AI in medicines regulation and submissions
Medicines regulators have set out what developers, applicants and medical writers are to explain, supervise and manage when AI is used in developing a medicine, in its product information or in regulatory work, and how the European medicines regulatory network plans its own use of AI.
Regulators' frameworks and guidance for AI in medicine development
In August 2021 the International Coalition of Medicines Regulatory Authorities (ICMRA) published a report setting out recommendations to help regulators address the challenges that the use of AI poses for global medicines regulation. The report was based on a horizon-scanning exercise in AI by the ICMRA Informal Network for Innovation working group, led by the European Medicines Agency (EMA). The network's goal is to identify challenging topics for medicine regulators, to explore the suitability of existing regulatory frameworks and to develop recommendations to adapt regulatory systems so as to facilitate safe and timely access to innovative medicines. AI, which includes technologies such as statistical models, diverse algorithms and self-modifying systems, was increasingly being applied across all stages of a medicine's lifecycle, from preclinical development to clinical trial data recording and analysis, pharmacovigilance and clinical use optimisation. This range of applications brought regulatory challenges, including the transparency of algorithms and their meaning and the risks of AI failures and the wider impact these would have on AI uptake in medicine development and on patients' health. The report identified key issues linked to the regulation of future therapies using AI and made recommendations for regulators and for stakeholders involved in medicine development to foster the uptake of AI. Its main recommendations included that regulators may need to apply a risk-based approach to assessing and regulating AI, which could be informed through exchange and collaboration in the coalition; that sponsors, developers and pharmaceutical companies should establish strengthened governance structures to oversee algorithms and AI deployments closely linked to the benefit–risk of a medicinal product; and that regulatory guidelines for AI development, validation and use with medicinal products should be developed in areas such as data provenance, reliability, transparency and understandability, pharmacovigilance, and real-world monitoring of patient functioning.
Applications of AI and machine-learning components in drug development were growing exponentially by 2023, a trend expected to continue, and the growth increased regulatory agencies' engagement in ensuring their safe and effective use, including by developing guidelines and discussion papers. Regulatory submissions with AI or machine-learning components rose about tenfold from 2020 to 2021, and common types of analysis in them included outcome prediction, covariate selection, image analysis, modelling, dose selection and adjustment, endpoint and biomarker assessment, and post-marketing surveillance. In May 2023 the US Food and Drug Administration (FDA) issued a discussion paper requesting feedback on the use of AI and machine learning in the development of drugs and biological products, covering the landscape of their current and potential uses, considerations for their use and next steps for stakeholder engagement. One regulatory writer held in 2023 that AI- and machine-learning-based applications are not perfect, that their potential risks are enormous, and that continued public–private engagement, vigilance and oversight for all parties is essential to using them successfully.
On 19 July 2023 the EMA published for public consultation a draft reflection paper outlining its current thinking on the use of AI to support the safe and effective development, regulation and use of human and veterinary medicines, reflecting on principles relevant to applying AI and machine learning at any step of a medicine's lifecycle, from drug discovery to the post-authorisation setting. The paper holds that AI and machine-learning tools have the potential to effectively support the acquisition, transformation, analysis and interpretation of data across the medicinal product lifecycle: modelling approaches can replace, reduce and refine the use of animal models in preclinical development; in clinical trials, systems may support the selection of patients based on disease characteristics or other clinical parameters, and tools can support the data recording and analyses later submitted to regulators in marketing-authorisation procedures; at the marketing-authorisation stage, tools can draft, compile, translate or review data to be included in a medicine's product information; and after authorisation, such tools can support pharmacovigilance activities including adverse event report management and signal detection. The challenges it names include understanding the algorithms, notably their design and possible biases, and the risks of technical failures and the wider impact these would have on AI uptake in medicine development and health. It holds that a human-centric approach should guide all development and deployment of AI and machine learning, and that the use of AI in the medicinal product lifecycle should always comply with existing legal requirements, consider ethics and ensure due respect of fundamental rights. Where an AI or machine-learning system used in a medicine's development, evaluation or monitoring is expected to affect the medicine's benefit–risk balance, the EMA advises developers to seek early regulatory support, for example through qualification of innovative development methods for human medicines, or scientific advice. In May 2024 it still urged companies in that position to contact the agency earlier in the process. The paper is part of the initiatives of the joint Big Data Steering Group of the Heads of Medicines Agencies (HMA) and the EMA to develop the European medicines regulatory network's capability in data-driven regulation, and it was developed in liaison between that group, the EMA's human medicines committee (CHMP) and its Committee for Veterinary Medicinal Products. Its public consultation was open until the end of December 2023. A joint HMA/EMA workshop on the topic was scheduled for 20–21 November 2023, and stakeholders' feedback was to be analysed and considered for finalising the paper and for future guidance as relevant.
