Scholarlink

<p>Emergency medical services (EMS) professionals make high-stakes decisions in austere environments. To support out-of-hospital emergency care, we developed an AI model that leverages lexical data to predict receipt of lifesaving intervention (LSI) in the field. Using a large, nationwide EMS data set, we extracted free-text, clinician-generated chief complaints and receipt of out-of-hospital LSI from the electronic health record. In our derivation cohort (CY2023; <i>n</i> = 9,171,818), we used random forest machine learning (ML) models to predict receipt of seven distinct LSI categories from chief complaint words. Models were temporally validated in a separate cohort (CY2024; <i>n</i> = 9,838,357). Models demonstrated good-to-excellent performance for six LSI categories (AUROCs ranged from 0.929 for thoracic interventions to 0.821 for cardiovascular interventions). Model performance was consistent across training and validation cohorts, indicating temporal robustness. These findings suggest that AI approaches relying on natural language data could support timely, informed out-of-hospital emergency care.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing out-of-hospital emergency care via lexical machine learning modeling of chief complaints

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月16日

查看全文
  • Aaron C. Weidman,
  • Remle P. Crowe,
  • Ali Treichel,
  • Francis X. Guyette,
  • David D. Salcido

摘要

Emergency medical services (EMS) professionals make high-stakes decisions in austere environments. To support out-of-hospital emergency care, we developed an AI model that leverages lexical data to predict receipt of lifesaving intervention (LSI) in the field. Using a large, nationwide EMS data set, we extracted free-text, clinician-generated chief complaints and receipt of out-of-hospital LSI from the electronic health record. In our derivation cohort (CY2023; n = 9,171,818), we used random forest machine learning (ML) models to predict receipt of seven distinct LSI categories from chief complaint words. Models were temporally validated in a separate cohort (CY2024; n = 9,838,357). Models demonstrated good-to-excellent performance for six LSI categories (AUROCs ranged from 0.929 for thoracic interventions to 0.821 for cardiovascular interventions). Model performance was consistent across training and validation cohorts, indicating temporal robustness. These findings suggest that AI approaches relying on natural language data could support timely, informed out-of-hospital emergency care.