FDA Digital Health clears first multi-modal triage Pre-Sub.

The Pre-Sub contents: what Aidoc actually submitted

Aidoc filed its Pre-Submission package with the FDA on 14 January 2026 under the Q-Sub procedure that the agency uses for early-stage device clearance dialogue. The submission is structured around what Aidoc internally calls the BriefCase MM platform — a multi-modal clinical triage capability that ingests four input modalities concurrently and produces a structured triage output. The four modalities, as named in the submission package: diagnostic imaging (CT, MRI, X-ray, and ultrasound), structured laboratory data (CBC, comprehensive metabolic panel, troponin, D-dimer, and lactate), continuous vital signs (heart rate, blood pressure, respiratory rate, oxygen saturation, and core temperature), and unstructured clinical narrative (chief complaint, history of present illness, and triage nurse notes). The output, as Aidoc has documented it, is a five-level acuity stratification — analogous to the Emergency Severity Index (ESI) but with explicit reasoning attribution for each modality's contribution to the final acuity score — combined with a suggested next-step recommendation that the triage clinician must affirmatively accept, modify, or reject before the recommendation enters the patient's clinical record.

The Pre-Sub package runs to 847 pages including all appendices, and is the longest single Pre-Sub the Digital Health Center of Excellence has accepted on an AI/ML SaMD product to date. The page count is not vanity. Multi-modal validation requires demonstrating that the integration layer — the model architecture that fuses the four modalities into a single output — is not introducing failure modes that the individual modality validations would not have caught. Aidoc's validation methodology is built on what its Regulatory Affairs team, led by VP of Global Regulatory Strategy Dr. Jessica Krasner, has documented as a three-tier analytical approach: first, per-modality validation against established reference standards (the standard SaMD validation approach the agency has been processing since 2018); second, paired-modality validation testing each two-modality combination against ground truth (a less common methodology that adds substantial validation cost); and third, full four-modality integration validation against clinical outcomes in the operational deployment environment. The third tier required Aidoc to commit to a post-market surveillance protocol that the agency's pre-publication guidance had recommended for high-risk AI/ML SaMD products but had not yet treated as a Pre-Sub gating requirement. Aidoc's voluntary acceptance of the post-market surveillance commitment is, according to two clinical informatics directors who have reviewed the submission under a research-use agreement, the structural decision that moved the Pre-Sub through to favourable review.

The submission also contains what Aidoc calls a reasoning attribution layer — a structured output element that decomposes the platform's final acuity recommendation into per-modality contribution weights. The clinical premise is that a triage clinician receiving a high-acuity recommendation needs to know whether that recommendation is driven primarily by the imaging finding, primarily by the laboratory result, or by a convergent signal across multiple modalities. The reasoning attribution layer is technically novel: it is the first formal Pre-Sub submission to include modality-attributed reasoning as a labelling-relevant output element rather than as an internal engineering feature. The agency's feedback letter explicitly endorses the reasoning attribution layer as a contributing factor in its favourable assessment, on the documented basis that the layer materially supports the agency's principle that AI/ML SaMD products in high-acuity clinical contexts must produce outputs that a clinician can interrogate and override. The endorsement is significant because it establishes, for the first time at the Pre-Sub level, that reasoning attribution is a recognised regulatory-quality structural feature rather than an engineering nice-to-have.

The agency's documented questions: where Aidoc must respond

A favourable Pre-Sub feedback letter is not unconditional. The Digital Health Center of Excellence's letter to Aidoc, dated 6 May 2026 and signed by Dr. Troy Tazbaz with concurrent signatures from the Office of Product Evaluation and Quality Director Dr. Michelle Tarver and the Center for Devices and Radiological Health Acting Director Dr. Jeffrey Shuren, documents fourteen specific questions that Aidoc must address in its formal De Novo submission. The questions are grouped into four analytic categories that, taken together, define the agency's current intellectual posture on multi-modal AI/ML SaMD review. Understanding the questions is more informative than the letter's headline outcome.

The first category, comprising five of the fourteen questions, concerns the integration layer architecture. The agency is requesting documentation of how the model architecture handles input modality absence — that is, the failure mode where a patient arrives with one or two modalities unavailable (a common clinical reality in emergency triage). The agency wants to see the platform's behaviour validated under partial-modality conditions with documented outputs that either decline to produce a recommendation or produce one with explicit confidence-degradation labelling. Aidoc's submission addressed this concern partially in the original package; the agency is requesting expanded validation, specifically asking for documentation of platform behaviour across all 15 possible non-empty subsets of the four modalities. The expanded validation is non-trivial — it requires running the platform through 15 distinct validation cohorts and documenting failure-mode behaviour for each — but it is a known-cost expansion that Aidoc's Regulatory Affairs team has signalled it can complete by Q4 2026.

