THE APPLICATION OF DRUG DEMAND PATTERN ANALYSIS FOR INVENTORY FORECASTING FROM THE CENTRAL PHARMACY TO THE IN-PATIENT PHARMACY: A CASE STUDY OF A UNIVERSITY HOSPITAL

Abstract

Inventory management in tertiary care hospitals is challenged by demand uncertainty, contributing to drug shortages, overstocking, and frequent after-hours emergency requisitions. This study aimed to develop optimized drug demand forecasting models and evidence-based inventory policies for injectable medications in an inpatient pharmacy department. A 26-week retrospective dataset of 197 non-refrigerated injectable drugs was analyzed. Items were classified into four demand categories using Average Demand Interval (ADI) and Squared Coefficient of Variation (CV²): Smooth (159 items), Intermittent (28 items), Lumpy (4 items), and Erratic (6 items). Simple Exponential Smoothing (SES) was applied to Smooth and Erratic groups, while Croston's Method was employed for Intermittent and Lumpy groups. Smoothing parameters (α) were individually optimized to minimize the Adjusted Mean Absolute Percentage Error (AMAPE). Safety Stock (SS), Reorder Points (ROP), and maximum inventory capacity (Q_t) were calculated using a lead time of L = 1/7 week (1 day), a review period of T = 4/7 week (4 days), and a 90% service level. The median AMAPE was 44.1% for Smooth items, 50.2% for Erratic items, and 50.0% for Intermittent and Lumpy items combined. Optimal α values were consistently low for Smooth and Erratic items (median α = 0.2 and 0.1, respectively), while Intermittent and Lumpy items under Croston's Method required higher α values (median 0.9 and 0.6), enabling rapid updating of sparse demand estimates. Among all groups, SES achieved the highest forecasting accuracy for Smooth items, while Croston's Method provided the greatest structural improvement for Intermittent and Lumpy items through separate estimation of demand size and inter-demand interval. The Excel-based decision support tool automates requisition calculations incorporating commercial pack sizes and storage constraints (Q_t), facilitating a transition from experience-based to evidence-based medication management and effectively reducing stockout risks and emergency requisitions.