Myriad Genetics

Artificial intelligence is a feature layer on a laboratory testing business whose value stands without it. What Myriad sells is sequencing and expression testing run in its own laboratories: hereditary cancer panels, pharmacogenomics, prenatal and carrier screening, tumor profiling and residual disease monitoring.

Two learned components sit on top. RiskScore is a polygenic risk score, a weighted sum of breast cancer risk variants with weights set by each woman's genetic ancestry, combined with the Tyrer-Cuzick clinical model to estimate lifetime breast cancer risk; it is offered with MyRisk to women whose panel finds no pathogenic variant. Prolaris + AI adds PATHOMIQ's digital pathology model to the Prolaris expression score for men on active surveillance. Both change the answer a clinician receives, and both ride on an assay that is sold and reimbursed without them.

Ask which of your ordered tests return a model derived score and which return a laboratory result alone.

The score reaches an ordering clinician as part of a laboratory report, and the controls around it are not described. RiskScore returns a lifetime breast cancer risk estimate that informs screening and prevention decisions, and Prolaris + AI returns a risk of upgrading that informs whether a man stays on active surveillance. In both cases a clinician receives the result and decides, which is review before action.

What is not published is the mechanism: who signs out the model derived score in the laboratory, which results the system will not report, how a score is handled when the input falls outside the validated range, and how a corrected score reaches a patient already managed on the earlier one.

Ask which conditions suppress a RiskScore or Prolaris + AI result, and how corrected scores are communicated after a report is issued.

The approach and the supplier are named, without version and update practice. RiskScore's construction is published: 149 ancestry informative variants to estimate ancestry, a weighted sum of 93 breast cancer risk variants with ancestry specific weights, and combination with the Tyrer-Cuzick model. Prolaris + AI is described as Myriad's cell cycle progression expression score combined with clinical and pathological features and PATHOMIQ's independent AI powered digital pathology analysis, producing a Gleason upgrade risk score.

What is not published is version discipline: which RiskScore version a report reflects, how variant weights are updated as reference data grow, and how changes to PATHOMIQ's model reach the Prolaris + AI result.

Ask for the current version of each model, its change history, and how ordering clinicians are told when a score's method changes.

The chain is short and the outside party is named. RiskScore is built and run by Myriad on its own laboratory data, with no outside model provider. Prolaris + AI names its one outside model supplier, PATHOMIQ, whose digital pathology technology is integrated into the test.

What is not published is the terms between them: whether PATHOMIQ receives slide images or patient data, whether it retains or trains on them, and under what agreement. The privacy notice says outside service providers act as business associates without naming them, and no subprocessor list is published.

Ask whether PATHOMIQ receives identifiable images or data, under what retention and training terms, and for a list of the outside parties that process genetic data.

Published research on both models, short of independent validation. Myriad reports a retrospective study in Genetics in Medicine (2024) of more than 130,000 women tested with MyRisk between 2017 and 2019, linking test results to medical claims, in which RiskScore improved predictive accuracy roughly two fold over Tyrer-Cuzick alone and reclassified about one woman in five; better calibration and discrimination are reported without the figures on the release. The all ancestry version was developed on more than 189,000 women and validated on more than 89,000 (ASCO 2021 abstract).

Prolaris + AI was trained on 998 active surveillance candidates and validated on 296, all with Gleason 3+3 or 3+4 disease, and its Gleason upgrade score was a significant predictor of upgrading at the next biopsy. That work is a preprint (medRxiv, 2026). The studies are company led and none is a prospective trial of either score.

Ask for the full RiskScore calibration and discrimination figures, and for the Prolaris + AI validation once peer reviewed.

The training position is stated broadly and in Myriad's favor, and retention is left general. The privacy notice says Myriad may use de identified data to develop and improve its genetic tests, contribute to its variant classification program and conduct research alone or with others, and that once data is de identified it may be used or disclosed for any lawful purpose. It says identifiable information is retained as part of the medical record for as long as services and legal requirements need, without a schedule for samples or sequence data, and it names no de identification method.

A grant to develop products on patient data, however clearly worded, describes what Myriad may do with the data rather than how it is protected. Safety engineering for RiskScore or Prolaris + AI, meaning guardrails and how a faulty score would be detected and handled, is not published.

Ask how long samples and sequence data are kept, which de identification method applies, and whether your patients' data trains the risk models.

Covered entity status is stated and a substantive privacy document reaches the patient. Myriad's HIPAA Notice of Privacy Practices, last updated 30 July 2026, says its testing subsidiaries are each covered entities under HIPAA, describes uses and disclosures, and says business associates inside and outside the Myriad family are bound by contract and by law to protect the information they receive. A separate privacy notice covers the website.

As a covered entity laboratory, Myriad contracts with health systems as a provider of testing rather than as their business associate, so no business associate agreement for customers is published, and the notice does not list which outside parties receive genetic data.

Ask which Myriad entity holds your patients' results, and which outside parties receive identifiable genetic data under business associate terms.

