An Interview With...
Gary Barker
Executive Director of Pharmacovigilance, PPD™ FSP Solutions, Clinical Research
Thermo Fisher Scientific
How are leading pharmaceutical organizations currently implementing AI technologies in their pharmacovigilance (PV) workflows, and what specific tasks are most targeted for automation or augmentation?
Leading pharmaceutical companies are predominantly applying AI to case intake and early case processing in PV, where rising Individual Case Safety Reports (ICSRs) volumes and diverse report formats have made manual handling increasingly unsustainable. AI – including large language models (LLMs), which can read and extract information from both structured and unstructured text – is used to support data intake, extraction, and transformation up to the point where information can be integrated into the safety database, while downstream steps like case assessment and medical review largely remain human-led.
Within this intake-to-database span, AI most commonly automates or augments tasks such as: parsing semi-structured and unstructured sources, including CIOMS-I forms (internationally recognized adverse event reporting forms), internal or partner adverse event (AE) forms, emails, literature, and web screenshots; detecting duplicates; flagging cases with missing or conflicting information. Ultimately, the integration of AI in these spaces is essential for better data entry consistency and reduced cycle times.
Recognizing that no single extraction method is optimal across every source format, organizations are turning to hybrid strategies that combine validated structured extraction for high-volume source types and human reviewed LLM-enabled unstructured extraction for diverse, lower frequency sources.
Although there are other areas of PV where AI is being explored and, in some cases, implemented (e.g. literature, signal detection and aggregate report writing), these are usually much less labor-intensive activities than case management, meaning positive return on investment can be harder to achieve. Though not usually explicit, it is often the case that the functionality, efficiency promises and cost models of these systems assume the user to be a mid-size or large pharmaceutical company (i.e., high task volume, sizeable legacy dataset and sophisticated data input formats). For smaller companies, the cost of these new technologies can therefore easily outweigh the cost of the existing processes, though this may start to change if vendors move away from offering separate systems to bundled suites of PV tools.
PPD Functional Service Partnership (FSP) Pharmacovigilance solutions, for example, leverage the PPD SafetyNet Suite, a comprehensive group of PV technology platforms, to improve consistency, quality, and reduce cycle times. There are external factors at play too; the owners of some of the large literature databases (the source for all PV literature processes) are imposing ever more stringent, and in some cases restrictive, licensing conditions regarding use of AI in conjunction with data extracted from these sources.
Can you share a concrete example where AI has demonstrably improved adverse event (AE) case processing in global PV operations?
Case management experts report reduced cycle times, improved processing consistency, and earlier duplicate detection compared with manual workflows, enhancing intake efficiency across high-volume global portfolios. AI also identifies “problematic” cases—those with conflicting or missing data—so they can be routed promptly to human reviewers, supporting timely triage and regulatory compliance.
At the operational level, AI has clearly enhanced global surge management and overall scalability. During the COVID-19 vaccine rollout, organizations faced extraordinary AE volumes, with insufficient human PV case processors in the global labor market to sustain legacy manual processes. This situation forced accelerated AI deployment to keep up with reporting obligations worldwide. In ongoing, routine operations, these same AI capabilities de-risk so-called black swan spikes – case boluses from regulators, class-action litigation, or product-safety publicity – by stabilizing throughput and reducing the impact of volume fluctuations on case cycle times and compliance.
Pharmaceutical companies often employ an FSP model to leverage global teams, ensure follow-the-sun coverage, and standardize AI-enabled workflows. This model empowers them to right-size case processing workforces and more easily scale PV systems to new products and markets, while surfacing real-time performance metrics with minimal extra effort.
What are the most significant operational or technical challenges organizations face when integrating an AI-based case intake system into legacy PV infrastructure?
On the technical side, the biggest issue is the wide variety of intake formats: E2B XML, clinical database outputs, multiple AE forms, emails, literature abstracts/articles, and web screenshots. This makes it difficult to build one extraction pipeline that is accurate and stable across all sources. Structured extraction tends to be more accurate but rigid, while unstructured extraction offers greater flexibility but needs more human review, so companies usually need to design and validate a hybrid approach that fits their actual source mix. Engaging an experienced FSP partner addresses these challenges by providing specialized systems and automation expertise, designing optimized extraction pipelines, and ensuring that AI outputs remain accurate and auditable across diverse source types.
On the operational side, key hurdles are cost, change management, and regulatory confidence. In addition to an organizational mindset-shift, AI brings significant setup and running costs – configuration, training data, validation, integration and specialist programmers – that can erode expected efficiency savings if not modeled carefully at system level. At the same time, PV is conservative, and anxiety about regulatory authority inspector reactions and the lack of detailed AI-specific guidance have slowed adoption. FSP partners working with multiple companies address operational challenges by offering scalable infrastructure, standardized processes, and human-in-the-loop oversight, while also supporting staff training and governance to ensure confidence in AI outputs and regulatory compliance.
Successful integration therefore depends on strong change management: training staff, supporting a culture that embraces new tools, and deliberately balancing automation with human oversight so people remain in control of complex decisions and avoid manual workarounds that fragment processes and undermine efficiencies.
How do current regulatory frameworks in key markets (such as US, EU/UK, and Japan) address the use of AI in PV, and where do you see regulators demanding more evidence or transparency for AI-driven decision-making?
Influential international regulators are still developing their thinking on AI in PV, and that uncertainty has made companies cautious. A barrier to implementing AI in PV has been concern about how regulatory authorities might respond, stemming from limited AI-specific guidance and minimal inspection experience with AI in PV systems.
