Add a bookmark to get started

abstract structure
5 September 20238 minute read

FDA is embracing AI for its own purposes. Are you keeping pace?

The US Food and Drug Administration (FDA) is increasingly integrating artificial intelligence (AI) into its operations. FDA’s Center for Devices and Radiological Health (CDRH) has been leading the conversation on AI within FDA, by virtue of its oversight of AI-based medical software and being home to FDA’s Digital Health Center of Excellence. Other agency centers are following suit, grappling with industry’s use of AI [1] and exploring AI’s potential to support FDA’s own activities. This piece highlights some of the ways these other FDA centers are embracing AI and explores what that means for the sector.

Center for Drug Evaluation and Research

In addition to publishing discussion papers seeking feedback on the regulation of AI in drug manufacturing and in the development of drug and biological products, the Center for Drug Evaluation and Research (CDER) has been working to facilitate AI's effective incorporation into regulatory decision-making processes. To assist in this mission, CDER formed a dedicated AI steering committee in 2020, which has since reportedly been tracking AI-related activities, identifying emerging best practices, and promoting the appropriate use of AI within the center.

According to HHS reporting, CDER has over 20 different AI use cases in the works, including the following:

  • The Office of Surveillance and Epidemiology (OSE) has a number of AI projects on the books. Earlier this year, OSE started using a tool with AI capabilities to assist human review and comparison of drug labeling to identify safety-related changes occurring over time. OSE also has plans to start building a prototype software application to support the human review of FDA Adverse Event Reporting System (FAERS) data. The algorithm, developed using annotated FAERS reports, is intended to semi-automatically categorize FAERS reports (as being medication related and with an identified type of medication error) based on natural language processing of report free text. Additionally, OSE has various AI initiatives relating to electronic health records (EHR) and FDA’s Sentinel system, including to develop a set of algorithms to augment assessment of mortality through probabilistic linkage of alternative data sources with EHRs and to develop algorithms to detect and reduce coding differences between healthcare systems in Sentinel.
  • The Office of Clinical Pharmacology (OCP) AI Analyst platform is trained to auto-author clinical study reports from source datasets. The platform has been trained and tested with hundreds of New Medical Entity submissions and over 1,500 clinical trials. It can be used to assess the strength and robustness of analytical evidence for supporting drug labelling languages. In 2022, the OCP initiated the RealTime Analysis Depot project aiming to routinely apply the AI platform to support the review of New Drug Application and Biologics License Application submissions.
  • The Office of Strategic Programs (OSP) is using AI to optimize its Opioid Data Warehouse (ODW) dictionaries. The ODW integrates disparate data sources into a centralized cloud environment to enable analytics, like finding patterns and trends in opioid dispensing, abuse and diversion, and morbidity and mortality. OSP maintains data source dictionaries and documentation to facilitate understanding of ODW data characteristics and considerations. OSP is using AI for opioid terminology identification and novel synthetic opioid detection, using the technology to analyze publicly available social media and forensic chemistry data to identify novel referents to drug products in social media text.
  • The Office of Biostatistics also reports having a couple of AI initiatives in flight. One project, focused on machine learning and data augmentation, aims to support detection of data anomalies under Abbreviated New Drug Applications (ANDA) by modeling the complex patterns of pharmacokinetic (PK) data to help FDA’s assessment of the adequacy of data submitted. In addition to making the ANDA regulatory process more efficient this project is also expected to have public health research and drug development utility.[2]  The other project uses unsupervised machine learning to detect and identify data anomalies in clinical trial data at country, site and subject levels. Such information could be used to assist FDA with site selection for inspections.

Center for Biologics Evaluation and Research

According to Director Dr. Peter Marks, the Center for Biologics Evaluation and Research (CBER) specifically recognizes AI's potential in enhancing post-marketing surveillance for biologics and vaccines. When speaking at a recent public event Dr. Marks remarked that AI, in particular natural language processing, would be very useful to analyze adverse event reports efficiently and prioritize which reports are flagged for human review. Since 2022, CBER has been using AI tools to help process voluminous docket comments submitted to the center. This includes automation of processes to transfer, deduplicate, summarize and cluster docket comments. CBER’s Biologics Effectiveness and Safety (BEST) system employs a suite of applications and techniques to improve the detection, validation and reporting of biologics-related adverse events from EHRs, including AI solutions to detect potential adverse events and extract the important features for clinicians to validate. A recent FDA contract notice suggests that CBER is looking to procure more AI-based tools for its BEST system.