The second category, four questions, concerns the labelling language for the reasoning attribution layer. The agency endorsed the layer in principle but is requesting specific labelling commitments around how per-modality contribution weights are displayed to the triage clinician. The agency's documented concern is that contribution weights presented as numerical percentages — Aidoc's current implementation — risk implying a false precision about the model's internal reasoning that the underlying architecture does not actually support. The agency is requesting Aidoc consider, in its formal De Novo submission, either a categorical labelling scheme (high / medium / low contribution) or a relative-rank labelling scheme that does not imply quantitative precision the architecture cannot deliver. This question is consequential because the labelling language is what frontline clinicians will actually see. Aidoc's product team has signalled internally that the categorical-versus-quantitative question is open and that the company's clinical advisors are split. The agency's position is influential but not yet final.

The third category, three questions, concerns the post-market surveillance commitment. The agency is requesting clarification on the data-sharing architecture that will support the post-market surveillance protocol, including how Aidoc will maintain HIPAA-compliant data flows from deployed clinical sites back to its analytic infrastructure, how it will handle institutional review board (IRB) approvals across the deployment network, and how it will treat post-market data that suggests material performance drift relative to the validation cohort. The post-market commitments are the most operationally consequential elements of the entire Pre-Sub package; they will define the cost structure of Aidoc's compliance function for the life of the product. The agency's questions in this category are exacting but consistent with the published post-market surveillance guidance the Digital Health Center of Excellence finalised in March 2025.

The fourth category, two questions, concerns the model update and retraining policy. The agency is requesting Aidoc commit, in its formal De Novo submission, to a documented model update governance protocol that specifies which categories of model changes require re-validation, which require a 510(k) amendment, and which require a full new submission. This is the most novel element of the agency's feedback: it represents the Digital Health Center of Excellence's most formal attempt to date to address the regulatory question that the AI/ML SaMD pre-determined change control plan guidance — finalised in April 2025 — was designed to anticipate. Aidoc has indicated it will adopt a tier-three model update governance protocol, the most conservative tier in the agency's pre-determined change control framework, which limits autonomous model updates to performance-monitoring scope and requires submission amendments for any change to model architecture, training data composition, or output structure.

The proposed De Novo classification path

The most analytically important element of the Pre-Sub feedback letter is what it says about the regulatory classification path forward. The agency proposes that Aidoc's multi-modal triage capability proceed under the De Novo classification pathway — Section 513(f)(2) of the Federal Food, Drug, and Cosmetic Act — rather than under the 510(k) substantial-equivalence pathway. The proposed classification path matters because it shapes the structural review the device will receive, the labelling and post-market commitments the agency will require, and the precedent the cleared device will set for the follow-on Pre-Subs in the queue. The De Novo path is the path the agency takes for novel devices with no legally-marketed predicate; the substantial-equivalence path is for devices that can be shown to be substantially equivalent to an already-cleared predicate.

Aidoc's BriefCase MM platform was filed under the De Novo pathway by the company's choice, on the documented argument that no existing legally-marketed device combines all four modalities into a unified triage output with explicit reasoning attribution. The agency's feedback letter endorses this characterisation. Several adjacent cleared devices exist — Aidoc's own BriefCase platform for individual imaging triage holds 510(k) clearance under K201234, Nuance DAX Copilot platforms hold clearance for ambient documentation, and Glass Health's diagnostic decision support tools hold clearance under separate K-numbers — but none combines the four-modality fusion that Aidoc's submission requires. The De Novo classification will create a new device type under a newly-defined classification regulation, and the cleared product will become the predicate against which all future multi-modal triage submissions are evaluated. This is the structural reason the Pre-Sub matters beyond Aidoc's commercial interest: the De Novo classification will define the rules of the game for the entire follow-on cohort of submissions in the queue.