In May 2024 the EMA was considering the responses to the consultation and was not sure whether firmer or clearer guidance would follow. One of its staff described the agency's stance as AI positive: AI can help effectively support the acquisition, transformation, analysis and interpretation of data within the medical product lifecycle if used correctly, and that condition is a big caveat.
Much of the agency's focus had been on managing risk, because the unpredictability of AI introduces risks that were not in the system before. One proposal was a plan for AI that defines the likely risk of failure, works out what to do if the system fails and monitors the system to make sure it is not failing. The use of AI must continue to follow all existing regulatory guidelines: doing things slightly differently does not lift the rules, and additional rules will come into play because of the AI itself.
On 14 January 2026 the EMA announced that it and the FDA had jointly identified ten principles for good AI practice in the medicines lifecycle, giving broad guidance on AI use in evidence generation and monitoring across all phases of a medicine, from early research and clinical trials to manufacturing and safety monitoring, for those developing medicines and for marketing authorisation applicants and holders. The principles are to underpin future AI guidance in each jurisdiction and support international collaboration among regulators, organisations setting technical standards and other stakeholders, and over time additional EU guidance will supplement them. Guideline development in the EU was already under way, building on the EMA AI reflection paper, which the announcement dates to 2024. The use of AI across the medicines lifecycle had increased significantly in recent years, and the European Commission's Biotech Act proposal emphasised that AI holds great promise as a tool to accelerate the path from innovation to safe and effective medicines. The new pharmaceutical legislation accommodates broader use of AI in regulatory decision-making and creates additional possibilities for testing innovative AI-driven methods in a controlled environment. In the agency's view, realising these benefits requires AI to be expertly managed, including the mitigation of risks, and as AI continues to evolve a principles-based approach will help regulators, pharmaceutical companies and medicines developers harness it while ensuring patient and animal safety and regulatory compliance. The initiative builds on collaborative work following the FDA–EU bilateral meeting of April 2024 and aligns with the agency's mission to promote the safe and responsible use of AI, as set out in the European medicines agencies network strategy to 2028 and the joint HMA–EMA multiannual data and AI workplan. With ethics at the forefront, the agency will continue to explore opportunities for global convergence on AI with international public health partners.
Explaining AI in submissions and product information
As one EMA staff member put it in May 2024, if AI is used in developing a medicine, evaluating the medicine includes evaluating the AI, so the AI has to be explained in the documents submitted, and medical writers play an important role in that. If a product involves AI in the use of the medicine itself, that has to be reflected in the summary of product characteristics. So medical writers need to know not only how to write about scientific subjects but also how to write about technical matters for non-technical audiences.
In one medical writer's understanding, the exact rationale for why an AI produces a specific output cannot be determined, and medical writers are not coders. Asked in May 2024 what level of explanation the EMA expected, an agency speaker answered with the one its reflection paper set: enough detail to allow the committee to evaluate the product, so that where AI is an important part of it, sufficient detail has to be given. How much detail that is depends on how much AI is used and on the risks: where there are risks, enough explanation of what they are has to be given, even where a risk is complicated. It also depends on the document: product information for doctors can probably be less technical and a package leaflet might be made very simple, with the detail coming where the product is assessed.
In May 2024 there was not a great deal from the agency specifically for medical writers on AI, and no formal guidance yet. Its AI policy staff's initial advice was that medical writers should be able to explain AI, which means expanding their vocabulary and understanding beyond science and regulatory matters to the AI used in developing, and perhaps in using, medicines. Writers should also supervise AI adequately, including applications used for drafting, compiling, translating or reviewing product information. There was great interest across the agency in knowing how the documents it receives are generated or modified by AI, and it was keen to learn how AI is being used by medical writers and by companies preparing documents. The Dutch medicines agency also wanted to know, because it too receives many regulatory documents from companies. Such information might feed back into guidance.
