Bringing doctors together

or

A free, private network for verified physicians

  • Discover your community

  • Stay current

  • A profile that represents you

  • Find your people

About RoonPrivacyTerms

    Overview

    Establishing a passive registry of patients with heart failure enabled by natural language processing and language models.

    Nov 13, 2026DOI: 10.1016/j.cardfail.2026.10.009Journal ArticleJCF

    Written by: Gerard T PortelaNils KrügerSebastian SchneeweissKerry NganBowen GuJoyce LiiJie YangMuthiah VaduganathanRishi J Desai

    Summary

    BACKGROUND: Significant budgetary and resource requirements for data collection remain a barrier to launch and maintain disease registries. Automated data collection from electronic health records (EHRs) is a possible solution, but certain data elements critical for research endeavors in heart failure (HF) are not routinely available as structured data fields.METHODS: Using EHRs from a large tertiary care system in Boston, we address this challenge by using natural language processing (NLP) and language models. First, patients with HF were identified using diagnosis codes and NLP of physician notes. Among the patients included, ejection fraction (EF) and New York Heart Association (NYHA) class were extracted using rule-based NLP. Large language models were used to extract social determinants of health (SDoH). EHRs were deterministically linked to insurance claims data from Medicare and Medicaid and state vital statistics files for a comprehensive view of healthcare receipt and outcomes.RESULTS: A total of 69,007 patients with HF were included, among whom mean age was 75.8 years (SD, 12.8) and 51.5% were female. 38,620 patients (56.0%) were classified into HF with reduced, mildly reduced, or preserved EF with average EF of 30.8% (SD, 7.0), 45.0% (SD, 1.7), and 61.9% (SD, 7.3), respectively. 215,627 valid NYHA class mentions were extracted for 16,147 patients (23.4%) and 67,432 patients (97.7%) had some element of social history recorded.CONCLUSIONS: The infrastructure and processes we developed can facilitate a diverse portfolio of future research activities aimed at characterizing and improving outcomes of patients with HF.
    JOJournal of Cardiac Failure

    Journal of Cardiac Failure

    Official Account
    https://onlinejcf.com/
    jcf