The U.S. Food and Drug Administration was slammed recently for overhyping its disclosure of a set of rejection notices it had sent to drug companies. As reporters quickly discovered, most of the “released” 200 complete response letters (CRLs) had been in the public domain for years, and the FDA had merely put them into a single downloadable folder.
Although the FDA’s action was criticized for its meager impact, it highlights an important opportunity. The already-public information about approved drugs is spread across various agency web pages and often buried in PDF documents, making it hard to efficiently find and analyze. Improving how the FDA organizes and disseminates these materials may be unsexy, but if done at scale, it would be a big step toward the era of “radical transparency” that Commissioner Marty Makary has promised.
To understand the potential value of upgrading how the FDA stores public data and documents, imagine you want to get a list of all medicines the agency has approved, even though one or more of their pivotal trials had failed. Maybe your company’s clinical study just tanked, and you want to know the odds you can still gain a green light from regulators. Or you might be an academic researcher interested in the policy question of whether the FDA’s approval bar is too lenient, too strict, or just right. Or perhaps you’re a member of the public who just read a Facebook post about a newly approved drug with mixed or negative data, and you want to do your own research to understand why and how often this happens.
Currently, this can take dozens of person-hours of hands-on labor — which I know because I did it myself in 2023 for a STAT report. First, I manually compiled a list of all 236 drugs approved by the FDA’s Center for Drug Evaluation and Research (CDER) from 2018 to 2022 from its year-end PDF reports. Then, I manually input each drug’s name into the agency’s public-facing Drugs@FDA website, clicked through until I reached the list of documents from the original assessment, downloaded the “summary review” PDF, and skimmed it to figure out how many pivotal trials were positive versus negative. Finally, when I had identified the 16 drugs that were approved based on mixed or wholly negative data, I carefully read the summary and clinical review documents to understand the FDA’s rationale in each case.
My experience highlights two main transparency bottlenecks at the FDA.
Failed clinical trials don’t always spell doom for a new drug. Here’s why
First, there is no “one-stop shop” to efficiently identify approved drugs that meet certain criteria and download their key attributes, full review packages, labels, and all other public documents. The Drugs@FDA interface allows only single-drug queries by name, and although some of my analysis could have been accelerated if I’d known how to use the site’s application programming interface (API), it still doesn’t give access to the full trove of public information and materials on all approved medicines.
Second, most documents that the FDA releases about approved drugs are available only as PDFs, so they can’t be efficiently analyzed with computational tools like large language models (LLMs) without significant additional processing.
The FDA should take two meaningful steps in the near term to make it easier to access and analyze public information about approved drugs.
First, the agency should integrate all public information and documents about approved drugs into the Drugs@FDA database and upgrade the interface to enable advanced searches, downloads, and analyses. There is no reason why the work I did to identify and analyze all drugs approved in recent years with negative trial data should be so laborious. From my initial forays, I’ve concluded it would be a bit of work but doable to upgrade Drugs@FDA so one could easily identify all drugs that meet specified criteria and bulk-download relevant documents and data. Bringing all public information about approved medicines into an updated Drugs@FDA would also help an individual user interested in a specific drug easily pull all relevant information.
Second, the agency should publish machine-readable versions of all public materials related to approved drugs instead of just PDFs. There are no significant technical barriers to converting reviews, press releases, advisory committee transcripts, and other agency documents into formats that are easier to analyze with LLMs and other natural language processing algorithms. This action would hasten the development by both the government and third parties of new tools to interrogate the FDA’s vast trove of public information about marketed prescription medicines.
FDA publishes rejection letters sent to drugmakers, with a big caveat
Importantly, these same technology improvements would also help the agency’s internal experts be more effective and efficient, which is another of Makary’s stated goals. Organizing the FDA’s collection of public documents and making them machine-readable would help reviewers compare new applications to relevant precedents and facilitate the deployment of artificial intelligence tools internally to drive further productivity gains. These upgrades could also boost the quality and consistency of reviews by enabling insights and best practices to spread more easily across the FDA’s divisions.
Although Makary’s early effort to increase transparency was underwhelming to many critics, it provides a hint of what he could accomplish during his tenure if he is serious about this endeavor. Fast-tracking technology upgrades to make it easier for professionals and members of the public to find and analyze all public documents related to approved drugs would be a relatively straightforward way for the FDA to score a quick win with significant lasting impact.
Frank S. David, M.D., Ph.D., is managing director at the consulting firm Pharmagellan and professor of the practice in the Tufts University Department of Biology. Pharmagellan advises companies, investors, and other entities that operate in and around the drug development industry.
