The FDA wants clinical trials to stream into review instead of arriving as paperwork

The Food and Drug Administration is testing whether drug review can work less like a stack of paperwork and more like a live dashboard. In two cancer trials already under way, the agency says it is now receiving near-real-time signals on how patients are doing, instead of waiting for the usual batch reports that arrive long after the fact.

That is the real story inside the FDA's new "real-time clinical trials" push. It is not AI discovering a new medicine. It is the regulator trying to change tempo. If the experiment works, review could start earlier and move faster because agency staff would see key safety and efficacy signals while a trial is still running, not months later when a sponsor packages them for submission.

According to the FDA press release, the agency launched two proof-of-concept studies and opened a request for information for a broader pilot program it wants to start this summer. One pilot uses AstraZeneca's TRAVERSE study, a Phase 2 trial in previously untreated mantle cell lymphoma, a rare blood cancer, with MD Anderson Cancer Center and the University of Pennsylvania. The other uses Amgen's STREAM-SCLC study, a Phase 1b trial in limited-stage small cell lung cancer, though the FDA said site selection there is still in process.

The AstraZeneca example matters because it is not a shiny new trial built for the announcement. On AstraZeneca's own clinical trial registry, TRAVERSE started on Dec. 13, 2023, has enrolled 108 patients, and is not expected to hit its primary completion date until July 2028. The FDA says it has already received and validated signals from that study through Paradigm Health, the vendor handling the data layer. That makes this less of a concept sketch and more of a live retrofit: the agency is wiring itself into an existing study to see whether earlier visibility changes review speed.

The phrase "real-time clinical trial" invites a lot of fantasy. The actual setup is narrower. Reuters reported that the FDA would receive aggregated safety and efficacy signals, such as adverse event rates or tumor response percentages, rather than raw patient records. That is an important distinction. The agency is not proposing to sit inside every hospital database. It is asking for a faster feed of decision-grade summaries.

Even the AI framing is more bureaucratic than glamorous. The Federal Register notice describes the summer effort as an "AI-enabled optimization" pilot for early-phase trials. Paradigm Health said in a company press release distributed by PR Newswire that its system pulls data from electronic health records and other sources, checks it against FDA-defined criteria in real time, and sends only the signals needed for regulatory decisions. The software's job is not to replace regulators. It is to compress the lag between what happens in a trial and what the regulator can see.

That lag is large enough that the FDA is now treating it as a competitiveness problem. Reuters reported that Commissioner Marty Makary said paperwork and administrative work account for 45% of the time between early testing and an approval submission. He also said China passed the United States in Phase 1 trial volume around 2021. Those claims deserve some caution because they come from Makary's own push for reform, not an independent scorecard. But they explain why the FDA is willing to experiment with the plumbing of trial review instead of just talking about modernization.

The upside is obvious. If reviewers can spot a safety issue, a response trend, or a trial-conduct problem earlier, sponsors might fix problems sooner and approvals might move faster at the margin. The skeptical case is obvious too. Two oncology pilots do not prove the system will scale across messy, multi-site studies in other diseases. And because the FDA says it is reviewing aggregated signals rather than source records, the sponsor and its software layer still control how much of reality gets turned into a readable stream.

So the near-term question is not whether AI is about to reinvent drug development. It is whether the FDA can turn one part of regulation from a batch document business into a streaming one without losing trust in the underlying evidence. If that sounds less cinematic than "AI for trials," good. It is also a lot more consequential.