台灣地區腸病毒急診症候群偵測的系統偵測方法評估、與傳統偵測之比較及其在公共衛生的應用

Syndromic Surveillance System for Detecting Enterovirus Outbreaks: Evaluation and applications in public health

馬曉恬(Hsiao-Tien Ma)

指導教授 ： 金傳春

國立臺灣大學/公共衛生學院/流行病學研究所/碩士(2007年)

https://doi.org/10.6342/NTU.2007.02774

摘要

症候群偵測系統(syndromic surveillance system, SSS)因每日資料之即時性與系統彈性，近年多被廣用在偵測呼吸道與腸胃道疾病。而腸病毒(enterovirus, EV)自1998年大流行以來，已成為台灣地區的地方性流行病，尤其三歲以下的易感受者若感染某些型別的腸病毒，易引起重症、死亡及後遺症，因而造成孩童父母的高度關切。為了有效利用症候群偵測系統進行台灣腸病毒之監測，期達「早期」預防之宏效，。本研究主要目的為：(一)評估適用與檢定症候群偵測系統腸病毒病例流行之統計方法；(二)比較腸病毒症候群偵測系統與其他傳統偵測系統；(三)利用症候群偵測系統中的「類腸病毒」病例進行流行病學因子與臨床嚴重度的流行趨勢分析，以期能日後即時掌控不同高危險族群、不同病徵之腸病毒流行。

本研究包含兩部份： (一)全國性醫院急診(emergency department, ED)症候群偵測系統(Nationwide ED-SSS)，及(二)台北市某醫院急診症候群偵測系統(Taipei ED-SSS)。做法上，先以國際疾病分類第九版碼(international classification of disease codes, ICD-9 碼)篩選全國性急診症候群偵測系統中2004年4月1日至2005年3月31日的「類腸病毒」病例，再以累積和管制圖(cumulative sum control chart, CUSUM)與指數加權平均移動管制圖[exponential weighted moving average加權係數為0.6 (EWMA0.6) 與0.9 (EWMA0.9)]進行異常流行警示之分析，並比較腸病毒症候群偵測系統與其他傳統偵測系統在敏感度、特異性與時效性之差異

關鍵字

腸病毒 ； 急診症候群偵測系統 ； 國際疾病分類法 ； 異常警訊統計方法 ； 累積和管制圖方法 ； 指數加權平均移動管制圖 ； 氣候因子 ； 波以松迴歸時間序列預測模型

並列摘要

Syndromic surveillance systems (SSS) with better timeliness and flexibility in acquiring daily information have been widely used to detect the outbreaks of influenza-like illness (ILI) or gastrointestinal (GI) disease in the recent years. Since 1998, enterovirus (EV) has become an endemic disease in Taiwan. Particularly, children younger than 6 year-old are the major susceptible hosts and children younger than 3 year-old would have the highest probability to develop more severe syndrome, chronic sequelae and fatality, particularly when they are infected with certain serotypes of EV.

To minimize health threat of severe cases arising from large-scale outbreaks of EV, we established an EV-like hospital emergency-department based syndromic surveillance system (ED-SSS) for early detection of the outbreaks. The specific aims of in this study were: (1) to evaluate the most suitable statistical methods used to detect the aberrations of EV outbreak, (2) to compare ED-SSS with the traditional surveillance systems in Taiwan, (3) to find out the important epidemiological characteristics of EV-like cases and trend analysis of severe EV cases for future monitoring various clinical syndromes in different high risk groups once outbreaks of EV occur.

There were two kinds of data sources in this study: (1) Nation-wide Emergency Designated Hospitals’ ED-SSS for Real-time Outbreak and Disease Surveillance (Nationwide ED-SSS) from Apr 1, 2004 to Mar 31, 2005 and (2) Taipei Emergency Department-based infectious disease Syndromic Surveillance System in Taipei (Taipei ED-SSS) from Oct 1, 2005 to Feb 14, 2007. Our approaches were to use the 9th edition of international classification of disease codes (ICD-9 code) to define the EV-like cases first, then applied three methods, such as cumulative sum control(CUSUM) chart and exponential weighted moving average(EWMA0.6 and EWMA0.9) to detect the aberration of each EV-like epidemic. Subsequently, the best performed statistical method in detection outbreaks from weekly data, were used to evaluate the sensitivity, specificity and timeliness of currently developed ED-SSS versus traditional EV surveillance systems. We also investigated the incidence rate, epidemiological characteristics and clinical severity of EV-like cases in ED-SSS. Furthermore, the standard chief complaints in Taipei ED-SSS were used to test the different combination of definition of EV-like cases, and compared them with the EV-like cases defined by ICD-9 criteria. Finally, a multiple Poisson regression time series model of daily EV-like cases obtained from Taipei ED-SSS with simultaneous controll of gender, age, hospital, weekend (Saturday, Sunday) and holiday (Chinese lunar new year days) effects and meteorological data (daily mean temperature, and relative humidity) was developed to prospectively estimate EV-like cases in the future.

