Cohorta · Populate

Synthetic patients that behave like the real world

Cohorta builds coherent virtual-patient populations for an indication, a region or a rare subgroup, so you can test a hypothesis on the people a historical trial left out.

Example cohortExample

chr-nsclc-eu-v3 · second-line NSCLC

Virtual patients
20,000
Age 75+
18% (amplified from 9%)
EGFR variant, rare
4% (amplified)
Coherence checks
41 of 41 pass
Lineage
3 sources · v3
Ready for simulation in InsilonVersioned
Example. Values are illustrative.

Capabilities

What Cohorta does

Conditioning

Indication, region and subgroup

Condition generation on diagnosis, geography, line of therapy or a named subpopulation.

Amplification

Rare groups made visible

Oversample paediatric, rare-disease and under-represented ancestry groups with stated weights.

Grounding

Real omic and EHR variability

Distributions come from your data and licensed sources, processed on GPU with RAPIDS and Parabricks.

Coherence

Physiologically consistent profiles

Constraints keep labs, comorbidities and genotypes consistent within each patient.

Lineage

Versioned for review

Every cohort records its sources, weights and generator version.

Privacy

Built for protected data

Generation runs in confidential-computing environments; outputs are checked for re-identification risk.

How it works

Inside Cohorta

  1. Profile

    Describe the population and subgroups.

  2. Ground

    Fit distributions to your data and priors.

  3. Generate

    Sample a coherent cohort on GPU.

  4. Check

    Compare to held-out real data before use.

Cohort packs

Example cohort-pack catalogue

Planned therapeutic-area packs. Pack names and IDs are examples; the first packs will be chosen with design partners.

Showing 9 of 9 example packs

Second-line NSCLCOncology
Regions
EU, US
Strata
Adults; EGFR and ALK variant strata
Pack ID
chr-nsclc-eu-v3
HER2-low breast cancerOncology
Regions
EU, US, JP
Strata
Adults; menopausal status strata
Pack ID
chr-brca-h2l-v1
Type 2 diabetes with CKDCardiometabolic
Regions
US
Strata
eGFR 30–60; age 75+ amplified
Pack ID
chr-t2d-ckd-v2
Heart failure, preserved EFCardiometabolic
Regions
EU
Strata
Obesity and AF strata
Pack ID
chr-hfpef-v1
Duchenne muscular dystrophyRare disease
Regions
Global
Strata
Paediatric; exon-skip eligible strata
Pack ID
chr-dmd-v1
PhenylketonuriaRare disease
Regions
Global
Strata
Paediatric and adult; PAH genotype strata
Pack ID
chr-pku-v1
Moderate to severe psoriasisImmunology
Regions
EU, US
Strata
Prior biologic exposure strata
Pack ID
chr-pso-v2
Ulcerative colitisImmunology
Regions
EU, US, CN
Strata
Steroid-dependent strata
Pack ID
chr-uc-v1
Paediatric asthma, 6 to 11 yearsPaediatric
Regions
US, EU
Strata
Weight-band and CYP2D6 strata
Pack ID
chr-ped-asthma-v1

Specification

Specification

Items marked planned are in design with pilot partners.

ItemDetail
OutputCohort of 104 to 106 virtual patients with demographic, genomic, comorbidity and biomarker profiles
InputsSDTM, OMOP CDM, VCF, imaging-derived features
ValidationHeld-out comparison of marginals and correlations, reported per cohort
PrivacyRe-identification risk checks; confidential computing (H100)
InterfacesAPI, Python client; cohort export planned
StatusIn development; first packs with design partners

Questions

About Cohorta

Are these patients real?

No. They are synthetic profiles drawn from distributions fitted to real data. No virtual patient corresponds to a real person.

How do you know a cohort is realistic?

Each cohort is compared to held-out real data, and the comparison is published with it. If a subgroup is thin in the source data, the report says so.

Can we use our own data only?

Yes. A cohort can be grounded only in your data, or combined with licensed sources you approve.

See Cohorta in a briefing

We walk your team through Cohorta on a public example, then discuss where a pilot could test it against outcomes you already know.