QuantHealth

The simulation is what is sold. The technology page describes a drug representation built from a proprietary knowledge graph (more than 5 million datapoints from 30 or more sources), patient journeys drawn from records on 350 million lives, and an outcome prediction model that estimates each simulated patient's response to a given treatment and aggregates them into a virtual trial result.

Protocol optimization, outcome prediction, probability of technical success, enrollment prediction, asset due diligence and portfolio prioritization are all outputs of that model. The knowledge graph and a curated database of results from more than 180,000 trials are inputs to it rather than separate products, and nothing on the site is offered without the simulation.

Outputs inform decisions people make, and the limits of the simulation are not published. The results go to clinical development, business development and portfolio teams as virtual trial readouts, enrollment curves and probability of success estimates; nothing is actioned on a patient.

The home page says the platform can predict how any patient will respond to any therapy, and the technology page that it can test any therapy on any patient with virtually unlimited statistical power, which describes scope rather than a stated limit. What is missing is where the model should not be relied on: drug classes, novel mechanisms or populations it has not been validated on, and the confidence shown with a prediction.

The architecture is described in general terms with nothing identified. The technology page describes a knowledge graph mapping drugs to targets, pathways, cells, anatomy and disease; representation of a novel drug by its connections in that graph; patient journeys decomposed and recomposed with a treatment at a chosen point; and an outcome prediction model, together called a digital drug and digital twin in one platform. The model family, training method, validation design and any version history are not published, and no methodology paper is listed on the site.

The model is the company's own, and the parties around it are partly named. The privacy policy names Amazon Web Services as cloud host (us-east-1, replicated) and lists service providers, analytics providers and advisers by type; the company has announced a data partnership with a healthcare data aggregator for part of its real world training data. The provenance of the rest of the 350 million lives, the knowledge graph's 30 or more sources, and the terms governing sponsor data once it arrives are not enumerated.

Vendor reported accuracy with examples and no published method. The evidence page reports more than 600 simulated trials with up to 90 percent predictive accuracy, 85 percent accuracy on primary endpoints for predictions made before readout without sponsor data, and three phase III examples with simulated and actual results side by side; case studies cover head and neck cancer, inflammatory, metabolic and respiratory disease, and one reports $31.4 million saved for an unnamed top ten pharmaceutical company.

No denominator, definition of accuracy, study design or peer reviewed validation is published, and the customers behind most figures are unnamed. Named endorsements come from Sanofi and Accenture Ventures.

General assurances, with the training data's terms unstated. The privacy policy (10 May 2025) covers personal data processed on behalf of customers as well as website visitors, keeps it only as long as necessary under a data retention policy that is not published, and excludes de-identified and aggregated data from its scope.

The model is trained on records for 350 million lives with more than 10,000 data points each from structured and unstructured sources; nothing published states how those records were de-identified, under what consent or license they entered a commercial training corpus, or whether sponsor data supplied for a simulation is kept separate and excluded from training. Nothing is published on safety engineering.

A substantive privacy document reaches the service, and no HIPAA position is stated. The privacy policy (10 May 2025) applies to personal data processed on behalf of customers as well as website visitors, names Amazon Web Services as host, sets retention by purpose and describes rights under GDPR; it does not apply to de-identified data, which is what the training corpus is described as. No business associate status, agreement or statement that the service takes no protected health information is published.

A trust center exists and nothing independent is readable. The site footer links a trust center, which returns a browser challenge to a plain read, so any attestation it lists is not visible from outside and earns nothing here. The site and privacy policy (10 May 2025) name no SOC 2 report, ISO/IEC 27001 certification or penetration test, and the policy's security statement is a general commitment to commercially acceptable protection. What the company holds is commercially sensitive for sponsors: protocol variants and simulated outcomes for unreleased assets.

No device claim is made and the product is scoped to development decisions. Trial simulation informs sponsors' protocol, portfolio and business decisions and touches no patient's care, so no device pathway applies and none is claimed. Where simulated results might support a regulatory interaction, such as justifying a design or an external comparison, no position on regulatory use or engagement with FDA or EMA on the method is published.

No evaluation of how the simulations behave across patient groups is published. The company markets simulation of any population or subgroup at single-patient resolution, and its case studies include identifying likely non-responder subgroups, which puts subgroup accuracy at the center of the product.

No governance framework, subgroup performance, or description of how the real world training population compares with trial populations by age, sex, race or disease stage is published, and real world data carries its own gaps in who is recorded.

Predictions can be checked against readouts, and nothing stands behind them. The evidence page publishes simulated against actual results for three phase III trials (IMVOKE010, TROPION-Breast01 and ELOQUENT1) and says predictions made before readout, without sponsor data, reached 85 percent accuracy on the primary endpoint, with up to 90 percent across more than 600 simulated trials.

No count of prospective predictions, definition of a correct prediction, or false negative rate is published, so the figures are not an error rate with a denominator. A sponsor may decide not to run a trial on a negative simulation, a decision no later readout corrects. No warranty, indemnity or remediation commitment is published.

There is no clinical record integration; the product works on the company's own data and sponsors' protocol inputs. On the sponsor's side, enrollment prediction combines protocol cohort definitions with site level data, and sponsors can supply their own data, but no clinical trial management, data capture or data warehouse system, standard or delivery format is named. The company's data framework integrates biomedical, clinical and epidemiological sources on its side.

A single hosted service with its primary location committed and replication left open. The privacy policy (10 May 2025) says the company runs its cloud infrastructure in the United States on Amazon Web Services us-east-1, with replication across other regions for resilience, and relies on standard contractual clauses for transfers out of the EEA and UK. The replica regions, tenant isolation and where sponsor data rests are not stated.

No price is published, and the posture is discoverable. The platform is sold to pharmaceutical and biotech sponsors for clinical development, business development and portfolio decisions through a meeting request, and value is stated as customer savings ($31.4 million for one unnamed top ten pharmaceutical company) rather than cost. No fee, unit of charge or contract basis is published.

Coverage is named broadly with validation shown for part of it. The evidence page claims coverage across therapeutic areas, clinical phases and treatment modalities, and case studies span oncology (head and neck, breast, multiple myeloma), inflammatory, metabolic and respiratory disease. The simulated against actual comparisons published are three oncology phase III trials; performance in other areas, earlier phases and novel modalities rests on vendor figures without method. Use extends beyond trial design into business development and portfolio decisions.