目的：因醫院藥品品項與特性差異大，既有的藥品需求耗用預測模式未能有效調控庫存，因此本研究提出一個預測系統，以需求耗用特性配合存貨政策監控調整庫存，提高存貨週轉率以降低庫存成本，提昇醫院管理藥品等存貨機制。方法：本研究依個案醫院13週藥品耗用量建立不同藥品耗用分類程式，並分別依四種時間序列法做藥品耗用需求預測模擬。結果：本研究發現藥品耗用「穩定類」宜採用趨勢預測法，而藥品耗用「波動類」 宜採用加權移動平均法，可使醫院庫存資金積壓減輕，較現行使用之存貨管理系統達6%之改善成效。同時本研究發現個案醫院應將現行安存天數14天降為10天以下，如此可使存貨金額降低約占現行個案醫院總存貨金額之14%，故本研究結果 成效為每月可減少20%之藥品庫存金額，提高存貨週轉率，達到降低存貨持有成 本之實效。結論：藥品需求（耗用）預測模式對於醫院主管監控調整庫存實為重要，因此為求提升預測能力，應該將藥品以耗用類型分類，再運用時間序列模型進行分析。以個案醫院為例，本研究證實可以獲得很好的管理效益。

Objectives: Due to large differences in the characteristics of both drug usage and demand in hospitals, most of the pharmaceutical items forecasting models failed to effectively stock hospital inventory. In this study, we propose a forecasting system based on consumption characteristics and inventory policies to help executives monitor and adjust certain pharmaceutical inventory to improve inventory turnover rate, reduce inventory costs, and overall improve the hospital inventory management mechanisms. Methods: This study developed category program for different medicines by recording medicines consumed for 13 weeks in case hospital. Each category, author simulated demand forecast of medicines by means of four time series analysis respectively. Results: The results show that the trend forecast model works for the ”steady consumption type” and that the weighed moving average forecast model works for the ”wave consumption type.” This result illustrated cost down capital tied-up and reaching 6% improvement than existing inventory management. With this demand forecast system, authors suggest case hospital to reduce inventory safety days from 14 days to 10 days instead. This will lower the cost of inventory management to 14% of what is current for the case hospital. With our proposed model, case hospital can reduce 20% of medicines inventory cost per month and reduce target inventory carrying cost. Conclusions: The drug demand (consumption forecast) model for hospital executives to monitor inventory adjustment is important. To improve the model's predictive capability, medicine should be classified as consumed type, then use time series models for analysis. This study uses the case hospital to confirm that the forecast model proposed can help this hospital gain significant benefits.