Historically, adverse events (AEs) were reported through tightly controlled and clearly defined human channels like call centers or nurse lines. But today, patients interact across many more touchpoints – phone, yes, but also chat, text, portals, email, etc. – and that means any of those channels could be the first place a patient mentions an AE.

It also means that AEs can be brought up when a patient is communicating with AI, and no one else is on the line. And with agentic communications on the rise, teams across pharmaceutical organizations are rightfully concerned about whether AI agents are equipped to identify and appropriately act on AEs. 

Here’s what patient support leaders, compliance officers, and brand teams need to know about AE reporting in the age of AI.

Can AI agents recognize adverse events?

The answer, in short, is yes – with a caveat. AEs are rare; finding them is akin to finding a needle in a haystack. But AI agents can detect AEs, as long as they have carefully designed systems that have deep healthcare expertise and product-specific knowledge. This is in part because what a patient says in describing an AE may be very different from the medical terminology used by professionals. For example, the agent would need to be able to understand that “There was buzzing in my ears” equates to a patient saying their ears were ringing, and to know that’s a possible side effect of the therapy in question. 

With modern large-language models (LLMs), this is possible. Models can be trained to have contextual understanding of the domain in which they’re operating. Advances over the last five years especially have made this more possible than ever. But as we know, detection alone doesn’t equal compliance. In order to ensure AEs are appropriately handled, AI agents need to be able to not just detect that an adverse event has occurred, but to make sure the appropriate parties are notified immediately.

How do AI agents log, flag, or hand off AEs?

The real importance of AE detection isn’t just identifying that an adverse event may have occurred, it’s ensuring that it is appropriately documented and necessary next steps are taken. This is where agentic AI is critical – the AI agent can take action autonomously, to ensure the proper procedure is initiated. And while different AI platforms may handle this step differently, at Infinitus, our AI agents follow our customers’ lead and log or hand off AEs in whatever way fits best into their existing workflows. 

Today, our AI system flags potential AEs and immediately escalates them to a human team for review. If an AE is confirmed, Infinitus immediately notifies the designated team at our customer, while also ensuring the full call recording is available in our portal to be listened to.

How can we be sure an AI agent doesn’t miss or misclassify an AE?

Ultimately, this concern is why questions about AEs are some of the most frequently asked by pharma teams today. AI agents are designed to detect AEs with a strong emphasis on recall, meaning they prioritize not missing any genuine events. However, the key to this lies in having at least a small amount of labeled data to guide model development for each customer’s specific needs. 

At Infinitus, our human-in-the-loop approach, combined with continuous error  analysis and model improvements, creates a robust safety net, which gives our customers confidence that our AI agents are both thorough and trustworthy in detecting adverse events. 

How does Infinitus handle adverse event reporting?

Infinitus has designed our system to a) identify all potential adverse events, and b) keep the false positive rate as low as possible. As such, our safety adverse-event guidance engine (SAGE) employs sophisticated natural language processing (NLP) and machine learning models to identify potential AEs that occur during our AI agents’ calls.

SAGE also leverages keywords and phrases provided by customers, enabling the system to tailor its scanning capabilities to detect events relevant to customers’ specific products. It analyzes the surrounding contextual cues where any designated keywords appear, differentiating between a genuine adverse event and a coincidental use of a word.

Because of Infinitus’ real-time API connections and native integrations with leading CRMs like Salesforce, customers are notified instantly when a potential AE is detected. These integrations allow SAGE’s detection capabilities to plug directly into existing workflows, so safety reporting can occur without disrupting established processes.

If you’re interested in learning more about our AE reporting capabilities, including how we’re helping some of the leading pharma companies stay compliant, reach out to us today.