In Vivo

Methodology

Clinical Pipeline & Recruitment Atlas

Python builds the registry census and reviewed AI extracts offline. The browser reads static, versioned JSON.

Open app

Current scope

The census contains 11,729 ClinicalTrials.gov interventional studies with an industry lead sponsor, at least one drug or biological intervention, and a phase from Early Phase 1 through Phase 3. Included studies are currently Not yet recruiting, Recruiting, Enrolling by invitation, or Active, not recruiting. No start-date cutoff is applied.

Entity model

The NCT record is the trial unit. A normalized intervention label is a provisional, case-folded drug or biologic name assigned to an experimental arm; placebo and comparator-only labels are excluded. This is not yet a complete molecular-entity or synonym normalization. Therapeutic-area rows count each label or trial once in every matching area, so their totals can exceed global distinct counts.

Therapeutic areas

A versioned keyword and MeSH rule set maps registry conditions to 15 named areas plus Other / unclassified. The current rules classify 89.6% of studies into at least one named area. Multi-condition studies may appear in more than one area.

Recruitment measures

Median trial age is elapsed time from the registry start date to the observation date among studies that have started. Recruiting 24+ months is the share of currently open-enrollment studies whose start date is at least 24 months before the cutoff. These are current-snapshot measures, not time to recruitment close. Historical recruitment transitions have not been reconstructed.

Offline AI use

LangGraph orchestrates structured eligibility extraction, deterministic source validation, an independent model judge, and approval or review routing. The reviewed 25-record pilot reached 137 of 138 scored fields correct (99.3%) and 22 of 23 scored records exact (95.7%). Two source-ambiguous records were excluded from scoring. Eligibility enrichment is shown only for this evaluated depth sample; the full census is not presented as AI-enriched.

Limitations

The Atlas does not reconstruct historical status changes, infer recruitment-close dates, predict trial success, classify regulatory outcomes, or use patient-level data. Candidate normalization and therapeutic-area mapping are provisional and versioned. Registry records remain the source of record, with direct evidence links in trial details.