Artificial Intelligence Meets Pharmacovigilance

Artificial intelligence (AI) refers to computer systems designed to carry out tasks typically associated with human intelligence, such as learning, reasoning, problem-solving, perception, and decision-making. AI encompasses a machine's ability to mimic human behavior and perform functions typically associated with human thinking and mental processes. Machine language (ML) is a subset of AI that enables a computer system to learn and improve from experience. ML uses algorithms to analyze large quantities of data and make informed decisions.

PV is the science of collecting, detecting, and assessing adverse events associated with pharmaceutical products for the ongoing monitoring and understanding of their safety profiles. PV is a compliance critical science and is highly regulated with diverse regulatory requirements in different parts of the world. Evolving regulations and novel therapies based on genes, tissues, cells, and organ transplants are broadening the scope of tasks and expertise required in PV, necessitating the need to generate robust and more extensive post-marketing evidence. Given that these therapies often involve complex biological systems and interactions, PV professionals are required to track and evaluate a much broader spectrum of adverse events (AEs) and other associated outcomes such as genetic modifications or immune responses that are usually difficult to predict and monitor. The emergence of the need to monitor and assess real world data collected through social media platforms, electronic health records, medical claims, medical imaging, large patient databases, and data from wearable technologies and mobile applications is resulting in rising case volumes. An individual case study report (ICSR) is a documented report of a single patient’s experience with one or more suspected AEs associated with the use of a medicinal product or device. It serves as a fundamental unit of data collection in PV, helping to monitor and evaluate the safety of a product. Currently, ICSR collection and processing are mainly done manually, involving PV professionals to perform numerous routine tasks before data can be made available for assessment and aggregate analysis. The adoption of practices such as use of AI-driven platforms to automatically extract, classify, and code data from ICSR sources; enabling automated triage and case prioritization; use of AI/ML tools for early identification of safety signals by analyzing ICSRs alongside other data sources and fostering Human-AI team where routine tasks are automated, and human focus shifts to critical analysis and risk mitigation can significantly optimize PV process performance. Early detection of safety risks can help companies implement measures to mitigate the risk before it has a significant public health impact.

AI-driven predictive analytics can help identify which ICSRs are likely to require follow-up, providing valuable insights for performing efficient and focused follow-ups to gather additional case information from consumers and healthcare professionals. AI models can analyze and predict when additional information is needed for a robust case assessment. This proactive approach enhances the ICSR management process, ensuring that essential data are collected to fully address safety concerns and facilitate better decision-making. Artificial intelligence tools can also be trained to identify safety information from large volumes of scientific literature and social media posts, which otherwise is a resource-intensive task.

Advanced AI-driven tools have already shown effectiveness in detecting adverse events from a wide range of data sources.  These technologies offer a scalable approach to managing the increasing volume of ICSRs, thereby aiding PV professionals to make timely and informed decisions. The primary aim of PV is to minimize the occurrence and severity of medicine-related risks by promptly analyzing suspected adverse reaction reports and extracting relevant health data to uncover potential safety signals early.

With the ability to efficiently process and analyze large datasets, generative AI (type of AI that is designed to create new content, such as text, images, audio, video, or even code, by learning patterns from existing data and generates original outputs that resemble the training data, often enabling creative or human-like production) has numerous applications with significant potential in PV, offering advantages such as improved detection of relevant adverse events, risk prediction based on data trends, and efficient drug and device development.

AI can be used to read and extract relevant text/information from various sources, such as scientific literature, adverse event forms, social media, medical records, discharge summaries, and analyze unstructured data from the source documents to create a structured ICSR in the safety database with data organized in appropriate fields to support generation of reports in various formats such as ICH E2b, CIOMS-I, US FDA MedWatch, line listings, summary tabulations to allow for standardized reporting of the data to the regulators and other stakeholders.  AI can also perform duplicate ICSR search in the safety database to identify if the AE report received is an initial ICSR or follow up to an existing ICSR. AI can be used to automatically code drugs and clinical terms (including events, medical histories, and diagnostic data) in the safety database using drug and medical coding dictionaries. The identification of adverse events and the determination of their seriousness criteria ensure timely detection. Adverse event seriousness is also an important criterion in determining reporting timelines. The increase in the volume of safety reports necessitates the exploration of solutions that facilitate the rapid assessment of reporting timeline requirements for case prioritization. AI can determine the seriousness of adverse events in spontaneous, solicited, and medical literature reports. The evaluation of the causal relationship between the drug or device and the event is currently a manual process that relies on clinical judgment and expertise. Artificial intelligence techniques can be leveraged to automate causality assessment, taking into account case attributes such as patient demographics, co-morbid conditions, temporality, medical history, concomitant medications, and analyzing past -event data in the safety database.

