根據衛生署疾病管制局資料指出，台灣地區從民國93~95年期間，公私立醫院住院感染病例數為78,886~79,968件，其中醫學中心感染密度增加47％；區域醫院感染密度增加27％。由於院內感染率代表住院病人之醫療品質，也是醫院評鑑的重點之一。所以目前社會大眾對於醫療品質的要求日益提高，要如何做好院內感染管制防範未然，則是醫院感染控制人員努力之方向。
本研究將針對目前醫院院內感染的現況，提出有效的改善方針，提出兩個主要目的︰（1）應用決策樹、分類迴歸樹與邏輯斯迴歸建構院內感染預測模型。（2）運用管制圖監控院內感染密度、感染率與感染人數之變化。以期能夠協助院內的感染控制人員能辨別出高感染風險因子，以降低院內的感染率。
研究發現分類迴歸樹所建構的模型ROC值80.17%為最高，顯示病患年齡、性別、就診科別、住院天數、侵入性裝置、使用營養劑、機械式換氧與發燒等變數為影響院內感染的重要因子。並藉由感控月報資料的監控因子建立MCUSUM管制圖，經由本研究測試與實際的感控狀況相符，能有效的監控院內的感染率。本研究所建構的模型尚需加入臨床的應用驗證，更能符合實務的醫院應用。

According to the statistics from the Center for Disease Control of the Health of Department, there were 78,886~79968 nosocomial infections in medical centers and regional hospitals between 2004 and 2006 in Taiwan. The mean hospital wide nosocomial infection density increase 47% and 27% for medical centers and regional hospitals, respectively. The nosocomial infection rate represents the quality of inpatient, and is emphasized by hospital accreditation. The common people are asking for high healthcare quality. The healthcare workers intend to reduce and prevent nosocomial infections.

The study put forward a plan for improving the rate of nosocomial infection. The study has two purposes.（1）Using C5.0 algorithm, classification and trees (CART) and logistic regression to construct a prediction model for nosocomial infection.（2）Using control chart to monitor three surveillance indexes that includs monthly incidence, monthly incidence density and monthly infection counts. The study hopes to help healthcare workers find risk factors and monitor nosocomial infection rate.

The results of the study show CART model has the highest Receiver Operating Characteristic Curve (ROC) 80.17%. Patient’sgender, age, specialties, length of hospital stay, invasive medical device, total parenteral nutrition, mechanical ventilation and fever were found to be important factors for nosocomial infection. The MCUSUM control chart monitors outcome of three indexes that match with reality. The model needs more clinical test to satisfy clinical need.

一、緖論 1
1.1 研究背景與動機 1
1.2 研究目的 3
1.3 研究範圍與限制 3
1.4 研究流程 4
二、文獻探討 5
2.1 醫院感染控制現況分析 5
2.1.1 院內感染風險因素 8
2.2 資料探勘 9
2.2.1 決策樹 9
2.2.2 分類迴歸樹 11
2.2.3 Logistic迴歸模型 12
2.3 多變量管制圖 14
三、研究方法 18
3.1 研究對象及範圍 19
3.2 預測變數 19
3.3 資料前處理與轉換 20
3.4 建構預測模式 22
3.5 預測模式的績效評估 23
3.5.1 正確率 23
3.5.2 作業特徵曲線 24
3.6 管制圖的建立 25
四、結果分析 28
4.1 訓練樣本與測試樣本資料量的選定 28
4.2 建構決策樹C5.0、CART與Logistic迴歸測模型 30
4.2.1建構決策樹C5.0模型 30
4.2.3 Logistic模型 32
4.3 預測模式測試與說明 33
4.4 建構監控模式 35
4.5 討論 39
五、結論與建議 42
5.1 結論 42
5.2 未來研究建議 43
參考文獻 44
附錄一 營養劑品名表 47
附錄二 抑制劑品名表 47
附錄三 第一線抗微生物製劑品名表 48
附錄四 病患基本資料表 49

