Nimblemind Multi-Agent System Automates Heart-Failure Feature Engineering
The Nimblemind Multi-Agent System (nMAS) streamlines feature engineering for heart-failure studies by utilizing electronic health records (EHRs). This innovative system, crafted by researchers and outlined in a preprint on arXiv (2608.06366), tackles the 39-45% of time data scientists dedicate to EHR feature engineering, which is vital for the 6.7 million adults in the U.S. suffering from heart failure. By merging fragmented EHR data with clinical insights, nMAS provides a sustainable alternative to traditional approaches. In tests involving 500 simulated patient records, it produced 132 structured features and 70 rubric-scored features, all confirmed for accuracy and compliance. This system is designed to alleviate the workload of data scientists, facilitating quicker heart-failure data analysis. The preprint is submitted as 2608.06366v1.
Key facts
- nMAS is an evidence-linked, rubric-grounded pipeline for automated heart-failure feature engineering.
- EHR feature engineering accounts for 39-45% of data scientists' workload.
- Heart failure affects an estimated 6.7 million U.S. adults.
- nMAS was evaluated on 500 dummy patient records from nine EHR source tables.
- nMAS generated 132 structured and 70 rubric-scored aggregated features.
- Features were verified for structural integrity, rubric compliance, and provenance.
- A restricted LLM audited the generated features.
- Adding aggregated features improved model performance.
- The system integrates fragmented EHR data with guideline-based clinical reasoning.
- The preprint is available on arXiv with identifier 2608.06366v1.
Entities
Institutions
- arXiv