The results showed that modified day-of-week effect CUSUM (mC) was the most suitable method in accuracy and timeliness using daily data of ED-SSS whereas the EWMA 0.6 was the best one in specificity and timeliness using weekly data of ED-SSS to detect the EV-like outbreak in Taiwan. Besides, the daily data counts from both Nation-wide and Taipei ED-SSS were capable to detect EV-like abnormal signals at two or three days earlier than the sentinel physician surveillance system.

Using nation-wide ED-SSS to analyze epidemiological characteristics of EV-like cases, we found that EV-like cases of 7-15-year old young adults in nation-wide ED-SSS and of 4-6-year old children in Taipei ED-SSS appeared one week earlier than other age groups would play roles in initiating transmission of EV-like cases. Geographically, the 2004 EV-like cases in TaoYuan, HsinChu and MiaoLi located in northern Taiwan and in TaiChung, ChangHwa and NanTou located in central Taiwan emerged much earlier than those in other areas. Further analysis found that incidence rates of EV-like cases in suburban and rural were much higher than those in urban areas (53/106 vs 30/106, p＜0.05) and the highest incidence of EV-like cases was in Tainan, Chiayi and Yunlin in southern Taiwan. And the higher the percentages of 4-9 year old children, the numbers of EV-like cases were more likely to increase [risk difference (RD) in 4-9 vs >10 yrs. = 0.0086). Most importantly, severe EV-like cases were significantly highest in central Taiwan (40.5%, 15/37, p＜0.05), 1-3-year-old children [51.4% (19/37), p＜0.05]. Interestingly, there was a phenomenon that more percentage of ≦3 year-old severe EV-like cases was observed in the later period of the 2004 epidemic.

Chief complaint (CC) in ED-SSS was not suitable to define EV-like cases, because the most commonly three CC in EV-like cases ≦6 year-old in Taipei ED-SSS were fever (69.5%), cough (18.1%) and sorethroat(13.0%), which were quite similar to the commonly symptoms/signs (s/s) [fever, cough and vomit] found in ≦6 year-old children with non-EV illness. To solve this non-specific problems, using more typical s/s of EV cases (skin rash, oral ulcer, poor appetite, vesicles) serving as screening criteria of CC=defined EV cases, more unexpected EV-like cases increased in non-epidemic periods (November-December and January-March) and thus were not suitable to define CC-based EV-like cases.

To investigate the effects of daily meteorological factors on daily case numbers of EV-like cases, we developed the Poisson time series model for Taipei ED-SSS using data from Oct 1, 2005 to Sep 30 to 2006. Temperature and relative humidity were the two most interesting prediction factors to estimate the future cases of EV. EV-like cases increased 1.45-fold, 1.72-1.73-fold and 1.27-fold if the mean temperature ranged 20-30℃ at 4 days ago, >30℃ at 3-4 days ago, and higher relative humidity ≧75% yesterday, respectively.

This study verified that it is feasible to timely find EV-like outbreak by applying ED-SSS properly plus selecting the best statistical method to detect aberrations, according to the data format of EV-like cases Additionally, the orders of occurring EV-like cases were very different in various age groups and geographical areas. Therefore, integrating these epidemiological characteristics (eg. the earlier risk group to spread the virus to other groups and areas) with the most appropriate aberration detection algorithms, it is very likely that EV-like case would be identified early enough even when the number of cases is small and the transmission could also be blocked efficiently to avoid large-scale outbreak. Future studies can extend this valuable experiences from Taipei to nationwide for better understanding the factors contributing to geographical variations in EV-like cases after establishing the prediction model using multivariate Poisson regression time series model of daily EV-like cases Furthermore, SSS should cover out-patient departments and local clinics to upgrade the effectiveness in prevention and control of infectious diseases.

並列關鍵字

Enterovirus ； emergency-based syndromic surveillance system ； international classification of disease codes(ICD9 code) ； aberration detection algorithms ； umulative sum control ； exponential weighted moving average ； meteological factor ； Poisson time series mode

參考文獻

gastrointestinal syndromic surveillance in New York City: what have we found?"

Chang, L. Y., C. C. King, et al. (2002). "Risk factors of enterovirus 71 infection and

Das, D., D. Weiss, et al. (2003). "Enhanced drop-in syndromic surveillance in New York

Ho, M., E. R. Chen, et al. (1999). "An epidemic of enterovirus 71 infection in Taiwan.

Kaufman, Z., W. K. Wong, et al. (2007). "Evaluation of a syndromic surveillance