Natural-language cohort queries on a hospital data platform: accuracy and provenance in a real-world evaluation

In an adult C71 brain-tumor cohort (n = 1,594), 15 of 25 natural-language prompts produced exact, definition-aligned, or traceable results, with 4 unanswered and 6 plausible-but-incorrect.

• Why it matters: Routine clinical data access is technically hard across heterogeneous hospital systems, so the study tested whether a commercial natural-language query tool can deliver accurate cohort outputs with traceability.
• What they did: An on-premises NL interface (AIRE) was deployed over two years across three tertiary-care hospitals, then 25 prompts were compared against reference SQL and checked for matching executed queries in exported logs.
• The result: Across initial and repeat evaluations, 22 of 23 logged responses were correct or traceable, while all eight unlogged responses were incorrect, enabling verification of displayed cohort results.

AI summary of the abstract · may contain errors · not medical advice · read the paper.