透過您的圖書館登入
IP:3.227.252.87
  • 學位論文

加護病房中抗藥性金黃色葡萄球菌帶菌者之偵測系統

Prediction System of Methicillin-resistant Staphylococcus aureus Carrier in Intensive Care Unit

指導教授 : 徐建業
共同指導教授 : 李友專(Yu-Chuan Li)

摘要


摘要: 抗藥性金黃色葡萄球菌(Methicillin-resistant Staphylococcus aureus; MRSA)是一種常見的院內感染致病菌。因為傳播迅速,且對多種抗生素有抗藥性,治療困難,遂成為醫療上感染控制的重要課題。而加護病房中,入住病患皆為危急之重症者,故其MRSA的管控更形重要。以今日感染症之研究,MRSA的感染起源常為帶菌者所傳播,其中尤以鼻腔帶菌為大宗。如何主動發現加護病房入住病患中之MRSA鼻腔帶菌者,乃成為是否能成功減少加護病房中MRSA感染之鑰。然而因為鼻腔帶菌者,多無MRSA感染之相關臨床症狀,所以雖然在感染管制上有其重要性,但在偵測實務上,卻是十分困難。 人工智慧自1950帶發展以來,便衍生出多種資訊技術協助決策。其中,啟發於神經生物學之類神經網路,因其平行運算及高度容錯之特性,擅長於大量駁雜資料之分類。醫療上因臨床資料收集不易且種類繁多,不易以一般方式處理,是以自1985年起,類神經網路在醫療決策的應用便風起雲湧,至2007年已達每年600篇論文之多。 本研究之目的即嘗試利用多層感知器架構的類神經網路,用以偵測加護病房入住病患中之MRSA鼻腔帶菌者,以資感控人員進行隔離及治療之措施。 本研究與某醫院之感染管制小組合作,以該院成人加護病房入住病患為研究材料,計畫研究收集時期為2007年10月至2008年4月。進行研究之時,先與該院感染管制小組之專家進行文獻討論,以決定蒐集之變數及製作登錄表。再請感染管制護理師,於研究收集時期中,進行個案登錄及鼻腔採樣,並送交細菌室培養。以細菌室傳統培養皿之菌種鑑定,及其藥物敏感測試,做為判定是否MRSA鼻腔帶菌者之黃金標準。本研究共收集547人,以專責感染控制護理師回顧病歷,取得共47項之變數,經由資料探勘技術,縮減變數為22項。再將所蒐集之變數及MRSA鼻腔帶菌之結果,匯入Statistica7.0軟體,採行intelligent problem solver,進行類神經網路訓練,其後,再分別以特徵選取及專家意見,對變數重要性分別排序,接著以之為逐步削減變數之順序依據,並以Area under receiver operating characteristic curve大小做為類神經網路模式優劣判定之標準。得到最佳模式。其為一個三層的感知器結構:輸入層為14個輸入變數,一層15個神經元的隱藏層,輸出層一個神經元。其AUROC為0.7990。接著,以ICU中因隔離及MRSA傳染感染所造成的費用作計算,在閾值為0.3779時,費用最低,為$371,545,而其敏感度為0.4222,其特異度為0.9136。最後再以測試組資料測試,並以之與全隔離及全不隔離做比較。最佳模式花費$244,509,敏感度0.1250,特異度0.9025。其花費略大於全部都不隔離的$240,408,但遠低於全都隔離的$323,536。 本研究之預測模式,於測試組之預測力有限。本研究所依據之費用數據,係根據法國醫院針對MRSA感染在加護病房所造成之醫療花費,所以治療MRSA感染,與隔離費用之比例,僅為$9,275相對於$1,480之6.27倍,所以預測模式之花費,在僅有ICU內之醫療花費之情形下,不如全不隔離。然而因為國內外皆缺乏對於MRSA感染之整體住院費用,及對社區造成感染衝擊的費用數據,本研究無法計入此二項費用,若能加上此二者之費用,則MRSA感染所造成之費用必然增加,相對於隔離費用不變,其比重自然傾向減少偽陰性,所以較之全部都不隔離,效益將會提升。醫療資料,尤以感染症之研究,其偽陰性及偽陽性之結果自不能以等價視之,而再生命不可逆的考量下,更必須把價值計入決策之後果評斷。本研究之14個變數模式雖較往昔研究之40-80個變數減少,然仍為臨床工作所難以負荷,其進一步降低所需變數字為延伸研究之課題。

