Bayesian prediction of emergency department wait time

Area of Science:

• Health Services Research
• Biostatistics
• Emergency Medicine

Background:

• Hospitals aim to provide accurate Emergency Department (ED) wait time information.
• ED wait time estimation is complex, depending on patient and ED status.
• Accurate predictions are challenging for prospective, low-acuity patients with limited pre-arrival information.

Purpose of the Study:

• To develop and evaluate a model for estimating ED wait time ranges for prospective low-acuity patients.
• To compare a novel Bayesian quantile regression approach with informative priors against standard methods.

Main Methods:

• Developed a Bayesian quantile regression model incorporating prior information from government statistics and expert opinion.
• Compared the proposed method with frequentist quantile regression and Bayesian quantile regression using non-informative priors.
• Validated the model on a test set of 1,024 low-acuity ED presentations (Categories 3-5).

Main Results:

• The proposed Bayesian model with informative priors outperformed non-informative Bayesian and frequentist methods on the Huber loss metric for median and 90th percentile predictions.
• Incorporating prior information significantly benefited the estimation of model coefficients.
• Demonstrated improved ED wait time quantile estimates.

Conclusions:

• Informative priors derived from expert opinion and government statistics enhance ED wait time prediction accuracy.
• The proposed Bayesian quantile regression approach offers a valuable tool for improving ED wait time estimations for prospective patients.
• This methodology has the potential to significantly improve patient experience and ED resource management.