DoMore Diagnostics

A single deep learning model is the entire company. Histotype Px Colorectal reads standard H and E stained slides and returns a risk group; there is no assay, sequencing step, viewer, or laboratory of its own. The company's positioning makes the centrality explicit: the entire value proposition is extracting prognostic signal from slides already produced, in a field where the alternative biomarkers require gene sequencing that decades of investment have failed to deliver adequately for this indication.

Produces a low, intermediate, or high risk stratification that informs an oncologist's adjuvant chemotherapy decision, so the human retains the treatment call. The three tier output rather than a binary flag is a meaningful design choice, since it surfaces genuine uncertainty in the middle group rather than forcing every patient to one side of a threshold.

The consequential direction here is de-escalation, identifying patients who may safely avoid chemotherapy, which means a false low risk classification carries a different and more serious cost than a false high risk one, and the vendor materials do not quantify that asymmetry.

Training scale is disclosed specifically at close to 100 million image tiles, the provenance is named as research led by the Institute for Cancer Genetics and Informatics at Oslo University Hospital, and the company is careful to distinguish the commercial product from its research predecessor, noting that the early version was called DoMore-V1-CRC during development and that the CE-IVD certified product is an updated version. Naming that distinction rather than claiming the research paper validated the shipping product is a real transparency point. Model architecture and per subgroup performance are not published.

No stewardship framework or governance terms were located, and no model, hosting arrangement or sub processor list was named. The lane wide point applies and takes a sharper form here because of the delivery route. Whole slide image files produced by scanners routinely embed a photograph of the slide label alongside the tissue data, and that label commonly carries the accession number and often the patient's name, so slides are identified records by default and treating them as anonymous tissue is unsafe unless the embedded layers have been deliberately stripped.

Where an algorithm is delivered through a third party platform, it should be explicit which party performs that stripping and at what point in the pipeline, because each can reasonably assume the other did it and neither is contractually obliged to a customer who never asked. A two party pipeline with an unassigned de identification step is worse than a single vendor with none, since a single vendor at least knows the answer.

Nothing published states what the vendor does with images it receives, whether customer slides contribute to model development, what retention applies, or what happens at contract end. One further question follows from the output type, since a model update could change how previously reported cases would have been classified, and a laboratory needs to know when that happens. Ask who strips the embedded layers and when, the retention position, and how model updates are notified.

The strongest evidence base of any pathology vendor added in this sweep. An early version of the algorithm was published in The Lancet, and a landmark study in The Lancet Oncology by Kleppe and colleagues demonstrated that the marker significantly improves prognostic accuracy when combined with TNM staging, validated within the QUASAR II trial with a reported hazard ratio of 9.47 comparing high versus low risk groups.

Validation inside a randomised controlled trial cohort is a materially higher standard than a retrospective institutional series, because trial cohorts have adjudicated outcomes and controlled treatment. Independent survey work on public evidence for digital pathology AI products lists this among the products where publication preceded regulatory approval, which is the right order.

Converted from Not Rated. No stewardship framework or data governance terms were located.

Two things specific to this product shape what to ask.

The first applies across this lane. Whole slide image files produced by scanners routinely embed a photograph of the slide label alongside the tissue data, and that label commonly carries the accession number and often the patient's name. Slides are therefore identified records by default, and treating them as anonymous tissue is unsafe unless the embedded layers have been deliberately stripped. Where an algorithm is delivered through a third party platform, it should be explicit which party performs that stripping and at what point in the pipeline, because each can reasonably assume the other did it.

The second concerns what the output is. This is not a detection or classification aid that a pathologist verifies against the slide in front of them. It is a prognostic risk assignment that informs whether a patient receives adjuvant chemotherapy after surgery, and its stated value is reducing overtreatment. That means its characteristic error is under treatment: a patient placed in a low risk group who was not low risk may have chemotherapy withheld in a curable cancer, and unlike a missed feature on a slide, that error is not checkable against the image. The clinical benefit and the clinical risk run through the same mechanism.

Ask who de identifies images and when, whether the vendor receives customer slides for support or model improvement, what retention applies to any images and derived risk scores, and how a laboratory is notified if a model update changes how previously reported cases would have been classified.

Converted from Not Rated. This is a scoping determination and it closes cleanly on the company's own stated market position.

The product carries a European in vitro diagnostic mark for clinical use in Europe, and the company states it is available for research use only outside the European Union. There is therefore no United States clinical deployment to which the health privacy rule would attach in the ordinary way, and the absence of a business associate posture is a correct consequence of where the product is sold rather than a gap in disclosure.

What governs instead should be understood rather than skipped. In Europe the operative framework is the general data protection regulation, under which health data is a special category requiring an explicit lawful basis, and the company is based in a country inside the European Economic Area, so the regime applies to it directly. Where the algorithm is delivered inside another vendor's platform, the controller and processor roles need to be assigned across three parties rather than two, and that allocation is not published.

Two situations would change the analysis and a buyer should watch for both. Research use of the product in the United States on identifiable specimens engages the research framework and, at a covered entity, an authorisation or waiver. And any move to seek United States clearance would bring the ordinary business associate questions with it.

Ask who is controller and who is processor across the algorithm supplier and the hosting platform, what the data processing terms say, and what changes if the product enters clinical use in a new jurisdiction.

Converted from Not Rated. No SOC 2, ISO 27001 or equivalent information security attestation and no trust centre were located.

