The U.S. Food and Drug Administration (FDA) has been ahead of the curve for a number of years on emerging issues and opportunities for the agency and regulated industry arising from the advent and evolution of artificial intelligence (AI). FDA’s website has had AI-related pages and publications posted for more than five years, and the agency has issued numerous guidances and related documents to assist researchers and the medical products industries in using AI in medical product discovery, testing and development. 

Most recently, FDA has turned its AI focus inward, in the form of a successful pilot program in which agency reviewers have used AI systems to help streamline the process of reviewing the enormous amount of complex data that is submitted by sponsors seeking approval of new drug applications (NDAs). The success of the AI review pilot program has prompted FDA Commissioner Martin Makary, M.D., M.P.H., to direct all FDA product centers to immediately begin deployment of AI-based scientific review programs, with a full integration target date of June 30, 2025. 

As the volume and complexity of data submitted in support of new drugs, biologics and medical devices grow, AI-assisted product application reviews have the potential to drastically improve efficiency and accelerate the review time needed for potentially life-saving new therapies and medical products. Anecdotal comments by one FDA product reviewer involved in the pilot program reported that the AI system “enabled me to perform scientific review tasks in minutes that used to take three days.” 

Other Potential Benefits of AI in Drug and Device Reviews

The use of AI and machine learning algorithms (MLAs) in the review of NDAs, BLAs, PMAs and other data-dense applications has many potential benefits, including:

• Summarizing clinical trial protocols and results
• Flagging anomalies in clinical trial results 
• Identifying safety signals
• Assessing statistical robustness
• Helping reviewers focus on critical data points
• Supporting benefit-risk assessments by synthesizing data from multiple sources including spontaneous reports, clinical trials and scientific literature. 

Cautions and Caveats

The use of AI in FDA’s core public health functions comes with a number of risks that must be carefully monitored and managed, many of which FDA has recognized and acknowledged. Key considerations include:

• The need to maintain scientific rigor, transparency and accountability at all stages (AI and human) of the regulatory review process 
• Adhering to high standards of validation 
• Avoiding ethical problems, including data privacy breaches, algorithmic bias and adverse impact on health disparities
• Maintaining public trust via transparent methodologies and clear communication with all stakeholders (patients, providers and industry) about how AI and MLA tools are used and how decisions are made. 

Questions Looking Ahead

As FDA’s use of AI gains traction and evolves over time, scientific reviews will presumably become both faster and more accurate. But if so, the impacts could go beyond just the process-oriented. For example:

• Will AI-assisted reviews reduce, or increase, the number of Complete Response Letters (CRLs) issued as the first FDA “action” on NDAs? 
  • Will sponsors be ready to go to market if they receive an unexpectedly fast final approval, or will their initial exclusivity periods be partially wasted by the inability to quickly launch and market newly approved products?
  • Will AI reviews lead to more onerous CRLs? Or will AI lead to fewer or less-burdensome follow-up requirements?
• What are FDA’s plans, and what are the possibilities, for using AI in reviewing and responding to development-stage meetings with FDA (e.g., pre-IND, EOP-2 meetings)? Will sponsors seek to “AI-proof” their meeting request packages to avoid unwanted surprises?
• What will AI-assisted review timing improvements mean in terms of the various User Fee programs? 
  • If FDA can and does improve its review efficiency by xx% using AI, can industry seek to negotiate a reduction in application-based user fees by a similar xx%? 
  • Or, can industry credibly ask for an xx% reduction in the application timing commitments, say from a 10-month standard review user fee clock, to a six-month standard clock? 
  • And what would be the impact on the market value of un-redeemed Priority Review Vouchers (PRVs)?
• Will FDA reductions in force be counterbalanced, in part, by efficiencies provided through AI-assisted scientific reviews?

In the realm of AI, FDA's pace of comprehension, adaptation and use appear to be in much closer parity to the pace of industry than historical precedents might reflect. While some may find this an unusual comparative dynamic, no FDA-regulated company can afford to pay short shrift to, or under-invest in, the rapidly expanding role of AI and how it is impacting, and will increasingly dominate, their pathways to success.