In this age of Generative Artificial Intelligence and Augmented Reality, new medical technologies push forward at a relentless pace. Yet true industry renaissance inches ahead more slowly, that is until it reaches its tipping point. With virtual twins of human experiences driving in silico clinical trials, we are now entering a new era of medical innovation.
Regulators, initially skeptical, have come to recognize the public health potential of modeling and simulation and the robustness offered by modern Generative AI methods. More importantly, they now understand how in silico clinical trials, or running a trial on virtual patient populations can accelerate medical device innovation and safely lower the barriers to regulatory approval.
The ENRICHMENT project, a 5-year collaboration led by Dassault Systèmes and the US Food and Drug Administration, launched in 2019 under the 21st Century Cures Act and has been completed on schedule. The outcome is a comprehensive description of what was learned, sort of a playbook that offers the structure of not only how to use virtual twins (sometimes called digital twins) to demonstrate safety or effectiveness of a new medical device for regulatory approval, but a complete methodology to create and analyze an entire population of patients in a “trial-before-the-trial.”
What is an in silico clinical trial (ISCT)?
The technologies behind ISCTs are not new. They’re quite mature and, in fact, have proven their merit in many other industries. Scientifically based virtual prototyping and simulation allow for experimentation and iteration in a failsafe environment.
We know that in other industries, rigorous testing on a virtual twin of a new car, or plane or even a bicycle precedes physical testing on a real prototype. Rarely, for example will an automotive company find the results of a crash test to differ from their predictions or that flight simulators improperly train pilots. This is not so in medicine. Overall clinic trial success rates remain in the 10-15% range, meaning 85-90% of the time, companies learn the hard way they got something wrong. An in silico clinical trail uses virtual twins or computer-based simulations to safely predict the effects of a drug, disease, medicinal device or intervention on an entire virtual twin population. They can be re-run until success is predicted and then tested in the clinic.
But is that really possible? This is exactly what the FDA hoped to find out.
Over the prior five years, the FDA had participated in the Living Heart Project, observing the development of the first virtual twin of a human heart. Over that time, they had seen examples virtual twins helping surgeons plan complex medical procedures, companies test new prosthetics designs and even have themselves created a family of virtual twins that predict MRI radiology safety. So their interest was high.
The ENRICHMENT project: Using Virtual Twins to accelerate regulatory success
After seeing its potential, the FDA wondered why the industry’s use of computational modeling and simulation remained low. They understood that industry needed to be careful when introducing new methods into the regulatory process where regulators were not experienced, yet the FDA could best learn by doing. A Catch-22. The ENRICHMENT project, which I had the privilege of co-leading, was shaped through discussions with the researchers at the Office of Science and Engineering Labs (OSEL) and ultimately the Director of the FDA Center for Devices and Radiological Health (CDRH), Dr. Jeff Shuren. They recognized that in silico clinical trials represented not just an increase in efficiency, but a paradigm shift in medical device development. Recreating the clinical experience to optimize treatment protocols and trial design could completely change the risk profile, reduce time to market, while supporting the current standards for safety and efficacy of approved devices.
Over the following five years, we worked together with others at the FDA, industry, researchers, and clinicians to systematically work through an ISCT in detail to define an approach the FDA could support. The findings have been submitted for publication.
This spring, I had the unique opportunity to present and observe our findings in a variety of settings. At CES, one of the first times we’ve presented findings from the project to a consumer-friendly audience, observers were struck by how virtual twins used Generative AI to simulate clinical trial outcomes, reducing the reliance on human testing, eliminating the need for a placebo group and speeding up the discovery process. The notion that virtual trials could precede actual ones to ensure efficiency and inclusivity was both fascinating and hopeful.
Highlighting our findings at an FDA sponsored Symposium in Washington, D.C. and moderating a DeviceTalks Boston panel on the potential impact of ISCTs on medical device innovation, our discussions dialed into the role of Generative AI as an accelerator in technology, the evolution of biomedical models, and the regulatory landscape that is gradually adapting to embrace these innovative approaches.
To me, these conversations acknowledged the FDA’s confirmation within the ENRICHMENT ISCT project, that properly trained and validated Generative AI methods are a significant enabler of innovation, offering a structured framework to establish regulatory credibility, and process that could dramatically reduce time to market, lower trial risk while ensuring safety and efficacy of the devices that are approved.
What are the regulatory implications of ISCT for medical device innovation?
With all of the buzz around the use of Generative AI in healthcare, the FDA’s approval to publish the ISCT playbook marks a significant advance in their regulatory framework, providing a structured approach to incorporate advanced simulation technologies into the development and approval process of new medical devices. This endorsement not only validates the work done by dozens of multidisciplinary experts over five years in the ENRICHMENT project, but also sends a strong message to industry that investments in improving the robustness and reliability of these model can pay dividends in future regulatory considerations.