[ad_1] When the FDA announced recently that two drug companies had begun streaming clinical trial data to agency reviewers in real time, FDA Commissioner Marty Makary, MD, MPH, reached for an analogy from his own training. As a surgical oncologist, he watched his patients in the operating room and ICU on continuous monitors. When vitals dropped, the team intervened. Nobody waited for a report in the mail. Why, he asked, should regulatory oversight of experimental drugs be any different? It's a good analogy. It also gives away more than the announcement intends. Real-time data is not the same thing as real-time oversight. A monitor that beeps in an empty hospital hallway is not a clinical safeguard, and a data feed that arrives at FDA without a defined plan for who acts on it, when, and how is not regulatory review. The challenge in continuous monitoring has never been getting the data. It has been deciding what to do about it. The announcement describes, in detail, a system that delivers signals faster. But right now, it says little about what happens when one arrives. That gap is something we should be watching as this initiative evolves. We have spent decades learning what goes wrong when monitors get installed before response systems mature, and the lessons are not subtle. What Is on the Public Record Two trials are underway. AstraZeneca's Phase II study in previously untreated mantle cell lymphoma is streaming data through a platform built by the company Paradigm Health. Amgen is preparing a Phase Ib study in small cell lung cancer. The agency has met with each company to establish the criteria for reporting signals in real time, and a broader pilot will open through a public comment process, with comments due May 29 and selection criteria expected in July. The data pipeline is well described: which trials, which company, which platform, what kinds of findings. What's much less clear is what happens after a finding arrives. Asked how reviewers should engage, FDA Chief AI Officer Jeremy Walsh told reporters the agency would "let the review committees decide what type of engagement that they feel like they need to have in order to get the answer they need." Makary's framing emphasized what reviewers will see, such as a fever as it develops or a tumor shrinking as the CT is read, rather than what they will do with that information. This initiative is not in a vacuum. Drug companies and investigators already follow rules requiring them to flag serious unexpected reactions to FDA within 15 calendar days, or faster for life-threatening events. Independent safety committees already review accumulating data. Real-time review doesn't replace any of that. It adds a new FDA-facing layer on top of it. What's not yet clear is how that new layer will talk to the existing one: which findings trigger which kind of FDA engagement, on what timeline, with what relationship to what companies are already doing. What We Already Know About Alerts Alerts without thresholds drift into noise. Alert fatigue in clinical settings is well-documented: when most alerts don't require action, clinicians habituate, and real signals get overridden along with the false ones. The lesson for regulators is not that real-time data is dangerous, but that data without a published action map loses meaning as the volume grows. Diffuse ownership produces no ownership. Clinical alert systems work when alerts are sent to a named person with a defined response window. They fail when alerts are sent to "the team" or "the unit," because everyone assumes someone else is handling it. Walsh's "review committees decide" is the regulatory version of an alert with no named recipient. Some reviewer, somewhere, will see the signal. What happens next is unspecified. What gets surfaced sets the agenda. Real-time platforms don't show reviewers everything. They show what the platform's validation logic flags as worth showing, in the order the logic decides. That's a reasonable design choice, as the alternative is drowning reviewers in raw data. But it means the platform's filtering decisions quietly become regulatory priorities. A finding that doesn't clear the validation bar, or that surfaces late, gets less attention by default. None of this is anyone's fault. It's a property of the system. Three Standards, Before the Pilot Scales The pilot is worth running, though it should not expand into a permanent program before the agency publicly specifies minimum response-side standards. Three are sufficient. A response plan, written down in advance. Before each trial begins, FDA and the company should agree on a document mapping classes of findings to categories of response: documentation only, reviewer acknowledgement, triage review, formal escalation to the company, or referral into the existing monitoring committee or institutional review board (IRB). The findings worth specifying include serious unexpected reactions, dose-limiting toxicities, stopping-rule triggers, pre-specified efficacy or futility boundaries, and clinically meaningful safety imbalances. The trial-specific version can stay confidential. A class-level template should be public. Named ownership. Each trial should identify the FDA division, accountable reviewer role, backup coverage, escalation pathway, and expected review window. Committee-level discretion is not a substitute for an owner. Public timing data. The agency should commit to publishing, across the pilot in aggregate, how long it takes signals to move through the system: from validated finding to acknowledgement, from acknowledgement to triage, from triage to action. It should also report what fraction of findings get classified as non-actionable, duplicative, batched, escalated, or sent back for closer inspection. Without these numbers, no one can tell whether real-time access is producing real-time decisions or just real-time watching. The FDA's own request for public input asks about decision timeliness. The pilot is the right place to determine an answer. One Objection Worth Addressing The strongest pushback on this initiative may be that the existing safety system already handles it: the 15-day reporting rules, the monitoring committees, the IRBs. But each was built around scheduled checkpoints, not a continuous feed. Take monitoring committees: they review accumulating data on a planned cadence, against stopping rules agreed on in advance. That's what makes their decisions defensible. An FDA reviewer watching the same data continuously, between those meetings, is doing something the existing setup wasn't designed to accommodate. The existing system is necessary but not sufficient. And whatever this pilot establishes now will be hard to undo later. The commissioner's analogy is the right one. Continuous monitoring in clinical care became useful over decades not because the sensors improved, but because the response infrastructure around them matured. Thresholds got specified. Pathways got named. Numbers got published. The systems that didn't develop that side produced more alerts, less accountability, and a documented pattern of harm. The FDA is at the start of that arc. The pilot is promising, and the case for real-time review is stronger than its critics have allowed. But continuous monitoring without a response loop is the most familiar failure mode of the tradition the agency is borrowing from. The way to avoid it is to specify the response side now, before the pilot becomes permanent infrastructure. Please enable JavaScript to view the comments powered by Disqus. [ad_2] Source link : https://www.medpagetoday.com/opinion/second-opinions/121080 Author : Publish date : 2026-05-01 19:25:00 Copyright for syndicated content belongs to the linked Source.