The uncertainty and ambiguity of not knowing how many patients will be discharged impact patient throughput in hospitals, causing concerns for responding to demand for admissions. Understanding the potential number of patients to be discharged can support caregivers, ability to concentrate on the range of interactions that patients require to ensure early discharge. Accurate forecasting of patients expected to be discharged by noon is beneficial in accommodating patients who need services and in achieving sustainable patient satisfaction.
Models to predict patient discharge before noon (DBN) were formulated using Holt's double exponential smoothing and Box-Jenkins' methods with the aim of achieving minimal errors in each model. The models are applied to 24 months of weekly patient discharge historic data in a medical observation unit and a short-stay clinical unit of a health care hospital system located on the East Coast of United States.
DBN prediction outcomes were more accurate when applying Box-Jenkins' method than Holt's method. Analysis revealed that the model of ARIMA(3,1,2) is most suitable for forecasting. Upon the outcomes of forecast error metrics, the study identifies the mean absolute percent error for the ARIMA model is 14%.
Box-Jenkins forecasting performance is superior in predicting DBN with the least forecast error. Predicted values are significant to decision-making interventions aimed at taking new patients, improving quality patient care, and meeting patient throughput performance goals.