1.王濟川、郭志剛，2005，Logistic回歸模型－方法及應用，五南圖書出版股份有限公司，台北市，中華民國。
2.李東穎，2006，應用管制圖偵測院內感染之群突發事件，國立陽明大學衛生資訊與決策研究所碩士論文。
3.林明瀅，2004，感染控制人員新思維，第14卷，第1期，頁55-59。
4.林金絲、黃忠智、許詩典等(2001)。加護中心住院病患使用相關導管引起的院內感染調查。感染控制雜誌，11(1)，11-14。
5.林佩燁，2002，某區域教學醫院門診病患之市場區隔策略：邊際貢獻之應用，國立陽明大學醫務管理研究所碩士論文。
6.吳怡慧，1998，新生兒加護病房低體重早產兒併發院內血流感染危險因子及其對醫療費用之影響，國立成功大學公共衛生研究所碩士論文。
7.吳儼庭，2005，應用類神經網路預測醫院之院內感染，國立陽明大學衛生資訊與決策研究所碩士論文。
8.施智源、陳瀅淳、劉美芳(2007)。美國疾病管制中心2004年院內感染定義中譯。感染控制雜誌，17(1)，11-14。
9.張上淳、盛望徽主持(民92)。因院內感染所造成感染個案之死亡率、合併症及致死原因之分析(行政院衛生署疾病管制局九十一年度科技研究發展計畫DOH91-DC-1043)。臺北市:台大醫院。
10.陳孟娟，(2001)。感染管制與醫療品質指標。感染控制雜誌，11(2)，115-118。
11.翁志男，2005，單一觀測值下監控多變量製程平均數向量之管制圖，國立清華大學統計學研究所碩士論文。
12.衛生署疾病管制局(2006)，93~95年傳染病統計暨監視http://www.cdc.gov.tw/internet-cdc/cdc_pdf/中華民國94年傳染病統計暨監視年報-中文版.pdf。
13.Arantes, A., et al. (2003). Use of statistical process control charts in the epidemiological surveillance of nosocomial infections. Revista de saúde pública, 37(6), pp768-774.
14.Balkhy, et al. (2006). Hospital- and community-acquired infections: a point prevalence and risk factors survey in a tertiary care center in Saudi Arabia. International Journal of Infectious Diseases, 10, pp326—333.
15.Breiman L., Freidman J.H., Olshen R.A., and Stone C.J.,(1993). Classification And Regression Trees, 1st Edition, Chapman & hall.
16.Barana, G., Erbay, A., Pinar.O., et al (2007). Risk factors for nosocomial imipenem-resistant Acinetobacter baumannii infections. International Journal of Infectious Diseases, 12, 16-21.
17.Carlos, A., Pena, R. and Moshe, S. (2000). Evolutionary Computation in Medicine: An Overview, Artificial Intelligence in Medicine, 19, 1-23,.
18.Ducel, G., Fabry, J. and nicolle, L. (2002). Prevention of hospital-acquired infections. A practical guide(2nd ed.).
19.Evans, R. S.,Burke, J. P. and Classen, D. C. et al (1992). Computerized identification of patients t high risk for hospital-acquired infection American Journal of Infection Control, 20(1), 4-10.
20.Frawley, W. J., Piatetsky, S. G., and Matheus, C. J. (1992). An overview in
Knowledge Discovery in Databases. AI Magazine, 58(13), 57-70.
21.Fridkin, S.K., Welbel. S.F., and Weinstein.R.A.(1997). Magnitude and prevention of nosocomial infections in the intensive care unit. Infectious Disease Clinics of North America,11(2), pp 479-96.
22.Gilles, C., Hilariob. M., and Saxc. H. (2004). An Application of One-class Support Vector Machines to Nosocomial Infection Detection .Medinfo,11(1),
pp 716-20。
23.Gustafson, E. T. (2000). Practical risk-adjusted quality control charts for infection control. Am J Infect, 28(6), pp406-414.
24.Hebert C. and Neuhauser D. (2004). Improving hypertension care with Patient generated run charts: physician, patient, and management perspectives. Quality management in health care, 13, 174–7.
25.Hayatiet F., Maghsoodloo S., Devivo M.J. and Carnahan B.J. ( 2006). Control chart for monitoring occupational asthma, Journal of Safety Research, 37,.7–26
26.Jarvis, W. R. (2001). Infection control and changing health-care delivery systems. Emerging Infectious Diseases, 7(2), 170-173.
27.Moran, C. J., and Bui, E.N., (2002). Spatial date mining for enhanced soil map modeling. Internation Journal of Geographical Iformation Science,16(6), 533-549.
28.Mayhall C.G., (1999). Surveillance of Nosocomial Infections. In: Hospital
Epidemiology and Infection Control (2nd ed.). Baltimore :Williams & Wilkins.
29.Montgomery, D.C.(2005). Introduction to statistical quality control,(5th rd). New York：John Wlley& Sons.
30.Piccoli A, Rizzoni G, Tessarin C, Calconi G, Filippini M, Dugo M et al. (1987).
Long-term monitoring of renal transplant patients by a CUSUM test on serum
creatinine, Nephron, 47, 87–94.
31.Pal, M., and Mather P. M., (2003). An assessment of the effectiveness of bescision tree methods for land cover classification. Remote Sensing of environment, 86, 544-565.

32.Quinlan J.R.,(1993). C4.5︰Programs for machine learning,Morgan Kaufmann Publishers.
33.Savas, L., Guvel, S., Onlen Y., Savas N., and Duran N. (2006). Nosocomial Urinary Tract Infections: Micro-organisms, Antibiotic Sensitivities and Risk Factors. West Indian Med J , 55 (3), pp189-193.
34.Sullivan, J. H. and Woodall, W. H. (1996). A Comparison of Multivariate
Control Charts for Individual Observations. Journal of Quality Technology, 28,
pp. 398-408.
35.Tracy, N. D., Young, J. C., and Mason, R. L. (1992). Multivariate Control
Charts for Individual Observations. Journal of Quality Technology, 24, pp.88-95.
36.The Quality Indicator Study Group (1995). An approach to the evaluation of quality indicators of the outcome of care in hospitalized patients, with a focus onnosocomial infection indicators. Am J Infect Control 1995; 23: 215-22.
37.Weinstein, R.A.(1998). Nosocomial infection update. Emerging Infectious Diseases, 4(3), 416-20.