Evaluation of a hospital admission prediction model adding coded chief complaint data using neural network...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Emergency Medicine

Background:

• Accurate hospital admission prediction is crucial for emergency department (ED) patient flow.
• Existing prediction models often rely solely on basic triage information.

Purpose of the Study:

• To evaluate the impact of incorporating coded chief complaint (CCC) data into neural network-based hospital admission prediction models.
• To determine if CCC data enhances prediction accuracy compared to models using only triage information.

Main Methods:

• Retrospective derivation and validation cohort study of adult ED visits.
• Neural network methodology with Levenberg-Marquardt back-propagation.
• Comparison of models with and without CCC data using sensitivity, specificity, and area under the curve (AUC).

Main Results:

• Models incorporating CCC data showed improved performance in both derivation and validation cohorts.
• AUC increased from 0.840 to 0.860 in the validation cohort with CCC data.
• Net reclassification index and integrated discrimination improvement were significant with CCC data.

Conclusions:

• The addition of coded chief complaint data modestly but significantly strengthens hospital admission prediction models.
• Neural network methodology is effective for developing these enhanced prediction tools.
• CCC data provides valuable insights for improving ED resource allocation and patient management.