The End of Article 82 of the MDR? Why AI and Software Are Now Taking Center Stage
The planned deletion of Article 82 of the MDR could have far-reaching consequences for software and AI in drug trials. The more digital systems evaluate clinical endpoints or generate evidence, the more important the question becomes: is it an investigational device or an investigational product? It is precisely this distinction that could become the central regulatory challenge in the future.
- What’s at stake with Article 82 of the MDR and what the Commission’s proposal really means
- The Previous Function of Article 82 of the MDR
- Software in Clinical Trials: The Real Gray Area
- Why the Deletion of Article 82 of the MDR Exacerbates the Problem
- What this means specifically for sponsors and manufacturers
- Conclusion
- FAQ: Frequently Asked Questions About Article 82 of the MDR, AI Software, and Clinical Trials
Teilen Sie diesen Beitrag:
Teilen Sie diesen Beitrag:
What’s at stake with Article 82 of the MDR and what the Commission’s proposal really means
On December 16, 2025, the European Commission published its long-awaited proposal to simplify the EU legal framework for medical devices. The proposal to amend the MDR (EU) 2017/745 provides, among other things, for the complete deletion of Article 82 of the MDR. This is the article that has thus far served as a regulatory safety net for so-called “other clinical investigations.”
It is important to emphasize that this is currently a legislative proposal. It still needs to be approved by the European Parliament and the Council, which is why it is realistic to assume that any changes will not take effect until the end of 2026 at the earliest—and even then, they may be in a modified form. Nevertheless, now is the right time to address the implications. This is because regulatory strategies for ongoing and planned clinical trials are being established today.
In addition to the deletion of Article 82, the proposal includes new provisions for combined studies: If a clinical trial of a medical device is conducted simultaneously with a clinical trial of a medicinal product, a single marketing authorization application under the Clinical Trials Regulation (EU) No. 536/2014 will suffice in the future. This sounds like a simplification—and in some cases, it is. For others, however, it merely shifts the crucial question: namely, whether the software used in a drug trial can even be classified as a medical device under investigation.
The Previous Function of Article 82 of the MDR
Article 82 of the MDR previously governed so-called “other clinical investigations”—that is, studies that were not primarily intended to demonstrate a medical device’s compliance with the MDR. In practice, this created a regulatory gray area that distinguished between three scenarios: the purely research-based tool without a clinical authorization context; the fully regulated clinical trial under Article 62 of the MDR, with all its requirements regarding study design, ethics committees, and regulatory notification; and national research law, which varies greatly depending on the Member State.
In Germany, Article 82 of the MDR was implemented in the Medical Device Act Implementation Act (MPDG), specifically in Section 53 of the MPDG. On this basis, the BfArM established clear procedural guidelines for the notification of other clinical trials. In practice, this gray area has gained particular significance for digital trial ecosystems and AI-supported software solutions, precisely because it has facilitated the regulatory classification of software that, while used in clinical settings, is not the primary focus of the trial’s evaluation.
If Article 82 is repealed, this buffer will also disappear. What remains is a choice between the strict path set forth in Article 62 of the MDR and national research law, without the flexibility that Article 82 has offered until now.
Software in Clinical Trials: The Real Gray Area
In modern drug trials, digital tools are no longer a niche phenomenon. AI-powered image analysis, remote monitoring platforms, ePRO systems for the electronic collection of patient-reported outcomes, and software systems for patient stratification or the algorithmic evaluation of clinical endpoints—all of these are firmly established in modern clinical trial infrastructure.
The key point is that the study in question is often not intended to assess the software’s compliance with the MDR. The primary focus is on a regulatory objective related to pharmaceuticals, such as demonstrating the efficacy of an active ingredient. Nevertheless, the software used can have a significant impact on the clinical evidence. It interprets image data, classifies patients, influences which endpoints are measured, and thus ultimately determines the study’s conclusions.
The situation becomes particularly sensitive when CE-marked software is used outside its originally approved intended purpose. This is not a theoretical scenario: In clinical practice, CE-certified software products are frequently used in contexts that go beyond their original intended use, such as when an AI system approved for routine radiological diagnostics is now used to evaluate primary study endpoints. This is precisely where the regulatory gray area begins, one that is likely to become even more confusing with the repeal of Article 82.
Test Product or Test Aid? The Decisive Choice
This brings to the forefront a question of classification that has thus far been discussed primarily in theoretical terms: Is the software in question still merely a diagnostic aid, or is it already a diagnostic device within the meaning of the MDR? This distinction is fundamental from a regulatory standpoint, and it has become even more urgent with the repeal of Article 82 of the MDR.
A study tool is used solely to conduct the study and is not itself the subject of the clinical question. Classic examples include standard software for data entry and collection, laboratory equipment for routine measurements, or IT infrastructure tools that support the study process. Their safety and performance are not under scrutiny. They are tools, not sources of scientific knowledge.
The situation is different when software determines clinical endpoints, interprets image data, classifies patients, or directly influences the quality of the evidence generated. In this case, the validity of the study effectively depends on the functionality of the software. This raises the suspicion that the software is an investigational device, with all the regulatory consequences that entails.
The real challenge lies in the fact that there is no clear legal distinction. Neither the MDR nor the MPDG defines “testing tools” as a separate regulatory term. The distinction is made on a case-by-case basis and depends on whether the clinical question of the study has a recognizable connection to the evaluation of the software in question or not. This question is anything but trivial, especially when it comes to AI systems with learning algorithms and adaptive behavior. If a model is further trained during the course of a study or adjusts its output based on feedback, an additional question arises: what is actually being evaluated—the software at the start of the study or the software at the end of the study?
Why the Deletion of Article 82 of the MDR Exacerbates the Problem
In addition, the Commission’s proposal also calls for a revision of the classification rules for medical device software (MDSW). Rule 11 in Annex VIII of the MDR is to be amended so that more software falls under Class I, meaning it does not require a Notified Body assessment. This sounds like a simplification, but it carries a risk: If fewer software products are classified as high-risk and regulated accordingly, there could be a growing temptation to use software in clinical trials that falls below the radar of MDR requirements. At the same time, the EU AI Act regulates the classification of high-risk AI in the medical sector from a different angle, leading to a complex coexistence of two regulatory regimes that raises its own questions.
Particularly when it comes to software and AI, the distinction between Articles 62 and 82 of the MDR is therefore likely to be less decisive in the future; rather, the more critical question will be: “Test product or test aid?” This question must be answered earlier in the study design process than has been the case to date.
What this means specifically for sponsors and manufacturers
Regulatory authorities such as the US Food and Drug Administration (FDA) or the European Medicines Agency (EMA) play a crucial role in determining when biological tests are sufficient and when clinical trials are required. These agencies issue guidelines based on the type of product, materials and intended use. Manufacturers are required to follow these guidelines and conduct the appropriate tests before a product can be placed on the market.
Conclusion
The debate surrounding Article 82 of the MDR has long since ceased to be confined to academic research. The real regulatory challenge lies in addressing the use of software and AI in clinical settings within modern drug trials, and this challenge exists regardless of whether the current Commission proposal enters into force as is or in a modified form.
The more digital systems are involved in the generation of clinical evidence, the more difficult it becomes to draw a line between a methodological tool and an independent investigational device. Those who fail to address this issue systematically operate in a regulatory gray area that poses risks to data usability, approval, and patient safety.
The planned reform of the MDR could make the distinction between “investigational product” and “investigational device” the most important regulatory decision when planning clinical trials with a digital component. And that is a good reason to address this issue now.