Simulation-guided clinical trials: Is EU law ready?

Before enrolling real patients, clinical trials can be “test-flown” on a computer, running thousands of virtual versions to assess how reliable, fast, and safe the chosen design will be.

The law today does not require trial simulations: Neither Regulation (EU) 536/2014 on clinical trials nor the forthcoming Pharma Package oblige sponsors to simulate a study’s design, although the Pharma Package mentions in silico methods as a way to replace animal testing. Meanwhile, the intellectual property status of simulation tools, algorithms, and resulting data remains largely uncharted in EU regulatory discourse.

Looking ahead, a “risk-based” simulation requirement for more complex trial designs could make trials safer, faster, and cheaper, provided it is calibrated so as not to penalize academic and noncommercial research, and provided that IP frameworks offer adequate incentives and protections for innovators and data generators alike.

What are clinical trial simulations?

A recent paper titled “Design, simulate, refine: simulation-guided clinical trials for accelerated drug development”, published in Nature Reviews Drug Discovery. The core idea is simple and powerful: before enrolling a single real patient, thousands of virtual versions of the same study can be “flown” on a computer to understand how the chosen design will behave, how likely it is to reach a correct answer, how long it will take, and how many subjects will end up on ineffective treatment arms. The authors describe this as a “test flight” that lets clinical teams anticipate risks and choose, transparently and with documented justification, among several possible designs.

In some respects, the approach outlined in the paper is almost engineering-like. In fact, in the field of engineering, the use of simulations has for many years been a standard procedure applied consistently and meticulously in the design process. Bridges are subjected to simulated stresses (e.g., strong winds, earthquakes, and heavy traffic) to determine whether they can withstand them; cars are tested in both virtual and real wind tunnels to verify their aerodynamics; the hydrodynamic study of ship hulls is always conducted first in silico. Similarly, drug research today relies heavily on simulations: for example, the interactions between the pharmacologically active molecule and its target are simulated using docking studies, as are those between monoclonal antibodies and receptors.

In practical terms,when designing a trial, teams face many possible designs (sample sizes, allocation ratios, and adaptations) and uncertain assumptions (treatment effect and recruitment rate). Pairing one candidate design with one assumption set forms a scenario. For each scenario, computer simulations imitate the trial’s execution across many replications, producing anticipated decisions and performance metrics. These outputs feed discussions among sponsors, investigators, regulators, and patients, guiding design choice and possibly triggering a new simulation cycle in an iterative process.

Though not legal in nature, the discourse touches on key issues that pharmaceutical and clinical-trial law cannot ignore. Here we read it through a European legal lens, examining the regulatory framework from Regulation (EU) 536/2014 and the forthcoming Pharma Package through to the International Council for Harmonisation (ICH) guidelines and European Medicines Agency (EMA) soft law. We also explore an additional dimension: the intellectual property implications that simulation tools, data, and algorithms inevitably raise.

Beware: “simulation” does not mean just one thing

To reason correctly in legal terms, three concepts that are often confused must be carefully distinguished:

• Trial design simulation, in which the patients remain real. What is simulated on a computer is the behavior of the study design, in order to measure its “operating characteristics,” e.g., statistical power, false-positive risk, expected duration, and exposure to ineffective doses. As has been aptly observed, it is preferable to “sacrifice” computer-generated patients rather than real ones while calibrating the design.
• In silico trials, in which virtual patients replace or supplement real data.
• In silico methods as an alternative to animal testing (New Approach Methodologies), in which computational models are used at the preclinical stage to reduce or replace animal testing.

The distinction matters: obligations, evidentiary standards, and ethical implications change radically depending on which of the three concepts one has in mind. Whether the law should require simulation before a trial starts concerns mainly the first concept. It is also the first concept that raises the most acute intellectual property questions, as we shall see below.

Does the Pharma Package require trials to be simulated before they start?

The short answer is: no.

