YouScript

The engine is a patented computational model over a curated evidence base rather than a learned system, and no machine learning claim was located anywhere in the company's material. What it computes is genuinely more than a lookup: cumulative interactions across a whole regimen, drug to gene effects, and phenoconversion, where one drug alters the metabolic capacity a patient's genotype would otherwise predict.

That last mechanism is the reason a pairwise interaction table gives the wrong answer for a polypharmacy patient, and modelling it requires pharmacological computation rather than a database join. The asset underneath is nonetheless the curated corpus of more than 19,000 references across 2,000 medications, which places this squarely in the moat is the dataset pattern. Graded C on the mechanism, with the accuracy of the company's own description credited: it calls this an evidence based patented algorithm and does not reach for an artificial intelligence label.

Output reaches a prescriber or a dispensing pharmacist who decides, and the product never touches an order. One design choice is worth more than the workflow position: guidance is severity coded across four levels, change, consider, monitor and no change, so the system grades the strength of its own recommendation instead of firing an undifferentiated alert.

That is a direct structural answer to the override problem that defines this category, since a clinician who can see which alerts the system itself considers actionable has a basis for triage that binary alerting denies them. The product also surfaces alternative medications within the same class or indication rather than only flagging a problem, which makes the recommendation actionable rather than obstructive.

Held at B because no acceptance rate, override rate or alert burden figure is published, and because the four level scale itself has no published validation showing that the levels correspond to real differences in risk.

Two partial transparency routes exist and neither substitutes for published validation. The algorithm is patented, and patents are public documents, so a determined reader has a route to the method that most vendors in this category do not offer even in principle.

And the product is designed to explain itself at the point of care, with the company stating that it is the only software providing a biologic basis for why an interaction occurs rather than asserting that one does, which is the practical form of letting a clinician check the reasoning.

Against that, retrieval located no description of how cumulative risk is computed across a regimen, no thresholds separating the four severity levels, no calibration data and no validation of the phenoconversion predictions specifically, which is the most technically ambitious thing the engine claims to do and the hardest for an outside party to assess.

This vendor holds the only prospective randomised controlled trial located in this category. Elliott and colleagues, publishing in PLOS One in February 2017, randomised polypharmacy home health patients to pharmacogenetic profiling reviewed through this tool against usual pharmacist guided medication management.

Rehospitalisations per patient were 0.25 against 0.38 at 30 days and 0.33 against 0.70 at 60 days, and emergency department visits were 0.25 against 0.40 at 30 days and 0.39 against 0.66 at 60 days. The honest reading requires both timepoints: neither 30 day result reached significance, at P equals 0.21 and 0.16, while both 60 day results did, at P equals 0.007 and 0.045. The paper published the null results alongside the positive ones, which is the standard this index looks for.

A separate University of Utah led study in the Journal of Medical Economics reported a 71 percent reduction in emergency department visits, a 39 percent reduction in admissions and 1,132 dollars in estimated savings over four months, and that figure is the one company material leads with.

Held at B rather than higher because the trial was a pilot with a control arm of 53 patients, because employees of the vendor's then parent laboratory are co authors on it, and because the laboratory supported the second study with materials and genotyping.

The data handled here is a category apart. A pharmacogenomic result is a genetic finding that does not change over a lifetime, cannot be revoked once disclosed, and carries implications for a patient's biological relatives who never consented to anything. The platform ingests those results from any laboratory alongside a full medication list.

Retrieval located no retention schedule, no de identification statement, no description of whether genotype data is retained after an interaction check completes, and no discussion of the additional protections that genetic information attracts beyond ordinary protected health information.

Graded C rather than lower because the laboratory agnostic architecture means the vendor receives an interpreted result rather than raw sequence data, which is a materially smaller footprint, and because the interaction check is a point in time query rather than continuous population monitoring.

Two retrieval passes across the company site, its news pages and the parent group's material surfaced no HIPAA statement, no business associate agreement terms, no execution path, no subprocessor list and no compliance page. The grade describes published posture rather than contractual reality, since integrations with Epic, Cerner, Allscripts and Telus Health, and deployments in home health and long term care, all require executed agreements.

The absence is more consequential for this vendor than for most in the lane because the data involved includes genetic results, which attract protections beyond the ordinary privacy rule in several United States jurisdictions and under separate legislation again in Canada, where the Telus Health integration operates.

