Veracyte

Machine learning built every product here. Each classifier was derived by training over RNA whole transcriptome data to find the small gene set that predicts an outcome, and the company describes that method consistently across prostate, bladder, thyroid, lung and breast.

Held at B because the assay is co equal with the model, as it is for the other laboratory records in this index. Once trained, each test is a fixed panel, 22 genes for prostate and 219 for bladder, so what a customer buys is a measurement and a locked scoring function rather than a system that continues to learn. The intelligence was spent at development time and is now embedded in a laboratory process.

Low autonomy by design. The classifier returns a risk score; a urologist or oncologist decides what to do with it, in conversation with the patient.

The consequence of that score is nonetheless substantial, because the decision it feeds is whether to treat more or less aggressively, and both errors are serious in prostate cancer: undertreating a dangerous tumour, or subjecting a man to hormone therapy and its effects when he would have done well without it. The guidelines that recommend the test also define how it should be used, which is a stronger external constraint on interpretation than most products here have.

The mechanism is stated with enough precision to evaluate. The method is named as RNA whole transcriptome analysis with machine learning, the gene counts are published, 22 for prostate and 219 for bladder, the sample types are specified, and what the score means is described in clinical rather than statistical terms.

Publishing the panel size matters, because it tells a reader the classifier is a fixed, auditable object rather than an evolving black box, and it makes independent replication possible in a way an unnamed model does not.

What is not published is the training cohort composition, the algorithm class, or the score thresholds and how they were chosen. Those sit in the literature rather than in company material, which is a reasonable place for them and does mean a buyer must go and read.

Nothing identifies any party in the chain: no hosting arrangement, no sub processor list, and no retention schedule, encryption detail or consent position was located in this pass. The stewardship question specific to this company is the research database built alongside the clinical business, holding genomic profiles from very large numbers of patients and made available to external researchers.

That is a scientifically valuable asset and it was assembled from people who came for a diagnostic test, not for a research programme, so what they were told about secondary use is the question rather than whether the arrangement is lawful.

This index recorded the same issue on the liquid biopsy records, and the same practical answer applies: for a laboratory the governing document is the requisition and the consent form, not a vendor agreement, so an ordering institution asking its vendor for terms is asking the wrong party for the wrong document.

The content compounds it in the way genomic material always does, since a sequence is durably identifying, cannot be revoked once disclosed, and carries implications for biological relatives who were never asked. Ask what the requisition and consent form actually say about research use, who may access the research database and under what approval, whether a patient can decline while still receiving the test, and what happens to profiles if the business changes hands.

The strongest guideline standing of any record in this index, and it refines the bar this index has been using.

Guideline inclusion alone already earns an A here, on the basis that a body with no commercial interest adopting a test outranks any volume of vendor publication. This goes two steps further. The national oncology guidelines assign Decipher Prostate a graded evidence level, Level I, and identify it as the only gene expression test to reach it. And they do not merely list the test: they recommend using its score to decide whether to add hormone therapy after prostatectomy, which means the guideline specifies the clinical action the result should drive.

Behind that sits a Phase 3 randomised trial with a median of twelve years of follow up showing that benefit from hormone therapy tracked the classifier's risk groups, more than 85 studies covering over 200,000 patients, linkage of results to national cancer registry outcomes, and seven further Phase 3 randomised trials sponsored by the National Cancer Institute currently underway.

The distinction worth carrying to other records is that these are three different things and only the last is present here. Guideline inclusion outranks vendor studies. A graded evidence level within the guideline outranks bare inclusion. And a guideline that names the treatment decision the score should drive outranks both.

Graded on an honest basis. No retention schedule, encryption detail or consent position was located in this pass.

The stewardship question specific to this company is the research database it has built alongside the clinical business, holding genomic profiles from very large numbers of patients and made available to external researchers. That is a scientifically valuable asset assembled from people who came for a diagnostic test, and what they were told about secondary use is the question. It is the same issue this index recorded on the liquid biopsy records, and the same answer applies: for a laboratory it lives in the requisition and the consent rather than in a vendor agreement.

Graded on an honest basis. No compliance documentation was located in this pass.

As with the other laboratory records here, the business associate frame does not fit well. A laboratory receiving a specimen on a clinician's order and reporting a result is generally a covered entity in its own right rather than a business associate of the ordering practice, so the instrument this axis usually turns on is not the governing one.

Recorded honestly and provisionally: the dedicated trust and security search this index requires was not run in this pass, and no attestation was encountered incidentally.

As with the other listed companies in this index, the authoritative source is the annual securities filing, which carries mandatory cybersecurity risk management and governance disclosure. It was not checked. Read it before quoting this grade.