The Pharma Package is the proposed new directive replacing Directive 2001/83/EC (the Community code relating to medicinal products for human use) and the proposed new regulation replacing Regulation (EC) No 726/2004 (laying down procedures for the authorization and supervision of medicinal products and establishing the European Medicines Agency). Procedurally, the provisional trilogue agreement was reached on 11 December 2025, the compromise texts have been published, and formal adoption is expected in autumn 2026, to be followed by transitional periods of up to three years.

It is important to highlight that the Pharma Package concerns marketing authorization, post-market obligations, the EMA’s institutional role, and the incentive system, and does not govern clinical trials. Clinical trials are currently governed, and will continue to be so also after the adoption of the Pharma Package, by Regulation (EU) No 536/2014 of the European Parliament and of the Council of 16 April 2014 on clinical trials on medicinal products for human use (the Clinical Trials Regulation (CTR)), which has been fully operational since 31 January 2023 through the Clinical Trials Information System (CTIS).

Moreover, where the Pharma Package texts refer to in silico methods, they do so as an alternative to animal testing (preclinical stage), not as an obligation to simulate the clinical trial design. In other words, the European push toward computational methods (consistent with Article 13 of the Treaty on the Functioning of the European Union (TFEU) and the broader New Approach Methodologies (NAMs) agenda) is real, but it is aimed at replacing animal models (also consistently with the recent EU Commission Communication C(2026) 3497), not at mandating the simulation of human studies.

Thus, simulating a trial’s design today remains an optional methodological tool, not codified as a general legal obligation either under the legislation currently in force or under that about to be formally adopted.

What EU law provides today

The CTR requires a study to be scientifically sound (Article 25) and the rights, safety, and well-being of subjects recruited in trials to be protected (Article 3). It also calls for a description of the statistical methodology in the protocol (Annex I, Section D), but it does not expressly require design simulations. Likewise, the ICH E6(R2) guideline on Good Clinical Practice (which the EMA regards as a reference standard for trial conduct within the EU) emphasizes scientific quality and risk-based monitoring but does not mandate simulation. Simulation, however, enters “through the window” via soft law and regulatory practice, especially for complex and innovative designs, including through:

• The joint EC/EMA/HMA document Complex Clinical Trials – Questions and Answers (first published 2022, updated 2023), which requires validation of the design methodology and, for Bayesian settings, expressly calls for validation “including via simulation.”
• The EMA Reflection Paper on Methodological Issues in Confirmatory Clinical Trials Planned with an Adaptive Design (CHMP/EWP/2459/02, 2007) and the ICH guidelines – in particular ICH E9(R1) on estimands and sensitivity analysis, and the draft ICH E20 on adaptive clinical trials, the latter of which devotes an entire section to the simulation plan and simulation report.
• The EMA Qualification of Novel Methodologies pathway (Article 1(28a) of the proposed new Regulation) and EMA scientific advice, which allow a simulation-based approach to be validated on a case-by-case basis prior to trial commencement.

Upstream, the fundamental rights framework already provides a normative impetus in favor of simulations where they reduce patients’ exposure to risk. Article 3 of the Charter of Fundamental Rights of the European Union (integrity of the person and the requirement of free and informed consent in the field of medicine), the WMA Declaration of Helsinki (2013, as amended: risk minimization and the duty to protect research subjects), and the Council of Europe’s Oviedo Convention on Human Rights and Biomedicine (1997) provide a solid axiological basis. If a preventive simulation avoids needlessly exposing dozens of patients to an ineffective arm, it is not merely “good statistics”: it is consistent with legal and ethical obligations that are already in force across the European Union.

Would it be appropriate to introduce an obligation?

The central question is whether, in the near future, legislation on trials should require simulations in order to make them safer, more effective, faster, cheaper, and more reliable. The arguments in favor are serious:

• Ethicality: fewer patients exposed to futile or ineffective arms, thanks to well-calibrated early-stopping rules.
• Efficiency: faster and cheaper studies, with fewer protocol amendments.
• Reliability: more robust designs, tested under many scenarios before launch.
• Systemic consistency: better assessment (via simulation) of the adaptive, Bayesian, and platform designs which agencies continue to move toward.

