Sign in
Module 2Synthetic mock data

LLM Cohort Builder

Define inclusion / exclusion logic across structured fields, ontology-backed conditions, medications (class or named regimen), procedures, the CA 19-9 biomarker, trials, and RECIST responses. Run queries to get population analytics — risk mix, mortality, drug utilization, care-intensity tiers — and stress-test trial screens against actual enrollment. Save reusable definitions for research workflows.

Describe your cohort

Natural language
Try:
Start from a recipe

Cohort definition

Ontology emphasis: pancreatic / EUS concepts (PDAC, CA 19-9, EUS-FNA).

U/mL
Additional serum biomarkers
CEAng/mL
Total bilirubinmg/dL
Molecular markers
Exclude (optional)

Identified mode is gated on authorized clinical workflows in production. Demo defaults to de-identified.

Renders the current filters as inclusion/exclusion criteria and opens the Trial Matcher pre-filled.

Export writes a versioned JSON of the current filters; import applies a saved definition to the form (unknown fields are dropped, and you still click Run).

Encodes the current filters into the URL (?c=…). Sharing or bookmarking the link re-opens the builder with the same definition pre-filled (you still click Run).

FHIR R4 (HL7): Group serializes the cohort definition; Bundle adds the de-identified sample as a List of Patient references. Pure JSON, no PHI.

Result

Configure filters then click Run cohort query.

Saved cohort definitions

Persisted to localStorage — production uses a versioned cohort store.

None saved this session.

Cohort worklist

On-demand re-query of saved definitions vs. their saved snapshot. Total deltas are exact; member-level new/dropped lists are over the bounded result sample.

No saved cohorts yet — save a definition below to add it to the worklist.

Cohort version history

No versions yet — save a cohort to start its history.

LLM document extraction

LLM: liveDeterministic preview

Per spec module 2 §"LLM extraction of cohort-relevant features from documents". The fields below are a deterministic preview: they run through the shared LLM abstraction (src/lib/llm), whose rule-based no-network backend extracts values and evidence spans from the sample document. Set ANTHROPIC_API_KEY to switch the same abstraction to live extraction. Full production build (PDF ingest, normalized variable mapping, honest-broker re-linking) layers on top of this same interface.

EUS-FNA pathology report · 2026-03-12 · synthesized

  • Diagnosisductal adenocarcinoma of pancreatic origin

    “Cytology is consistent with ductal adenocarcinoma of pancreatic origin.”

    90% confidence

  • Grademoderately differentiated

    “The tumor is moderately differentiated (G2).”

    90% confidence

  • Specimen sitehead of pancreas

    “Endoscopic ultrasound-guided fine-needle aspiration (EUS-FNA) of the head of pancreas was performed using a 25-gauge ProCore needle with 4 passes.”

    90% confidence

  • Tumor cellularity70% tumor cellularity

    “Approximately 70% tumor cellularity is present.”

    90% confidence

  • Mitotic rate5 mitoses per 10 high-power fields

    “Mitotic activity is 5 mitoses per 10 high-power fields.”

    90% confidence