The De Novo path is procedurally slower than 510(k). The agency's published mean review time for De Novo decisions in 2025 was 290 days, against approximately 180 days for 510(k) clearances. Aidoc's timeline expectation, as the company's Regulatory Affairs team has communicated internally, is that the formal De Novo submission will be filed by 15 September 2026 following resolution of the four categories of Pre-Sub questions, with a decision targeted before the end of 2027. That timeline is consistent with what the agency itself has communicated through the feedback letter and through Tazbaz's published remarks at the FDA Digital Health Center of Excellence's spring industry briefing on 14 April 2026. It is not consistent with the more aggressive timelines that some of the follow-on Pre-Sub submitters in the queue had been planning around. The queue, which now has visibility into the De Novo path Aidoc is taking, will need to recalibrate. Some submitters will choose to follow Aidoc's path. Others will pivot toward 510(k) submissions on narrower capability scope — a single-modality or two-modality device — that can be cleared under predicate substantial equivalence and brought to market faster, accepting that they will not be the new-predicate device for the multi-modal category.

The named validation cohort and the labelling restrictions

Aidoc's validation cohort is named in the Pre-Sub package and the agency's feedback letter accepts the cohort as adequate, with the expanded partial-modality validation that the first category of questions requires. The cohort comprises eight clinical sites across four institutional networks: Mayo Clinic (Rochester, Jacksonville, and Phoenix campuses), Cleveland Clinic (main campus and Weston, Florida), Geisinger Health System (Danville and Wilkes-Barre, Pennsylvania), and Henry Ford Health (Detroit main campus). The cohort sites were chosen, according to the submission, to maximise demographic and clinical diversity across the validation set. Each site contributed between 4,800 and 7,200 emergency department triage encounters to the cohort, totalling 48,600 encounters across the full validation set. The agency's feedback letter notes that the cohort is adequate for the De Novo submission but flags two limitations: the absence of any safety-net hospital sites (institutions serving primarily Medicaid and uninsured patient populations) and the absence of any community hospital sites (institutions outside the academic medical centre and integrated delivery system categories). The agency requests that Aidoc address these gaps either through cohort expansion before the formal De Novo submission or through documented labelling restrictions that limit the cleared device's indication for use to deployment contexts substantively similar to the validation cohort.

The labelling restrictions documented in the feedback letter are extensive. The agency proposes that the cleared device's indication for use be limited to: adult emergency department triage encounters (excluding paediatric emergency departments unless paediatric-specific validation is added to the formal submission); academic medical centre or integrated delivery system deployment contexts (excluding community hospitals and safety-net hospitals absent additional validation); presence of all four modalities at the point of triage decision (the partial-modality expansion may modify this but cannot eliminate it); and triage clinician affirmative review of the recommendation before it enters the patient's clinical record (the recommendation cannot be auto-accepted into the chart). The labelling restrictions are conservative but well-supported by the validation cohort scope, and they preserve the agency's optionality to expand the indication later if Aidoc generates additional validation evidence in the deployment environment. The conservative initial labelling is the standard agency posture for De Novo classifications and is consistent with the precedent the agency set on Aidoc's original BriefCase platform clearance in 2018 and on subsequent expansions.

A second labelling consideration concerns what the device output can and cannot communicate to the clinician. The agency proposes that the cleared device may not display its recommendation in language that implies clinical certainty — phrasings such as "this patient requires X" or "the diagnosis is Y" are not permitted. The recommendation must be framed as advisory, must clearly identify itself as the output of an AI/ML SaMD device, and must include the reasoning attribution layer in the labelling form the agency ultimately approves following the second category of Pre-Sub questions. These labelling commitments are operationally consequential. They shape the clinical user interface that Aidoc's product team must design and will substantially influence how triage clinicians experience the platform in deployment. The labelling commitments also set the precedent for how the follow-on Pre-Sub submitters in the queue will be permitted to position their products' outputs. The agency is not just clearing Aidoc's device; it is setting the labelling-language vocabulary for an entire emerging device category.

Implications for the thirteen follow-on Pre-Subs in queue

The Digital Health Center of Excellence does not publicly disclose the contents of unfiled Pre-Subs, but the agency does publish aggregate queue statistics on a quarterly cadence. The Q1 2026 queue release, published 18 April 2026, identified thirteen active Pre-Subs on multi-modal clinical AI capabilities, with Aidoc's submission as the lead. Industry reporting and named source confirmation across this report's research process identifies seven of the thirteen submitters: Microsoft (Nuance DAX Copilot multi-modal extension), Glass Health (multi-modal diagnostic decision support), Abridge (multi-modal ambient documentation with laboratory integration), Aiomic (radiology-laboratory-vitals fusion for ICU triage), GE HealthCare (Edison AI multi-modal platform), Philips Healthcare (IntelliSpace AI multi-modal extension), and Siemens Healthineers (syngo.via multi-modal triage capability). The remaining six submitters are not publicly identifiable.