Yet there are equally real counter-considerations that argue against an indiscriminate obligation:

• “Garbage in, garbage out”: a simulation is only as good as the assumptions that feed it. There must be a degree of alertness to “biased comparisons,” where a seemingly rigorous analysis is built (even unintentionally) to favor the preferred design. A poorly conceived obligation would risk producing false scientific rigor.
• Proportionality and access to research: quality simulations require expertise and resources. A general obligation would weigh above all on academic, nonprofit, and SME sponsors, risking an advantage for large groups and a conflict with the CTR’s goal of facilitating noncommercial research.
• Regulators’ assessment capacity:assessors often cannot re-run all the simulations. An obligation would make sense only if paired with “good simulation practice” standards, reproducibility, and code transparency.
• Not all trials need it: for a simple fixed design, classical statistical formulae are enough. Requiring simulation everywhere would be disproportionate and unreasonably increase costs.

The reasonable synthesis would be a risk- and complexity-based obligation, not a blanket one, with simulation reports only being mandatory for adaptive, Bayesian, seamless, and platform designs, but strongly encouraged for others. This should be paired with the codification of a good-simulation-practice standard inspired by the paper’s five principles – validity, transparency and reproducibility, thoroughness, efficiency, and comparability – and a strengthened EMA qualification pathway. As to the legislative “vehicle,” the proper home would be the CTR (and the CTIS portal), not the Pharma Package, which governs a different phase of a medicine’s life cycle.

Finally, one boundary must be kept firm: requiring the design to be simulated is not the same as accepting in silico evidence that replaces patients: these are two issues that should not be conflated.

Intellectual property dimensions of trial simulation

The regulatory analysis above would be incomplete without considering the intellectual property framework surrounding simulation tools, algorithms, and data. Trial design simulation generates valuable intangible assets at multiple levels, each governed by distinct legal regimes.

Patentability of simulation methods and tools

Under the European Patent Convention (EPC), computer-implemented inventions are patentable provided they produce a “further technical effect” beyond the mere execution of an algorithm.

A simulation tool that optimizes a clinical trial design may qualify where it demonstrably improves a technical process, for instance, by reducing patient exposure to ineffective doses through an adaptive randomization algorithm tied to pharmacokinetic modelling. Conversely, a purely statistical method for comparing trial designs, unlinked to a specific technical application, risks falling within the exclusions of Article 52(2) EPC (mathematical methods and methods for performing mental acts).

Patent applicants should therefore frame their claims around the technical contribution, i.e., the concrete improvement in trial safety, efficiency or dosing, rather than the abstract mathematical method. National Courts and the Unified Patent Court (UPC) will provide additional venues for enforcement and interpretation.

Data exclusivity and market protection

Under EU pharmaceutical law, clinical trial data submitted for marketing authorization benefit from a period of regulatory data protection. This is currently eight years of data exclusivity plus two years of market protection under Directive 2001/83/EC, Article 10, although the Pharma Package proposes to revise and modulate these periods.

A question not yet fully explored is whether data generated by simulations – as opposed to data collected from human subjects – should benefit from the same or analogous protection. If simulation outputs are used to support an authorization application (e.g., as part of a Bayesian prior or to justify a particular design choice), their regulatory status vis-à-vis data exclusivity remains to be clarified. Furthermore, the sui generis database right under Directive 96/9/EC may protect the compilation of simulation datasets where their creation required substantial investment, although the Court of Justice’s restrictive interpretation and subsequent case law would require a careful assessment of whether the investment relates to the creation, rather than merely the verification, of the data.

Trade secrets and know-how

The algorithms, assumptions, calibration parameters and source code underlying simulation models are paradigmatic examples of trade secrets within the meaning of Directive (EU) 2016/943 on the protection of undisclosed know-how and business information (Trade Secrets Directive). To qualify for protection, the information must be secret, have commercial value because it is secret, and be subject to reasonable steps to keep it secret (Article 2). For simulation developers – whether pharmaceutical companies, Contract Research Organizations (CROs), or academic groups – this implies the need for robust confidentiality agreements, access controls, and contractual provisions governing the sharing of simulation code with regulators and collaborators.

