Predicting left-without-being-seen in an emergency department as a dynamic risk.

Area of Science:

• Healthcare Analytics
• Machine Learning in Medicine
• Emergency Department Operations

Background:

• High rates of patients leaving the Emergency Department (ED) without being seen (LWBS) negatively impact patient flow and care.
• Accurate prediction of LWBS risk is crucial for implementing targeted interventions and improving ED efficiency.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting LWBS risk in ED patients.
• To compare the accuracy of a dynamic ML model, which updates predictions over time, against a static model.

Main Methods:

• Two XGBoost classification models were developed using 18 months of ED visit data: a static model and a dynamic model updating predictions after 30 minutes.
• The models were tested on six months of subsequent data from a large academic medical campus.
• The study analyzed a cohort of 150,959 ED patient arrivals.

Main Results:

• The dynamic ML model achieved an AUROC of 0.86, outperforming the static model's AUROC of 0.80.
• The dynamic model reduced missed LWBS cases by approximately 50% compared to the static model.
• No increase in false positives was observed with the dynamic model.

Conclusions:

• Dynamically updating LWBS risk predictions over a patient's ED wait time significantly improves prediction accuracy.
• This approach is more effective than traditional static models in identifying patients at high risk of LWBS.
• Implementing dynamic ML models can lead to substantial reductions in missed LWBS cases and enhance ED patient management.