臺灣流行性感冒之監測與流行病學

Abstract

研究背景 流行性感冒(流感, influenza)是一高度傳染性疾病，能在短時間內迅速傳播到全世界，2009年的流感全球大流行即為實例。流感疫情在每年冬季流行，如何減少疾病負擔，是公共衛生一大挑戰，若在流行初期加強偵測流感輕症是一項有效的策略，即早介入公共衛生措施，將可避免流感後續大規模的流行。在臺灣，每年肺炎與流感(pneumonia and death, P&I)的死亡中約有90%(約5千人)為65歲以上的老人，面對如此高的健康威脅，衛生署自1998年始對此危險群免費施打流感疫苗，接種率自1998年的9.9%提升至2007年的49.1%，在嚴重急性呼吸道症候群(severe acute respiratory syndrome, SARS)爆發的2003年，接種率更劇增至68.4%，顯示此二十一世紀第一起新興傳染病流行對於臺灣的醫療、公衛體系、民眾健康行為的影響衝擊甚鉅，以往我國流感研究多自病毒學角度著眼，因此如何整合流行病學的結果，提升流感輕症疫情偵測的效率，並評估衛生政策對於流感預防的成效，有其公共衛生的重要性與社會責任的迫切需求。 研究目標 本論文的目標有三：(一)改進流感的偵測系統、研展早期察出流行異狀之方法與更佳的預測力；(二)了解臺灣流感的流行病學特徵，並以其結果改進偵測及(三)評估流感疫苗病毒株與流行株的吻合度與其他公共衛生努力對於老人流感死亡率的影響。 研究方法 做法上，在改進監測系統部分，使用免費、公開的資訊平臺與簡易資訊聚合(really simple syndication, RSS)技術，成功開發「庶民流感偵測」系統（http://www.flu.org.tw），已能自動地蒐集最新的流感相關新聞、政策宣導與衛生教育，志工也可以經由簡易的網路問卷，通報自己或周遭親友的類流感症狀，系統將動態更新侵襲率與其他流行指標於地圖或統計圖上。 在提升異常偵測的準確度上，以北市每日類流感的症候群偵測系統，並納入空間關係、前一天的類流感就醫數、季節性、週末與假日效果、氣象因子等因子於貝氏階層模式(Bayesian hierarchical model)，使用WinBUGS與R等兩免費軟體進行模式的建構與預測，並以超過警戒值的事後機率，量化疾病爆發的可能性。 為瞭解臺灣流感流行病學的特徵，使用2005-2007年的疾病管制局定點醫師監測資料，運用貝氏最大熵原理(Bayesian maximum entropy)，推估臺灣所有鄉鎮的流感就醫率，並針對流感高峰期，研究流感傳播的走向；再經由計算不同年代各年齡層肺炎與流感死亡率與流感重症比率的變化，比較季節性流感與新型流感全球大流行時的流行病學特徵，並蒐集臺灣、墨西哥、日本等地2009年新型流感H1N1確診病人的年齡分佈，進行國際比較。 在評估疫苗吻合與後SARS期公共衛生努力對於流感相關死亡率的影響，以負二項回歸模式(negative binomial)估計1999至2007年冬季與全年老人流感相關的肺炎與流感、呼吸道與循環系統與全死因死亡三類超額死亡率(excess mortality rates)，並對肺炎與流感死亡率超額的年、月，分析流感疫苗病毒株與主流行株的血球凝集素1 (hemagglutinin 1, HA1)胺基酸序列相同度與抗原決定位(epitope)。 研究結果 自2009年12月1日至2010年5月31日，共11,675人次瀏覽「流感，你在哪裡？」庶民監測網頁，每次瀏覽時間平均為15.54分，另有10,444人次使用流感小工具(Google Gadget)，主要的通報族群又恰是2009年流感的主要族群-學生。比較2009年流感大流行與季節性流感的流行病學差異，年輕成人與學童感染後導致重症(57.9%, 537/928)與死亡(31.4%, 11/35)的比率均較其他年齡層為高。 整體而言，在流感的空間傳播上，疫情高峰有自北向南傳播的趨勢。以貝氏階層模式應用在臺北市的每日症候群監測系統上，此決策規則更成功地在資料驗證期間偵測到高峰，並發現此法能在類流感病例升高前的1-2天即偵測到異常的「流行」訊號。因此建議當事後(posterior)機率超過70%時，需啟動公共衛生措施。 在估計大於65歲老人流感的相關死亡率上，發現愈高的H3N2分離率與疫苗病毒株的胺基酸不吻合度，與愈越高的肺炎與流感死亡率有統計正相關(p<0.05)；相反的，若愈高的H3N2疫苗株的胺基酸吻合度與後SARS公共衛生影響，是與有愈低的肺炎與流感死亡率相關(p<0.05)；此外，2003年以前冬季平均肺炎與流感超額死亡率均較2003年以後為高[mean ± S.D.: 1.44 ± 1.35 vs. 0.35 ± 1.13, p = 0.04]。進一步分析發現當A型流感的流行病毒株與疫苗株吻合時，冬季肺炎與流感的超額死亡率在後SARS時期（即2005-2007年）仍較SARS之前為低[0.03 ± 0.06 vs. 1.57 ± 1.27, p = 0.01]。將疫苗病毒株胺基酸吻合度與後SARS影響兩因子進行分層分析，發現老人冬季肺炎與流感超額死亡率在疫苗不吻合的後SARS時期較SARS之前為低或於SARS之前在疫苗吻合較疫苗不吻合為低。其中最重要的是在2003年5月，當SARS爆發醫院院內感染的高峰期，三項超額死亡率均達最高。此外，在H3N2流行病毒株與疫苗株不吻合的年代，均有較高的肺炎與流感超額死亡率，其疫苗株與主流行株之胺基酸序列相同度較低，且在抗原決定位B上的變異胺基酸也較多。 結論 本研究研發由下而上的庶民流感監測系統，發現發生率高的年輕族群可以使用創新的資訊工具參與疾病偵測的預防的工作，貝氏階層模式不僅可協助動態症候群偵測系統的偵測，也提供機率協助決策者評估是否要採取相關的公共衛生介入措施；另以負二項建構模式，觀察疫苗株與流行株間胺基酸的相似度與抗原決定位的胺基酸變異數三者齊頭並進，將可提供未來決策者有效估計肺炎與流感的超額死亡率，作為奠基評估公衛策略有效性的方法。全球流感監測網絡的完備、資訊分享、提早偵測到異常的「流行」警訊及提供流行機率，助決策者定奪是否需公共衛生措施介入，將有助於預防未來的流感全球大流行，並減輕流感的疾病負擔。未來整合流行病學、生物資訊、病毒學、免疫學、臨床醫學，將能讓醫療公衛人員正確掌握流感病毒的傳染力、致病力、毒力之分子變異，制定最完善的防疫政策。

