Castor says AI data extraction hit 99.16% acceptance in sponsor study

Castor said its Catalyst clinical data platform extracted source-record data with 99.16% reviewer acceptance in a Top-10 BioPharma study and is now live in five countries. The company says the human-reviewed, audit-friendly workflow is designed to speed clinical data entry without sacrificing regulatory controls.

Why it matters: - Clinical trial teams still spend heavy time re-keying source documents into electronic data capture systems by hand. - Castor is pitching Catalyst as a faster workflow that keeps a human reviewer in the loop and preserves auditability. - The platform’s reported acceptance rate suggests AI-assisted source data extraction can work in regulated research settings, where every value has to withstand review and inspection.

What happened: - Castor announced a milestone for Catalyst, its AI-driven clinical data platform that extracts data directly from source records under human review. - In a recent study for a Top-10 BioPharma sponsor, reviewers accepted 99.16% of the values Catalyst extracted and overrode 0.78%. - Catalyst is live in production across five countries. - The platform is being used in real-world evidence, post-market clinical follow-up and pivotal medical device trials.

The details: - Catalyst reads source records, extracts data and presents each value in a reviewer-friendly interface that keeps the original context. - Reviewers can keep, correct or reject each value before it enters the clinical record. - In the Top-10 BioPharma study, reviewers assessed 14,290 data points extracted from 285 source files across 158 participants. - The study submitted 14,282 values to the clinical database with no upload errors. - Each value was shown to a reviewer against its source before submission, so the low override rate reflects review rather than unchecked pass-through. - In internal EMR benchmarking of site-based chart review, Catalyst reduced the work from 39 minutes to 6 minutes per chart. - Catalyst uses a Google Gemini model with data residency in the United States and the European Union. - Personal data is redacted through Castor’s localized pseudonymization before anything is processed. - Castor operates under zero data retention, and no client data is used to train models. - Catalyst is GxP compliant and is designed to meet ICH E6(R3) GCP, 21 CFR Part 11 and EU Annex 11. - The platform includes a full immutable audit trail and mandatory human review.

Between the lines: - The regulatory angle is central. AI extraction in clinical research has to clear a high bar because the output becomes part of the clinical record. - Castor is positioning Catalyst as a compliance-first product, not just an automation tool. - The company is also using the five-country rollout and the 99%+ acceptance rate to signal that the approach is moving beyond pilots.

What’s next: - Castor said demand indicates research teams want faster data entry with human oversight. - The company appears to be pushing Catalyst further into clinical operations, real-world evidence and med tech trials. - More adoption will likely depend on whether the workflow continues to hold up under sponsor review and regulatory scrutiny.

The bottom line: - Castor is betting that AI can speed clinical data extraction without breaking the audit trail. The early numbers suggest sponsors may be willing to accept that tradeoff when human reviewers stay in control.