[Image from Christopher Burns on Unsplash]
In particular, natural language processing (NLP) could help health authorities transform adverse event reports from COVID-19 vaccines into structured data while reducing the need for manual data entry. But more importantly, the technique can accelerate adverse event detection while giving regulatory authorities the potential to predict and prevent future adverse events, according to Updesh Dosanjh, practice leader of technology solutions at IQVIA.
NLP could thus play a role in persuading the public that COVID-19 vaccines are safe and effective. As of October, only half of Americans said they’d be willing to get a COVID-19 vaccine, according to a Gallup poll. Yet achieving herd immunity would require that roughly 70% of the population receive the vaccine, according to Cornell University researchers’ estimate.
Further complicating matters is the fact that many COVID-19 vaccines require two doses, and the durability of COVID-19 immunity remains unknown. It is conceivable that boosters for a COVID-19 vaccine would be necessary, making mass COVID-19 vaccination a perennial challenge.
Traditional pharmacovigilance is ill-equipped for COVID-19 vaccines
The potential scale of mass vaccination efforts will quickly overwhelm traditional pharmacovigilance strategies, Dosanjh said.
Health authorities will likely recommend that seniors and people in high-risk groups be among the first to receive COVID-19 vaccines. Determining whether COVID-19 vaccines cause adverse events in that population is a prerequisite for efforts to roll out the vaccine to the broader public. “We’re not going to be able to use the traditional approach of rolling [the vaccine] out into a small population group, monitor them heavily, then into a larger population group and so forth,” Dosanjh noted.
The potentially unprecedented speed of vaccine deployment also doesn’t support traditional timelines for analyzing pharmacovigilance signals, which are hypothetical risks linked to a medicine with some supporting evidence. “A typical signal analysis can take three months to do,” Dosanjh said.
There’s little appetite for such an approach during a pandemic. “With a traditional signal injection process, there would be a good chance that every high-risk patient has taken the vaccine by the time you’ve gone through your typical signaling process,” Dosanjh said.
Added to that are the potential unknowns of delivering relatively novel vaccine types — based on, for instance, mRNA or chimpanzee adenovirus vectors — at scale.
“You’re talking about vaccines that have potential liabilities. It’s an unknown unknown,” Gary Nabel, chief scientific officer at Sanofi, told the Financial Times.
Because of the uncertainty, it is crucial to swiftly detect potential problems. “The Pfizer vaccine, for example, uses a two-injection scenario delivered three weeks apart,” Dosanjh said. It would be ideal to know if there’s a potential problem before a patient receives a second dose of the vaccine. “But traditional essentially manual data diving is not going to get you there, no matter how many people you throw at it,” he added.
Enter AI
Government agencies from the CDC and HHS to FDA have embraced data analytics techniques in the fight against COVID-19. The NIH is using AI to analyze thousands of research papers related to the novel coronavirus. And the Medicines and Healthcare Regulatory Authority in the UK also plans on employing AI to review suspected COVID-19 safety events.
Keeping tabs on the mass vaccination effort will require automation tools and artificial intelligence tools such as natural language processing (NLP), which can mine real-world data to assess the safety and efficacy of COVID-19 vaccines as they become broadly available.
NLP algorithms can read and normalize adverse event data. Using automated techniques for data ingestion allows an almost immediate analysis. NLP systems can find patterns in adverse events and identify anomalies. As the technique progresses, it can become more future-oriented. It will suggest phrases and contextual parameters that can uncover adverse event patterns in structured safety databases. The technology also supports simultaneous quantitative and qualitative data analysis, enabling fast adverse detection and monitoring.
Such abilities are crucial, Dosanjh said, because many healthcare organizations are more focused on vaccine delivery than adverse event reporting. And some of the most vulnerable patients — the elderly and those with chronic conditions — will be among the first to be vaccinated.
NLP could push regulatory science forward
As regulatory agencies embrace tools such as NLP to battle COVID-19, they lay the groundwork for using the technique for other regulatory science applications.
“I’m not going to say that NLP is going to work perfectly, but this is going to be a huge test case,” Dosanjh said. “I don’t think there’s any going back once you see even if it works 50%, it’s still 10,000 times better than whatever you are doing today.”
Regulatory agencies have unique advantages when deploying NLP, according to Dosanjh. “Organizations like the FDA are the ones who are trying to use NLP in the most innovative ways,” he said. Because of their charter to protect human health with an often limited budget, “they need to be innovative.”
NLP can help FDA get better at not just detecting adverse events, but predicting and preventing them, Dosanjh said.
Because of the sheer amount of data the agency has, they can compare adverse events between related drugs. “If you’re a manufacturer of a drug, all the data that you have belongs to your drugs. You don’t have access to other companies’ information about drugs,” Dosanjh said. “But the FDA does. They can say: ‘Because we know what happened in the past, we know what could potentially happen.’”
Such capabilities are exciting for regulators, Dosanjh said. New applications outside of vaccine-based adverse event reporting are likely. “You can go to generic manufacturers and say: ‘You do realize that once you see one [adverse event], you’re possibly going to get a whole bunch more,’” Dosanjh said. “‘And rather than wait until you’ve got them, can you do something about it today?’”
Filed Under: Drug Discovery and Development, Infectious Disease