The distribution model changes what this absence means, and the analysis is worth stating carefully rather than treating as a simple gap. Because the algorithm is delivered inside enterprise pathology platforms operated by other companies, the security a laboratory actually relies on day to day is the platform's. The laboratory contracts with the platform vendor, that vendor holds the images, and its certifications and controls are the ones governing the environment where analysis happens. A laboratory assessing this algorithm should therefore assess the host platform's security posture as the primary control, which is a different diligence exercise from assessing a standalone vendor.

That does not make the algorithm supplier irrelevant. Three points of contact remain. Model development and validation require access to real slides from real laboratories. Support and troubleshooting require someone examining the case that failed. And model updates are delivered by the supplier into a regulated diagnostic pathway, which makes the integrity of that delivery channel a security question in its own right.

One credential is implied but not equivalent. A European in vitro diagnostic mark requires a quality management system, which addresses design control, traceability and post market surveillance. It is not an information security assessment, the same distinction this index applies throughout this lane.

Ask what security examination the supplier itself has undergone, how model updates are authenticated and validated before reaching a live diagnostic environment, and what access the supplier retains to customer images.

CE-IVD marked and reported as the first CE marked product using AI to predict patient outcome from image analysis, which is a category first rather than a routine certification. No FDA clearance was located. One detail a buyer should note: the company's own product page states the mark is CE-IVDD, the older directive, rather than IVDR, the same distinction flagged against Visiopharm's DeepBio prostate APP and Owkin Dx, and IVDD carries a lower evidence bar than IVDR. Graded C on the established basis of real European authorization without US clearance, with the caveat that the underlying clinical evidence here is unusually strong for a C on this axis.

Validation within a randomised trial cohort provides controlled conditions and adjudicated outcomes, and publication preceding regulatory approval indicates evidence was generated before commercial claims. No formal governance framework, monitoring commitment, or demographic subgroup analysis was located. The gap matters for a de-escalation tool, since a risk model that systematically underestimates in any patient group would direct those patients away from chemotherapy they would have benefited from.

One disclosure habit here is the distinction this index has had to draw against several other vendors, and this company draws it itself. Training scale is disclosed at close to a hundred million image tiles with provenance named to a research institute at a university hospital, and the company is careful to separate the commercial product from its research predecessor, noting that the early version carried a different name during development and that the certified product is an updated version.

Naming that distinction rather than letting the research paper stand as validation of the shipping product is a real transparency point, because a published method is not a published product and most vendors are content for a reader to conflate the two. What holds it below the top grade is the nature of the output and what cannot be checked.

This is not a detection aid a pathologist verifies against the slide in front of them; it is a prognostic risk assignment that informs whether a patient receives adjuvant chemotherapy after surgery, and its stated value is reducing overtreatment.

Its characteristic error is therefore under treatment: a patient placed in a low risk group who was not low risk may have chemotherapy withheld in a curable cancer, and unlike a missed feature on a slide that error is not checkable against the image. The clinical benefit and the clinical risk run through the same mechanism. Architecture and per subgroup performance are unpublished and no commitment attaches. Ask for the low risk group's observed recurrence rate in independent follow up.

Designed to sit inside existing hospital pathology workflow rather than requiring a dedicated environment, with the company stating the product adds value across the various digital pathology platforms available rather than tying to one. Third party product profiling describes integrations or partnerships with digital pathology platforms including PathAI, Sectra, Proscia, and Paige, three of which are separately indexed here. That multi-platform posture mirrors Primaa and Mindpeak and is the pragmatic route for an algorithm-only vendor. No published connector list or scanner compatibility matrix was located.

Converted from Not Rated. The prior note had the shape right: where the algorithm runs through a partner digital pathology platform, residency follows that platform rather than this vendor. That is now confirmed and can be stated as the deployment model rather than inferred.

This company supplies an algorithm rather than operating a platform. Its product is integrated into enterprise digital pathology systems built by others, with one such partnership announced in detail and further platform integrations reported. Those platforms are what laboratories actually deploy, contract for and run, and the hosting, tenancy, residency and retention terms a laboratory relies on are theirs. For an algorithm supplier that is a coherent and sensible model, and it is the right answer for a small company that would otherwise be asking hospitals to adopt separate infrastructure for one test.

One question decides everything else and it is not answered publicly. Where an algorithm is described as integrated into a platform, two quite different architectures are possible: the algorithm executes inside the platform's environment, so images never reach the algorithm supplier, or the platform transmits images to the supplier for analysis and receives a result. Both are routinely described using the same word. The first leaves the supplier outside the data path entirely; the second makes it a processor of identifiable images.

Ask which of the two applies for the specific platform under consideration, whether any image or derived data reaches the algorithm supplier in normal operation, and what the arrangement is during validation, support and model updates, which often differs from steady state.

No rates are published, but the economic argument is quantified more precisely than almost any vendor in this index, and in the buyer's own terms. The company states that most patients in this indication are cured by surgery alone and that adjuvant chemotherapy provides no benefit to 96 to 98 percent of stage II and 80 percent of stage III patients, meaning the value case is avoided chemotherapy cycles and avoided toxicity, both of which a payer or health system can cost directly.

Turnaround is also framed against the comparator, delivering risk information immediately after surgery rather than the weeks to months current markers require. Third party listings describe enterprise diagnostic, oncology, and platform-partner pricing available on request.

A single indication precisely specified: outcome prediction in stage II and III colorectal adenocarcinoma to inform adjuvant chemotherapy decisions following surgical resection. That focus is the source of its evidence depth, since the QUASAR II validation is specific to exactly this decision.

It does not extend to other tumour types, to metastatic disease, or to diagnostic as opposed to prognostic tasks, and the Histotype Px naming suggests intended expansion to further indications not yet certified.