FDA Policy on When Software that Uses Artificial Intelligence and Machine Learning qualifies as a Medical Device

Artificial intelligence dates back to the 1950s. In the 21st century, it’s uses are rapidly advancing to address a wide range of concerns including healthcare. As AI and other related technologies become more and useful to hospitals, doctors, and patients; there will likely be new laws and regulations that govern how and when AI software can be used and what restrictions and best practices apply.

The FDA states that AI and machine learning may have the ability to transform healthcare in numerous ways starting with the ability to provide new insights from the huge amount of data the healthcare industry churns out every day. Manufacturers of medical devices are aiming to use AI and machine learning to create products which improve the way health care is delivered and the benefits to patients.

The FDA is reviewing the necessary regulatory framework for the product lifecycle of these new devices with the twin aims of:

• Ensuring the software is safe and effective
• Allowing for changes to healthcare delivery based on the examination of the mountains of new data that AI and machine learning make possible

Health Capital states that the most AI laws and regulations are just general laws. There are no federal statutes that specifically govern AI. The term is also noticeably absent from the 21st Century Cures Act, which was passed in 2016 to usher in a new, more industry-friendly era of drug and device regulation.

The Definitions of Artificial Intelligence and Machine Learning?

The FDA defines Artificial Intelligence as:

“The science and engineering of making intelligent machines, especially intelligent computer programs (McCarthy, 2007). Artificial intelligence can use different techniques, including models based on statistical analysis of data, expert systems that primarily rely on if-then statements, and machine learning.”

The FDA defines machine learning as:

“An artificial intelligence technique that can be used to design and train software algorithms to learn from and act on data. Software developers can use machine learning to create an algorithm that is ‘locked’ so that its function does not change, or ‘adaptive’ so its behavior can change over time based on new data.”

How AI and machine learning are being used to develop new medical devices

A few of the medical devices that are using AI and machine learning are:

• Imaging systems that can give diagnostic information for patients with skin cancer
• Smart electrocardiograms (ECGs) to help indicate the likelihood of a heart attack

And then there are diagnostics and medical devices that use Big Data for predictive analytics and to incorporate personalized medicine into healthcare.

Among other healthcare systems, the National Institutes of Health states that wearable health technology, such as an Apple Watch, allows people to monitor their heart rhythms using EKG patches. NIH researchers have discovered that AI can help “distinguish reliably between harmless rhythm irregularities and potentially life-threatening problems.”

AI “learns” to recognize patterns and to form rules. The NIH study used a powerful computer that “could classify 10 different types of irregular heart rhythms, including atrial fibrillation (AFib). In fact, after just seven months of training, the computer-devised algorithm was as good—and in some cases even better than—cardiology experts at making the correct diagnostic call.”

“In machine learning, computers rely on large datasets of examples in order to learn how to perform a given task. The accuracy improves as the machine “sees” more data. The Stanford University research team that NIH referenced did more than just use standard machine learning. “The team’s real interest was in utilizing a special class of machine learning called ‘deep neural networks, or deep learning.’ Deep learning is inspired by how our own brain’s neural networks process information, learning to focus on some details but not others.”

“The findings suggest that artificial intelligence can be used to improve the accuracy and efficiency of EKG readings.”

More examples of FDA-approved medical devices that use Artificial Intelligence (AI)

A few of the new AI systems that have been approved by the FDA include:

• IDx’s IDx-DR is an AI software system for the autonomous detection of diabetic retinopathy in adults who have diabetes. The algorithm analyzes images taken with the Topcon NW400 retinal camera and uploaded to a cloud server. Within minutes the software provides doctors with a binary result, either indicating that more than mild diabetic retinopathy is present and that the patient should be referred to an eye care professional, or that the screen is negative and should be repeated in 12 months. Approved in April, the software is notable in that it was the first AI-based diagnostic system to be authorized by the FDA for commercialization in the US that can provide a screening decision without the need for clinician interpretation.
• Imagen’s OsteoDetect uses an AI algorithm to scan X-ray images for a common type of wrist bone fracture, known as distal radius fracture. The software can be fed images of adult wrists in the posterior-anterior and medial-lateral position, and using these highlights regions with potential fracture. OsteoDetect — which received De Novo clearance in May — is intended for use by primary, emergency, urgent and specialty care practitioners alike, but should be accompanied by a standard clinical review.
• The cloud-based DreaMed Advisor Pro is a diabetes treatment decision support product that analyzes data from continuous glucose monitors, insulin pumps and self-monitoring to determine an insulin delivery recommendation.” Through an event-based learning process, the software incorporates a number of components into its recommendations, including basal rate, carbohydrate ratio and correction factor. Dosage recommendations are delivered directly to the monitoring clinician, who can push the adjustment to a patient’s diabetes management devices with the click of a button. FDA granted DreaMed’s algorithm a De Novo approval in June.
• The Coronary Calcium Scoring algorithm from Zebra Medical Vision offers a coronary artery calcification score from a patient’s ECG-gated CT scan. Clinicians can use this score to flag patients at high risk of cardiovascular disease sooner, thereby allowing for quicker and more effective care. The July 510(k) clearance was the first for the Israel-based company, which also holds a number of algorithm clearances in the EU.

