A prediction framework for pharmaceutical drug consumption using short time-series

The increasing pharmaceutical expenditure in many countries has raised concerns regarding the sustainability of healthcare services. To address this issue, accurate forecasting of pharmaceutical demand is crucial for healthcare planning and policy development. This paper proposes a novel prediction framework that integrates different types of historical data and simulates a part of the generative process that produces pharmaceutical consumption, considering both exogenous and endogenous factors, such as per capita consumption trends and population dynamics. The output of the framework is a distribution of likely values, enabling the use not only of the central value for making a prediction but also of the explicitly stated uncertainty, which is crucial for decision-makers in such a critical and complex context. The reliability and consistency of the framework are ensured through backtesting and comparing the predicted results with actual data.