Inventory Models from Medicine Demand Forecasting with Dummy Variables and Interaction Terms of Multiple Regression

Abstract

The objective of this study is to forecast the demand for both generic and original medicines and to construct their inventory policies at a hospital using these forecasts. To construct an inventory policy, both uncertainties in demand and lead time are accounted for. We consider three inventory models: Models 1 and 2 assume normal distribution, but different calculations are used to compute the order-up-to level and the reorder point. Model 3 does not assume normal distribution but uses the empirical distribution of the lead time and fits distribution to the demand using the R Studio program.  In our multiple regression model for the generic medicines, our independent variables include both numerical variables, such as price, and categorical variables, such as group of doctors, section in hospital, classification of medicine and vendor of medicine. After applying the proposed inventory policy to the six-month actual data, we find that the cycle service level (CSL) exceeds 99% for 8 out of 10 generic medicines, and the inventory cost is reduced by 60%, compared to the cost given the current policy. Furthermore, we quantify the effect of these new generic medicines to the original medicines and obtain the appropriate forecasts and the corresponding inventory policies for the affected original medicines. In addition to the proposed regression models, time series forecasting models can be used for original medicines, whose historical data are available. In this case study, we use the innovations state space model for exponential smoothing (ETS), the autoregressive integrated moving average (ARIMA) and the neural network autoregression (NNAR) models. Our result reveals that the CSL exceeds 99% for 9 out of 10 original medicines, and the inventory cost is reduced by 47%, compared to the cost given the current policy.