ORCID
Mary E. Helander: 0000-0002-5185-6867
Document Type
Article
Date
8-27-2024
Keywords
logistic regression, binary classification, EMS survival, missing data, MLB method, threshold optimization, machine learning evaluation, NEMSIS, medical emergencies, regression imputation
Language
English
Disciplines
Emergency Medicine | Other Medicine and Health Sciences | Public Health | Social and Behavioral Sciences
Description/Abstract
Objectives: Recent studies address missing survival outcomes in the National Emergency Medical Services Information System (NEMSIS), a system that stores and shares standardized EMS data from U.S. States and Territories. This study aspired to improve imputed EMS outcome accuracy and establish benchmarks for common medical emergencies treated in prehospital care. Methods: This retrospective cohort study of ground-transported EMS patients with definitive end-of-EMS-event outcomes, based on dataset years 2017 to 2023, predicted the probability of “dead” or “alive” status using logistic regression. K-fold cross-validation assessed over-fitting and internal generalization. Threshold optimization and machine learning classification metrics enabled comparison with a recently proposed imputation method (“MLB” method) and its NEMSIS implementation. Published NEMSIS case definitions framed accuracy benchmarks computed for common medical emergencies. Results: Based on 1,324,681 patient events, regression imputation improved the accuracy of mortality outcome imputation (“dead” category accuracy, True Positive Rate, TPR) by 15 percentage points over the MLB method, and by 7 points over its NEMSIS adaption: It correctly identified 2,806 and 1,266 more patient deaths, respectively, while eliminating prediction dependency on a retired NEMSIS element (eDisposition.12). TPR for out-of-hospital cardiac arrest with attempted resuscitation, opioid overdose, and STEMI, improved by 19 to 25 (8 to 13) percentage points over MLB (NEMSIS case definition). The study computed accuracy benchmarks for 40 medical emergency categories, including all case definitions published by NEMSIS. Conclusions: This work improves and establishes benchmarks for imputed EMS outcome accuracy and contributes to the utility of NEMSIS datasets for clinical studies of EMS interventions and operations. Assessment and reporting of TPR is recommended whenever applying any survival imputation method to NEMSIS data, including the NEMSIS case definition for patient death. More attention by EMS clinicians to complete documentation of target NEMSIS elements can improve future studies that consider patient survival as an outcome.
Recommended Citation
Helander, Mary E. “Dead or Alive?” Regression Imputation of the Binary End-of-Event Outcome Indicator for the NEMSIS Public Research Dataset. Preprint: August 27, 2024.
Source
submission
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Included in
Emergency Medicine Commons, Other Medicine and Health Sciences Commons, Public Health Commons, Social and Behavioral Sciences Commons
