การจัดกลุ่มยาเพื่อการพยากรณ์อุปสงค์ในโรงพยาบาล

Abstract

This independent study aims to categorize medications and select the most important drug groups for a hospital, employing the forecasting method with the lowest error rate for these essential drugs. The researcher utilized demand data for 1,501 drugs from 2018-2019 from a private hospital in Chiang Mai, Thailand. The drug categorization was conducted using a combined approach of Vital-Essential-Desirable (VED) Analysis and k-Means Clustering, a machine learning technique. The attributes used for clustering included Annual Drug Expenditure (ADE), average monthly usage, and variance. At least 3% of the total drugs in the essential category were then forecasted using statistical methods such as Simple Moving Average, Exponential Smoothing, Linear Regression, and machine learning methods like Artificial Neural Networks and Support Vector Machines. The forecasting results were evaluated using Mean Absolute Deviation (MAD) and Mean Absolute Percentage Error (MAPE). The study's findings show that out of the 1,501 drugs, the VED Analysis classified them into 213 Vital (14.19%), 588 Essential (39.17%), and 700 Desire (46.64%) drugs. The essential group was further analyzed using k-Means Clustering, resulting in drugs with high annual expenditure, high monthly movement, and high variance. A total of 45 drugs were selected for forecasting. The forecasting results indicated that machine learning methods, specifically Artificial Neural Networks and Support Vector Machines, provided better Mean Absolute Deviation for 22 (48.89%) and 6 (13.33%) drugs, respectively. Additionally, the Mean Absolute Percentage Error was more accurate for 23 (51.11%) and 12 (26.67%) drugs, respectively. Meanwhile, traditional statistical methods such as Simple Moving Average, Exponential Smoothing, and Linear Regression also performed well for certain drugs, sometimes yielding better results than the machine learning methods. The study concluded that forecasting essential drug groups through clustering is beneficial for prioritizing drugs and selecting the appropriate forecasting methods. It suggests initially using machine learning techniques and comparing them with other methods to determine the most suitable forecasting approach for each drug, thereby optimizing resource utilization in the analysis.