Last week, medical device and IVD manufacturers—from startups to large corporations—met in Konstanz to develop their AI strategies. In today’s Executive Corner, I’ll share the method we used.
Regulatory Updates
a. EU
Update to the GeDIG Report
First, I’d like to thank Dr. Malte Schmieding. In response to my update on the German Act on Data and Digital Innovation in Healthcare (GeDIG), he points out that:
The current version of Section 303e(2)(9) of the German Social Code, Book V (SGB V) already explicitly permits the use of data from the Health Data Lab to develop and monitor the safety of medical devices, including AI development.
The amendment primarily changes the terminology from “artificial intelligence systems” to “AI models and AI systems,” aligning it with the terminology used in the EU AI Act and Book X of the German Social Code (SGB X).
The message for manufacturers is clear: you don’t need to wait for the GeDIG to collect data for post-market surveillance through this route. And you shouldn’t wait, either.
Draft implementing regulation (5 pages): The draft implementing regulation requires applicants and health data access bodies to use standardized templates for the health data access application, data permit and health data request. These cover applications for data access (Article 67), data permits (Article 68), and requests for purely anonymized statistical results (Article 69).
Annex (59 pages): This document contains the actual templates:
Annex I: Application for data access
Annex II: Data permit
Annex III: Request for statistical results
Anyone applying for data—for example, a manufacturer seeking to train an AI model—must describe, among other things, the purpose, datasets, risks and safeguards, and the secure processing environment.
Important: Applications will be made public! There are, however, exceptions for trade secrets and intellectual property. The regulation will not apply until March 2029.
Can you already anticipate changes to your products, and would you like to avoid having to obtain a new marketing authorization? If so, I recommend taking a closer look at this approach: Plan changes in advance and obtain FDA clearance for them before implementation.
The FDA already lists around 240 products for which it has approved a PCCP, ranging from Apple’s heart rhythm feature to PCR tests.
A PCCP requires three things:
A robust quality management system
Clearly specified changes
Predetermined test methods with acceptance criteria
My tip: Limit the scope to a few changes that can be readily validated, and discuss your PCCP with the FDA early through a Q-Submission. We can help you with this.
Executive Corner: Developing an AI-Strategy in One Day
In the last issue of the Medical Device Briefing, I shared tips on how to introduce AI from the bottom up when “the top” provides no direction—or no useful direction.
This week, I’d like to offer executives some ideas on how to define and implement an AI strategy for their area of responsibility.
Last week, participants in the grand finale of our AI Lab had the exclusive opportunity to develop their AI strategies with us. We’re now opening up access to this knowledge and these methods to all manufacturers, service providers, and Notified Bodies.
don’t want to prepare your AI strategy day yourself;
would value an external, impartial perspective; and
want to be confident that you can get started with a clear roadmap, including the necessary decisions, assigned responsibilities, tasks, and deadlines.
I couldn’t resist trying this: in his YouTube Short, Nate B. Jones suggested asking AI how intelligent you are. He argued that the interesting part isn’t the result itself, but the reasoning behind it.
“Tell me how intelligent I am in terms of percentile. What’s my IQ percentile, based on your interactions with others and what you’ve observed about me?”
I had to try it myself! I used an extended conversation with my AI about IEC 62366-3, since the AI needs a basis for its assessment.
It evaluated me against five criteria. Among them were “abstraction discipline” and “epistemic flexibility.” If the latter term is new to you (it certainly was to me), it means having “zero ego defense when confronted with your own mistakes.”
Now for the results that mattered more to me: I asked the AI where I could improve. Among other things, it recommended that I look into the “curse of knowledge” and become more aware of the consequences of the “authority bias trap.” Ouch!
Since then, I’ve been asking the AI for advice on how to turn this new awareness into action. Hopefully, you’ll be able to follow my progress here in the Journal. 🙂
My tip: If you’d like a generative AI to assess your potential for growth, ask it to do so based on a conversation in which you’ve worked through a new topic or a complex document together.
b. How to Use AI Even When IT Isn’t Ready Yet
At two of the 50 companies taking part in my “Zero to Hero” program, IT hadn’t approved the use of the AI tool in time—or at all. This was despite the provider’s contractual commitment not to use the data for training. We weren’t even working with confidential data, just a sample case file.
So what can you do? I came up with a workaround that IT was happy to approve. If your company hasn’t given you access to a powerful AI model yet, you can use this fallback option, too:
My tip: Arrange access to a virtual Windows machine via DaaS (Desktop as a Service)—essentially, a computer that “runs in your browser”—and work in this separate environment.
We also provide the same secure infrastructure to everyone taking part in our Zero to Hero on demand program. Over seven weeks, you’ll build your own Technical Documentation review system. It’s like paint-by-numbers—no IT or AI expertise required.
The best way to find out how the program works is to read more here.
IEC 60601-1 Safety of Medical Electrical Equipment
November 10-11, 2026
Learn how to successfully implement the requirements of the IEC 60601 family of standards in practice and ensure standard-compliant documentation.
In this seminar, you will get all the knowledge to speak to investigators and auditors on an equal footing. You will develop a comprehensive understanding of the basic concepts of the IEC 60601 family of standards, and learn how to successfully implement the relevant requirements in practice using various exercises and how to create standard-compliant documentation.
AI Power for Medical Device Development and Approval – Practical Workshop
April 27-28, 2027
Learn in just two (half) days how to use large language models (LLMs) specifically in developing medical devices – from the initial idea to post-market surveillance. This practical workshop will teach you specific methods that will enable you to drastically reduce the time needed to develop and document your medical devices while remaining compliant with the law.
/Johner_Institute_Logo_RGB.png){kind=logo}