AI: Streamlining translation in Pharmacovigilance
May 24, 2021
May 24, 2021
In an ongoing and continued effort to enhance pharmacovigilance (PV), we have been working with the science and technology company, Merck KGaA, Darmstadt, Germany, to develop solutions that increase the speed and accuracy of drug safety case report intake. Our work began in 2015, when we helped establish two centralized hubs for intaking their case reports to maximize the global footprint. These hubs improved efficiencies in the operating model, but we felt we could go further still.
By employing the benefits of the latest AI technologies, we worked toward a proof of concept solution to help generate additional efficiencies for translating drug safety case source documents. Together we formed a project team of experts on machine learning (ML) and pharmacovigilance to pilot an AI-powered translation assistant using existing Google technology. Our preliminary POC (proof of concept) targeted translating reports from Portuguese to English where Accenture’s team of medical translators reviewed 100% of the AI-translations for accuracy, making modifications to both ensure quality and to help the platform learn.
<<< Start >>>
This combination of Human + Machine honed the accuracy of the tool and in a very short timeframe we saw a dramatic consistency of the intake process.
<<< End >>>
To power the translation tool, our cross-discipline Applied Intelligence team set up a secure, private, cloud environment and developed a simple user-interface for team members to enter, retrieve, and edit results if needed. The system automatically analyses and learns from modifications made by our team of medical translation professionals, which further improves the first-time quality of the AI translation. We also continue to use human efforts to review 100% of the translations, ensuring meticulous quality control.
The results of the AI Translation Assistant pilot for the first language (Portuguese) went above and beyond expectations. As the output accuracy improved with the help of ML, we saw a reduction of translation efforts required by at least 30%. Excited by these results, we and Merck KGaA, Darmstadt, Germany agreed and deployed five more languages.
This project truly demonstrates how infusing intelligent technologies can bring significantly improved capabilities to pharmacovigilance. This “Human + Machine” model, in which a skilled team uses and engages with AI to amplify their capabilities while fine tuning the technology, has proven to be a model for the future and embodies Intelligent and Resilient Operations. The accuracy of the AI Translation Tool continues to improve as the case volume grows across languages. More samples make better models and better models make less effort for the team, allowing them to work through more cases faster and greater agility.
We are excited about this ongoing partnership with Merck KGaA, Darmstadt, Germany and look forward to scaling the deployment of the AI Translation Assistant into more languages.
To learn more about this innovative approach to PV, please visit www.accenture.com/PVServices.