TLDR: A European implementing rule makes every drug company selling in Europe mine the European adverse-reaction database inside its own signal process, and artificial intelligence is entering that process, so pharmacovigilance hiring shifts toward data-literate roles.
A European Union implementing regulation turned EudraVigilance monitoring into standing work for every marketing authorisation holder
Pharmacovigilance (PV), the work of detecting, assessing and preventing harm from medicines after they reach patients, changed its legal footing in 2025. On 22 July 2025 the European Commission adopted Commission Implementing Regulation (EU) 2025/1466, published in the Official Journal of the European Union (EU) the next day, amending Implementing Regulation (EU) No 520/2012 on the performance of pharmacovigilance activities. Most of its changes apply from 12 February 2026. Two of them applied from entry into force, which the European Medicines Agency (EMA) dates to August 2025, and those two are the ones that reshape safety teams.
The first concerns monitoring. The 2012 text required marketing authorisation holders (MAHs) to monitor the EudraVigilance database, the EU system for suspected adverse reactions, only “to the extent that they have access to that database”. The amended Article 18(2) drops that qualifier: MAHs “shall monitor the data available in the Eudravigilance database and use it” together with their other safety sources. The second change deletes Article 21(2), which had required an MAH that found a new signal in EudraVigilance to validate it and “forthwith” inform the Agency and national authorities through a standalone notification.
The EMA’s questions and answers on the regulation, revised on 28 January 2026, spells out the consequence. The signal-detection pilot that had run since 22 February 2018 on a limited list of active substances is terminated. Validated signals from every source, EudraVigilance included, now run through the marketing authorisation holder’s own signal-management process under good pharmacovigilance practices (GVP) Module IX, and EudraVigilance data belongs in answers to requests from the Pharmacovigilance Risk Assessment Committee (PRAC). The EMA’s January 2026 questions and answers had scheduled an update of GVP Module IX for the second quarter of 2026 to reflect the amended regulation; employers should check the current revision of the module on the EMA website.
| Provision | Before | After 2025/1466 |
|---|---|---|
| EudraVigilance monitoring by marketing authorisation holders (Art. 18(2)) | To the extent of access | Required, combined with other sources |
| New signal from EudraVigilance (Art. 21(2)) | Standalone notification to the European Medicines Agency and authorities | Deleted; handled in the holder’s own signal process (GVP Module IX) |
| Signal-detection pilot | Limited substance list, since Feb 2018 | Terminated |
| Subcontracted third parties (new Art. 13(1a)) | Paragraph absent | Audited by or for the holder by risk; open to inspection |
Swiss groups with an EU marketing authorisation holder staff this work across Basel, the Lake Geneva region and the EU
The regulation binds the legal entity that holds the marketing authorisation in the European Economic Area (EEA). Many Swiss-headquartered pharmaceutical and biotech groups place that authorisation in an EU subsidiary, while the global safety function, the safety database and the signal-detection team sit at headquarters in Basel or in the Lake Geneva region. EU law adds a fixed point to that design: the qualified person responsible for pharmacovigilance “shall reside and operate in the Union”, under Article 104(3) of Directive 2001/83/EC. The signal work itself can run from Switzerland, provided the EU entity can show that its own process covers EudraVigilance data and that its Qualified Person Responsible for Pharmacovigilance (QPPV) oversees it.
In Edward Galle’s view, this split shapes three hiring decisions for Swiss groups. The first is location: whether signal scientists and PV data analysts sit with global safety in Switzerland, with the EU entity, or in both, and how the written process shows which team owns each product. The second is reporting line: whether the people who run EudraVigilance screening report to the QPPV office, to global safety science, or to a data function, because inspectors read the organisation chart alongside the standard operating procedures. The third is language and mobility: teams in Basel tend to work in English and German, teams around Geneva and Lausanne in English and French, and an EU-based QPPV office may need a candidate willing to sit in the EU entity’s country.
Edward Galle’s recommendation, based on its search work for Swiss life-sciences employers, is to write these three decisions down before opening a requisition. A job description that names the product portfolio, the EudraVigilance responsibilities, the reporting line and the working languages attracts candidates who already understand the split, and filters out generalists who would need months to learn it.
Owning the signal in-house creates recurring data work that scales with the portfolio
The EMA explained at a November 2025 information session that MAHs should set monitoring at a “frequency proportionate to the risk, safety profile and characteristics of the product”, and that the requirement applies to all MAHs with medicinal products authorised in the EEA. Each product therefore needs a monitoring rhythm, a documented rationale for it, data pulls, statistical screening, validation decisions and an audit trail that an inspector can follow. The standalone notification that this process replaces was a single reporting step, while a documented signal-management process needs people, methods and records all year.
