Deep 6 AI

Natural language processing over unstructured clinical data is the whole product and the company quantifies why it matters: it states 92 percent of trial inclusion and exclusion criteria benefit from unstructured data, and that 15 to 20 percent of eligible patients are found through unstructured data alone. Mining physician notes, pathology reports, genomics results, and labs is described as accessing up to 80 percent more patient data than systems relying on coded fields. Without the language models the platform would be a structured query tool that misses most eligible patients.

The system identifies and surfaces candidates with proof of matching, and human research staff validate and recruit. The company describes a distinct validation step between identification and recruitment, and separately enables treating physicians to refer patients to studies they are eligible for based on targeted study information and matching evidence.

Providing the matching rationale rather than a bare score is what makes both the coordinator validation and the physician referral decision meaningful. No published detail describes confidence thresholds or how borderline matches are handled.

The approach is described concretely at the data level, covering AI and NLP applied to structured and unstructured EMR data in real time across notes, pathology, genomics, and labs, with quantified claims about how much additional patient data that unlocks. What is not published is model architecture, matching accuracy, or precision and recall figures.

This is a category wide gap: peer reviewed literature has specifically noted that companies in trial matching, naming this one among others, have not published performance data outside small studies or restricted cohort abstracts, which is a fair criticism to record.

Nothing identifies any party in the chain: no model or model family, no hosting arrangement and no sub processor list was located in two passes, and no retention or training terms were found beyond a stated commitment that the platform operates within a secure compliant collaborative environment. The word doing the most work there is collaborative, and it is the substance of this record.

The ecosystem model deliberately connects providers, sponsors and contract research organisations, so the product's value proposition is that parties who would not otherwise share data can act on the same picture of a patient population.

Each of those parties has a different relationship to the patient and a different interest in the data: a provider holds it under a treatment relationship, a sponsor wants to find and enrol, and a research organisation operates under contract to the sponsor. Where patient level data flows between them, at what identifiability, and under whose authority is therefore the central question rather than a peripheral one, and nothing published maps it.

The input compounds it, since the platform reads full unstructured records including notes, pathology and genomics rather than a structured extract. Ask what each party in the ecosystem can see at what identifiability, what authority governs each flow, what is retained after a match, and for a sub processor list.

Operational claims are specific but vendor reported, with sites stated to find more than 25 percent more patients than traditional recruitment methods and one referenced study citing recruitment acceleration up to tenfold alongside increased trial population diversity. Named partnerships with oncology development organizations indicate real commercial traction.

The significant caveat is external and worth recording plainly: published academic literature has observed that this company and its peers have not released performance data beyond small studies or restricted abstracts, so independent verification of matching accuracy does not exist publicly.

The company states the platform operates within a secure HIPAA compliant collaborative environment, which is a direct if brief commitment. The data surface is broad, spanning full unstructured records including notes, pathology, and genomics across health system populations, and the ecosystem model deliberately connects providers, sponsors, and CROs, so where patient level data flows between those parties is the question a buyer should map. No retention or training data terms were located.

HIPAA compliance is stated explicitly for the platform environment. No separate business associate agreement commitment was located, which is what would raise this, particularly given the multi party ecosystem structure where data visibility extends beyond the originating health system.

Converted from Not Rated. No attestation scoped to this platform and no trust centre was reached across two searches, one of which returned only unrelated companies' trust pages and generic framework explainers, which is a retrieval pattern this index treats as weak evidence rather than none.

The material change since this row was first written is ownership. This platform was acquired in March 2025 by a company listed on a United States exchange. That does two things to this axis and a buyer should hold them apart.

It opens a route. A domestic filer's annual report must carry a cybersecurity governance item describing processes for assessing and managing cybersecurity risk, board oversight and accountable management. That item is the source of graded disclosures elsewhere in this index and it exists here as a matter of law. It was not retrieved in this pass, so this row does not rely on it, and it should be read before anyone does.

