Predicting emergency department volume using forecasting methods to create a "surge response" for noncrisis events.

OBJECTIVES This study investigated whether emergency department (ED) variables could be used in mathematical models to predict a future surge in ED volume based on recent levels of use of physician capacity. The models may be used to guide decisions related to on-call staffing in non-crisis-related surges of patient volume. METHODS A retrospective analysis was conducted using information spanning July 2009 through June 2010 from a large urban teaching hospital with a Level I trauma center. A comparison of significance was used to assess the impact of multiple patient-specific variables on the state of the ED. Physician capacity was modeled based on historical physician treatment capacity and productivity. Binary logistic regression analysis was used to determine the probability that the available physician capacity would be sufficient to treat all patients forecasted to arrive in the next time period. The prediction horizons used were 15 minutes, 30 minutes, 1 hour, 2 hours, 4 hours, 8 hours, and 12 hours. Five consecutive months of patient data from July 2010 through November 2010, similar to the data used to generate the models, was used to validate the models. Positive predictive values, Type I and Type II errors, and real-time accuracy in predicting noncrisis surge events were used to evaluate the forecast accuracy of the models. RESULTS The ratio of new patients requiring treatment over total physician capacity (termed the care utilization ratio [CUR]) was deemed a robust predictor of the state of the ED (with a CUR greater than 1 indicating that the physician capacity would not be sufficient to treat all patients forecasted to arrive). Prediction intervals of 30 minutes, 8 hours, and 12 hours performed best of all models analyzed, with deviances of 1.000, 0.951, and 0.864, respectively. A 95% significance was used to validate the models against the July 2010 through November 2010 data set. Positive predictive values ranged from 0.738 to 0.872, true positives ranged from 74% to 94%, and true negatives ranged from 70% to 90% depending on the threshold used to determine the state of the ED with the 30-minute prediction model. CONCLUSIONS The CUR is a new and robust indicator of an ED system's performance. The study was able to model the tradeoff of longer time to response versus shorter but more accurate predictions, by investigating different prediction intervals. Current practice would have been improved by using the proposed models and would have identified the surge in patient volume earlier on noncrisis days.