In February, medical imaging software company Arterys Inc. touted 510(k) clearance for its Artyrys Oncology AI suite, a web-based platform that helps clinicians analyze ARIs and CT scans for signs of potential liver and lung cancer. The tool uses deep learning algorithms to expedite interpretation of these images.

Note: Premarket approval (PMA) is the FDA process of scientific and regulatory review to evaluate the safety and effectiveness of Class III medical devices.  Whereas 510(k) clearance (“Clearance,” not “Approval”) is the FDA-designated pathway for less invasive, less risky medical devices for which a “predicate device” that is the “substantial equivalent” already exists on the market.

And also note that we have AI-powered consumer products which, while they may not necessarily meet the legal and regulatory definition of a “medical device,” might be allowed under the “General Wellness” FDA guidance.

How AI and machine learning are classified by FDA

The FDA states that,

“Adaptive artificial intelligence and machine learning technologies differ from other software as a medical device (SaMD) in that they have the potential to adapt and optimize device performance in real-time to continuously improve health care for patients. The International Medical Device Regulators Forum (IMDRF) defines “software as a medical device” as software intended to be used for one or more medical purposes that perform these purposes without being part of a hardware medical device.

“The FDA under the Federal Food, Drug, and Cosmetic Act (FD&C Act) considers medical purpose as those purposes that are intended to treat, diagnose, cure, mitigate, or prevent disease or other conditions.”

Mobile Health News states that most AI products are being classified as medical devices through the 510 (k) pathway. This pathway essentially allows the developer to file a premarket submission based on a showing that the AI medical devices is substantially equivalent to or as effective as an existing marketed device that is not subject to premarket approval.

Additionally, some products are being approved through De Novo clearances. The last option, which is usually quite cumbersome and expensive, is to seek premarket approval.

The FDA states that:

“the De Novo pathway for novel medical devices allows the FDA to conduct a rigorous review of new technologies so that patients have timely access to safe and effective medical devices to improve their health,” said FDA Commissioner Scott Gottlieb, M.D. “At the same time, the FDA is modernizing its 510(k) pathway, which is used for clearance of low- to moderate-risk devices that are substantially equivalent to a device already on the market. The De Novo pathway provides a vehicle for establishing new predicates that can reflect modern standards for performance and safety and can serve as the basis for future clearances.

The FDA admits that its traditional models for medical device regulations weren’t designed for AI and machine learning health technologies. It admits that many new AI and machine learning software applications may require premarket review.

A potential new FDA approach to AI and machine learning

On April 2, 2019, the FDA published a discussion paper – “Proposed Regulatory Framework for Modifications to Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device (SaMD) – Discussion Paper and Request for Feedback” that discusses the FDA’s thoughts on a new approach for reviewing artificial intelligence and machine-learning software for premarket review. The paper requests feedback from the public.

 “In this framework, the FDA introduces a “predetermined change control plan” in premarket submissions. This plan would include the types of anticipated modifications—referred to as the “Software as a Medical Device Pre-Specifications”—and the associated methodology being used to implement those changes in a controlled manner that manages risks to patients —referred to as the “Algorithm Change Protocol.”

“In this approach, the FDA would expect a commitment from manufacturers on transparency and real-world performance monitoring for artificial intelligence and machine learning-based software as a medical device, as well as periodic updates to the FDA on what changes were implemented as part of the approved pre-specifications and the algorithm change protocol.”

The FDA hopes the new regulatory proposals will help them and manufacturers review and monitor AI/ML software from its premarket development though its performance after the product reaches the market. An additional hope is that FDA regulatory oversight will be able to assure patient safety while encouraging the “iterative improvement power” of AI and ML.

The FDA currently approves AI and ML software as a medical device through de novo certifications, 510(K) submissions, and premarket approval. The major problem with approving AI devices is that that they change/adapt after they are delivered to customers or to health care providers. The FDA has developed a framework for how this new technology can be reviewed from development through the post-market distribution. It is seeking feedback from the healthcare community and plans to have new regulations for medical device approval soon – though it made need Congressional approval for these changes.

At the Cohen Healthcare Law Group, our healthcare and FDA Attorneys help physicians and healthcare entrepreneurs navigate choppy and uncertain legal waters. To review key aspects of your medical practice’s legal issues, contact us today.