Split across the portfolio, which is worth stating precisely rather than averaging.

The flagship classifiers are laboratory developed tests offered under clinical laboratory certification, validated by the laboratory and not reviewed by the Food and Drug Administration. Their standing comes from the guidelines and the literature instead, which in this case is substantial. Elsewhere in the portfolio the breast cancer assay is a regulated in vitro diagnostic that has been through review and is available internationally.

So a reader comparing this company with the artificial intelligence prostate tests indexed beside it should understand that the two took different routes to the same guideline. One holds a De Novo marketing authorisation for a locked algorithm reading tissue images; this one holds Level I guideline evidence for a laboratory developed classifier reading RNA. Both are recommended for overlapping decisions in the same disease. The status of federal oversight of laboratory developed tests has been contested, so a buyer should confirm the current position rather than assume it.

No performance breakdown by ancestry, age or comorbidity was located, and in this disease that omission is pointed rather than routine.

Prostate cancer shows one of the widest documented outcome disparities in oncology, with substantially higher mortality among Black men, and the biology of that difference is actively debated. A 22 gene expression classifier trained on cohorts that may not reflect that population, then used to decide who receives less intensive treatment, is precisely where a performance gap would do harm, because the error that follows a falsely low score is treatment withheld.

The company has the means to answer this better than almost anyone. Its results have been linked to national cancer registry outcomes data, which is the mechanism that would let ancestry stratified performance be measured at scale, and no such analysis was located in company material. The literature may contain it; this pass did not retrieve it, and it should be sought before the grade is quoted.

Publishing the panel size is doing more work than it appears to, and it is the reason this sits above the floor. The method is named as whole transcriptome analysis with machine learning, the gene counts are published for each assay, the sample types are specified, and what the score means is described in clinical rather than statistical terms.

A published gene count tells a reader the classifier is a fixed, auditable object rather than an evolving black box: a fixed panel can be replicated, can be compared against another laboratory's result on the same specimen, and cannot quietly acquire new inputs between one patient and the next. That is a structural guarantee of a kind most vendors in this index cannot offer. Held at C because the material a clinician would need to interrogate a result sits elsewhere.

The training cohort composition, the algorithm class, and the score thresholds together with how they were chosen are not in company material, and thresholds matter most, because a threshold is where a continuous score becomes a treatment decision and where a small shift moves patients between management pathways. Those details are in the literature, which is a reasonable place for them and does mean a buyer must go and read rather than being handed them. No warranty, indemnity or remediation commitment was located. Ask for the thresholds and their derivation, and performance in patients unlike the development cohort.

Not described, and largely peripheral to how the product works. A specimen is sent, a report comes back to the ordering clinician, and the integration question is whether that report arrives in the chart as a discrete result or as a document.

For a test whose score is meant to drive a specific treatment decision recommended in guidelines, discrete capture matters more than it would for a narrative report, because a score buried in a scanned document cannot trigger anything downstream. No interface standard or record integration was located.

A genuine architectural choice rather than a default. Alongside its own certified laboratories, the company holds exclusive global access to a specific analysis instrument, which allows hospitals and laboratories elsewhere to run its tests locally rather than shipping specimens to California.

That matters for reach and for data: a test performed in a local laboratory keeps the specimen and often the data within that jurisdiction, which is the practical answer to residency questions that central testing cannot give. Held at B because the split between centrally run and locally run testing is not quantified and no hosting or retention detail is published.

No list price is published, and listing produces more usable information than a private laboratory offers. Segment revenue, test volumes and market penetration are all disclosed, so an outside reader can derive an approximate revenue per test and watch it move.

The more important commercial fact for a patient is coverage rather than price. These are reimbursed tests, and this index has repeatedly found that reimbursed diagnostics are the most price transparent segment because a published rate exists. A buyer or clinician should establish the coverage position for the specific indication being ordered, since coverage differs between the biopsy and post surgery settings and between the tests in the portfolio.

One method applied across many cancers, which is the company's actual proposition. Commercial tests cover prostate, bladder, thyroid, breast and lung, plus interstitial lung disease, with lymphoma and renal classifiers in development and a tumour informed test for minimal residual disease in the pipeline.

The setting is the decision point after a diagnosis is made but before treatment intensity is chosen, which is a narrow moment and a consequential one. Reach is national and international, supported by the ability to run tests in local laboratories. Held at B because everything sits within diagnostics: nothing addresses treatment delivery, monitoring or the wider care pathway, and the depth is in prostate with the other indications materially smaller.