Can AI Replace Signal Detection?

Can AI Replace Signal Detection? The simple answer we have is, not yet, and not entirely. For sure, Artificial Intelligence (AI) is, or will be, capable of rapid scanning of humongous Pharmacovigilance databases, including FAERS, VigiBase, and EudraVigilance, and identifying the usual patterns of adverse events specific to certain drugs. But once the patterns are out, the human judgment becomes essential. AI can surface the hypotheses and the rest, i.e., validation and contextualization, is still dependent on PV scientists. In short, the industry is exploring AI-assisted Signal Detection, not AI-led Signal Detection. 

Here is a quick backstory of AI entering the picture in signal detection.

A 2020 study (Preventive Adverse Drug Events: Descriptive Epidemiology) found that a substantial portion of Adverse Drug Events (ADEs) is preventable. The reasons cited included inadequate drug monitoring and delayed recognition of the safety issues as key contributors. Insufficient drug monitoring was also noted in another study as the root cause of 60% of preventable ADEs. These failures are primarily attributed to poor signal detection in pharmacovigilance systems.

It might be misinterpretations in the analysis of large data sets, or loose ends in human intervention.  Sometimes, these kinds of misses can cause serious safety issues; hence, the process demanded automation that not only automates signal handling but also pushes for clearer, more precise identification of safety challenges. There came the integration of Artificial Intelligence (AI) in signal detection.

Rooting back to the initial question, can AI replace Signal Detection? We can say it can spot signals and assist PV experts in redefining drug safety workflows, but overtaking human-led signal detection might take some time.

The Role of Artificial Intelligence (AI) in Signal Detection

As many say, leveraging machine learning (ML), natural language processing (NLP), and data mining enables processing structured/unstructured reports. And Pharmacovigilance signal detection, too, can leverage the same. The structured/unstructured reports include spontaneous reports, electronic health records (EHRs), literature, and real-time social media conversations. The reports consist of disproportionate analyses, unclassified narratives, duplicate case handling, and many clustered events, which safety teams may have to spend many man-hours reviewing. The AI models are expected to automate all these processes, giving safety teams breathing room for evaluation and risk management.

Predictive surveillance: AI enables predictive surveillance. It helps detect pharmacovigilance signals by predicting which patient groups, drug combinations, or comorbidities are likely to experience adverse drug reactions. In a way, AI continuously learns from new data and adopts a proactive, personalized approach to drug safety.

AI in Signal Detection and the Gaps

Although AI systems excel in many areas of pharmacovigilance signal detection, there are many grey areas to look after, such as data quality issues – data-bias, unreported data, the absence of external validation, and the black box behaviour, which impose challenges to regulators and qualified persons for pharmacovigilance (QPPVs) to understand signal decisions.

Falling back to the same question, can AI replace signal detection?

With the opportunities and challenges discussed, AI can work more effectively as an integrated capability rather than a plug-in tool. Either for mapping PV processes, curating multiple datasets, and building narratives that withstand health authority reviews, AI in pharmacovigilance signal detection can be a strategic opportunity for life sciences organizations and CROs. In simple terms, the same 30% rule in AI can be applicable here.