The provision of patient care in hospital pharmacies requires forecasting across a range of operations planning horizons to support adequate drug availability. This means that demand forecasts need to (1) be aligned across various planning horizons to support inventory management, (2) accommodate volatile drug delivery lead times and (3) be robust against erratic demand patterns including varying levels of intermittency. These are the challenges observed at the UK-based hospital pharmacy in the study. In response, we propose constructing forecasts that leverage the different time scales intrinsic to hospital pharmacies’ inventory management. Using temporal hierarchies, we mitigate the challenge of intermittency and volatility in drug demand, while also enabling coherent decisions across planning time scales. Across the range of service requirements, lead times, and from ‘Drug’ to ‘Drug by Dispensary’ planning, the proposed approach, based on temporal hierarchies, ranks consistently well, reducing modelling risk and supporting automation. It ranks statistically best on pinball loss when planning for 95–99% service at the Drug by Dispensary level for all lead times from 1 to 20 days, a key challenge for the hospital. The approach is model-agnostic allowing hospital pharmacies to adopt either state-of-the-art forecasting models, or adjust to existing software and modelling capabilities.
Barrow et al. (Fri,) studied this question.