FDA’s February 2024 draft guidance on data monitoring committees puts it in one sentence. Unblinded interim data and the results of comparative interim analyses “should generally not be accessible by anyone other than DMC members or the statisticians performing these analyses and presenting them to the DMC” [FDA, 2024].
Twenty-six months later the agency announced two live trials that report data signals to FDA while they run. The vendor operating the data layer says those signals reach the sponsor too.
The Problem
The debate over real-time clinical trials has been a debate about speed. Commissioner Marty Makary framed it that way on April 28, 2026: for 60 years “key data signals can take years to reach the FDA,” and under the new approach “FDA scientists can view safety signals and endpoints in real time as a trial progresses” [FDA, 2026]. AstraZeneca opened TRAVERSE, a Phase 2 trial in treatment-naive mantle cell lymphoma running with MD Anderson and the University of Pennsylvania. Amgen opened STREAM-SCLC, a Phase 1b in limited-stage small cell lung carcinoma. The agency said it had already received and validated signals from the AstraZeneca trial through Paradigm Health, proving the technical framework holds [FDA, 2026].
Nobody disputes that the plumbing works. The unanswered question is who sits on the other end of the pipe, and what they can see.
That is not procedural housekeeping. It decides whether the trial counts.
The Insight
FDA’s DMC guidance makes the case better than any critic could. “Knowledge of unblinded interim comparisons from a clinical trial is rarely critical for those conducting or sponsoring the trial. Such knowledge can bias the outcome of the trial by inappropriately influencing trial conduct or the approach to analyses” [FDA, 2024]. The guidance ties that directly to 21 CFR 314.126(b)(5), the regulation requiring sponsors of controlled studies to take appropriate measures to minimize bias. The firewall is not professional etiquette. It is a condition of the trial being adequate and well-controlled in the first place.
Then came June 2026. FDA reissued its draft guidance on demonstrating substantial evidence of effectiveness, the document that replaces the 1998 standard when final. Its section on trial conduct lists what shapes the strength of evidence: maintenance of blinding, quality of data collection, adherence to protocol, completeness of follow-up, and “the maintenance of confidentiality of interim results while the trial is ongoing” [FDA, 2026]. A few lines later the same guidance says poor execution “can render a trial of any design not adequate or not well-controlled and, therefore, unable to contribute to substantial evidence of effectiveness” [FDA, 2026].
Set those two documents beside the April announcement and the gap gets specific. FDA has told sponsors, in writing and twice, that who sees interim results during a trial factors into whether that trial can support an approval. It has not published how the real-time arrangement preserves it. No charter appears in the record. No statement of whether the FDA staff watching signals accrue are the same reviewers who will later judge the marketing application. No description of what a signal contains, or whether any of it is comparative.
That last one carries the most weight. Paradigm Health, the vendor, describes its platform as capturing data from electronic health records, algorithmically evaluating FDA-defined reporting criteria, and transmitting the critical signals needed for regulatory determinations to the trial sponsor and the FDA [Paradigm Health, 2026]. That is a vendor describing its own product, not an FDA protocol document, and it deserves to be weighted that way. It is also the most specific public account of the data flow that exists, and it names the sponsor as a recipient.
“FDA has told sponsors twice, in writing, that who sees interim results decides whether a trial counts. It has not said who is watching its own.”
A chartered DMC solves this problem the boring way. An independent committee sees unblinded interim data under prespecified rules, an independent statistician prepares the analyses, and everyone else stays blind. The February 2024 draft spells out what the charter must cover, down to who besides the DMC and the unblinded statistician gets access [FDA, 2024]. That machinery is decades old, unglamorous, and exactly the sort of thing a speed initiative routes around without meaning to.
None of which makes real-time trials a bad idea. Early-phase development genuinely drags, and the hiatus between phases that FDA wants to eliminate is real dead time. Chief AI Officer Jeremy Walsh’s framing, that the agency should judge its processes “from the standpoint of a patient awaiting a potentially powerful treatment,” is the correct instinct [FDA, 2026].
The trouble is what the rest of the package does at the same time. The June guidance clarifies how a sponsor can rely on one adequate and well-controlled trial plus confirmatory evidence to meet the statutory standard, and it widens what confirmatory evidence can be: mechanistic data, natural history studies, and registry or real-world data sources, judged on “the reliability and relevance of the RWD source for its proposed use” [FDA, 2026]. It arrived inside Operation TrialBlazer, an HHS package that also revised the master protocols guidance and opened a request for information on an expedited IND pilot [FDA, 2026].
Fewer pivotal trials. More weight on the one that runs. More weight on evidence that medical affairs generates rather than clinical development. And a newly seated party watching that trial while it enrolls.
Medical affairs inherits all three. When a KOL asks at a 2029 advisory board why a confirmatory registry was reliable, or whether anyone at the sponsor saw interim comparisons during the pivotal study, field medical answers that question, not regulatory. The chain-of-custody detail nobody writes down in 2026 becomes a scientific exchange problem later, and by then the answer is either in the trial master file or it is not. It is the same failure mode as treating a peer-reviewed paper as a finish line rather than reading its limitations section, and the same reason what sits underneath an evidence claim matters more than the claim itself.
In Practice
| What FDA has published | Where it says so | What it leaves open |
|---|---|---|
| Unblinded interim data should reach only the DMC and its statistician | DMC draft guidance, February 2024 | Whether real-time signals are comparative or unblinded |
| Confidentiality of interim results affects the strength of evidence | Substantial evidence draft, June 2026 | Who at FDA sees the feed, and whether they review the application |
| Signals validated through a vendor platform in a live Phase 2 | Press announcement, April 2026 | No charter and no firewall specification in the record |
The Bottom Line
Comments on the substantial evidence guidance close September 22, 2026, docket FDA-2019-D-4964. FDA said it intended to complete selections for the broader real-time pilot in August. Those two dates sit a month apart, which means sponsors are being picked for a data-sharing arrangement while the guidance defining how their evidence gets judged is still an open draft.
The exposure lands unevenly. Large sponsors with regulatory affairs groups that actually read a 17-page draft will negotiate terms before signing anything. Smaller sponsors join because the commissioner announced it and the board wants the meeting. The clearest beneficiaries are the platform vendors now seated between the sponsor and the agency, holding a position in the evidence chain that no guidance describes and no charter governs.
So here is the question worth putting to any sponsor in that pilot, and to the agency running it. If the confidentiality of interim results is a factor in whether a trial is adequate and well-controlled, and FDA wrote that down in June, who has written down how these trials preserve it?