AI and the Future of Pharmacovigilance

As Data Explodes, Safety Scientists Need to Be More Tech Savvy​

Young people interested in a career in pharmacovigilance are learning about computer science and artificial intelligence (AI) in college. What has previously been a passive science is getting more active, as pharma companies and health authorities recognize the need for technology to ensure that medical treatments are safe and effective.

Bruce Donzanti, PhD former Sr. Grp. Director, Global Regulatory Pharmacovigilance Innovation Policy at Genentech, joined the Safety Signals Podcast for a lively discussion about the evolution of pharmacovigilance and where the field is headed.

“I have great faith that pharma…is starting to become more proactive,” Donzanti says. “We all know that pharmacovigilance for decades has been a reactive science. You wait for something to happen and then you try to figure out what’s going on and I really do think that’s changing…I have worked with several groups, both within my company and with other outside functions where several companies have been involved, where we have actually taken approaches of systems pharmacology and systems toxicology…for predictiveness, of being able to look at an early development, look at a molecule and actually predict, before you even get into phase two, what could potentially happen.”

The Importance of Proactivity

According to Donzanti, such approaches are forming the foundation for a future of pharmacovigilance based on data and driven by artificial intelligence technologies, such as machine learning. For the past few years, Donzanti and a colleague from Genentech have been teaching a course on machine learning in pharmacovigilance, along with the  deputy director of safety surveillance from the FDA, as well as the head of safety surveillance at the MHRA.

“Those of us who are still learning the field see the potential value of not only the use of AI, but the necessity to have to use it,” Donzanti says, adding that progress can only occur when pharma and health regulators work together. “Nowadays there’s a very strong openness to share ideas and try to come up with better solutions.”

AI Can Help Eliminate Bias

Critics of AI argue that machine learning algorithms make it difficult to interpret signals accurately, because the systems are opaque. Donzanti maintains that even without AI there are disagreements.

If done properly, AI can eliminate the human bias that causes problems in pharmacovigilance today, based on which companies experts have worked at and even where they went to school. “I work with a colleague of mine at Genentech and he always brings up the question, who’s really the black box? Is it a human being or is it an AI?” Donzanti says.

Getting to the Right Data to Find the Right Answers

Increased data volume and complexity, from both traditional and new sources, is making it more difficult to separate signals from the noise.

We need “better tools to help analyze the data,” Donzanti asserts. “What if you get what you think is a potential signal from spontaneous, and then you have another bucket of real world data that’s not quite giving you that same potential signal?”

Donzanti says that smart minds in pharmacovigilance, AI, and statistics must work together to develop the right algorithms, but it’s been difficult to get regulators to agree on the best approach.

Noting some global regulatory deharmonization in the regulations globally. Donzanti says that one camp wants pharma to look at everything so nothing is missed, while another camp prefers selecting a smaller subset of quality data to analyze.

“Some of the agencies, like the FDA, have been very proactive in saying we’ve got to get smarter,” Donzanti says. “We always thought the more data the better, and that’s actually not really turning out to be the case.”

If the best information can be identified early, or more rapidly due to AI, then it will be much easier to eliminate false positives and negatives and get to the truth. Noting the need for a better way to filter out quality data, Donzanti says that “just putting 100 more people in a room and trying to look at documents and bounce ideas off of each other isn’t going to work anymore.”

Now Is the Time to Work Things Out

“This is the time for all of us to work this out,” says Donzanti. “I think that’s the way we have to do it to get this right from the get go.”

Donzanti is excited about the future prospects of pharmacovigilance and anticipates a technological transformation on the clinical and marketing aspects of drug development as well.

“I think the future is very good because I think we’re getting smarter and smarter about how we use all of these data sources,” Donzanti says. “The AI technology is there, and with some fine tuning and education it will smoothly feed into what safety scientists need to look at.”

This post is based on an episode of the Safety Signals podcast. To hear more, check us out on Apple Podcasts, Spotify, Google Podcasts or on our website.

The views of the hosts and guests featured on Safety Signals are their own and do not necessarily reflect the views of Saama or the individual companies for which the guests may work.