Dovetail and IDM are building a clinical AI blueprint for the German Federal Ministry of Health
How do you get a clinical AI product through conformity assessment when its outputs aren’t fully predictable and its next version is already in development? Navigating the EU MDR and AI Act means working out how to evaluate that variability, demonstrate performance and manage continued change. Dovetail and IDM are working together to make those questions more concrete.
Through Germany's Federal Ministry of Health KI-Reallabore programme, Dovetail and IDM will write reference documentation for clinical AI, with templates, worked examples and guidance that other manufacturers can adapt.
The value is in being able to follow a decision across the file, from the first sign of a problem to the evidence supporting a change.
About the Partnership
IDM builds medical AI for hospitals and practices. For this project, the reference case is IDM's work on an agent that answers clinical questions from patient records. That gives the documentation a practical starting point. If the data or the system changes, how do you establish whether the answers remain reliable, and what evidence do you retain?
IDM holds the contract with the Ministry and brings the product and its clinical context. Dovetail writes the regulatory documentation together with IDM. The point is to show how to reason through those questions in a concrete case, without assuming that your product will follow the same design or regulatory pathway.
What to test when clinical AI changes
Our work will focus on managing changes and monitoring performance after launch. Those activities belong together. A monitoring finding can lead to an investigation, the investigation can reveal a need for a change, and that change needs its own evaluation and testing.
The planned examples will follow that sequence. Imagine a performance measure crossing a threshold set in the monitoring plan. The useful record is more than a note that something went wrong. It should show what the manufacturer investigated, what action it chose, how it evaluated the proposed change and which evidence supported the next decision.
We will develop the supporting data-governance, validation, change-planning and post-market monitoring documents, using synthetic data for the worked examples. Each document will come with a blank template, a completed example and notes on how to use it, so you can see both the structure and the reasoning behind the entries.
A starting point for your own documentation
The ministry intends to publish the reference material openly. If you are developing a clinical AI device, the aim is to give you an example you can examine and adapt before building your own documentation around it.
The limits matter too. A reference case cannot determine your device's classification or establish that your evidence is sufficient. A change plan is not blanket permission for future updates, and it does not replace the assessment of the initial product or any further assessment a change requires.
We want the result to make the next step more concrete. Instead of starting with an empty template, you will be able to inspect a worked example, understand why a decision needed particular evidence and identify what must be different for your product.
If you are planning changes to a clinical AI device and need to work through the documentation, talk to us. We will help you identify the questions to resolve and tell you honestly whether Dovetail is the right team to help.
Funded by the German Federal Ministry of Health from the Special Fund for Infrastructure and Climate Neutrality
Dr Spencer Todd is the CEO and co-founder of Dovetail (formerly FormlyAI), a regulatory partner for medical device companies. Before founding Dovetail, he spent years at the FDA and as a medical device consultant.