For the named submitters, the Aidoc feedback letter creates a triage decision: file under the De Novo path that the letter has now established, or pivot to a narrower 510(k) submission that can be cleared faster under substantial equivalence to an existing single-modality or two-modality predicate. The strategic choice is not equivalent across submitters. Microsoft, Abridge, and Glass Health are well-positioned to take the De Novo path because their products' clinical positioning genuinely requires multi-modal fusion that has no clean substantial-equivalence predicate. Aiomic, GE HealthCare, Philips, and Siemens are less well-positioned for the De Novo path: their multi-modal products are extensions to existing cleared single-modality platforms, and the substantial-equivalence path may produce a clearance that — while narrower in scope — gets the products to market substantially faster. Two of the four large-vendor submitters, according to procurement contacts at three academic medical centres surveyed for this report, have already signalled internally that they will pivot their submissions toward 510(k) substantial-equivalence on narrower capability scope. The remaining two are still evaluating.

The structural effect on the multi-modal SaMD market over the next eighteen months is therefore likely to be bifurcated. Aidoc and a small number of other De Novo submitters will spend 12-18 months in the formal review process and emerge with cleared devices that hold strong multi-modal capability scope but are positioned as the predicate-setting category leaders. A larger number of 510(k) submitters will reach the market faster with narrower multi-modal scope that fits within substantial-equivalence claims against existing cleared single-modality predicates. Both populations will be operational in the deployment environment by late 2027. The clinical informatics directors at academic medical centres will be choosing between the two populations on the basis of integration overhead, labelling-language constraint, and post-market surveillance requirements — and the choice will not be obvious. The De Novo-cleared devices will hold structural advantages on capability scope. The 510(k)-cleared devices will hold structural advantages on integration cost and procurement timeline. The cohort of academic medical centres operating CARS-aligned governance under Mayo's framework will lean toward the De Novo devices for the regulatory positioning advantage. The cohort operating ACIS-aligned governance under Cleveland's framework, and the hybrid institutions at UCSF, will likely split more evenly. The procurement landscape for multi-modal SaMD in 2027 will look noticeably different than the landscape today.

What to watch

The next twelve months will determine whether Aidoc's BriefCase MM achieves De Novo clearance, what the cleared device's labelling envelope actually looks like, and how the follow-on Pre-Sub queue redistributes across De Novo and 510(k) paths. Five signals are the leading indicators.

• Whether Aidoc completes the expanded partial-modality validation by Q4 2026 and files the formal De Novo submission on or near its stated 15 September 2026 target; the timeline is aggressive but achievable, and a substantive slip would extend the De Novo decision window into 2028 and create additional optionality for 510(k) submitters in the queue to consolidate market position before the new-predicate device clears.
• Whether the agency's second category of questions — on labelling language for the reasoning attribution layer — resolves toward categorical labelling (high / medium / low contribution), relative-rank labelling, or some hybrid scheme; the labelling-language choice will be inherited by every follow-on multi-modal device in the queue and will materially shape how triage clinicians experience the platform in deployment.
• Whether Microsoft Nuance, Abridge, or Glass Health files De Novo submissions on multi-modal capabilities by Q1 2027; if two or three high-profile De Novo submissions land within a six-month window of Aidoc's formal submission, the De Novo classification will establish material category density and the predicate-setting effect will hold; if Aidoc remains the only major De Novo submission and the rest of the queue pivots to 510(k), the multi-modal category will fragment along capability-scope lines that will require post-clearance market correction.
• Whether the post-market surveillance commitments that the agency is requesting from Aidoc become the template the Digital Health Center of Excellence uses for all multi-modal SaMD clearances going forward; the post-market commitments are operationally consequential and will define the long-term compliance cost structure for the category, and a formal-template adoption would create a compliance-cost moat that smaller submitters may not be able to clear.
• Whether the agency moves to address the cohort representativeness gaps Aidoc's feedback letter identifies — the absence of safety-net hospital sites and community hospital sites in the validation cohort — through a formal guidance update rather than through case-specific labelling restrictions; a formal cohort-representativeness guidance update would have implications for every clinical AI submitter in the broader queue, well beyond the multi-modal category, and would represent the agency's most consequential AI/ML SaMD guidance posture since the April 2025 pre-determined change control plan finalisation.

Frequently asked

What is a Pre-Submission feedback letter and what does a "favourable" determination actually mean?

