Tracing the Heart: An Evidence-Linked Pipeline for Heart-Failure Feature Engineering

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Soorya Ram Shimgekar, Michelle Hu, Dorisa Shehi, Daniel Kang, Roy Ka-Wei Lee, Koustuv Saha, Christian Poellabauer, Christopher Lee, Sajeev Singh, Piyum Zonooz, Navin Kumar, Zeeshan Ahmed, Priyadarshini Kachroo
• Venue: arXiv: cs.AI
• Published: 2026-08-06
• Collected: 2026-08-07
• URL: http://arxiv.org/abs/2608.06366v1
• PDF: https://arxiv.org/pdf/2608.06366v1
• DOI: UNKNOWN

Plain-Language Brief

Researchers have developed an evidence-linked pipeline called Nimblemind Multi-Agent System (nMAS) to automate heart-failure feature engineering for electronic health record data. The system generates structured and rubric-scored features that improve the accuracy of HFrEF and HFpEF phenotyping in clinical research. The evaluation was limited to a single-institution cohort, and external validation is needed.

What the Document Says

• The research focuses on electronic health record (EHR) feature engineering as a major bottleneck in clinical research and AI.
• Heart failure affects 6.7 million U.S. adults and requires integrating fragmented EHR data with disease-specific, guideline-based clinical reasoning.
• Existing rule-based and large language model (LLM)-based approaches offer limited automation for heart-failure feature engineering.
• The Nimblemind Multi-Agent System (nMAS) was developed to address this issue by providing an evidence-linked, rubric-grounded pipeline for automated heart-failure feature engineering.
• nMAS was evaluated on 500 dummy patient records from nine EHR source tables.
• The system generated 132 structured and 70 rubric-scored aggregated features.
• These features were verified for structural integrity, rubric compliance, and provenance.
• An audit of the features was conducted by a restricted LLM.
• Adding the aggregated features improved held-out AUROC for HFrEF and HFpEF phenotyping.
• The system showed promising results in improving the accuracy of heart-failure feature engineering, but evaluation was limited to a single-institution cohort.
• External validation is needed.

Key Concepts Mentioned

• Electronic health record (EHR) feature engineering
• Heart failure
• Nimblemind Multi-Agent System (nMAS)
• AUROC
• HFrEF and HFpEF phenotyping

Claims Made by the Document

• Existing rule-based and LLM-based approaches offer limited automation for heart-failure feature engineering.
• nMAS improves the accuracy of HFrEF and HFpEF phenotyping.

Evidence Used

• 39-45% of data scientists' workload is due to EHR feature engineering
• Heart failure affects 6.7 million U.S. adults
• nMAS generated 132 structured and 70 rubric-scored aggregated features
• Adding the aggregated features improved held-out AUROC for HFrEF and HFpEF phenotyping
• An independent LLM-based rubric assessment scored the features at 81.5% of maximum points.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Electronic health record (EHR) feature engineering is a major bottleneck in clinical research and AI, accounting for 39-45% of data scientists' workload. This is especially pronounced in heart failure, which affects an estimated 6.7 million U.S. adults and requires integrating fragmented EHR data with disease-specific, guideline-based clinical reasoning. Existing rule-based and large language model (LLM)-based approaches offer only partial automation with limited maintainability and evidence traceability. We developed the Nimblemind Multi-Agent System (nMAS), an evidence-linked, rubric-grounded pipeline for automated heart-failure feature engineering, and evaluated it on 500 dummy patient records from nine EHR source tables. nMAS generated 132 structured and 70 rubric-scored aggregated features, verified for structural integrity, rubric compliance, and provenance, and audited by a restricted LLM. Adding the aggregated features improved held-out AUROC from 0.895 to 0.963 for HFrEF and 0.870 to 0.910 for HFpEF phenotyping, and an independent LLM-based rubric assessment of evidence support and methodological soundness scored the features at 81.5% of maximum points. These results demonstrate the feasibility of automated, auditable feature engineering for complex cardiovascular EHR data, though evaluation was limited to a single-institution cohort and external validation is needed.