FDA Moved a Class of Cancer Tests From PMA to 510(k). Fewer Than Five Parties Commented.
On 17 August the FDA published a final order reclassifying in situ hybridization test systems used alongside approved cancer therapies from class III to class II. The coverage has called it an easing of the approval path, which it is. The more useful reading is in the comment record, in one special control that was quietly deleted, and in what happens to the evidence bar once the first predicate exists.
FDA’s final order, published at 91 FR 53184 and effective 16 September 2026, moves ISH test systems indicated for use with a corresponding approved oncology therapeutic product, product codes NYQ, MVD, OWE and PNK, out of premarket approval and into class II with special controls under a new 21 CFR 864.1890. Manufacturers will file 510(k)s rather than PMAs.
FDA acted on its own initiative rather than on petition, declined to convene a classification panel, and received comments from fewer than five parties, mostly from the device industry, all of them supportive. Two special controls changed between the proposal and the final order: surrogate samples became acceptable to supplement clinical specimens in precision studies, and the proposed requirement for reagent stability data was removed altogether.
This is not an isolated decision. CDRH announced in 2024 that it intended to begin reclassifying most high-risk in vitro diagnostics, and this order is one instalment. The commercial consequence sits in the predicate chain rather than in this order alone.
A reader can stop here with the full picture. The sections below are the detail.
The move is defensible, and the mechanism is the part worth watching
Nothing in the science is contentious. In situ hybridization has been in clinical use for decades, and FDA’s reasoning is that a mature technology with a well-characterised failure mode can be governed by special controls rather than individual premarket approval. The risk it names is specific: false positive and false negative results that may influence treatment decisions, including by delaying access to an appropriate alternative therapy. Special controls at 864.1890 answer that with nine design verification and validation requirements and three labeling requirements, covering cut-off justification, analytical sensitivity and specificity, precision across lots, readers and sites, linearity, specimen stability and clinical performance.
That is a real framework, not a waiver. What changes is who verifies it and when. Under PMA, FDA reviews the full evidence package for each device. Under 510(k), a manufacturer demonstrates substantial equivalence to something already on the market. The agency states plainly that this decreases regulatory burden and that 510(k) review is typically shorter. Both are true, and neither is the whole picture.
One special control was added, one was deleted, and the deletion is the quieter one
Between the June 2025 proposal and this final order, FDA made two substantive changes to the special controls, both in response to industry comments.
The first is visible and reasonable. A commenter argued that clinical specimens are sometimes unobtainable in sufficient numbers, particularly in rare tumours, and asked that comparable surrogate samples be permitted in precision work. FDA agreed in part, holding that precision studies remain fundamental while allowing surrogates, where it determines them appropriate, to supplement clinical specimens. The word supplement is doing work, and FDA kept the right to judge case by case.
The second change is a deletion. The proposed order carried a special control at 864.1890(b)(1)(viii) requiring device performance data demonstrating appropriate reagent stability. The final order removes it. FDA’s reasoning is that premarket notification requirements and the quality system regulation at 21 CFR part 820 already ensure reagents are appropriately assessed and labelled, so a dedicated special control is unnecessary. Specimen stability survives as a special control. Reagent stability does not.
Whether that matters depends on how much confidence you place in part 820 as a substitute for a device-type-specific requirement. It is a defensible call, and it is also the kind of change that disappears into a numbered list, made in response to comments from the industry that will now file the 510(k)s.
The access angleFDA’s stated expectation is that reclassification will let more manufacturers develop these tests, and that patients will benefit from increased access. That mechanism is credible. PMA economics have kept companion testing concentrated among a few large diagnostics firms, and a cheaper route opens the category to smaller developers. Where a targeted therapy exists but the test identifying eligible patients comes from one supplier at one price, access is constrained by the diagnostic rather than the drug. The trade is fewer per-device reviews in exchange for wider availability, which is coherent rather than a giveaway.
The word FDA declined to use is companion
One commenter asked the agency to insert the term companion into the identification language at 864.1890. FDA refused, saying the comment offered no supporting context and that the identification as drafted was clear enough. So the regulation describes tests intended to provide information related to the use of a corresponding approved oncology therapeutic product, and never calls them companion diagnostics.
Most coverage will use the term anyway, and in commercial substance it fits. But the distinction is not accidental. Companion diagnostic is a defined status tied to a specific therapeutic product’s labeling. The category FDA has created here is broader and more portable, which becomes relevant the moment somebody wants to use one of these devices as a predicate.
The predicate chain is where a one-off order becomes a standing policy
A commenter asked how substantial equivalence would work if a device’s intended use expanded to a new therapeutic product, a new clinical indication or a new clinical cut-off. FDA’s answer repays reading carefully. Devices could serve as predicates for a new clinical cut-off or a new indication that includes a new therapeutic product, provided the new indication falls within the predicate’s intended use and any technological differences raise no new questions of safety and effectiveness.
Read that alongside identification language that avoids naming a specific companion product, and the next few years come into view. Once a first device is cleared under 864.1890, others can be built against it, and a new cut-off or paired therapy may travel through 510(k) rather than triggering fresh evidentiary review. The clinical performance special control still applies to every submission, so this is not an open door. It is a lower threshold applied repeatedly, and the cumulative effect will exceed this single order.
The order also flags a gap without resolving it. FDA states that, at publication, it has not classified, cleared, approved or authorised any ISH test system of this type incorporating digital pathology, including AI or machine learning assisted algorithms. It points manufacturers toward predetermined change control plans and the Q-Submission programme for future AI-enabled modifications. So the framework built for a mature manual technology is the one algorithmic image analysis will eventually enter, and no such device has been through it.
What to watch
Three things. Which manufacturer files first under 864.1890, and whether it is an incumbent converting a modification or a genuine new entrant, since that answers whether the access argument is working. Whether FDA reverses its decision not to issue compliance guidance, which it left open and which the first ambiguous submission will likely force. And how many further high-risk IVD categories move under the 2024 CDRH programme, because treating each reclassification as routine gets harder the more of them arrive on comment records this thin.
The substantive judgement here is probably right. Fewer than five commenters, all of them supportive, all of them from the regulated industry, and no advisory panel, is a thin basis on which to be confident of that. Those two sentences are not in conflict, and holding both is the honest position on this order.