AI in Drug Safety: Building the Elusive ‘Loch Ness Monster’ of Reporting Tools
Whenever a patient takes a drug and has an unexpected side effect — from minor problems like a runny nose to more serious ones that require hospitalization — pharma companies are legally required to report this information, known as an adverse event (AE), to regulatory agencies.
Within the pharmaceutical industry, AE reporting is a critical and time-consuming part of ensuring the safe and effective use of medicines by patients. Pharmacovigilance employees process a growing number of cases from sources as varied as patients’ social media accounts to reports from investigators overseeing clinical trials. Pfizer’s Worldwide Safety organization processed approximately 1.4 million AEs globally in 2019 alone. Looking forward across the industry, the volume of AEs is expected to increase by 20% annually.
To keep up with the increasing volume of cases, pharmacovigilance experts at Pfizer partnered with industry experts to develop an innovative artificial intelligence (AI) platform that can handle the more repetitive tasks of AE case report intake and processing. Its creators are careful to note that these tools will not be replacing humans, but in fact empowering those in case-processing to turn their time and attention to activities that require more critical thinking. “It’s freeing up our experts to do more of the human interactions and the types of tasks that have the most beneficial impact for patients,” says Bhavin Patel, Worldwide Safety.
Pfizer’s Drug Safety Unit in Rome, Italy, is the first location to use the platform in real life. The tool focuses on the first part of AE case processing — the “intake” phase, where information is received from patients, doctors and others. In coming years, says Patel, they hope to implement an “end-to-end” tool that can help with case processing, follow-up with patients, medical review and reporting of this information to regulatory authorities.
The ‘Loch Ness Monster’
While industries such as finance and banking have used AI tools for years to automate intake tasks, it’s been challenging for pharma to develop similar tools because AE case reporting requires a high level of decision-making usually by healthcare professionals. To experts in pharmacovigilance, it can be like the elusive hunt for the Loch Ness Monster, says Patrick Caubel, Global Head of Worldwide Safety at Pfizer. “It feels like everybody is talking about artificial intelligence, but nobody has seen it. There’s a lot of buzz around these tools but I would say at an industrial scale, it’s far less easy to implement,” he says.
Caubel and his team began developing these tools in 2014, conducting a pilot to test out AI’s performance on these tasks, as well as to narrow down the best platform. In phase one of the tool, the AE model can make basic intake decisions for a case, such as whether it’s a valid report and whether it is fatal or life threatening.
Sorting and identifying “safety signals,” — instances where there are multiple reports of adverse events with a specific drug — is one of the most important activities in pharmacovigilance. “By automating time-intensive intake activities this tool is going to allow our healthcare professionals to really focus on activities that are most meaningful for patients — signal detection and investigation,” says Patel. “In the end, we will be able to collect better information and optimize the understanding of our medicines to enable patients and providers to make better decisions about which treatments will deliver the best benefit.”