研究背景與目的:台灣自1995年實施全民健保後，財務問題日漸嚴重，欲改善醫療資源浪費，健保局於2002年實施總額給付制度；本研究使用台灣全民健康保險資料庫針對五大癌症患者，研究其主要影響醫療支出因素，並以2002年為界分析實施總額給付前後病患在死前三年的醫療支出變化。
研究方法:從健保資料庫中篩選全部五大癌症患者及其醫療支出總額，並建購決策樹模型(Decision tree)討論醫療支出影響變數及分類規則，在藉由分量迴歸(Quantile regression)分析總額給付前後其病患在死亡前三年(分為12季)醫療支出變化。
研究結果: 決策樹分析影響醫療支出的主要因素有死亡標記、年齡、性別、居住都市等，其中已死亡標記影響最大，證明死亡會影響其在生前的醫療支出，其模型預測能力達七成，可幫助分類出高醫療支出患者；從分量迴歸中得知，年齡、性別、居住城市、醫院規模、65歲以上人口、死亡距離皆對死前輸出有顯著影響，其中死亡距離越近則醫療支出會上升。總額給付制度實施後，死亡前一季的醫療支出影響相較於實施前幅度變小，但在二到十一季的幅度皆大於實施前，女性的影響也相對增加，而醫院規模的影響則相對下降。
研究結論:相較於本研究其他因素，死亡為影響醫療支出變化最大的因素，死亡患者為高醫療支出的機率為未死亡患者的2.621倍。無論總額給付前後，死亡距離越近其醫療支出皆會上升，在總額給付前，其醫療支出會在死前一季大量提升，在總額給付後此情形有趨緩的現象，取而帶之的是總額給付後在死前二到十二季的影響皆較大，故並沒有降低整體醫療支出的跡象。

Background and objective: Since 1995, Taiwan began to implement National Health Insurance system (NHI). However, Taiwan encountered financial problems for health care expenditures’ rapid rise. To solve this problem, Taiwan changed the way of payment from Fee-for-service system to global budget system in 2002. This paper focuses on the influential factors of the health care expenditures for patients with top 5 cancers and the change of health care expenditure before and after the adoption of global budget payment system.
Method: This paper uses the National Health Insurance Research Database (NHIRD) as data source to finds the patients with top 5 cancers. And use the decision tree to compare the importance of different factors. Last, using the quantile regression to analyze the change of health care expenditures before and after the global budget payment system is applied.
Results: In the decision tree, the death mark is the main variable to affect the health care expenditures. The predictive capability of decision tree is up to 70%. In quantile regression, the influence factors include age, gender, occupation, hospital size, time to death, etc. 〖BD〗_q have a downward trend form 〖BD〗_1 to 〖BD〗_12. After implanting global budget payment system, the influence of women and 〖BD〗_1 become stronger. However, the influence of center is weaker.
Conclusion: Compare to the other factors in this study, death mark is the most important variable to affect the health care expenditures. The results showed the health care expenditures would rise when the patients’ death is getting close. In other words, health care expenditures is massively increasing in the last quarter before the application of global budget payment system, but .this situation became stable after the introduction of global budget payment system. In contrast, the influences of 〖BD〗_2~〖BD〗_11 are bigger after the global budget payment system is carried out. Therefore, there is no indication for the reduction of health care expenditures after the global budget payment system is adopted.

1.Schneider, E. and Guralnik, J., 1990. The aging of American：Impact on Health Care Cost.Journal of the American Medical Association,263(17),2335-2340.
2.Getzen, T., 1992.Population Ageing and the Growth of Health Expenditures.Journal of Gerontology：Social Sciences,47(3),98-104.
3.Mendelson, D. and Schwartz, W., 1993.The Effect of Aging and Population Growth on Health Care Cost.Health Affairs,12(1),119-125.
4.Felder, S.,Meier,M.,& Schmitt, H.(2000). Health care expenditure in the last months of life. Journal of Health Economics, 19,679-695.
5.Zweifel, P., Felder, S., &Meiers, M. (1999). Ageing of population and health care expenditure: A red herring? Health Econ., 8, 485–49
6. Fayyad, Usama; Piatetsky-Shapiro, Gregory; Smyth, Padhraic (1996). "From Data Mining to Knowledge Discovery in Databases" . Retrieved 17 December 2008.
7.Berry M J A, Linoff G. 2000. Mastering Data Mining: The Art and Science ofCustomer Relationship Management. John Wiley & Sons, Inc., New York.
8.Hand D J, Mannila H, Smyth P. 2001. Principle of Data Mining. MIT Press,USA.
9.Roiger R J, Geatz M W. 2003. Data Mining: a Tutorial-BasedPrimer.Addison-Wesley, Minnesota, USA.
10.Han J, Kamber M. 2007. Data Mining:Concepts and Techniques, Second Edition.Morgan Kaufmann Publishers, San Francisco, California, USA.
11.Fayyad U, Piatetsky-Shapiro G, Smyth P. 1996. From Data Mining toKnowledge Discovery in Databases. AI magazine, AAAI press Fall:37-54.
12.J.R. QUINLAN.1985.Induction of Decision Trees.Centre for Advanced Computing Sciences, New South Wales Institute of Technology, Sydney 2007, Australia
13.Leo Breiman (Author), 1984. Classification and Regression Trees (Wadsworth Statistics/Probability) Paperback – January 1, 1984.
14.Mosteller, F.,Tukey, J.W., 1977. Data Analysis and Regression. Addison-Wesley Publishing Co, Reading, MA.
15.Kass, G. V. (1975). Significance testing in automatic interaction detection. Ph.D. thesis, South Africa: University of Weiwaterstrand.
16.McCarty, J. A., Hastak, M. (2007).Segmentation approaches in data-mining: a comparison of RFM, CHAID, and logistic regression. Journal of Business Research 60, pp. 656-662
17.Koenker, R., Bassett, G., 1978. Regression quantiles. Econometrica 46 (1), 33–50
18.Ulf-G. Gerdtham.1993.The impact of aging on health care expenditure in Sweden. Health Policy, Volume 24, Issue 1, April 1993, Pages 1-8
19.Emanuel, E. J., & Emanuel, L. L. (1994). The economics of dying: The illusion of cost savings at the end of life. The New England Journal of Medicine, 330(8), 540–544.