Multi-Agent EHR Feature Engineering Produces 202 Auditable Heart-Failure Features, Lifting HFrEF AUROC From 0.895 to 0.963
The Nimblemind Multi-Agent System (2608.06366, submitted 2026-08-06) attacks a bottleneck the authors quantify at 39-45% of clinical data scientists' workload: EHR feature engineering. Run over 500 dummy patient records drawn from nine EHR source tables, nMAS generated 132 structured and 70 rubric-scored aggregated features, each verified for structural integrity, rubric compliance, and provenance, then audited by a restricted LLM. Adding the aggregated features raised held-out AUROC from 0.895 to 0.963 for HFrEF and 0.870 to 0.910 for HFpEF phenotyping, and an independent LLM rubric assessment scored evidence support and methodological soundness at 81.5% of maximum. The authors flag the limit themselves — a single-institution cohort with no external validation.
↳ Follow the thread