In addition to our patient-facing and provider-facing agents, we have voice AI agents that call payors on behalf of healthcare providers to collect detailed patient benefits and prior authorization data. On these long and complex phone calls, our AI agents operate on a multi-modal, multi-model platform adhering to each customer’s unique standard operating procedures (SOPs) and industry regulations. On each call, these agents can gather over 200 call outputs, including plan details, network status, cost-share for the patient, prior authorization requirements, and more.

Within these long phone conversations, reportable events like an injury, side effects, safety findings, or product complaints resulting from medical interventions related to drugs can be disclosed. At Infinitus, we understand that vigilant detection and reporting of potential adverse events (PAEs) are critical for pharmacovigilance and patient safety. Organizations supporting pharmaceutical companies must have a robust PAE detection process so that their customers can continue to maintain regulatory compliance and demonstrate a commitment to patients’ safety. By identifying and managing PAEs, pharmaceutical organizations and their partners  can work together to develop safer and more effective medications.

How Infinitus handles potential adverse events 

While our purpose-built conversation AI system handles the call navigation and patient data collection, our SAGE (Safety Adverse-event Guidance Engine) system employs sophisticated natural language processing (NLP) and machine learning (ML) models to scan every call and identify potential adverse events.

Detecting PAEs in benefits verification and prior authorization calls is challenging due to their rarity. No matter how rare, if an event is discussed, we must flag it. To overcome this high-recall problem, we utilize advanced models that go beyond simple keyword detection. This approach ensures accurate PAE identification while effectively minimizing false alerts.

Accurately detecting specific events or intentions within conversation cannot purely rely on keywords. For instance, consider the task of identifying patient termination events, such as when a patient has passed away. A simplistic approach might involve creating a glossary of keywords and phrases commonly associated with this intent, such as “passed away,” “deceased,” “expired,” “not alive,” or “dead.”

However, relying solely on keywords presents significant limitations and can introduce considerable noise into the detection process. For example, the word “dead” can frequently appear in speech-to-text (STT) transcriptions as a result of background noise, leading to false positives. Furthermore, we must account for potential transcription errors, where words might be dropped or mistranscribed. An example of this is “deceased” being incorrectly transcribed as “they ceased”

To overcome these challenges, a more sophisticated system like SAGE is required. SAGE goes beyond mere keyword matching by incorporating advanced contextual understanding, and uses multi-modal inputs (text and audio). Contextual understanding is improved by jointly aligning text and audio, which corrects the aforementioned issues caused by STT errors.

SAGE can also leverage keywords and phrases provided by customers, tailoring its scanning capabilities to detect events relevant to their specific products. Crucially, it analyzes the surrounding contextual cues where these keywords appear, enabling it to differentiate between a genuine event and a coincidental use of a word. This allows for a much more accurate and robust detection of critical information, minimizing false positives and ensuring that relevant events are not missed due to transcription anomalies or ambiguous language.

When our AI detects a potential PAE, we follow a rapid and reliable four-step process:

1. AI detection: The SAGE system automatically flags a potential PAE (e.g., a report that a patient is deceased and their plan is terminated).

2. Rapid human verification: The flagged call is immediately escalated to one of our human experts for review. Our human experts listen to the call and review if the PAE report is accurate. We are typically able to flag tasks needing additional customer review in as little as 15 minutes after a call completes. 

3. Immediate customer notification: Once our expert confirms that a PAE has likely occurred, we instantly alert the customer’s designated team via email, so they can initiate internal procedures for PAE analysis. 

4. Secure portal review and data release: The customer team can log in to our customer portal to securely listen to the call recording and perform the appropriate next action. This data (whether a task needs additional review from our customers ) is also available via our APIs.

Human-in-the-loop: collaborative oversight

At Infinitus, we maintain a steadfast conviction that AI enhances human intelligence rather than superseding it. AI undertakes laborious tasks, thereby enabling people to apply their expertise and sophisticated decision-making capabilities – and to do more fulfilling work.

Our PAE detection system (SAGE) identifies calls containing PAEs for human review, thus liberating teams of reviewers from scores of routine, time-consuming verifications and allowing them to concentrate their expertise on crucial, intricate situations. Furthermore, human involvement in this process serves to validate the quality of our AI system’s outcomes, thereby fostering increased confidence in the system.

How to get started 

If PAE detection and compliant escalation are critical to your workflows, let’s talk. Our team can walk you through SAGE in the context of your SOPs, data flows, and audit needs.