By: Raj More

Introduction

Twenty-five years ago, pharmacovigilence (PV) was all about case management. Physicians and nurses provided information about adverse drug events which was processed and reported to regulatory agencies. The work was largely paper-based and the focus was compliance. Skip ahead to the present, and the use of advanced technologies like artificial intelligence (AI) have been an absolute gamechanger for PV.

Based on techniques like Generative AI, Large Language Models (LLMs) and Agentic AI, AI-based tools have dramatically improved data processing, pattern recognition, and automation capabilities within the PV workflow. This new toolkit has shifted a system with a historically reactive posture to one with proactive and predictive capabilities for signal detection and management, risk management, aggregate data analytics and insights. It is even changing business expectations.

Some AI technologies are now becoming indispensable to the pharmaceutical industry and safety organizations in managing the growing volume and complexity of medical data, the rapid introduction of new drugs and improving operational speed and efficiency within an increasingly complex regulatory environment.

Given that pharmaceutical companies and regulatory agencies are at different stages of realizing the full scope of benefits from AI tools, it seems a good time to survey the field as it currently stands and provide some tips to help safety organizations unlock the full potential of AI applications. It is also worthwhile to look ahead at some emerging trends and a selection of factors that will influence the continued development of these remarkable tools.

AI-Based Tools for PV Activities

Legacy PV infrastructure and management systems were based on manual processes that were slow, costly in human resources and prone to human error. They used traditional statistical methods that depended on structured data, which could increase the risk of missing important adverse drug events (ADEs). These methods and processes were increasingly overwhelmed by the explosive growth of new structured and unstructured data sets following the health science sector’s digital transformation, in addition to the growing body of research and medical literature.

AI has undergone significant progress in the development of new applications that are highly effective in automating tasks, providing actionable insights through analytics, accelerated reporting and supporting regulatory compliance, and introducing predictive opportunities to proactively identify potential adverse drug events and emerging safety signals. These software products and platforms use models trained with modern AI techniques to provide this incredible range of valuable new capabilities and efficiencies, as evidenced in the following examples.

A new generation of AI models has played a key role in improving core surveillance activities, reducing false positives thereby delivering efficiency gains, and identifying potential false negatives that might otherwise be overlooked.

Deep Learning, which is a subset of ML, has been used to create architectures like neural network models that are inspired by the way the human brain works. These models can be trained on unstructured data for tasks such as analyzing, classifying, and extracting meaningful information from case files or medical literature. Their performance can be continually improved through retraining or fine-tuning and, in some applications, by incorporating new data and feedback from users.

Powered by ML, Natural Language Processing can convert unstructured data into searchable data that is useful for screening. Given the wide range of text sources used in monitoring drug safety (including electronic health records, clinical notes, literature articles and case reports), this capability enables tasks like case analysis and comparison and can support case data capture, translation, summarization, support reporting and regulatory compliance. It has also become helpful for analysis of complex real-world evidence, such as data obtained from publicly available social media platforms.

Now, agentic AI technologies are creating next-generation operating models that drive even further applications for intelligent systems already used by industry and health science agencies. AI agents provide an intelligent layer that organizes and manages workflows and software functions across multiple platforms, assisting with case intake, product and event coding, follow-up query management, PV document authoring, surveillance, signal detection, risk oversight and a host of other activities. They are also designed to support the scalability of operations.

“AI has undergone significant progress in the development of new applications that are highly effective in automating tasks, providing actionable insights through analytics, accelerated reporting and supporting regulatory compliance, and introducing predictive opportunities to proactively identify potential adverse drug events and emerging safety signals.”

Tips for New Technology Adoption

Case management remains a major PV pain point and the greatest driver of operational costs. Large pharmaceuticals are making big investments in monitoring and reporting capabilities. However, technology adoption rates have varied as companies juggle increasing requirements and tight operating margins. The best advice for those in the early stages of adoption is to clarify their corporate priorities, such as analytics and reporting or case processing cost optimization. Then once they have a roadmap, they can concentrate their investment in the most appropriate solutions within their budget for those top corporate priorities.

To build the necessary capabilities, pharmaceuticals can either improve or build new systems internally, or optimize the cost by purchasing commercial solutions from vendors. A single across-the-board system replacement tends to be difficult as well as very expensive. Modular software products are often preferable as they allow companies to take a more economical, staged approach to system updates. This incremental approach breaks updates into bite-sized chunks that better address specific pain points and business priorities and deliver value much faster than one “big bang” project.

Companies require additional resources before embarking on a technology update. Internal subject matter experts are critical, as a well as someone designated as the champion of the implementation process. Someone must also be designated to serve as the internal business process owner after the system goes live. This person (or people) must have an in-depth understanding of the new system and be able to explain the system, reporting processes and supporting data to regulatory inspection agencies.

