Integrating the interpretable machine learning Score For Emergency Risk Prediction (SERP) with emergency department...

Area of Science:

• Emergency Medicine
• Machine Learning in Healthcare
• Clinical Triage

Background:

• The Patient Acuity Category Scale (PACS) is a standard tool for emergency department (ED) triage.
• Objective mortality endpoints are crucial for accurate risk prediction in ED patients.
• Existing triage systems may benefit from enhanced predictive capabilities.

Purpose of the Study:

• To integrate a machine learning (ML)-based Score for Emergency Risk Prediction (SERP) with the PACS.
• To evaluate the effectiveness of the combined SERP-PACS models in predicting 30-day mortality in ED patients.
• To compare the performance of enhanced triage models against the standard PACS.

Main Methods:

• A retrospective cohort study of ED patients from a tertiary hospital (January 2018 - December 2019).
• Developed two enhanced triage models (PACS+ model 1 and PACS+ model 2) by incorporating SERP into PACS.
• Assessed predictive performance using area under the receiver operating characteristic curve (AUC) and decision curve analysis (DCA).

Main Results:

• The study included over 194,000 ED visits across derivation and test cohorts.
• PACS+ model 1 (AUC 0.828) and PACS+ model 2 (AUC 0.812) significantly outperformed standard PACS (AUC 0.722) in predicting 30-day mortality.
• PACS+ model 1 demonstrated superior clinical utility across various risk thresholds.

Conclusions:

• Integrating ML-based SERP with PACS enhances 30-day mortality prediction accuracy in ED triage.
• The developed PACS+ models offer a more precise risk stratification tool for emergency care.
• This approach has the potential to improve patient outcomes through more informed triage decisions.