Abstract
Initial data analysis and model fitting were conducted using the first 2 years of data (January 1, 2005, to December 31, 2006), here after referred to as the training set. Daily patient counts for each facility for the period January 1, 2007, to March 31, 2007, were held out of the modeling process to empirically evaluate the post sample forecast accuracy, here after referred to as the validation set. Forecast accuracy was assessed at horizons ranging from 1 to 30 days in advance. Data from each facility were treated as individual time series and analyzed and evaluated separately.