- 學位論文
CUSUM管制圖應用在腸病毒疾病監控上
Applying CUSUM Control Charts on the Surveillance of Entervirus Diseases
摘要
腸病毒是一群病毒的總稱，台灣地區位於在亞熱帶，全年都可能有感染病例出現，但仍以夏季為主要流行季節。自從1998年，台灣地區爆發腸病毒大流行造成全台恐慌，腸病毒的嚴重性也開始被重視。站在防疫的角度，如果能即時發現疫情擴散速度異常昇高的狀況，便能提早針對疫情做更有效的控制，將疫情所造成的危害與衝擊減至最低。 本研究針對腸病毒提出就診人數的預測模式以及疫情監控的機制。資料來源是台灣全民健康保險資料庫，本資料以16萬人為基準，根據第九版國際疾病分類號(ICD-9 codes)計算2003至2007年每日感染類腸病毒症狀的病患人數。 我們利用 ARMA (Autoregressive Moving Average) Regression 模式來建立預測模型，由於每日就診人數為季節性時間序列資料，因此ARIMA模式可適用。建模資料為2003及2004年每日感染類腸病毒症狀的人數，利用2005年的資料來驗證，並應用此預測模型於2006及2007年的資料上。 我們利用累積和管制圖來建立疫情監控機制，方法是以2003至2004年所建立的預測模式來計算2005至2007年的預測誤差然後建立預測誤差的單邊累積和管制圖，管制上限値滿足管制內平均連串長度等於50 ( )。 我們利用2005年的資料來針對我們的監控方法做驗證，結果顯示我們的監控方法可以在疫情擴散的初期即時發出警訊，只有少部份警訊(兩天)為受到星期天影響所造成的錯誤警訊，並利用我們的監控方法持續監控2006至2007年的資料，發現有同樣的偵測能力。 我們利用模擬的資料做為2005年的觀察值，我們在分開的三個期間產生異常高的觀察值並設定三種不同等級的增加量，針對管制內平均連串長度等於50，90和180三個不同設置下的累積和管制圖進行敏感度分析，發現在管制內平均連串長度等於50的設置下累積和管制圖有較好的表現。
並列摘要
There are several serious Enterovirus (EV) outbreaks, for example, the EV infection caused 78 deaths and 405 severe cases in Taiwan in 1998. According to the historical statistics from Center for Disease Control in Taiwan (Taiwan-CDC), the periods of May to June and September to October are the two major epidemic peaks for EV in Taiwan. Traditional infectious disease surveillance systems have lacked active surveillance because it is based on the compulsory reporting of specific and diagnosed diseases to Taiwan-CDC. Therefore, in this study we consider a syndromic surveillance for EV-like surveillance. We perform our method on the Ambulatory Care Expenditures by Visits (ACEV) data obtained from the Taiwan National Health Research Institute (NHIRD). We use 160 thousand people as a unit to survey the daily EV-like cases during year 2003 to 2007 in terms of the 9th edition of international classification of disease codes (ICD-9 codes). We build an ARMA (Autoregressive moving average) Regression model by using the data of year 2003 and 2004 and implement one-day-ahead forecasts for year 2005 to 2007. The EV-like cases time series contain trend, seasonal and cyclical components which result in substantial autocorrelation. Therefore, the ARIMA model has been adopted to handle auto-correlation issues. We use CUSUM charts to detect the aberrations on EV-like cases. A CUSUM chart of forecast errors is build by using the residuals form the ARMA Regression model to monitor the forecast errors of year 2005 to 2007. And we design the parameters of the CUSUM chart with the desired in-control average run length ( ). Finally, verify our surveillance method by using the data of year 2005 and we compare the actual cases with its forecasts to check each signal during 2005. The results show that our method can timely signals on the initial of outbreak and only have few false alerts, which caused by the effect of weekend. Then, we adopt our method to monitor the data in year 2006 to 2007, and the results show that our method has a good performance in 2006 and 2007, either. To verify our method we simulated the in-control and out-of-control data in year 2005 and we compare the detection ability of CUSUM between three levels of ARL0, (that is, , and ) at three levels of shifts (1 , 2 and 3 of residuals).