Recent experience with COVID-19 vaccine case surges and broader use of LLMs has started to ease that anxiety as both companies and authorities share lessons learned, but formal guidance remains relatively high-level.
Although there are ongoing inter-regulator initiatives, such as the various collaborations among FDA (U.S. Food and Drug Administration), MHRA (U.K. Medicines and Healthcare products Regulatory Agency), and Health Canada (Canada’s federal health regulatory authority), the resulting publications are understandably high level. Regulators have openly shared the challenges of regulating this type of technology, in that it evolves so quickly that specific guidance can become out of date before it is even published. CIOMS guidance, for example, typically takes four years to develop and publish, and the ongoing work of “Working Group XIV - Artificial Intelligence in Pharmacovigilance” which started in 2022, already reportedly has run into challenges.
Within this evolving landscape, regulators are clearly looking for companies to retain evidence and transparency wherever AI affects safety outputs. Transparent, explainable AI systems should be used, with automated steps kept human-readable for oversight, allowing PV professionals to maintain control over case assessment and medical review. Inspectors will expect evidence of how AI is configured, validated, deployed, and supervised. For example, PPD FSP Pharmacovigilance solutions pair AI platforms with experienced specialists who document workflows and align processes with evolving regional expectations. This approach provides clients with inspection-ready AI-enabled PV solutions, enhancing their existing capabilities.
What best practices are emerging to ensure ongoing data quality, model validation, and auditability of AI-driven outputs within PV compliance frameworks?
Emerging best practices fall into two main areas: technical configuration and governance. Across both, engaging with an FSP partner empowers organizations to access specialized systems and expertise across AI, PV, and data management to ensure the success of these practices.
Technical configuration: Organizations use a hybrid extraction approach to maintain data quality and rationalize validation effort. High-volume, structured inputs (like CIOMS-I or internal AE forms) rely on tightly configured, validated structured extraction, while lower-volume or unpredictable sources (emails, literature, web screenshots) are processed via LLM-supported unstructured extraction with additional human review. This ensures formal validation focuses on the structured data streams while conventional human quality checking ensures the validity of unstructured data handling.
Governance: Human oversight remains essential. AI workflows and outputs should remain human-readable and retain PV professional control over case assessment and medical review. There may also need to be some monitoring and control over how the LLM learns to ensure that its evolution remains positive. Organizations also need to invest in training, continuous improvement, and process design to avoid manual workarounds to any perceived system limitations that could compromise audit trails and efficiency.
Can you discuss the importance of human oversight in AI-enabled PV workflows, particularly in contexts such as signal review or case assessment?
Human oversight is critical because AI currently reliably handles only certain stages of PV workflows. AI can support data intake, extraction, and transformation, but downstream activities, case assessment, medical review and many aspects of signal management require nuanced clinical judgment and should, for now at least, remain the responsibility of PV professionals.
Oversight also mitigates “black box” risks. Automated steps should be human-readable, allowing reviewers to understand, verify, and, if needed, override system decisions, something that we can expect to be scrutinized during regulatory inspections. With this structure, AI accelerates processing and improves consistency, while humans retain accountability for safety conclusions, regulatory submissions, and ultimately patient protection.
Looking ahead, what advances or regulatory changes are needed to fully realize the potential of AI in global PV while maintaining patient safety and compliance?
To unlock the full potential of AI in global PV while staying compliant and protecting patients, two things need to move in parallel: better and more accessible tools, and clearer rules.
On the technology and operations side, AI platforms need to be easier and cheaper to acquire and configure, so companies can handle many different data formats without weeks of custom programming for every new form or data source. Tools for different PV tasks (e.g. case management, literature, signal detection, safety writing) should also be bundled – potentially through engagement with FSP partners. For example, PPD FSP Pharmacovigilance solutions utilize the PPD SafetyNet Suite to support every pharmacovigilance process, which enhances consistency and quality and creates additional efficiency, flexibility, and insights. Today, high setup and maintenance costs, separate tools, plus the need for large training datasets, still limit adoption, especially for smaller companies. As tools mature and ROI thresholds drop, AI-driven PV tools should become realistic for small and mid-sized organizations, not just the largest. At the same time, companies will need simple, well-tested approaches to using AI - such as hybrid structured/unstructured extraction and built-in human checks - so automation improves speed and consistency without undermining control.
On the regulatory side, a major barrier has been anxiety about how regulators will react, driven by the lack of clear, AI-specific guidance and limited inspection history for AI in PV. Looking ahead, regulators will likely need to share examples of what good looks like for AI in safety: expectations for validation, best practice for process transparency (avoiding “black box” tools), what level of human oversight is expected, and how to differentiate AI used for core GxP decisions from AI used purely for efficiency. In parallel, companies will need to invest in training and culture - so teams understand the tools, know how and when to override them, and can demonstrate to inspectors that AI is being used thoughtfully, with patient safety and data integrity remaining at the core of the PV system.
Gary Barker began his career in pharmacovigilance (PV), working closely with customers and leading teams responsible for delivering clinical trial PV services. This experience ignited his passion for drug development and ensuring patients have access to safe, effective medicines. With 20 years of industry experience, he combines operational expertise with a commitment to building strong, trusted partnerships. Relocating to the Asia Pacific region, Gary focused on developing and expanding PPD’s capabilities across Japan, Korea, and the broader region. He also led the implementation and growth of the Philippines PV operations, building a team of over 200 healthcare professionals to support global PV accounts.