Signal detection is the process of examining evolving adverse event profiles of a drug or device to identify possible safety signals. The use of AI in signal detection and management can enable faster and more accurate detection and evaluation of safety signals.

The automation and machine learning models can optimize PV processes and provide a more efficient way to analyze information relevant to safety. However, more research is needed to identify if this optimization has an impact on the quality of safety analyses. It is expected that its use will increase in the near future, particularly with its role in the prediction of ADRs. While AI has immense potential in PV, it also presents certain limitations and challenges to its successful implementation.

Data privacy and security: The use of AI involves processing large amounts of sensitive health information, raising concerns about data breaches and privacy. Ensuring compliance with data protection regulations (e.g., GDPR, HIPAA) is essential.

Data quality and quantity: Extensive training of the AI algorithm with substantial quantities of relevant and high-quality data is required in order for the algorithm to process and analyze data accurately. A lack of a high-quality PV dataset to train the algorithm will be a deterrent to harnessing all the required benefits from AI-based PV.

Scientific acumen and precision: PV data assessment is not a standardized or straightforward process, as it involves multiple decision-making points that vary depending on the amount and type of information available, the product, disease condition, and patient characteristics. It requires expert evaluation by trained health care professionals. The AI tool has to be strong enough to think and work like a trained PV professional, identifying, assessing, and flagging safety risks with quality and consistency in a real-world setting.

Human intervention: The intention of AI should not be to replace PV professionals, but rather to support them in their consistent decision-making, enabling them to better handle the overwhelming amount of data that would otherwise be manually curated and monitored for ongoing surveillance requirements. Through this supported decision-making, PV professionals may have more time to apply their knowledge in assessing cases rather than spending it on transactional tasks to capture pertinent data within a safety database. It will change their work from volume-based to value-based. Intellectual services may be key to more proactive decision-making, which is necessary to meet regulatory requirements and enhance patient safety.

Evolving regulatory landscape: Regulatory authorities are working to establish clear guidelines on the use of AI in PV, including algorithm validation, data interoperability, and data privacy. It will take considerable time for the use of AI in PV is likely to earn the complete trust of regulators and companies worldwide.

High implementation and maintenance cost: For large-scale applications based on database-wide analyses, AI will have slow processing speeds and high cost, due to which conventional machine learning methods are better suited to perform the task. The development, implementation, and maintenance of AI systems in PV involve significant financial and technical investments. This includes the cost of data curation, model training, infrastructure upgrades, and staff retraining. Smaller organizations might find difficulty justifying these investments without clear returns in terms of efficiency or compliance benefits.

Successful AI integration into PV processes and realizing its benefits requires thoughtful consideration of use case scenarios, strengthening technical capacity, upskilling, and aligning people, as well as implementing strong governance frameworks in an already highly regulated space to ensure controlled and effective use of AI.

A significant aspect is the adoption of an augmented PV approach, where AI enhances human decision-making and efforts, rather than solely controlling the end-to-end PV processes. Furthermore, it is essential that openly available training datasets and clear guidelines for AI implementation are available. Stepwise implementation strategies, supported by real-world case studies, will help establish best practices and inform policy development.

Conclusion

Artificial intelligence presents a significant opportunity to modernize PV operations. AI also creates an opportunity to transition from traditional PV to proactive real-time surveillance strategies and early detection of safety risks. Automation and machine learning models can aid in PV processes, providing a more efficient means of assessing information relevant to safety. It can be usefully applied to certain aspects of ICSR processing and evaluation, but the current AI models require careful human supervision. AI should be viewed as a tool to supplement human decision-making, enabling faster, more informed judgments in the realm of drug and device safety. Practical experience with stepwise implementation of AI for ICSR processing and evaluation will likely provide important lessons that will inform the necessary policy and regulatory framework to facilitate widespread adoption and provide a foundation for further development of AI approaches to other aspects of PV.

Conclusively, AI alone may not be able to overcome all challenges, but by forming a human-AI team, PV can be advanced. AI-augmented PV should be considered as a novel approach to support and enhance human decision making.

References

1. U.S. Food & Drug Administration. Using Artificial Intelligence & Machine Learning in the Development of Drug & Biological Products Discussion Paper and Request for Feedback; May 2023 (Revised February 2025)
2. Artificial intelligence in pharmacovigilance CIOMS Working Group report Draft; 1 May 2025
3. European Medicines Agency. EMA/CHMP/CVMP/83833/2023: Reflection paper on the use of Artificial Intelligence (AI) in the medicinal product lifecycle; 09 September 2024
4. Ball R, Pan GD. “Artificial Intelligence” for pharmacovigilance: ready for prime time? Drug Saf. 2022;45(5):429–38.

--Issue 06--