Payne, ThomasTSAI, MING-TSE2018-01-202018-01-202017TSAI_washington_0250O_18174.pdfhttp://hdl.handle.net/1773/40818Thesis (Master's)--University of Washington, 2017Emergency Department (ED) overcrowding has become common across the globe. Among many proposed measures, ED length of stay (LOS) remains the most commonly reported outcome resulting from overcrowding. Predicting patients’ ED LOS, especially as early as at presentation, could provide valuable information for both patients and providers: it could not only improve resource allocation, but also could facilitate decision-making. In addition, understanding the influence of each associated predictor enables better operational management on this complex and harmful situation. In this paper, data available at patient presentation were identified based on operational data and patients’ demographic data from an ED, and subsequently predictive modeling was attempted. Overall, the resulting model suffered from high bias, but it performed well in the subgroup of ED LOS between 1 hour and 8 hours. In addition, it was able to capture the trend of ED loading. Furthermore, influential predictors were identified, which serve to inform future more sophisticated modeling.application/pdfen-USCC BY-NC-NDemergency departmentlength of staymachine learningpredictive modelingBioinformaticsHealth care managementInformation scienceTo Be AssignedThesis