What is the FDA’s approach to regulating AI/ML in SaMD (software as a medical device)?
Article Context:
As the quote goes, "Necessity is the mother of invention." Ironically, we don't know who invented that quote. It became an amalgam of people from Plato to a 19th-century theologian.
According to the FDA's site, their first Medical Device approved was in 1976 through their CDHR division (Center for Devices and Radiological Health) per their 21CFR Parts 800-1299. As we know, medical devices fall into three categories, from low risk to high risk: Class I, II, and III
Per the Int’l Med Device Regulators Forum (IMDRF), the need for regulating SaMD, this new form of med device, came about as long ago as 2013, again through an amalgam of regulated elements within the med device world.
"SaMD" even reaches as far back as the 60s with Lockheed's consideration of getting the first computerized information-handling system in the hospital sphere.
And do we know who the first company was to be approved by the FDA as a SaMD? I'm sure there was a hybrid that led up to that, and my guess is it's archived somewhere, but my research didn't come up with a specific name, but rather several SaMD firsts:
- Pacemakers
- X-rays and MRIs
- Bedside monitors
- Urine analyzers
- Glucometers
- Spirometers
The FDA had a tall order to decipher regulatory compliance with this new hybrid. And companies in the SaMD space had to figure out how to manage regulatory affairs per the FDA.
Thanks to SaMD and other inventions, Artificial Intelligence (AI) and Machine Learning (ML) are pervasive in nearly everything we do now. Are Alexa, OK Google, and Siri listening and learning? Absolutely. Is their AI helping us with more targeted options for buying, listening, and purchasing… even routing us to avoid traffic? Definitely. Conversely, have you ever looked up airfare, car rentals, and hotels, only to find out after your third or fourth search that the prices went up for some reason…on the same day? Hmmm. What changed that? Supply and demand…AI/ML interpreted your multiple searches (and others) as increased demand, so they bump up the price a bit more or a lot more. This "encourages" you to buy quicker. Maybe using Incognito or an anonymous site helps? What are your thoughts?
Would SaMD do the same? Not unless the developing company wanted to violate HIPAA compliance and all kinds of regulatory violations.
For some guidance, per Deloitte, here are some Is and Is Not examples:
What is and is not SaMD?
SaMD | Not SaMD |
|---|---|
| Software that can determine the proper drug dose for a patient, given personalized patient data | Software that operates a pacemaker |
| Software that can detect and diagnose a stroke by analyzing MRI images | Software that drives or controls an infusion pump's motors |
| Software that can track the size of a mole over time and determine the risk of melanoma | Electronic health record systems |
| Software that draws on data from other digital devices to determine risk factors associated with epileptic seizures | Software in the machines that assemble medical devices |
In one of my previous blogs, I mentioned that Apple, which started in 1976, hit it big with its desktops, but even bigger more significant with their iPhone innovation. That iPhone, coupled with the right software, became one of the forefront integrators of SaMD. Suddenly, eyes were opened to a world of information, whether on our wrist or transported electronically to the doctor. Abruptly, our eyes opened to a universe of data, accessible on our wrist and to be shipped electronically to our primary care physician.
Thanks to necessity, we can live healthier and longer and invent more!
FAQ's
How does FDA regulate AI/ML in medical devices?
The FDA has developed a regulatory framework for AI/ML medical devices based on the same principles as traditional medical devices through a risk-based approach. Specifically, the review process will assess the safety, efficacy and intended use of the medical devices. Existing medical device application pathways will be used for AI/ML devices, i.e., 510(k), De Novo or PMA. The FDA will also focus on how these algorithms perform and how they will be maintained over their lifecycle compared with traditional medical devices.
What is the FDA’s AI/ML SaMD Action Plan?
The FDA’s AI/ML SaMD Action Plan provides a framework for supporting development of innovating while also ensuring the safety of patients. Key areas of attention include establishing regulatory clarity for developers, monitoring real world performance, sharing knowledge about good practices in the use of machine learning and enhancing stakeholder engagement. These objectives will help to modernise oversight of AI-based software medical devices.
Why does the FDA propose a Predetermined Change Control Plan for AI/ML SaMD?
The pre-determined change control plan by the FDA enables the manufacturer to perform pre-approved algorithm updates on their AI/ML SaMD product, without the need for a new regulatory submission for each update. This supports continuous learning systems and allows for an orderly approach to maintaining the safety and effectiveness of the products. It facilitates innovation and regulatory oversight in a timely manner.
How does the FDA ensure transparency for AI/ML SaMD?
The FDA supports transparency by requiring manufacturers to use clear labelling and provide information about the product’s performance and how users can access it. Manufacturers must provide an explanation of how their algorithms function, what their limitations are, and what their intended use is. The FDA also requires manufacturers to provide information on how they will monitor newly approved products and information about any software updates that could impact the clinical performance of their product.
Does every AI/ML medical device need FDA approval before marketing?
No, not all AI or ML -based devices require written approval prior to marketing them, as regulation is different based on the classification, risk level and intended use of that type of device; however, lower risk types of devices may be exempt from that authorization process while moderate and higher risk types of devices will generally require clearance or approval from the Food and Drug Administration (FDA) prior to entering the marketplace.
Submit the form below, and our expert will reach out to assist you!
