Table 4 Area under the receiver operating characteristic of MIMIC-IV-ED data validation for predicting the adverse outcomes.

From: Development and validation of a transformer model-based early warning score for real-time prediction of adverse outcomes in the emergency department

 

TEWS (full model) (95% CI)

Vital sign only TEWS (95% CI)

MEWS

No transfer learning model

Vasopressor use

0.847 (0.837–0.857)

0.882 (0.873–0.890)

0.828 (0.817–0.839)

Respiratory support

0.759 (0.753–0.763)

0.837 (0.833–0.841)

0.706 (0.699–0.713)

ICU admission

0.777 (0.774–0.780)

0.815 (0.812–0.819)

0.688 (0.684–0.692)

Septic shock

0.872 (0.857–0.889)

0.905 (0.891–0.918)

0.874 (0.857–0.887)

Transfer learning model (1% of data)

Vasopressor use

0.870 (0.862–0.878)

0.889 (0.880–0.898)

-

Respiratory support

0.875 (0.871–0.879)

0.851 (0.847–0.855)

-

ICU admission

0.851 (0.848–0.853)

0.846 (0.843–0.848)

-

Septic shock

0.903 (0.892–0.915)

0.878 (0.860–0.897)

-

Transfer learning (5% of full data)

Vasopressor use

0.882 (0.8742-0.890)

0.885 (0.876–0.894)

-

Respiratory support

0.897 (0.894-0.900)

0.870 (0.866–0.874)

-

ICU admission

0.863 (0.861–0.865)

0.855 (0.852–0.858)

-

Septic shock

0.901 (0.895–0.908)

0.911 (0.898–0.922)

-

  1. TEWS transformer-based early warning score, MEWS modified early warning score, AUROC area under the receiver operating characteristic curve, CI confidence interval, ICU intensive care unit.
  2. *P < 0.05 for all comparisons with MEWS.