A recognized certification is named without the artifact. Myriad's IT security page displays a HITRUST r2 two year certification badge alongside its laboratory credentials (CLIA, CAP, ISO 15189, ISO 13485) and PCI and SOX compliance, and describes security built into its systems and third party vendor screening.

The HITRUST letter, its date and its scope are not published, so a buyer cannot tell whether it covers the systems that compute RiskScore or receive pathology images. The laboratory standards address testing quality rather than information security, and no penetration testing statement, vulnerability disclosure program or SOC 2 report is published.

Ask for the HITRUST letter with its scope and expiry, and confirmation that it covers the scoring systems and any partner processing.

The lead products are offered as laboratory developed tests, and that position is stated. Myriad's May 2026 release describes Prolaris + AI as a laboratory developed test available in the United States outside New York, with New York approval expected; RiskScore is reported within MyRisk from Myriad's CLIA certified and CAP accredited laboratory. No FDA clearance is claimed for either model.

Myriad does hold an FDA approved companion diagnostic, MyChoice CDx, which identifies homologous recombination deficiency for treatment selection; that approval covers that assay and not the risk models. The models inform screening and surveillance decisions, so the laboratory developed test route is the governing position rather than a gap.

Ask whether Myriad plans an FDA submission for either model and how New York approval is progressing for Prolaris + AI.

Results are compared across the groups RiskScore serves, with the weaker group visible. Polygenic scores are known to perform differently by ancestry, and Myriad's published summary of its all ancestry RiskScore (ASCO 2021, posted on Myriad's site) states the method: ancestry informative variants estimate each woman's fractional African, East Asian and European ancestry, and the weights on the breast cancer risk variants follow that composition. It then reports results by group: women in the top decile carried roughly double the average risk in each group except Black and African women, where the increase was 44 percent, and at least 95 percent of the risk variants were common enough in each self reported population to contribute.

No per group counts, discrimination or calibration figures are given in that summary, and Prolaris + AI publishes no results by race or ancestry, a gap for a test used in a disease with known racial differences in outcomes.

Ask for per group counts and calibration for RiskScore, and for Prolaris + AI results by race.

Myriad is the covered entity laboratory, the model derived score is part of the result it reports, and its Notice of Privacy Practices gives the patient a right to have that record amended. The notice says a patient who believes the information Myriad holds is incomplete or incorrect may request an amendment, a route the patient exercises against Myriad directly and that reaches the reported result.

No error rate is published for either model at its reporting threshold, and no warranty, indemnity or remediation commitment attaches to a RiskScore or Prolaris + AI result. Results reach a clinician who decides, which is the main protection against a wrong score.

Ask for the false reassurance and false alarm rates of RiskScore at the thresholds that change screening advice, and what Myriad does when a score is later corrected.

Named record systems with two way ordering and results, without a named live site. Myriad's integrations page describes bidirectional order and results integration with Epic, where all Myriad offerings are available through the Aura network and Myriad is a founding member, and names athenahealth, eClinicalWorks, NextGen Office, Greenway, Practice Fusion and OncoEMR through Flatiron Health. Results return to the record visible across the organization, with variant level data for genomic care planning and VCF files for research.

No health system is named as live on these pages, and how RiskScore appears in the chart, as a discrete value or inside the report document, is not described.

Ask for reference sites live on your record system and whether RiskScore and Prolaris + AI results arrive as discrete data.

A single service model with location implied. Myriad runs testing and scoring in its own accredited laboratories, headquartered in Salt Lake City and licensed in states including California, Maryland, New York, Pennsylvania and Rhode Island, and returns results by portal and record integration. The digital pathology step in Prolaris + AI draws on PATHOMIQ's technology, and nothing published says whether slide images are processed by Myriad or by its partner, or where.

No hosting provider, data location commitment or retention location for sequence data and slide images is published.

Ask where sequence data and pathology images are stored and processed, and whether any processing for Prolaris + AI happens outside Myriad.

A pricing basis with figures is published for patients, without institutional tiers. Myriad's MyRisk cost page says most eligible patients pay nothing after insurance or financial assistance, that 95 percent of commercial and federally funded plans cover hereditary cancer testing, and that financial assistance places patients at 0, 100 or 249 dollars by household income and size. The virtual care route charges 39 dollars for the test kit at order and asks patients to call if the estimate exceeds 249 dollars; interest free payment plans start at 15 dollars a month. No separate charge for RiskScore is published. A self pay route at a fixed price is offered without the figure, and Prolaris + AI pricing is not published.

Ask for the self pay price of MyRisk and Prolaris + AI, and for any institutional or client bill rates a health system program would pay.

Coverage is named with validation behind the two models. RiskScore is for breast cancer risk in women tested with MyRisk, and its all ancestry version was developed and validated across African, East Asian and European continental ancestry components. Prolaris + AI is scoped to men with Gleason 3+3 or 3+4 prostate cancer who are candidates for active surveillance, matching its validation cohort, and is available in the United States outside New York pending state approval. The wider test menu covers oncology, women's health, prenatal care, urology and mental health prescribing, where the assays rather than a model carry the result.

Ask which patient groups RiskScore is not reported for, and whether Prolaris + AI has been evaluated outside the active surveillance population it was trained on.