Patient safety signals in non-traditional channels: AI and NLP uncover critical information

The clinical research industry has experienced a seismic shift over the past few years, from the pandemic to decentralised trials to technological innovations. Concurrently, the internet's growing convenience and accessibility are a driving force in patient behaviour. Patients are increasingly using non-traditional channels, such as social media, to voice their concerns and seek medical guidance.

In fact, a recent study found that 85% of patients tend to seek health information online using social media. At the same time, the increase in social media use around the globe, which has tripled in the past decade, is a serious indication of shifting trends in patient behaviour. As these trends evolve, clinical research professionals are turning to new avenues to gather patient feedback. With novel technologies, such as artificial intelligence (AI) and natural language processing (NLP), clinical trial stakeholders can better distill and detect drug safety concerns voiced by patients.

Traditional PV landscape: Manual processes and fractured systems fall short

The traditional pharmacovigilance (PV) landscape has relied heavily on direct feedback from healthcare providers (HCPs), patient registries, and regulatory databases. The sourcing and collating of this information through structured data capture networks frequently leads to underreporting or the omission of valuable details that provide more context to the adverse event (AE). Additionally, traditional reporting methods may rely on manual or fractured systems that are quite time-consuming.

These efforts alone may not adequately examine or address the true nature of drug safety events. Research shows that improving visibility into patient safety events has been a chronic issue in the life sciences industry. The latest metrics show that traditional methods may not be sufficient for detecting the full spectrum of AEs. In fact, the Federal Drug Administration (FDA) estimates that only between 1% to 10% of adverse drug reactions (ADRs) reach the Adverse Event Reporting System (FAERS). In other words, almost 90% of ADRS remain unreported. The drastic dissonance between the number of reported AEs and the true number of occurring drug safety events underscores the need for an innovative approach to pharmacovigilance strategies.

Introducing AI and NLP to PV efforts in evolving channels

As patient behaviours evolve, PV experts realise that new solutions built to work in tandem with the shifting digital landscape are essential to protecting patient safety. Compared to traditional channels, such as registries and databases, online channels provide unstructured feedback from patients. Information sourced from social media platforms, online forums, and discussion groups provides authentic patient conversations that provide more detail into potential AEs.

The difficulty in understanding information across these non-traditional channels is learning to navigate unstructured data. The challenge of standardising and operationalising information across diverse formats, as opposed to a database, can complicate the process of AE identification. Navigating the nuances of human language variations, such as emojis, slang, and colloquialisms, complicates the process of patient safety detection across these new channels. However, the use of AI and NLP has provided PV teams with the resources to analyse and organise varying forms of unstructured data.

Leveraging AI and NLP in social media monitoring

This technology, with its continuous learning capabilities and vast ontologies, provides the ability to identify signals that traditional methods may not have noticed. Through AI and NLP, organisations are able to develop algorithms, word banks, and specific terms or patterns that automatically identify patient safety events. The use of conceptual models, designed to connect medical terminologies with safety language, can identify patterns and word proximity to classify AEs.

For instance, after the rollout of COVID-19 vaccinations, healthcare providers’ limited patient contact hindered their ability to assess the safety and potential side effects of vaccinations. To remedy this disconnect, experts increased their PV efforts across social media. With the help of AI algorithms and pattern recognition, organisations effectively identified patient safety signals related to myocarditis. The effectiveness of AI and NLP in safety event recognition not only aids in uncovering patient concerns, but also in the workload and efficiency of PV efforts.

In one use case, the utilisation of AI and NLP allowed for the review of 7.7 million social media posts across 300 data sources, 91 languages, and 38 countries. Through the use of this technology in examining social media sources, one organisation was able to identify over 100,000 events critical to the reporting process. Traditional AE reporting has relied on manual efforts across disparate systems in order to align safety events. With the implementation of AI and NLP, organisations can both improve strategies to detect safety events and cut down on the time-consuming process of uniting fractured systems and protocols.

AI and NLP have been critical to the collection and analysis of unstructured data across growing channels online. As the use of non-traditional channels expands across social media, online forums, chat rooms, and discussion groups, PV efforts are shifting to accommodate evolving patient tendencies. Adapting strategies to expand AE detection efforts aligns with the goal of increasing patient safety. Diversifying the analysis of new channels through AI and NLP will allow organisations to improve patient safety efforts and drug performance moving forward.