Automated feature engineering for clinical data is feasible when grounded in clinical guidelines and evidence trails, but requires careful validation and auditing to ensure reliability in real-world healthcare settings.
Researchers built an automated system (nMAS) to extract and engineer features from fragmented heart-failure patient records in electronic health records. The system combines multi-agent AI with clinical guidelines to generate interpretable features, reducing manual work that typically consumes 39-45% of data scientists' time.