The European network's workplans for AI
On 18 December 2023 the EMA and the HMA published an artificial intelligence workplan to 2028, setting out a collaborative and coordinated strategy to maximise the benefits of AI to stakeholders while managing the risks. The workplan is to help the network embrace the opportunities of AI for personal productivity, automating processes and systems, increasing insights into data, and supporting more robust decision-making to benefit public and animal health. It was prepared under the Big Data Steering Group and adopted by the EMA's Management Board at its December 2023 meeting. The agencies present it as keeping the network at the forefront in benefiting from AI in medicines regulation. By then pharmaceutical companies increasingly used AI-powered tools in the research, development and monitoring of medicines, and national competent authorities were starting to use and develop AI tools. The workplan focuses on four key dimensions:
- Guidance, policy and product support: continuous support to products in development and the development and evaluation of guidance for the use of AI in the lifecycle of a medicine, beginning with the public consultation on the AI reflection paper, open until the end of December 2023, with preparations to support the implementation of the EU AI Act starting in 2024.
- AI tools and technology: frameworks across the network to use AI tools to increase efficiency, enhance understanding and analysis of data, and support decision-making, in full compliance with data protection legislation.
- Collaboration and training: initiatives to continuously develop the capacity and capability of the network, its partners and stakeholders to keep ahead of the evolving field of AI.
- Experimentation: actions to ensure a structured approach to experimentation across the network, acknowledging its fundamental role in accelerating learning and gaining new insights.
Because AI technology, including its ethical and policy aspects, is fast evolving, the Big Data Steering Group will regularly update the workplan. Regulators, medicine developers, academics, patient organisations and other interested parties will be informed and engaged throughout its implementation.
Three drivers lay behind the EMA's AI workplan: the need to regulate the application of AI, which produced the reflection paper and might produce guidance later; improving the efficiency of internal processes, within the agency and across the European medicines regulatory network; and data analysis, where AI can do much more with the wealth of data the agency deals with. AI was at the top of the agenda at the most recent in-person meeting before May 2024 between the agency's executive director and her US counterpart.
On 5 September 2024 the EMA and the HMA published high-level principles and recommendations for all staff across the European medicines regulatory network using large language models in their work. Large language models, a category of generative AI, can support regulators in querying the extensive documentation they routinely receive, automating knowledge and data mining, and acting as virtual assistants in everyday administrative tasks, but they also present challenges such as variability in results, irrelevant or inaccurate responses (hallucinations) and potential data-security risks. The principles aim to build staff's understanding of the capabilities and limitations of these applications so that they can harness them effectively and avoid pitfalls and risks. They cover safe input of data, critical thinking and cross-checking of outputs, and knowing whom to consult when concerns arise; they stress familiarity with the tools and continuous learning; and they encourage agencies to support their staff by defining governance on the use of large language models, specifying permitted use cases, providing training and monitoring risks. They are one of the deliverables of the multiannual AI workplan to 2028, and a living document to be regularly updated.
On 7 May 2025 the EMA and the HMA published a joint workplan, "Data and AI in medicines regulation to 2028", setting out how the network plans to leverage large volumes of regulatory and health data and new tools to encourage research and innovation and to support regulatory decision making, for better medicines that reach patients faster. It lays out a roadmap for managing, analysing and sharing data across the network while adhering to high security and ethical standards. It also provides a framework for coordination on new EU legislative initiatives, notably the pharmaceutical legislation, the European Health Data Space, the Interoperable Europe Act and the AI Act. A new strategic advisory group, combining the former Big Data Steering Group and the Network Data Board, will oversee its implementation. The workplan translates the objectives of the European medicines agencies network strategy to 2028 into concrete deliverables:
- strengthening the network's data analytics capabilities to generate high-quality evidence with established and novel methods
- continuing the CHMP's clinical study data pilot to clarify the benefits and practicalities of accessing individual patient data from clinical trials
- further expanding the Data Analysis and Real World Interrogation Network, DARWIN EU, to deliver evidence that helps fill knowledge gaps and understand the use, safety and benefits of medicines
- reviewing methodologies, including biostatistics, modelling and simulation, AI and pharmacoepidemiology, and lesser-used data types, including genomic data, synthetic data, digital twins data and patient experience data, to establish a shared understanding and position their future use
- enabling efficient discovery, access and use of the network's data assets through cataloguing and strengthening data quality, starting with real-world data, adverse drug reaction data and medicinal product master data
- advancing and harmonising the implementation of the Product Management Service, the network's source of product master data for all EU medicinal products, supporting EU-wide use cases
Master data, the core data needed for the network's operations, is essential for increasing the interoperability of data assets and systems.
The workplan's key initiatives on AI include supporting the EMA's scientific committees and the pharmaceutical industry in evaluating AI through the medicines lifecycle, developing guidance on AI in clinical development and in pharmacovigilance, fostering EU-wide and international collaboration, and providing the network with training on AI and a framework for sharing and collaborating on AI tools. Their aim is to facilitate safe and responsible use of AI that benefits public and animal health.