A tension may arise between the call for transparency and reproducibility of simulations, on the one hand, and the holder’s interest in keeping proprietary algorithms undisclosed, on the other. Regulatory frameworks could address this by requiring the submission of simulation reports and results to competent authorities while allowing the underlying code to remain confidential, subject to inspection under secure conditions – a model already familiar in the pharmaceutical sector for manufacturing process data.

Rights over simulation-generated data: the Data Act and the GDPR

The Data Act (Regulation (EU) 2023/2854), which applies from 12 September 2025, introduces new rules on access to and use of data generated by connected products and related services.

While clinical trial simulations are primarily software-based rather than IoT-driven, the Data Act’s principles (in particular the concept that users of data-generating services should have rights of access to the data those services produce) may influence contractual negotiations between sponsors, CROs and simulation-tool providers over ownership of and access to simulation outputs.

Where simulation inputs include real-world patient data or historical clinical data, the General Data Protection Regulation (GDPR) (Regulation (EU) 2016/679) applies in full: adequate legal bases (typically Article 6(1)(e) or (f), combined with Article 9(2)(j) for health data), purpose limitation, data minimization, and, where data originate from clinical trial participants, transparency under Article 13 or 14. Anonymization, pseudonymization, and synthetic-data techniques will play a growing role in enabling data-rich simulations while maintaining GDPR compliance.

Further intersections to watch

AI Act. Mention has been made of the use of generative artificial intelligence to program the simulators. Where code and models are generated or assisted by AI, questions of validity, traceability, human oversight, and liability arise that Regulation (EU) 2024/1689 (the AI Act) requires to be taken into account. Depending on the risk classification, simulation tools incorporating AI components could be subject to transparency and documentation obligations under the Act’s tiered framework. The proposed interplay between AI-generated simulation code and the reproducibility standards deserves further exploration.

Animal testing and NAMs. Since the Pharma Package promotes in silico methods at the preclinical stage to reduce animal testing (in line with Article 13 TFEU and the broader European policy on NAMs) systemic consistency would suggest recognizing equal dignity for design simulation at the clinical stage. The two applications share the same logic: using computational methods to reduce avoidable harm, whether to animals or to human subjects.

Data governance and the EHDS. Realistic simulation assumptions require historical and real-world data. Beyond the GDPR considerations discussed above, the European Health Data Space Regulation (EHDS) – Regulation (EU) 2025/327 – will create a framework for the secondary use of health data for research and innovation. Simulation developers will need to navigate the EHDS access conditions and permitted uses alongside the GDPR’s requirements.

Cyber Resilience Act. Where simulation tools are placed on the market as software products (whether standalone or embedded in broader clinical-trial management platforms), the Cyber Resilience Act (Regulation (EU) 2024/2847) will impose security-by-design and vulnerability-handling obligations. This is particularly relevant where simulation platforms process sensitive health or proprietary data.

Looking ahead

Simulations do not replace human judgment: they inform it, making intuitive and transparent what would otherwise remain a “black box.” EU law does not require them today; but a proportionate embrace of them, in the right legislative vehicle, with clear methodological standards, and without crushing noncommercial research, could make trials safer for patients, faster, cheaper, and more reliable, without sacrificing rigor or fair access.

At the same time, as simulation tools become more sophisticated and commercially valuable, the intellectual property framework must keep pace: patent applicants, trade-secret holders, data generators, and regulators alike need clarity on how the EPC, the Trade Secrets Directive, data exclusivity rules, and the Data Act interact in this rapidly evolving space. The goal should be an ecosystem that incentivizes innovation and openness in equal measure, protecting the legitimate interests of those who invest in developing simulation capabilities while ensuring that the resulting insights benefit patients, science, and the public interest.

It is a direction worth discussing now, as Europe’s regulatory framework enters a phase of profound renewal.