The 2025 EudraVigilance annual report, presented on 12 March 2026, records 1,765,581 individual case safety reports (ICSRs) collected and managed in 2025, and a database holding 31,253,245 ICSRs on 17,898,709 individual cases at year end. In 2025, 1,174 MAHs at headquarter level sent reports to the post-authorisation module, and 20,331 individual MAH users were registered. The EMA reviewed 1,201 potential safety signals, and PRAC prioritised and assessed 60 confirmed signals.
| Signal workload, 2025 | Figure |
|---|---|
| Individual case safety reports collected and managed in EudraVigilance | 1,765,581 |
| Individual case safety reports held in the database, end 2025 | 31,253,245 |
| Marketing authorisation holders reporting to the post-authorisation module | 1,174 |
| Potential signals reviewed by the European Medicines Agency | 1,201 |
| Confirmed signals assessed by the Pharmacovigilance Risk Assessment Committee | 60 |
| Signal evaluations completed by Swissmedic | 153 |
Millions of reports become about a thousand potential signals at the agency, and a few dozen confirmed ones. Every MAH now runs its own version of that funnel on its own products, and the people who can design the screening, read a disproportionality result and document whether a cluster forms a real signal sit at its narrowest, most valuable point. Switzerland shows the same pattern at national scale: Swissmedic’s Safety of Medicines Division completed 153 signal evaluations in 2025, according to Vigilance News edition 36 of May 2026, most of them triggered by international sources.
Artificial intelligence is entering signal detection, and regulators expect a human to stay accountable for it
The same EudraVigilance annual report names artificial intelligence (AI) and digital transformation as a central theme of 2025, with pilots on automatic literature review, an adverse drug reaction reference dataset and automatic case adjudication. Machines take on more of the intake and triage, while the regulatory accountability for each safety conclusion stays with people.
The Council for International Organizations of Medical Sciences (CIOMS), the Geneva-based body whose working groups shape global pharmacovigilance practice, published the report of its Working Group XIV, Artificial Intelligence in Pharmacovigilance on 4 December 2025. The group warns that misguided automation focused exclusively on efficiency and financial targets risks eliminating the value added by humans in the loop. The report’s executive summary sets out seven guiding principles: a risk-based approach, human oversight, validity and robustness, transparency, data privacy, fairness and equity, and governance and accountability. It distinguishes a human in the loop, where the decision is the end result of a human-machine interaction, from a human on the loop, where the machine returns a result that a human then checks.
Swissmedic’s May 2026 Vigilance News describes a natural language processing model developed in the QUADRATIC study at the Ente Ospedaliero Cantonale in Ticino, which scores Italian-language hospital discharge summaries for possible adverse drug reactions and ranks them for daily review. A review interface under development at the Regional Pharmacovigilance Centre in Lugano lets experts confirm or reject each flag, stores each decision against the model’s prediction, and leaves the final judgement to the human expert. In validation, the ranking surfaced substantially more summaries with confirmed reactions than the earlier rule-based system.
The amended regulation adds a governance layer to this. Its new Article 13(1a) requires subcontracted third parties to be audited by or on behalf of the MAH according to the risk of the activity, and allows authorities to inspect them. When case intake, literature screening or signal analytics run on a vendor platform or an AI model, someone inside the MAH has to specify, validate and audit it.
Four digital pharmacovigilance profiles gain weight as signal work moves in-house
The companion guide to pharmacovigilance and drug safety careers explains the classic ladder from case processing to the QPPV. The regulation and the arrival of AI add a parallel, data-facing track. Edward Galle sees four profiles in it, each defined by the work the rules now create:
- Signal detection scientist. Designs the product-by-product monitoring rhythm, runs disproportionality screening on EudraVigilance and internal data, and writes the validation rationale. Typical entry: pharmacy, medicine, epidemiology or life sciences plus case-processing experience.
- PV data analyst or statistician. Builds the queries, dashboards and statistical methods behind the screening, and makes the outputs reproducible for an inspector. Typical entry: biostatistics, data science or epidemiology.
- Safety systems and automation specialist. Configures the safety database, the EudraVigilance gateway and any machine-learning intake, and owns the validation documentation. Typical entry: computerised-system validation, regulated information technology or clinical data management.
- PV vendor and AI oversight lead. Specifies, audits and monitors outsourced and AI-supported activities under the risk-based audit duty. Typical entry: PV quality assurance, audit or a QPPV office.
These profiles sit where pharmacology meets data, which is the same hybrid territory Edward Galle covers on the commercial side of life sciences. They also inherit the scrutiny that regulated industries apply to any automated decision, a theme explored in Edward Galle’s analysis of AI in regulated-industry talent acquisition: a named, competent person signs each conclusion that a tool helped to reach.
