FDA Prepares to Regulate AI in Medical Devices

New and advanced AI innovations are increasingly touching medical devices.  According to the Roll Call report, the FDA is already struggling with the reality of regulating all living things in a much better way. Looking forward, traditional ways of regulating medical devices are unlikely to work for the latest innovations powered by AI.

At present, the FDA is evaluating a specific approach to evaluate companies manufacturing medical devices, which includes the maker’s organizational culture instead of simply focusing on the equipment or device. The FDA is all set to assess various machine learning practices, data that AI systems are trained to manage, and fine-tuning of algorithms. The performance assessment and monitoring processes would be a part of the evaluation process too.

However, according to the report, the history of the agency dealing with medical device regulation is still found to be waiting in certain cases. In addition, fast-track approvals for medical devices are also under scanner after consumers reported injuries for surgical staplers, mesh, and breast implants. Contrarily, making AI algorithm changes could require a review from FDA under all future requirements.

The Future of AI

The Food and Drug Administration is already in the process of preparing for a future of AI-enabled medical devices. They are planning to roll out the guidelines to assess and certify medical devices, considering them to be living objects that cannot be regulated in the same way as any other old-fashioned equipment.

In fact, the agency is already pivoting to innovative ways to assess devices by artificial intelligence and machine learning as traditional approaches are not applicable to new machines, says Bakul Patel, Director, FDA Digital Health Division.

In an interview, Patel said, “artificial intelligence and machine learning-based devices can be created quickly, and iterations of the product can be rapidly fielded.” He also pointed out, “It’s a living thing to some degree, so having a concept of authorizing it to go to the market and waiting for things to happen and then doing review seems to be outdated.”

The agency released a white paper in early April, which outlined a complete product lifecycle approach focusing on processes, quality control, organizational culture, and testing of the marker of medical devices instead of a typical static assessment of particular equipment. It is now seeking feedback and comments on the proposals.

The white paper released by FDA reads that this approach “would provide reasonable assurance of safety and effectiveness throughout the lifecycle of the organization and products so that patients, caregivers, health care professionals, and other users have an assurance of the safety and quality of those products.”

The Troubled Past

The oversight of the FDA’s medical device approval process, however, is believed to be erroneous. The International Consortium of Investigative Journalists, after investigating for a year, found that FDA approved implant devices too early. They also failed to stop these devices from being inserted into patients, which includes breast implants, surgical staplers, and mesh. They could not even stop these from being sold and used in spite of several complaints received from doctors and patients.

The investigation reported the fast-track approval process of the FDA, for devices, called the 510 (K), as the major reason for faulty devices on sale. The new approach of the FDA, which involves artificial intelligence-driven devices, extends the fast-track process of the agency.

Zach Rothstein, AdvaMed’s associate vice president for technology and regulatory affairs cited that presently, the agency is also testing new processes, solely intended to apply to the latest and emerging devices having the capability of evolving, learning, and modifying software algorithms for better result. He also added that the group is preparing comments and feedback on the FDA’s proposal.

In his statement, Rothstein said, “right now under FDA statute and rules, there isn’t a mechanism for the agency to let that happen. All products that use machine learning and neural networks are locked, in the sense, they’re not evolving anymore, and the FDA is attempting to come up with a pathway to allow products to learn in a post-market environment.”

Using a locked algorithm in an artificial intelligence-enabled device means that making any new changes to the algorithm based on new or added information collected from real-life would “likely require FDA premarket review for changes beyond the original market authorization,” as per the proposal of the agency.

Approval Expected Soon

The proposal by the FDA to find innovative ways of assessing continuously adapting artificial intelligence devices is considered o be an extension of the process that started almost five years ago. At that time, the agency issued its first guidance on mobile medical applications.

Since that time, the agency has been steadily moving forward towards the assessment of the overall quality of software maker rather than certifying each software iteration.

The FDA has a precertification pilot program in place, which took off in 2017, for software makers functioning as medical devices. Approved software makers under the same process could potentially launch some new products without the premarket review of the FDA or even win approval by following a few simple steps.

sepStream® is one of the leading artificial intelligence-enabled solutions that are comprehensive, reliable, and easy to use. We provide EHR/EMR/practice management suite, which ensures 100% satisfaction.