採用資料探勘技術建立不同醫院層級門診服務量預測模式

Application of Data Mining Techniques in Predicting Volumes of Outpatient Services of Different Hospital Levels

黃興進(Hsin-Ginn Hwang) ； 陳啟元(Chi-Yuan Chen) ； 周宣光(Shrane-Koung Chou) ； 高正雄(Chung-Siung Kao)

《智慧科技與應用統計學報》 3卷2期 (2005/12) Pp. 95-115

https://doi.org/10.29807/JTITAS.200512.0006

摘要

本研究主要在找出醫學中心、區域醫院及地區醫院三個層級之門診服務量影響因素，並建立門診服務量預測模式。首先，透過文獻探討與專家會議的方式，找出影響門診服務量之因素；再以健保局某分局三年之真實資料，建立該區域不同醫院層級之門診服務量預測模式。經迴歸分析後，在醫學中心層級，影響門診服務量之顯著相關因素有診療費、藥費及0-9歲病患人數；R^2=0.399；在區域醫院層級，因素有診療費、藥事服務費、0-9歲病患人數，以及季節；R^2=0.545；在地區醫院層級，因素則有診療費、50-59歲病患人數、30-39歲病患人數、醫師人數、藥事服務費、0-9歲病患人數，以及季節；R^2=0.663。以前述顯著相關因素做為類神經網路之輸入神經元，所得預測模式之R^2在三個醫院層級分別為0.144、0.776及0.845。觀察類神經網路預測模式與迴歸分析預測模式在三個醫院層級的預測結果，類神經網路預測模式於地區醫院層級有較佳的表現。

關鍵字

資料探勘 ； 迴歸分析 ； 類神經網路 ； 預測

並列摘要

The purpose of this research is to find out factors affecting volumes of outpatient services and to build predicting models for annual health insurance payments to hospitals of three different levels, including medical center level, district hospital level, and regional hospital level. By starting with a literature review, a prototype of assumed model was built and then reviewed by the experts from medical field and academia. After that, data from a subunit of the bureau of national health insurance were adopted for doing the study. To build predicting models, regression analysis and neural network approach were employed. With regression analysis, it was found that the factors which influence the volume of outpatient services in the following week at the medical center level were treatment fee, medicine fee, and # of 0-9 year-old patient, for which R^2=0.399 was obtained. At the district hospital level, the factors were found to be treatment fee, dispensing fee, # of 0-9 year-old patient, and season for which R^2=0.545 was obtained. At the regional hospital level, the factors were found to be treatment fee, # of 50-59 year-old patient, # of 30-39 year-old patient, # of doctor of outpatient services, dispensing fee, # of 0-9 year-old patient, and season for which R^2=0.663 was obtained. Then neural network approach was employed. By using the factors found in the regression model at the medical center level as input neurons, we built a neural networks predicting model for the medical center level and obtained R^2=0.144. By repeating the procedure, the predicting models at the district hospital level and at the regional hospital level were obtained, having R^2=0.776 and 0.845 respectively. After the predictions using the neural network models and the regression predicting models at three hospital levels were reviewed, neural network approach appeared to perform better at the regional hospital level.

並列關鍵字

Data mining ； Regression analysis ； Neural networks ； Forecasting ； Prediction