Short-term forecasting of hospital discharge volume based on time series analysis

摘要

The hospital bed is a scarce resource, which makes it difficult for hospital administrators to manage. Predicting the total number of discharged inpatients (or available beds) from one specific department can improve the management and allocation of hospital bed resources. The objectives of this study are to a) predict the daily inpatient discharges of Nephrology department, b) provide decision support for the hospital beds manager and c) find the most appropriate forecasting model. The discharges data were obtained from the Nephrology of the West China Hospital (WCH) from 2014 beds. Data mining have different techniques can support demand prediction. In this study, we choose three models to prediction, respectively, Autoregressive Integrated Moving Average Model (ARIMA), Long Short-Term Memory model (LSTM) and Random Forests (RF). Normalized mean squared error(NMSE) and mean absolute percentage error(MAPE) are utilized to assess the accuracy of results. The findings indicate that RF, the only multivariate model among the three models, performs best, and the results can be used to aid in strategic decision-making on inpatient beds resource planning in response to predictable discharges.