並列摘要


Methicillin-resistant Staphylococcus aureus (MRSA) is an important pathogen of nosocomial infection. It spreads very rapidly and extensively. It can cause an infection that is resistant to a variety of antibiotics. The control of MRSA infection, therefore, becomes a serious issue in hospital, especially in intensive care unit (ICU) where patients with critical illness and compromised immunity were admitted. MRSA most commonly colonizes the anterior nares and transmits by contact. To find the MRSA nasal carrier and isolate him on admitting to ICU is a key point to prevent from MRSA nosocomial infection in ICU. However, it is difficult to find the MRSA nasal carriers because they are usually asymptomatic. Artificial intelligence develops since 1950s. There are many kinds of information technique to help human to make decision. Artificial neural network (ANN) is one of these smart techniques. It is inspired by neurobiology and is good at dealing with tremendous and variable data because it computes parallel and is fault tolerate. Medicine data is usually heterogenous and difficult to collect completely. Therefore, ANN is applied in medical field extensively since 1980s. The goal of this study is to establish an ANN with multilayer perceptron (MLP) structure to detect the MRSA nasal carriers while they are admitted to ICU. We collected those who was admitted to ICU of one hospital in Taipei from October 2007 to April 2008. We defined the risk factors of MRSA nasal carriage with discussion with the infection control specialists. Then infection control nurses collected the nasal swab speciment and 47 clinical parameters of those patients who were admitted to ICU in the study period. We defined the MRSA carriage as nasal swab culture yields in Oxacillin plate. The total study population comprised of 547 cases. The number of variable was reduced to 22 from 47 by data mining technique in variable reduction. We used intelligent problem solver of Statistica7.0 to construct the ANN and adopt the structures by their area under receiver operating characteristic curve (AUROC). We compare the effect of reducing variables furtherly according to feature selection and expert opinion respectively. The best model got is a MLP ANN with 14 input neuron, 15 neurons in hidden layer, and one output neuron. Its AUROC is 0.7990. The cut-off value of threshold of classification is 0.3779 to produce lowest coat and 0.2023 to produce highest prediction accuracy. While applying in testing set, the cost, sensitivity, and specificity are 244,509, 261,478, 0.1250, 0.4167, 0.9025, 0.6429 for the two cut-off values respectively. The cost of total isolation is 323,536 and 240,408 for non-isolation. The performance of prediction model is not satisfactory in this study. Futhermore, the cost of prediction model is more expensive than non-isolation. It maybe attributed to the inaccuracy of cost in isolation and treatment of MRSA infection. Besides, there are only 4 statiscally significant variables in univariate analysis. This means the there is few difference between MRSA carrier and non-carrier. Accordingly, it is difficult to select carrier from the background population. With the increasing cost of treatment of MRSA infection, or the number of ICU beds, or length of stay in ICU, the cost of prediction model will be lower than non-isolation. This study is limited in one hospital in the study period because the MRSA prevalence is variable in different hospital and different period. It needs to be retrained if we apply this model in other hospital in the future. Besides, further reduction of variables and trial of other classification model will be the next step to extend this study.

參考文獻


10. 王宗曦, et al., 台灣醫學中心及區域醫院2001年-2003年金黃色葡萄球菌之抗藥性情形. 感控雜誌, 2006. 16: p. 1-8.
9. 張靜美, 林金絲, and 陳依雯, 某醫學中心個加護中心院內感染金黃色葡萄球菌之調查. 感控雜誌, 1999. 9: p. 245-54.
1. Jevons, M., Gelbenin-resistant staphylococci. British Medicine Journal, 1961(1): p. 124-5.
2. Voss, A. and B. Doebbeling, The worldwide prevalence of methicillin-resistant Staphylococcus aureus. International Journal of Antimicrobiotic Agents, 1995. 5: p. 101-6.
3. Diekema, D., B. Bootsmiller, and T. Vaughn, Antimicrobial resistent trends and outbreak frequency in United States hospitals. Clinical Infectious Disease, 2004. 38(1): p. 78-85.

被引用紀錄


吳娟(2012)。運用資料探勘技術預測末期病人短期存活時間〔碩士論文,元智大學〕。華藝線上圖書館。https://doi.org/10.6838%2fYZU.2012.00047

延伸閱讀