Study Shows How Query-Level Transparency Lets Researchers Verify AI-Generated Cohort Results in Briya's AIRE
In a real-world evaluation on brain tumor data at one of Europe's largest university hospitals, 22 of 23 query-backed responses matched hand-written SQL or traced to a specific clause
NEW YORK, Nov. 18, 2026 /PRNewswire/ -- Briya, developer of AIRE, the scientific AI platform that helps researchers turn fragmented healthcare data into rigorous, defensible evidence, today announced a new preprint posted on medRxiv by researchers at Charité – Universitätsmedizin Berlin. In a real-world evaluation against hand-written SQL across a broad range of cohort analyses, 22 of 23 responses that came with a logged query were correct or traceable to a specific clause in the generated SQL across initial and repeat testing, including complex operations that benchmark studies consider difficult. The authors noted that the platform's query-level transparency allowed them to verify which answers were backed by properly executed database queries.
As research teams adopt AI to query hospital data, verifying its methodology and analysis is critical. A language model can return a plausible number with a confident methodology note, but without visibility to the query behind it, a researcher can't tell a correct result from a convincing wrong one. Charité's evaluation tested exactly that.
"Routine hospital data is a powerful foundation for research, but small definitional choices can change who ends up in a cohort," said Dr. Dmitriy Desser of the Department of Neuroradiology at Charité. "Seeing the query behind each result in AIRE is what allowed us to identify and explain those differences. Without that transparency, this evaluation would not have been possible."
Building patient cohorts for clinical research typically requires weeks of custom database programming. Within the AIRE platform, to evaluate whether natural-language AI can accurately accelerate that workflow, the researchers posed 25 natural-language research questions about adult patients with malignant brain tumors (ICD-10 C71), re-posed 12 of them in a new conversation thread, and evaluated AIRE against hand-written reference SQL on Charité's own hospital data platform. AIRE and its language model ran entirely on a server inside Charité's network, under the hospital's GDPR-compliant pseudonymization, governance and ethics framework, so no patient-level data left the hospital. The team published every prompt, generated query and result so others can audit them.
Because AIRE co-builds the methodology with the researcher, it displays the generated SQL to execute the method alongside the results of the analysis, asks clarifying questions when a definition is ambiguous and maintains an exportable log of every query enabling transparency around the analysis and answer behind every question. The authors also noted how AIRE offered alternative code-set definitions and disclosed restrictions when the query was not supported by the data.
"Throughout AIRE's development, our own testing taught us where AI-generated analysis can go wrong, how important traceability is, and what it takes to monitor it," said Dr. Or Shaked, Executive Director, RWE Medical Research at Briya. "Charité's study took that work to external experts in a real hospital setting, and it pinpointed what matters most: a number is only as reliable as the visibility to the query behind it."
"We're proud of what this study shows and how it was done," said Jason Hein, Chief Commercial Officer at Briya. "Trust in AI for scientific research is earned through transparency and the scrutiny of the scientists who rely on it. The more we put AIRE in front of customers and peers like Charité, the more scrutiny we invite, and transparency is what lets AIRE stand up to it. That's how we intend to lead this market."
The preprint, Natural-language cohort queries on a hospital data platform: accuracy and provenance in a real-world evaluation, is available at https://doi.org/10.64898/2026.11.09.26367645.
Dr. Desser and Dr. Shaked will present the findings in a webinar on November 18, 2026. Registration is available at https://briya.com/webinar-building-patient-cohorts-with-ai.
Conflict of interest and governance disclosures: Briya provided the platform free of charge, its staff installed it and performed the data mapping, and Briya funds 20% of the lead author's position. One co-author is a former Briya employee. A pre-submission version of the manuscript was shared with Briya to verify factual statements about the product. Briya had no role in the design of the query battery, the reference SQL, the analysis or the interpretation, and the authors retained full control over the content and the decision to publish.
About Briya
Briya is the company behind AIRE, the scientific healthcare AI platform purpose-built to expand healthcare research capacity and accelerate the generation of rigorous, defensible evidence. AIRE removes the operational complexity between fragmented healthcare data and evidence generation, while preserving scientific rigor and keeping researchers in control. Unlike generic AI or traditional analytics platforms, AIRE combines methodology-aware Scientific AI, a unified environment for institutional, client, third-party and Briya data, and a privacy-preserving, governed architecture designed for healthcare research. Its advanced healthcare NLP and no-code analytics support workflows from feasibility and proposal development, through cohort design, chart review, analysis, validation and reporting, with access to Briya's global real-world data network when additional data are needed. The result: researchers across life sciences, healthcare organizations, and research service firms spend less time navigating data and administrative bottlenecks and have more capacity and more room for science.
Visit us at www.briya.com and follow Briya on LinkedIn
Media Contact:
Ellie Hanson
FINN Partners for Briya
[email protected]
SOURCE Briya

Share this article