We’re solving the greatest challenge in the AI era: TRUST.
Verantos is pursuing pioneering work in pharmacoepidemiology: how to achieve high-validity evidence generation at scale.
15+
peer-reviewed publications and conference posters1B+
clinical narratives processed17,000
clinical concepts modeled and phenotyped600%
increase in available data elements per patientModel error becomes measurement error. And measurement error becomes biased evidence.
Advanced approaches to generating high-validity real-world evidence in asthma. Kilpatrick K, Cahill K, Chandran U, Riskin D. Epidemiology, 2025. 18 protocol-defined features · n = 3,481 · FDA award U01FD007172.
Sensitivity alone can be gamed by flagging everyone. PPV alone, by flagging almost no one. F1 is the measure that requires both to be right.
Verantos is the only company to have published methods for measuring data reliability, doing so based on FDA-sponsored work.
Three threads run through our science.
Verantos research is not a collection of one-offs. Every study reinforces the same argument: real-world evidence becomes high-validity only when data reliability is measured, innovations such as AI phenotyping are used, and methods are applied at scale.
1
Data reliability, defined and measured
Through the FDA-sponsored TRUST and VERIFY projects, Verantos operationalized the FDA Real-World Evidence Program’s reliability dimensions of accuracy, completeness, and traceability, and showed how each shapes the validity of downstream evidence. The methods are peer-reviewed and reproducible.
JAMA Network Open · 20252
AI phenotyping that meets regulatory-grade thresholds
Across cardiovascular, asthma, and migraine cohorts, Verantos has repeatedly shown that traditional structured-data phenotyping falls short of regulatory-grade accuracy, while AI applied to unstructured EHR notes meets or exceeds it, closing the gap that historically blocked EHR-based RWE from regulatory use.
JAMIA · BMJ Open · Epidemiology3
Applied evidence at registry scale
The methods are not academic. They are deployed in Verantos Pragmatic Registries and in studies, generating evidence that clinical, payer, and regulatory stakeholders can act on.
Pragmatic and Observational Research · ISPOR
The reference standard is read twice.
Models are only as trustworthy as the data they’re trained on. Two clinicians independently label every record in the reference sample, and neither sees the other’s read. Adjudication and inter-rater reliability thresholds gate the reference standard before any model sees it. Failures route back for adjudication.
FEV1/FVC 0.64 post-BD
two courses in 12 months
40 pack-yr · wheeze · reversibility
The trained model reads the same records and is scored against a held-out reference standard: per-variable sensitivity, specificity, PPV, and F1. Past that gate, the model reads alone, at scale.
Illustrative record — the note, both clinician reads, and the values are constructed to demonstrate the mechanism; they are not patient data or study results.
Validated findings, across therapeutic areas.
A quick read of what our publications have established: head-to-head comparisons, reliability measurements, and AI-based phenotyping at scale.
| Therapeutic area | Finding | Venue |
|---|---|---|
| Data reliability cross-therapeutic | Operationalized accuracy, completeness, and traceability in 120,616 patient records, contrasting traditional versus advanced data and technologies. | JAMA Network Open 2025 |
| Asthma 18 protocol-defined features | F1 score rose from 52.2% (traditional structured) to 94.7% (AI-enabled unstructured) across 3,481 patients, an 81.4% relative improvement. | Epidemiology 2025 |
| Migraine severity and outcomes | Defined and validated a 10-point migraine outcomes model, severity plus associated features, extracted from unstructured EHR with AI. Provides a reproducible framework for measuring soft outcomes that historically resist structured capture. | J Headache Pain 2022 |
| Cardiovascular heart failure | Structured-EHR phenotyping fell short of regulatory-grade accuracy; AI on structured and unstructured data met or exceeded thresholds across most concepts. | JAMIA 2019 BMJ Open 2023 |
TRUST and VERIFY: defining the rules of real-world evidence.
Verantos conducts FDA-sponsored demonstration projects that define how real-world data reliability is measured, and how that reliability shapes the validity of regulatory evidence.
TRUST (awarded 2020) defined the measurement methods and informed the FDA’s 2024 real-world data guidance; VERIFY (contracted 2024) extends them into label expansion.
Ready to advance your evidence program?
Speak with our team about how the Verantos pharmacoepidemiology stack can accelerate your timeline.