The pricing logic in Edward Galle’s article Digital Hires in Swiss Pharma Are Priced Against the Technology Market, Above the Pharma Median applies here as well. A PV data analyst or a safety systems specialist can work in a bank or a software company as easily as in a drug safety department, so employers benchmark these hires against the technology labour market as well as against the pharmaceutical grade of a case-processing team.
Interviews for signal and PV data roles test method, documentation and judgement
Edward Galle’s view, drawn from its search practice in Swiss life sciences, is that interviews for the new profiles work best when they mirror the inspection questions the role will face. Three exercises cover most of what an employer needs to see.
A screening exercise. Give the candidate a small, anonymised table of report counts and ask how they would set a monitoring frequency for the product, which disproportionality method they would apply, and what threshold would trigger a validation review. The answer shows statistical fluency and, more importantly, whether the candidate documents the rationale.
A validation memo. Ask for a one-page note on whether a cluster of reports forms a signal, using the Medical Dictionary for Regulatory Activities (MedDRA) terms provided. The memo shows clinical reasoning, coding fluency and the ability to write for an inspector.
An oversight scenario. Describe a vendor or an AI model that ranks literature or cases, and ask the candidate how they would validate it, monitor its performance and decide when to overrule it. The answer shows whether the candidate understands the difference between a human in the loop and a human on the loop, the distinction CIOMS draws.
Candidates who document signal and data work, and employers who staff oversight early, lead the 2026 pharmacovigilance market
Candidates already in drug safety can make the shift visible. A curriculum vitae (CV) that names the GVP modules worked under, the signal-management steps owned, the statistical methods used and any system or model validated reads as a signal-and-data profile, which is where the regulation adds work. Experience reviewing or overruling an automated output deserves its own line, because it is the evidence of human oversight that CIOMS and regulators describe. Data scientists and statisticians moving in from other sectors can lead with method and reproducibility, then show fluency in MedDRA coding and GVP Module IX vocabulary.
Employers, including Swiss-headquartered groups whose products are authorised in the EEA through an EU-based MAH, can map each product’s monitoring frequency and rationale, decide which steps stay in-house and which run on vendor or AI tools, and staff the oversight of both before the first inspection under the new rules. Candidates can submit a CV for confidential matching to drug safety and PV data roles, and current mandates appear on the jobs page.
This article has been reviewed by Haider Alleg, Managing Partner at Kainjoo SA.
References
- European Commission. Commission Implementing Regulation (EU) 2025/1466 amending Implementing Regulation (EU) No 520/2012 on the performance of pharmacovigilance activities. Official Journal of the European Union, L series, 23 July 2025. https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=OJ:L_202501466
- European Commission. Commission Implementing Regulation (EU) No 520/2012, Chapter III, as adopted. 19 June 2012. https://www.legislation.gov.uk/eur/2012/520/chapter/III/data.htm?view=plain
- European Parliament and Council. Directive 2001/83/EC, Article 104, as amended by Directive 2010/84/EU. Version of 20 January 2011. https://legislation.gov.uk/eudr/2001/83/article/104/2011-01-20?view=plain
- European Medicines Agency. Questions and answers on Implementing Regulation (EU) 2025/1466, EMA/243145/2025 Rev. 2. 28 January 2026. https://www.ema.europa.eu/en/documents/other/questions-answers-implementing-regulation-eu-2025-1466-amendment-regulation-eu-no-520-2012-conclusion-signal-detection-eudravigilance-pilot-marketing-authorisation-holders_en.pdf
- Savvaidou, E. (European Medicines Agency). Impact of the EudraVigilance signal detection pilot termination on MAHs. 5 November 2025. https://www.ema.europa.eu/en/documents/presentation/presentation-impact-eudravigilance-signal-detection-pilot-termination-mahs-ireu2025-1466-amendment-ir-eu520-2012-e-savvaidou-ema_en.pdf
- European Medicines Agency. 2025 Annual Report on EudraVigilance for the European Parliament, the Council and the Commission, EMA/371010/2025. 12 March 2026. https://www.ema.europa.eu/en/documents/report/2025-annual-report-eudravigilance-european-parliament-council-commission_en.pdf
- Swissmedic. Vigilance News, edition 36. May 2026. https://www.swissmedic.ch/dam/swissmedic/it/dokumente/marktueberwachung/vigilance/vigilance-news-mai-2026.pdf.download.pdf/E_Vigi_News36_260507.pdf
- Council for International Organizations of Medical Sciences (CIOMS). Artificial Intelligence in Pharmacovigilance: Report of the CIOMS Working Group XIV. Geneva, published 4 December 2025. ISBN 978-929036110-7, doi:10.56759/cdob6397. https://cioms.ch/artificial-intelligence-inpv/
