONS: GenAI shows great promise in detecting and describing discrepancies brought to light through the process of Comparative Review. Further research in the form of a proof of concept would be beneficial to determine the extent to which this process could be used without human intervention, but in an already time-restricted process, utilizing AI for this intermediate step could make the difference in timely, cost-effective documentation in the clinical trial process.