Multi-site benchmarking framework for scalable extraction of geriatric care constructs from electronic health records - npj Health Systems

A study published in npj Health Systems reports a real-world natural language processing (NLP) framework designed to extract geriatric and aging-relevant clinical concepts from multi-site, unstructured electronic health record (EHR) data. The system spans Comprehensive Geriatric Assessment and Age-Friendly Health Systems “4Ms” constructs, unifying functional, cognitive, behavioral, and social domains into a single interoperable workflow. Using a TRUST-based gold-standard development process, researchers benchmarked three clinical concept extraction approaches across multiple EHR systems. They built an expert-annotated dataset of 49,195 fully annotated sentences from 400 clinical notes across four sites, including UT Physicians outpatient data, Memorial Hermann inpatient records, Harris County Psychiatric Center, and MIMIC ICU data from BIDMC.