Office of Regulatory Affairs

The Office of Regulatory Affairs (ORA) is embracing AI to identify critical public health risks that require attention and to improve inspectional operations. By leveraging mobile platforms and analytics, ORA can more easily capture mobile data and generate automated reports. This approach enhances risk assessment and management, as AI can swiftly analyze vast datasets for potential hazards.

ORA’s specific integration of AI is most commonly associated with import screening, whereby the agency has integrated the technology into existing import data systems to inform decisions about food and drug sampling at the more than 300 US ports of entry. Now in its third phase, the AI import pilot program is scheduled to be completed in late fiscal year 2023 and has reportedly contributed to a 10 percent increase in the number of products evaluated and a 9 percent increase in compliance actions taken.

Center for Food Safety and Applied Nutrition

FDA’s New Era of Smarter Food Safety initiative, announced in July 2020, promises to help strengthen the agency’s public health mission through AI by improving FDA’s ability to quickly and efficiently identify products that may pose a threat. Through the AI import pilot program, FDA’s Center for Food Safety and Applied Nutrition (CFSAN) is exploring the use of AI algorithms to target high-risk seafood products offered for import. CFSAN is also developing standardized ontology-based metadata to leverage foodborne pathogen whole genome sequencing to predict breakout source tracking.

Office of Chief Scientist and Other FDA Research

Further advancing its regulatory science, FDA has multiple initiatives aimed at research on special topics, including various efforts focused on AI. These workstreams are too numerous to itemize in this article; here are a few illustrative examples of the ways in which FDA is working to advance its understanding and use of AI:

  • The National Center for Toxicological Research’s (NCTR) AI4TOX program aims to harness AI for applications including predicting toxicology data, analyzing preclinical data, and more efficient information retrieval.
  • A Yale University-Mayo Clinic Centers of Excellence in Regulatory Science and Innovation (CERSI) collaboration evaluates the applicability of AI-based adaptive strategies that allow mid-course modifications in clinical trials. The goal is to evaluate the use of AI to facilitate clinical trial design, which will inform methods and establish standards for future regulatory submissions.
  • A University of Maryland CERSI collaboration seeks to improve remote interaction tools, including automatic speech recognition technologies, like transcription and translation tools, to develop models that work well with different accents and specialized regulatory and scientific terminologies.
  • A Johns Hopkins University CERSI collaboration intends to advance tobacco product surveillance through the use of AI. The project was set up to use AI statistical computation to analyze millions of social medial posts with the goal of helping government agencies and public health advocates to better understand consumer perceptions and the tobacco product landscape.

What does this all mean?

As described above, FDA has been working toward a varied integration of AI within its operations, and forward-looking companies will want to do the same. Surveillance, primarily in a post-market safety context but also in relation to clinical data, is emerging as a key use of AI by FDA. Regulated industry is motivated to adopt similar AI tools to avoid being technologically outpaced by its regulatory body. It looks like AI is well on the way to being a standard part of the life science surveillance toolkit, and staying up to date with AI advancements in this area could help regulated entities demonstrate a proactive commitment to compliance.

The agency seems to have accepted the utility of AI in various use cases in drug manufacturing and development, even if it is not yet clear on how it is going to regulate those uses. Regulated product developers should implement AI solutions in thoughtful and responsible ways to realize the benefits of AI, while guarding against reputational and legal risks. Because this is a continuously evolving area, industry should provide feedback to FDA to help shape the regulatory framework around AI and remain flexible as new and updated guidance is issued.

To keep up to date on FDA’s AI evolution, please contact your DLA Piper relationship partner or the authors of this article.


[1] For example, there has been an uptick in drug and biological product submissions that reference AI - only one such submission in 2017, but 175 such submissions in 2022.

[2] The models may be used to understand and quantify variability in drug response to guide stratification and targeting of patient subgroups, and to provide drug and dosage insights for such subgroups.