The field of pharmacovigilance has evolved significantly in recent years. While regulatory authorities have long favored technologies such as artificial intelligence and machine learning for monitoring patient safety, the trend has accelerated during the pandemic. “I’ve never seen as much advancement as I have in the last two years,” said Marie Flanagan, director, offering management, Vigilance Detect at IQVIA.
Flanagan predicted that regulators’ interest in AI and ML for pharmacovigilance will catalyze growing interest across the pharma industry in 2022 and beyond. The pandemic has demonstrated to regulators and life science companies that the technologies can help make sense of exploding volumes of pharmacovigilance data points.
“If we look at just the EudraVigilance database [citing the European Medicines Agency database] in the past two years — as of last week, 64% of all information reported on COVID-19 vaccines came directly from consumers,” Flanagan said. “That represents a seismic shift towards consumers curating this information and delivering it to the regulators,” she added. “Previously, that was healthcare professionals or the manufacturers of the drug.”
IQVIA expects the trend to continue. “Patients’ and consumers’ input is going to change the landscape of safety reporting,” Flanagan predicted.
The explosion of social media conversations involving COVID-19 vaccines during the pandemic is another factor that has shifted expectations. IQVIA tracked 50 million social media conversations related to COVID-19 vaccines in four months.
“That kind of volume is indicative of how much information is out there and how much of that information needs to be analyzed,” Flanagan said.
In addition, conversations on social media platforms and online forums focused on seasonal flu vaccinations, COVID-19 drugs and other therapies have also surged during the pandemic.
The trend is “opening the doors for so much more volume and more sources of information to come in about existing marketed products and products that will be coming to the market shortly,” Flanagan said.
Regulators are also proving to be savvy using AI for pharmacovigilance. The Medicines and Healthcare products Regulatory Agency (MHRA) in the UK is “doing wonderful things with AI in their Yellow Card system, for example,” Flanagan said, referring to the system for tracking adverse drug reactions and other incidents.
For instance, the Yellow Card Biobank initiative aims to assess the role of genetic factors in adverse drug reactions.
Flanagan imagines that, in the future, such systems could enable patients to know if they are genetically predisposed to specific side effects of the drug.
Initiatives such as the Yellow Card Biobank point to a future where AI in pharmacovigilance does more than drive efficiency, Flanagan said. While accelerating workflows and reducing manual labor are worthy goals, “ultimately, we want to make sure that what we’re doing has a greater benefit to drug safety overall,” she said.
Advances in natural language processing (NLP) also broaden the scope of information used in pharmacovigilance.
“A classical set of ontologies for NLP would have been structured around traditional sources of information,” Flanagan said. For example, those sources could include adverse events described in scientific literature or reports from a doctor’s registry.
“You might have information in scientific literature that would have given you your adverse event you’d have, you know, your adverse event coming in through a very structured form from a doctor registry or a pharmacy.”
But NLP can now potentially integrate relevant information from various sources from the public domain. “So the places that we’re seeing that include social media, audio sources, voice notes — the audio from call centers is a burgeoning source in terms of safety information or potential safety signals,” Flanagan said.
The volume of data is set to explode further with the rise of AI-enabled call center agents.
Still, challenges remain in using data from such myriad sources holistically.
For audio-based data, for instance, slang and dialects remain a hurdle for NLP. For textual data on social media, there is the question of how NLP might deal with a description of adverse events laced with emojis.
IQVIA’s optical character recognition (OCR) is looking at how to transcribe emojis and emoticons and embed them into its analyses.
“We’re moving forward into an era where social media call centers and AI agents are going to be a force of delivering safety information,” Flanagan said.
The industry and regulators need to have the infrastructure and technology in place to cope with the transition. “That’s the future as we’re all seeing it unfold,” Flanagan concluded.
Filed Under: Data science, Drug Discovery and Development, machine learning and AI