Medicine consumption demand forecasting in French hospitals using Seasonal Auto-Regressive Integrated Moving Average (SARIMA) models

In managing medicine inventory in hospitals, an accurate forecast is essential for planning and optimizing down-stream operations. This study aims to investigate the forecast performance of SARIMA models on different time granularities and identify the characteristics of time series that affect the forecast performance. Forecasts for the consumption demand of 21 medicines used in 4 French hospitals were produced and benchmarked. As a reference to the forecast performance, the forecast performance of rolling moving average (rMA) models is used. The results revealed that higher time granularity yielded improved forecast performance and the statistical analysis demonstrated that the performance of SARIMA models was superior to rMA models on weekly and fortnightly time granularity levels. Then, the forecast accuracies in the context of 1) comparison between different medicines and 2) the same medicine consumed in different hospitals were examined.