Time and time again, responsibility for managing the implementation and subsequent operation of the new system is given to someone already bearing full-time responsibilities. Assigning someone this as an “off the side of their desk” kind of task can lead to problems. It helps that many new AI products are delivered as SaaS solutions. These systems are often easier to deploy and maintain, automate many time-consuming activities as well as reducing the effort required for testing and validation. Much of the training has also been automated, reducing the necessary investment in time and further streamlining implementation.

Emerging Trends Influencing the PV Transformation

More and more companies are implementing applications that improve their processes across all aspects of safety. Technology leveraging AI helps reduce operational costs, such as the four to six hours that used to be needed to process a medium-complexity case end-to-end. These tools are making suitable case processing workflows virtually “touchless”, requiring human effort primarily for Human-in-the-Loop (HITL) reviews, which is very appealing when budgets have not kept pace with workloads. They also improve the accuracy and efficiency of signal detection and reporting and are minimizing false positives, while helping identify potential false negatives, and providing explainable results to support regulatory review.

Looking at emerging trends, the first involves leveraging real-world data. Currently, pharmaceuticals focus on analysis of clinical trial data and post-market real-world data. Using AI tools to integrate these data sets allows the detection of more subtle safety signals or trends. On the regulatory side, agencies are keen to expand their use of real-world evidence (RWE). Data integration would also speed up the rate at which safety issues and trends are identified, which could translate into significant savings over the course of clinical development. By adding quantifiable value, PV evolves from a cost centre into a strategic contributor to corporate value.

Another emerging trend in Pharmacovigilance is the automation of PV document authoring. Recent advances in Generative AI and Large Language Models (LLMs) have made it possible to significantly accelerate the creation of aggregate reports, Pharmacovigilance Agreements (PVAs), Pharmacovigilance System Master Files (PSMFs), Risk Management Plans (RMPs), and other regulatory documents. However, document generation is only one part of the process. True enterprise-grade solutions must also support planning, workflow orchestration, HITL reviews, approvals, version control, audit trails, and GxP compliance throughout the document lifecycle.

A final trend being observed is the integration of product quality and drug safety. On both the regulatory agency and pharmaceutical sides, the integration between product quality and drug safety has increased significantly. Technologies such as AI are already supporting the integration of these functions within one single system that provides both capabilities.

Future Prospects for AI in PV

In the big picture, has the adoption of AI-based tools improved overall drug safety for the average person? Social media has certainly magnified public interest in drug safety while fueling safety concerns at a time when regulatory bodies and pharmaceutical companies are having to contend with a growing number of new drugs and the enormous volume of both spontaneous and clinical data. New AI technologies, processes and techniques are leveraging this data flow to improve safety outcomes.

The increasing interest in patient-centric safety is an emerging driver of the use of AI in PV. Drawing on rapid advances in genomics and biomarkers, the field of personalized medicine is leading the way in developing more personalized therapies, including certain immunotherapies for cancer. A similar approach will likely emerge in drug safety, with technology helping evaluate drug safety within the context of an individual patient. Patient health data is being collected from a growing range of sources, including wearable devices, chatbots and social media posts. By integrating this information into AI models, patients could conceivably receive information that helps them evaluate their individual risk of adverse drug reactions.

Looking down the road, there are possible speedbumps ahead. A potential future challenge to the further implementation of AI may be computing costs. The costs of AI technologies to augment systems used for signal detection and regulatory reporting can be high. These costs are currently being reduced through continued optimization by vendors establishing more efficient pricing models and building trusted relationships within the industry and with regulatory bodies. As major tech companies continue to invest heavily in AI infrastructure and business models evolve, there is a possibility that the overall cost of deploying AI solutions and prices paid by customers may climb significantly.

While significant advances in patient safety have been gained, PV must always contend with the extraordinary complexity of human physiology. The complexity only grows when factors such as multiple medications or a patient’s additional medical conditions are added to the equation. The cost of developing and testing a new product to understand its safety profile runs into the hundreds of millions of dollars. Even supplementing clinical trial data with real-world data cannot produce absolute certainty. While human trials can go a long way towards identifying potential safety issues, the complexity of human responses to medicines cannot be underestimated.

There is some hope in emerging technologies like quantum computing. This technology is already seeing early-stage applications, including quantum-enhanced algorithms. Quantum computing has the potential to significantly increase computational capability for massive combinatorial problems involving interactions among multiple drugs and the human genome. Once commercially viable, which many experts believe could occur over the next decade, it is expected to outperform classical supercomputers for certain classes of combinatorial and optimization problems, and allow more accurate prediction of complex interactions among drugs, the human genome, and genetic variations.

Conclusion

AI-based products draw on a broad range of architectures and techniques enhancing the speed, precision and efficiency of activities across the PV ecosystem. Different organizations are turning to different AI applications.

Even as pharmaceutical companies and regulatory authorities figure out the growing list of options for updating or re-designing their systems, these new tools are already transforming the PV sector. The benefits have been significant, as already discussed. The transformation is still, however, a work in progress. As the technology continues to evolve and its implementation matures, and notwithstanding its current and future challenges, the PV ecosystem will continue to evolve, enhancing efforts to ensure drug safety for patients.