Verantos Curation Platform.
An agentic platform that pairs clinical annotators with AI agents to label reference standard data, train extraction and clinical inference models, and apply them to medical records, producing fully structured, research-ready curated data sets.
Patient insights require understanding the entire medical record.
Pharmacoepidemiological studies depend on complex variables: eligibility criteria and outcomes that can’t be easily extracted and analyzed from traditional data sources alone. Producing them with research-grade quality requires curation: extraction, imputation, and clinical inference applied across the full medical record, along with demonstrated reliability in accordance with FDA guidelines.
Most of the clinical record never reaches a structured field. The signal that decides a study lives in notes, reports, and documents.
Complex variables used to select eligible patients and evaluate outcomes:
Complex variables require curation to become research-ready: extraction, imputation, or clinical inference.
Three integrated capabilities for producing curated data sets.
The Verantos Curation Platform pairs clinical annotators with AI agents to create reference standard data, train validated models, and apply them to medical records at scale.
1
Data labeling
Rapidly create reference standard data using clinical annotators and AI agents working in concert.
Two readers 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.
FEV1/FVC 0.64 post-BDA0.64B0.64fev1_fvc_post_bd = 0.64scored ✓
two courses in 12 monthsA2 exacerbationsB2 exacerbationsexacerbations_12mo = 2scored ✓
40 pack-yr · wheeze · reversibilityAasthma–COPD overlapBsevere persistent asthma routed back for adjudicationunscored
Where the two reads diverge, nothing moves forward: the record routes back for adjudication rather than into the reference standard.
2
Model training
Train new extraction, imputation, and clinical inference models using the reference standard data.
Every model is evaluated against the held-out reference standard and reported with the metrics regulators and reviewers expect.
3
Curated data sets
Apply validated models to medical records to produce fully structured curated data sets at scale.
Curation reads every source, not text alone: structured fields and unstructured narrative, with inferred insights where no single field holds the answer and recovery where the structured record is silent.
67F, 40 pack-yr hx1. Wheeze worsening over 6 days. Spirometry: FEV1/FVC 0.64 post-bronchodilator2, partial reversibility. Two courses of systemic corticosteroids in past 12 months3; ED visit in Nov. Impression: severe persistent asthma, inadequately controlled4.
Illustrative record — constructed to demonstrate the mechanism; not patient data.
Every value carries the source that produced it: a code, a note span, or the inference that joined them. And where the whole record is silent, the field stays honestly empty rather than quietly filled.
Full validation methodologyReady to advance your evidence program?
Speak with our team about how the Verantos pharmacoepidemiology stack can accelerate your timeline.