人工智能监管

Introduction Every drug sponsor wants to make the best use of Artificial Intelligence, but doing so requiresunderstanding where health authorities are drawing the lines. That’s always something of a movingtarget, especially in a field as new as AI. The Pink Sheet’s globe-spanning has produced one ruleof thumb, however: The closer Artificial Intelligence is to patient care, the more scrutiny it will get.Regulators seem content to allow firms to make product development decisions based on AI withoutsignificant oversight. But if AI is used to assess a clinical trial outcome or examine manufacturing process, authorities willwant to make sure that model has been thoroughly vetted – though it doesn’t necessarily need to becompletely explainable. Below is a sample of our coverage from Asia, Europe and the United States of the emergingframework for artificial intelligence regulation. EMA’s AI Principles Intended ToBe ‘Flexible & Long Lasting’ Eliza Slawther04 Oct 2024 There is “a lot of flexibility” in the European Medicines Agency’sreflection paper on the use of artificial intelligence during drugdevelopment, which is principles-driven rather than setting rigidrecommendations, says the agency’s Florian Lasch. Executive Summary The European Medicines Agency recentlyfinalized a reflection paper on the use ofartificial intelligence (AI) during the lifecycle ofmedicinal products to help manufacturers useAI in a safe and effective way. 1,342 Comments” – Pink Sheet, Oct. 2 2024.). “There is a lot of flexibility following this risk-based approach,” Lasch said, which could helpcompanies to define their metrics and measurethe performance of AI systems used in drugdevelopment. The approach taken by the EMA is a flexibleone that is “focused on principles, rather thanon specific recommendations,” Florian Lasch,a biostatistics specialist at the EMA, said on 1October during the TOPRA (The Organisation forProfessionals in Regulatory Affairs) Symposium2024. “With this approach, we hope that the principlescan actually live quite a long while, whereasthe specific application of the principles mightchange,” he added, pointing out that translatingAI principles into practice could be where thechallenge lies. He explained that the AI reflection paper,which was initially drafted in July last year andfinalized 14 months later based on a review of1,342 feedback comments from stakeholders,did not set any specific metrics or technicalrequirements for companies. (Also see “EMAFinalizes AI In Medicines Paper After Reviewing “We have interaction channels like scientificadvice that I would encourage every companyto make use of if any AI applications are to beused for real-world evidence generation,” Laschcontinued. EMA’s AI Principles Intended ToBe ‘Flexible & Long Lasting’ He said that companies also had theopportunity to engage in informal exchangeswith the regulator through innovation task forcemeetings and portfolio technology meetings. “At Sanofi, we think of AI as systems that usedata to learn and to adapt their actions or theiroutputs accordingly,” she said, noting that AIwas “not a replacement for human intelligence”and was “only as powerful as the data it isgiven.” EMA Working On New ToolsIn March, the EMA rolled out Scientific Explorer, an AI-enabled knowledge mining tool thatallows regulatory assessors in the EU to moreeasily find the information they need to helpinform their scientific decisions. (Also see “EURegulatory Assessors Get AI Boost In ReachingScientific Decisions” – Pink Sheet, Mar. 28 2024.).During a question-and-answer portion of thediscussion, Lasch was asked by an audiencemember whether the EMA planned to develop AItools for applicants, starting with an AI searchtool for its website. Philippe said that AI was “revolutionizingthe whole value chain” and helping Sanofi tobecome an “AI-powered biopharma.”For example, the company is using AI to speedup research by predicting clinical trial results, aswell as in manufacturing to improve processes,resulting in better yields of medicines, she said. “For [the] supply chain, we’re using generativeAI to redact the product quality reports,”Philippe added, explaining that such reportswere made by aggregating, synthesizing andsummarizing large amounts of data. He said that there was an ongoing “use casecollection” for the development of both inwardand outward-facing AI tools within the EMA,and that while he could not comment onspecific plans, a “lot of work” was going into thisprocess. She emphasized the importance of using AI ina responsible way to mitigate its potential risks,and said that Sanofi had implemented an AIframework “based on accountability.” “AI will most likely impact the way we work inthe next couple of years. It won’t replace us, butwe need to foster that culture of innovation andadapt to the new tools,” Philippe said, stressingthat it was important that AI regulations “areable also to evolve accordingly to [match] thefast pace of the techno