Do You Need Clinical Data to Get a CE Mark for Your AI Medical Device?

Do AI medical devices need clinical data for CE mark? Many teams building AI that predicts, measures, prioritises, or triages ask whether prospective clinical data is required under the EU MDR, or if rigorous retrospective validation will suffice. In this Insight, Dr Ankeet Tanna outlines when clinical investigations are essential, when indirect clinical benefit and non‑clinical evidence (e.g., Article 61(10)) can be enough, and how these strategies helped MyCardium AI secure CE marks for AI‑driven measurements on cardiac MRI and echocardiograms.

AI medical devices are unique: they handle the “thinking” and “decision-making” aspects of the clinical workflow - predicting, prognosing, prioritizing, and more. This distinction can make it challenging to determine the appropriate regulatory evidence needed to secure a CE mark.

A common question we get asked is whether AI devices require prospective clinical data to obtain CE marking, or if retrospective validation is enough. This Insight Article explores some strategies to consider. 

We’ve recently used some of these strategies to successfully help our client MyCardium AI in securing 2 CE marks: for AI-driven measurements on cardiac MRI and echocardiograms. So, you can be sure they have passed the muster of a rigorous audit.

Understanding Clinical Data and Clinical Benefit under EU MDR

To answer this, it’s important to consider two key concepts from the EU Medical Device Regulation (MDR): the definitions of clinical data and clinical benefit.

Under the EU MDR, most medical devices must demonstrate a clinical benefit, defined as:

“The positive impact of a device on the health of an individual, expressed in terms of a meaningful, measurable, patient-relevant clinical outcome(s), including outcome(s) related to diagnosis, or a positive impact on patient management or public health.”

To demonstrate clinical benefit, devices typically need to present clinical data, which is:

“Information concerning safety or performance generated from the use of a device, sourced from clinical investigations of the device concerned…”

A clinical investigation is defined as:

“Any systematic investigation involving one or more human subjects, undertaken to assess the safety or performance of a device.”

In essence, the regulations require most devices to gather clinical data through clinical investigations to prove clinical benefit to patients.

Challenges for AI Devices

However, this requirement can be complex for certain AI devices. Consider, for example, a radiology AI tool that measures cardiac anatomy on radiology images - like MyCardium.

In this case, the clinical benefit is indirect rather than direct. While the measurements must be accurate, it’s often difficult to link these measurements directly to improved patient outcomes. The accurate measurement may support a clinician’s decision-making, but ultimately, it is the treatment based on that decision that drives patient benefit.

This concept of indirect clinical benefit is explicitly recognized by MDCG 2020-6.

The more challenging question is whether a clinical investigation involving human subjects is required to demonstrate that indirect benefit. Using the AI measurement tool example, one could argue that recruiting patients is unnecessary if the device’s accuracy can be validated retrospectively using existing image datasets.

Thus, accuracy can be demonstrated without prospective clinical data or human subject recruitment.

Demonstrating Clinical Benefit Beyond Prospective Data

This does not exempt manufacturers from demonstrating clinical benefit. Often, they may rely on evidence of indirect benefits - such as workflow improvements, enhanced measurement precision, and established literature - to show downstream positive effects on patient care or public health.

Manufacturers must also demonstrate summative usability, proving that the device performs effectively in the hands of real users, as part of showing clinical performance.

Article 61(10) of the EU MDR: When Can Non-Clinical Evidence Suffice?

Article 61(10) of the EU MDR allows for exceptions where clinical data is not deemed appropriate:

“Where the demonstration of conformity with general safety and performance requirements based on clinical data is not deemed appropriate, adequate justification for any such exception shall be given based on the results of the manufacturer's risk management and on consideration of the specifics of the interaction between the device and the human body, the clinical performance intended and the claims of the manufacturer. In such a case, the manufacturer shall duly substantiate in the technical documentation … why it considers a demonstration of conformity with general safety and performance requirements that is based on the results of non-clinical testing methods alone, including performance evaluation, bench testing and pre- clinical evaluation, to be adequate.”

The key phrase here is “not deemed appropriate.” The justification must show that clinical data is genuinely inappropriate - not simply difficult to obtain.

It is important to note that this exception does not apply universally. For example, AI tools that predict future disease will eventually need to demonstrate that their predictions manifest in real-world clinical outcomes.

Key Takeaway

As with all emerging technologies, regulatory expectations for clinical evidence depend heavily on the device’s intended purpose and operating principles. For AI medical devices, the approach to clinical data must be tailored accordingly.

If you need guidance navigating these clinical requirements for novel AI technologies, Hardian is happy to help.