Tracing the Heart: An Evidence-Linked Pipeline for Heart-Failure Feature Engineering
Researchers developed the Nimblemind Multi-Agent System (nMAS), an evidence-linked, rubric-grounded pipeline for automated heart-failure feature engineering. Evaluated on 500 dummy patient records from nine EHR source tables, nMAS generated 132 structured and 70 rubric-scored aggregated features. Adding these features improved held-out AUROC from 0.895 to 0.963 for HFrEF and 0.870 to 0.910 for HFpEF phenotyping. An independent LLM-based rubric assessment scored the features at 81.5% of maximum points for evidence support and methodological soundness.
A multi-agent system called nMAS automates heart-failure feature engineering from EHR data, generating evidence-linked features that significantly improve phenotyping accuracy for HFrEF and HFpEF.
nMAS uses a multi-agent architecture with rubric-grounded scoring and evidence traceability to produce structured and aggregated features from fragmented EHR tables, achieving substantial AUROC gains in heart-failure subtyping.
Automated feature engineering with built-in evidence provenance could reduce the 39–45% workload burden on clinical data scientists and accelerate AI-driven clinical research, particularly in complex diseases like heart failure.
Reduces manual effort in clinical AI pipelines, potentially lowering costs and time-to-insight for healthcare providers and pharmaceutical research.
Next signals include validation on real-world EHR data, integration with clinical decision support systems, and extension to other disease areas requiring guideline-based feature engineering.