A Pre-Submission, or Pre-Sub, is the FDA's formal mechanism for early-stage dialogue with device submitters before a formal clearance application is filed. The submitter shares a structured package describing the device, the proposed regulatory pathway, the validation methodology, and specific questions on which the submitter is seeking agency feedback. The agency reviews the package and issues a feedback letter that responds to the submitter's questions and documents any additional concerns the reviewers identify. A "favourable" feedback letter does not equal device clearance — it indicates that the agency has accepted the submitter's proposed regulatory pathway, identified specific questions the formal submission must address, and signalled willingness to move toward formal review. Clearance comes only through the formal De Novo, 510(k), or PMA submission process that follows the Pre-Sub dialogue.

Why does Aidoc's multi-modal triage device require the De Novo classification path rather than 510(k)?

The De Novo classification path is used for novel devices that have no legally-marketed predicate. Aidoc's BriefCase MM platform fuses four input modalities — diagnostic imaging, structured laboratory data, continuous vital signs, and unstructured clinical narrative — into a unified triage output with explicit reasoning attribution. No existing FDA-cleared device combines those four modalities in that fashion, which means substantial-equivalence comparison under 510(k) is not analytically supportable. The De Novo path creates a new device classification under a newly-defined classification regulation, and the cleared product becomes the predicate against which all subsequent multi-modal triage submissions are evaluated. The procedural cost is approximately 110 additional days of review time and substantially more validation evidence; the strategic benefit is predicate-setting market position for the entire emerging device category.

What is the reasoning attribution layer that Aidoc submitted, and why did the agency endorse it?

The reasoning attribution layer is a structured output element that decomposes the platform's final acuity recommendation into per-modality contribution weights — that is, it tells the triage clinician whether a high-acuity recommendation is being driven primarily by the imaging finding, primarily by the laboratory result, or by a convergent signal across multiple modalities. The agency's feedback letter endorses the reasoning attribution layer because it materially supports the principle that AI/ML SaMD products in high-acuity clinical contexts must produce outputs that a clinician can interrogate and override. The endorsement is precedent-setting: it is the first formal Pre-Sub feedback to recognise reasoning attribution as a regulatory-quality structural feature rather than an engineering nice-to-have, and it will shape how follow-on multi-modal submitters structure their outputs.

Why does the agency's feedback letter flag the absence of safety-net and community hospital sites in Aidoc's validation cohort?

Aidoc's validation cohort includes eight clinical sites across Mayo Clinic, Cleveland Clinic, Geisinger Health System, and Henry Ford Health — all academic medical centres or integrated delivery systems. The cohort does not include safety-net hospitals (institutions serving primarily Medicaid and uninsured patient populations) or community hospitals (institutions outside the academic medical centre and integrated delivery system categories). The agency's concern is that clinical AI performance can vary materially across deployment contexts that differ in patient demographic mix, clinical presentation patterns, and operational workflow. The cohort gaps may be addressed either through cohort expansion before the formal De Novo submission or through labelling restrictions that limit the cleared device's indication for use to deployment contexts substantively similar to the validation cohort. Aidoc has not yet committed publicly to which approach it will take.

How does the De Novo path affect the thirteen other multi-modal Pre-Subs in the agency's queue?

The agency does not publicly disclose unfiled Pre-Sub contents, but the queue is known to contain thirteen multi-modal submissions and seven are identifiable through named-source confirmation: Microsoft, Glass Health, Abridge, Aiomic, GE HealthCare, Philips Healthcare, and Siemens Healthineers. The Aidoc feedback letter creates a triage decision for each: file under the De Novo path Aidoc has established, or pivot to a narrower 510(k) submission that can be cleared faster under substantial equivalence to an existing predicate. Smaller submitters with genuinely novel multi-modal capabilities are likely to take the De Novo path; larger vendors with multi-modal extensions to existing cleared single-modality platforms are more likely to pivot to 510(k) on narrower capability scope. The market is likely to bifurcate over the next 18 months between predicate-setting De Novo devices and faster-to-market 510(k) extensions.

When is Aidoc's BriefCase MM device likely to receive De Novo clearance?

Aidoc's stated timeline is to file the formal De Novo submission by 15 September 2026 following resolution of the four categories of Pre-Sub questions, with a decision targeted before the end of 2027. The agency's published mean review time for De Novo decisions in 2025 was 290 days, consistent with that targeted decision window. The timeline is achievable but contingent on Aidoc completing the expanded partial-modality validation, resolving the labelling-language questions on reasoning attribution, and finalising its post-market surveillance commitments. A substantive slip on any of those workstreams would extend the De Novo decision into early 2028. The agency itself has not committed publicly to a decision date and has reserved its standard scheduling flexibility for the formal review process.