- 學位論文
CUSUM管制圖應用在呼吸道症候群監控上
The CUSUM Control Chart for Respiratory Syndromic Surveillance
摘要
本論文目的在應用品管手法中的累積和管制圖來偵測呼吸道症候群的爆 發，傳統的傳染病偵測方法時效性上明顯不足，因此主動式的症候群偵測方式被 提出用來及早發現疾病疫情的異常，例如：CUSUM 管制圖，研究資料為2003 到2007 年健保資料庫之健保申報資料，健保申報資料是以16 萬人為樣本，呼吸 道症候群ICD-9 診斷碼則採用美國疾病管制局之症候群分類標準。2003 至2007 年中，平均一年有六十萬就診人次，高峰期多發生在10 月至隔年4 月。 累積和管制圖常常被應用在疾病監控上，但是基本假設為觀察值必須互相獨 立，然而本研究資料之間卻有相關性，因此先使用廻歸模式來建立統計預測模 式，然後建立其預測誤差的單邊上累積和管制圖，其ARL 為50，來進行呼吸道 症候群就診人次的監控機制。由於資料間具有相關性，假設廻歸模式的誤差項服 從ARMA 模式，而自變數包含星期一、星期日、12 月、國定假日與國定假日隔 天。本研究利用2003 到2004 年的資料建立廻歸模式，並使用2005 年資料來做 驗證及監控2006 年與2007 年的呼吸道症候群流行情形。 本研究建立出呼吸道症候群統計預測模式與偵測症候群爆發的累積和管制 圖，應用在2005 年時，呼吸道症候群流行的高峰期期間，所建立出的累積和管 制圖確實能偵測出疫情異常，但是在低峰的時候可能會受到颱風的影響而產生錯 誤的警訊。經由監控2006 到2007 年的資料中發現每年的1 月到4 月與12 月都 是呼吸道症候群容易爆發的月份，但是呼吸道疾病似乎有漸漸提早在10 月份的 時候爆發的跡象。在模擬驗證上，本研究所建立出的累積和管制圖偵測疾病爆發 的能力不論在呼吸道症候群流行的高峰期、低峰期或是由低峰期進入高峰期都沒 有顯著差異，大體上本研究所建立出的累積和管制圖確實能偵測出爆發。
並列摘要
The purpose of this thesis is to apply the cumulative sum (CUSUM) control chart of the quality improvement tools to detect the outbreak of the respiratory syndrome. The timeliness to identify the outbreak is not good for the traditional infection surveillance. The active detection “Syndromic surveillance” was developed to early detect the aberration of diseases. Research data are the 2003 to 2007 data on outpatient and emergency department visits obtained from the National Health Insurance Research Database (NHIRD) and the samples of the data are 160,000 patients. The ICD-9-CM codes of the respiratory syndrome are adopted according to the syndromic classification criteria of CDC, USA. The annual average number of the visits is 600,000. The peak period of the epidemic for respiratory syndrome is from October to April in the next year. The CUSUM charts are widely used in disease surveillance, but the basic assumption is that the observations are mutually independent. However, the data in this research are correlated. We use the regression model to establish the statistical forecasting model. We construct the one-sided upper CUSUM chart based on forecasting errors in which ARL is set 50 and then undertake the surveillance mechanism of the number of visits for respiratory syndrome. As a result of the correlated data, we assume that the error term of the regression model follows autoregressive moving average (ARMA) model. The independent variables are Monday, Sunday, December, the national holiday and the next day of the national holiday. In this study, we use the data in 2003 and 2004 to establish the regression model. The data in 2005 are used to validate and then we apply it to monitor the epidemic of the respiratory syndrome from 2006 to 2007. In this research, we establish the statistical forecasting model for respiratory syndrome and construct a CUSUM chart that detects the outbreak of respiratory syndrome. When the CUSUM chart is applied in 2005, it is indeed able to detect the aberration of respiratory syndrome during the peak period of the epidemic for respiratory syndrome in 2005. However, in the low peak period of the epidemic, the control chart could be affected by typhoons to have the false alarms. By monitoring the data in 2006 and 2007, January, February, March, April and December are the months that the respiratory syndrome would outbreak easily. The respiratory diseases seem to outbreak in October, which time is earlier than past year. In the verification with simulation, the detecting ability is non-significant for the epidemic peak period, the epidemic low peak period and the period from the epidemic low peak to the epidemic peak. The CUSUM chart we construct certainly can detect the outbreak in substance.