F2 Tech Notes

Buckle up, folks—our favorite regulatory sheriffs at the FDA have dropped another massive rulebook, this time titled “Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations.” If you’re still awake after reading that title, congratulations! This document is essentially a roadmap for manufacturers navigating the treacherous waters of AI-powered medical devices, ensuring these gadgets don’t turn into rogue robots making wild medical decisions.

The Big Idea: The Total Product Life Cycle (TPLC) Approach

The FDA is obsessed with this Total Product Life Cycle (TPLC) concept, which basically means manufacturers should keep an eye on their AI medical devices forever—from conception to post-market babysitting. This makes sense because AI has a mind of its own (almost), evolving over time like that one houseplant you keep forgetting to water. The guidance helps companies make sure their devices stay safe, effective, and don’t decide to randomly diagnose everyone with the plague.

The FDA’s Cheat Sheet for AI Device Manufacturers

Here’s what the FDA wants you to do (besides reading their 50-page document without crying):

Build a Quality Fortress 🏰

Document everything. No, seriously. The FDA wants meticulous records of your design, development, and maintenance phases so they can check if you’ve done your homework. Think of it as keeping receipts—because if something goes wrong, they will ask for them.

Describe Your Brainy Device 🤖

The FDA expects you to spell out exactly what your AI-powered marvel does, how it does it, and why. They’re not impressed with vague descriptions like “It just works.” So, be transparent about its purpose, the AI models it uses, and how it makes its life-changing (or mildly useful) decisions.

User-Friendly or Bust 🎛️

Ever used a product so confusing that you wanted to throw it out the window? The FDA doesn’t want that happening with life-saving devices. Clear labeling, intuitive design, and interfaces that don’t require a PhD to operate are a must.

Sniff Out the Risks 🚨

AI can be a bit of a wildcard, so manufacturers need to conduct risk assessments to ensure their devices aren’t accidentally introducing new hazards. Basically, figure out all the ways your AI could go rogue before it does.

Manage That Data Like a Pro 📊

AI feeds on data like a hungry teenager, but not all data is good data. The FDA insists on strong data management, ensuring that datasets are diverse, secure, and not riddled with errors that could make the device biased or ineffective.

Keep an Eye on It (Forever) 👀

Just because your device got FDA approval doesn’t mean you can ignore it. The agency wants manufacturers to set up post-market performance monitoring—because AI loves to learn and change, and sometimes, it learns the wrong things.

Bias Alert: AI’s Not-So-Secret Flaw

One of the biggest takeaways? AI can be super biased. If a device is trained on limited data, it might work great for some people and be completely useless (or even dangerous) for others. The FDA is nudging manufacturers to catch and fix these biases before their AI devices start playing favorites.

FDA’s Golden Advice: Talk to Us Early

Rather than surprising the FDA with an AI medical device they’ve never seen before (not recommended), manufacturers should chat with them early in the process. Think of it as showing your teacher a rough draft before turning in the final essay—this can help avoid unpleasant surprises later.

Final Verdict

The FDA’s latest guidance provides a structured game plan for AI-enabled medical devices, ensuring they’re safe, effective, and not just some high-tech gimmick. For manufacturers, following these rules is the key to getting their devices approved without unnecessary headaches.

CONTACT F2 LABS FOR HELP, THIS IS WHAT WE DO.

Want to discuss your project with us?

You can contact us at this link. Our phone number is 877-405-1580 and we are here to help you.

F2 Labs has many conversations every week with medical device manufacturers who are struggling to define Essential Performance for their device.

 
From our experience, the confusion around Essential Performance frequently stems from the fact that what a device manufacturer views as ‘essential performance’, does not align with the definition of ‘Essential Performance’ from IEC 60601-1.
 
They consider every function described by the product’s specification to be ‘essential’.  If the device does not work as intended – what is the point?  Who would want to buy or use it?   These specifications will need to be proven through bench testing typically performed by the manufacturer, on the manufacturer’s behalf by an accredited Medical Device Testing lab.  However – this is not the same as what 60601-1 intends when it defines Essential Performance.    
 
For the purposes of 60601 testing, the definition Essential Performance is this:

Essential Performance: those clinical functions which ‘if degraded or not present, could result in an unacceptable risk’.   

 Blending the two ideas of essential performance together muddies the waters, and often creates confusion in developing the correct Essential Performance for a device.

 Defining Essential Performance can be simplified by breaking it down into manageable pieces as follows:
 
1. Create a list of your device’s clinical functions (there could be one, none, or many).
2. Review each one individually and ask “If this function is not performing correctly (inaccurately or at all), could it result in an unacceptable risk?”

(a) If the answer is no, that particular function is not Essential Performance.
(b) If the answer is yes, that particular function is Essential Performance.

3. For each item that was answered ‘Yes’, define the exact requirement and define the method of monitoring.
 
That’s it – you could end up with one, none, or many. The rationale you use for making these decisions should be detailed in your Risk Management File.

Let’s follow the above steps using the simple example of a basic thermometer.
 
1. List of clinical functions:

(i) Measure temperature accurately.
(ii) Display that measurement on a cell phone app.

2. Review each clinical function individually, and ask the question – is this Essential Performance?

(i) Measure temperature accurately:

Yes – treatment decisions will be made based on this measurement (i.e. administer medication)    

(ii)  Display that measurement on a cell phone app:

No – if the result is ‘unavailable’, a different thermometer could be used, no treatment decisions would be made, etc.

3. For each item that was answered ‘Yes’, define the exact requirement and define the method of monitoring it.

(i) Measure temperature accurately.                     

(a) Define the exact requirement(s):  measure temperature within +/-  0.10 degrees                           
(b) Define the method of monitoring/measuring: use a calibrated thermometer or calibrated water bath to verify the measurement accuracy.

In this example, the Essential Performance and Method of Monitoring for this thermometer is:

 CONTACT F2 LABS FOR HELP, THIS IS WHAT WE DO.

Want to discuss your project with us?

You can contact us at this link. Our phone number is 877-405-1580 and we are here to help you.