It also raises a question the route does not answer. A parent's enterprise cybersecurity disclosure covers the consolidated registrant. It says nothing about whether a platform acquired part way through the period has been brought inside the parent's integrated security programme, which infrastructure it runs on, or whether any attestation names it. This index has repeatedly found assurance published at a parent level that on inspection described one constituent business.

Three things to establish. Which legal entity now contracts. Whether the platform still operates on its original infrastructure or has migrated to the parent's. And whether any attestation names this platform in its scope section, rather than naming the group.

No FDA pathway applies. Trial matching identifies candidates for research rather than informing diagnosis or treatment. The governing regime is human subjects research oversight, meaning IRB approval, informed consent, and Good Clinical Practice, with the physician referral pathway adding conventional clinical judgment on top.

Converted from Not Rated. No governance framework, model card or bias disclosure was located, and none now appears at the parent that acquired this platform in March 2025 either.

The domain relevant question here is more consequential than the axis usually implies, and it is worth stating precisely because it applies to every trial matching product in this category. The output of this system determines who is offered access to an investigational therapy. The system surfaces candidates by mining the medical record, so it inherits whatever the record contains: who has been seen, who has been coded, whose notes are detailed enough to satisfy an eligibility criterion. Patients with thin documentation are systematically less matchable regardless of their actual eligibility, and thin documentation correlates with the same populations that trials already under enrol. A matching system can therefore increase recruitment volume and reproduce under representation at the same time, and the volume figure will look like success.

Retained from the prior assessment because it identifies the right measurement. A referenced study reports increased diversity in trial populations through identification of underrepresented patient groups, which points in the right direction. It is a volume claim rather than a comparison. What would substantiate it is a published analysis setting the demographic profile of candidates surfaced by the system against that of candidates found by manual screening at the same sites, and no such analysis was located.

That comparison is the single thing that would move this row, and this vendor is unusually well placed to produce it, because the deployments it describes run alongside existing manual screening at academic centres and designated cancer centres.

Two passes located no matching accuracy, no precision or recall figures, no model architecture, no evaluation methodology and no warranty, indemnity or remediation commitment. The absence is documented by outside parties rather than only observed here, which makes it firmer than a retrieval gap: peer reviewed literature has specifically noted that companies in trial matching, naming this one among others, have not published performance data outside small studies or restricted cohort abstracts.

That is a fair criticism to record and it applies across the category rather than singling this vendor out unfairly. What is described is the data level approach, covering language processing applied to structured and unstructured record data in real time across notes, pathology, genomics and laboratory results, with quantified claims about how much additional patient data that unlocks. Volume unlocked is not accuracy achieved.

The failure direction that matters is invisible by construction and worth stating plainly. A patient wrongly matched wastes a screening visit and is caught at consent. A patient who should have matched and did not is never told a trial existed, never appears in any denominator, and generates no artefact anywhere, so a site running this for a year sees the enrolments it produced and never the ones it missed. Ask for recall against a manually reviewed cohort, the false negative rate by tumour type or condition, and what governs a match a clinician disagrees with.

Direct EMR integration is the delivery mechanism, with the platform analyzing structured and unstructured record data in real time rather than working from periodic extracts. Real time operation against live records is the harder engineering problem and is what enables the stated ability to surface matches in minutes. Deployment runs on major cloud infrastructure. No named EHR integrations, certifications, or standards support were enumerated publicly.

Described as a cloud based platform deployed on major cloud infrastructure. No tenancy, hosting detail, or data residency terms are published, which matters for an ecosystem model where multiple organizations access a shared environment.

No published pricing. The ecosystem spans health systems, treating physicians, sponsors, and CROs, and as with peers in this category the economics likely flow primarily from the sponsor side while providers host the patients, but nothing about structure or rates is disclosed.

Functional coverage is broad across the trial lifecycle, spanning cohort building and feasibility at study design, site and principal investigator selection, patient recruitment, on study data gathering, and real world evidence generation, with an integrated recruitment module beyond identification. Oncology is the evident depth given named partnerships and product positioning, with fibrosis also referenced. Buyers span health systems, treating physicians, sponsors, and CROs.