Retrieval located no service organisation controls report, no HITRUST certification, no ISO 27001, no trust centre, no penetration testing cadence and no vulnerability disclosure policy for this product or for its parent group. The gap is notable given the ownership history, since the platform has passed through two publicly traded or venture backed genetics organisations that would each have maintained security programmes, and given that four named electronic health record integrations imply that partner security reviews were passed repeatedly. None of that is readable by a prospective buyer. For a platform holding genetic results, publishing an attestation would be the highest value single change available to this record.

This vendor states its regulatory position explicitly where most of the category leaves it to inference, publishing that the software has not been reviewed or approved by the Food and Drug Administration and cannot be used to diagnose or treat any disease or health condition. That directness earns credit.

The tension a buyer should weigh is that the product does more than the disclaimer implies: it interprets a genetic result against a specific medication and recommends alternative drugs, which sits closer to a therapeutic proposal than to a reference lookup, and pharmacogenomic interpretation has been an area of regulatory attention distinct from the underlying laboratory test.

Establish which claims the vendor considers covered by its disclaimer and which by the laboratory's own regulatory basis, because the interpretation layer and the test are regulated through different routes and a buyer contracts for both.

The exposure here is more technically specific than for any other vendor in this lane, and it is intrinsic to pharmacogenomics rather than a criticism of this company's conduct. The frequency of metabolising enzyme variants differs substantially between ancestral populations, and the reference data underpinning drug to gene guidance has historically been derived disproportionately from people of European ancestry, so the completeness of any interaction engine's coverage varies by whose genome it is reading.

A patient whose relevant variants are poorly characterised in the source literature receives a confidently formatted answer built on thinner evidence, and nothing in the interface distinguishes those cases. Retrieval located no statement of which populations the evidence base covers well, no performance reporting by ancestry, age or sex, and no bias assessment. A vendor whose entire proposition is genotype informed prescribing is the right party to publish that coverage map.

Integration breadth is real and named across four systems: Epic, Cerner, Allscripts and Telus Health, the last of which extends reach into Canadian practice and is unusual in this index. Delivery is offered three ways, embedded in the electronic health record, as a standalone application, and through an application programming interface, which lets an organisation adopt the analytics without an integration project and matters for the long term care and home health settings where the evidence was generated and where record systems are often lighter.

Held at B because no marketplace certification, no FHIR conformance statement and no public API documentation were located, and because the specific mechanism by which a pharmacogenomic result from an arbitrary laboratory reaches the engine, which is the crux of the laboratory agnostic claim, is not described publicly.

Delivery is cloud based with embedded, standalone and programmatic access described, and operations span the United States and Canada through the Telus Health relationship. Retrieval located no named hosting provider, no cloud region, no data residency commitment and no statement about whether Canadian patient data remains in Canada, which is a live question rather than a theoretical one given provincial health privacy legislation and the sensitivity of the genetic results involved.

The ownership history compounds it, since the platform has moved between three corporate parents in seven years and public material does not describe whether the underlying infrastructure moved with it or was rebuilt.

No price, unit, pricing basis, contract shape or implementation fee was located. The disclosure that matters more here is structural and it concerns independence rather than cost. The product's central commercial claim is that it is laboratory agnostic, meaning it will interpret a pharmacogenomic result from any testing provider, and that claim is what distinguishes it from decision support bundled with a particular laboratory's panel.

Yet the platform was incubated inside a pharmacogenetic testing laboratory, then owned by a genetics testing company that acquired that laboratory at the same time, and is now owned by a group whose portfolio includes a clinical testing laboratory. A tool that advises whether and how to act on a genetic test has been owned by parties selling genetic tests throughout its existence. That is commercially ordinary and it is exactly the kind of alignment a buyer should be able to read about rather than reconstruct, and no statement addressing it was located.

Coverage follows the medication list rather than a specialty, spanning more than 2,000 drugs, which means the tool applies wherever polypharmacy occurs rather than to a defined clinical service. Settings evidenced in practice are the ones where polypharmacy concentrates and where the published research was conducted: home health, where the randomised trial ran, long term care and programmes of all inclusive care for the elderly through a pharmacy partnership, and ambulatory practice through electronic health record integration.

Geographic reach covers the United States and Canada. Held at B because acute inpatient deployment is not evidenced, paediatric use is not addressed despite pharmacogenomics being clinically consequential in children, and the named customer base is thinner than a product with two decades of history would suggest.