The pace of drug therapy approvals has reached astounding highs. In 2023 alone, the FDA approved 55 new drugs. However, each treatment comes with the potential for adverse events. According to a study from Harvard, adverse events were identified in nearly one in four hospitalizations with adverse drug events accounting for approximately 40% of all adverse events. 

As pharmaceutical companies face the challenges of reactively responding to these events, they also have to ponder how the industry can transition from retrospectively analyzing adverse events to proactively identifying risk factors with any new treatment.

Seeing the whole picture

The importance of proactive pharmacovigilance is often dismissed during the testing phases due to the “rarity” of adverse effects. While adverse effects might take place on smaller scales during clinical trials, the impacts are amplified when taking that treatment to market. For instance, if an adverse event occurs in one out of 10,000 people during a clinical trial, that’s typically defined as a rare adverse reaction. Yet, when that treatment is delivered to one billion people, that “rare” event now affects 100,000 people.

The impact of an adverse event is far from the only flaw of reactive pharmacovigilance. Reactivity is not conducive to the goal of pharmacovigilance which is to detect, assess, understand, and prevent adverse effects before they occur. Yet, it’s estimated that only 1-10% of adverse events are reported.

Additionally, retrospective attempts to track adverse events result in confounding factors. Often clinicians are unable to tell if events are solely a reaction to the drug itself or a coincidence involving other factors. For example, adverse events could be induced by the progression of the disease that the patient is taking medication for instead of from the drug of interest. These third variables make assessing the effects of a drug too complex to determine a causal relationship.

The path to proactivity

Today pharmacovigilance data is a valuable information asset within organizations. It can be used to identify and inform drug discovery and portfolio prioritization. Those who can best leverage the data to evaluate safety have a leg up on their peers. We saw this during the COVID-19 pandemic; the commercially successful vaccines were those that were perceived to be less risky than those with heavy side effects.

Yet having a wealth of data is not enough; we must learn to use it to our advantage. AI will be a transformative power in helping us do so and take the first steps into proactive pharmacovigilance. Artificial intelligence provides healthcare professionals better visibility into the clusters of patients who might be more likely to experience adverse events and the differences in experiences between subpopulations. With the right tools, clinicians will be able to identify which drugs work for what patients and inversely know which patients shouldn’t be exposed to those treatments. Effective use of AI and data can help identify links between adverse effects and specific factors, like basic chemistry, genomics, pharmacokinetics, and more.

Additionally, Large Language Models (LLM) powering generative AI will become a fundamental asset as we leverage this data more broadly. For instance, clinicians/researchers could be able to query the data and hold a “conversation” with the system’s chatbot to ask about things such as the rarity of the adverse event, compare the effects of one drug against other similar drugs, and project the adverse event case volumes in the next year. Comparatively, today’s model relies on an intermediary—a data specialist or scientist—in order to answer the same questions. These additional layers of interactions can result in misunderstandings related to what’s being asked and take more time. By leveraging LLMs, bringing access to data closer to the people forming research questions, and enabling real-time responses, we can help eliminate the risk of information being lost in translation and speed up the process.

The future is not humanless

AI can empower drug manufacturers to experiment with touchless case processing: the receipt, content extraction, product and event coding, assessment, and archiving of the case with little to no human involvement aside from perhaps a modest-quality review step. This will especially be the case with non-serious cases that don’t provide a lot of new scientific information. There will also be tremendous opportunities to take repetitive non-value-added processes, like manual data entry, out of the pipeline and automate them.

Of course, people must be part of the process. There must be checks and balances to maintain control of the automated processes and ensure medical judgment. Soon we will move into automated processing or decision support, where healthcare professionals make the final judgment in consultation with the AI-processed data.