Background Influenza is a highly contagious infectious disease that can spread rapidly worldwide in a short period of time as it did in the 2009 influenza pandemic. Influenza epidemics tend to appear every winter in countries throughout the globe. Reducing the disease burden of influenza poses a great public health challenge. Intensive surveillance of mild influenza cases in the early stages of an epidemic is an effective strategy for implementing earlier public health interventions and thus for avoiding subsequent large-scale epidemics. In Taiwan, the elderly (> 65 years of age) accounted for nearly 90% (around 5,000 deaths) of total annual pneumonia and influenza (P&I) deaths. Faced with such a dire threat to public health, Taiwan implemented free influenza vaccinations for the elderly in 1998. The influenza vaccine coverage rate increased rapidly from 9.9% in 1998 to 49.1% in 2007, peaking in 2003 (68.4%) after the outbreak of severe acute respiratory syndrome (SARS). This peak suggests that the first new infectious disease to emerge in the 21st century had a significant impact on Taiwan’s medical/public health systems and health behaviors. Existing literature on domestic influenza has focused predominantly on virology. There is a need for new research that integrates virology with epidemiological findings to improve surveillance efficiency for mild influenza cases, evaluate the effectiveness of health policies, and meet urgent social and public health needs. Objectives This dissertation has three objectives: (1) to improve the influenza surveillance system and its capacity for early aberration detection and prediction; (2) to understand the epidemiologic characteristics of influenza in Taiwan, and to apply this understanding to improving surveillance; and (3) to evaluate the impact of public health efforts and the matching status of influenza vaccine and dominant circulating strains on influenza-associated mortalities in the elderly. Methods We used free, open platform, and really simple syndication (RSS) techniques successfully to develop the influenza citizen surveillance system (http://www.flu.org.tw), which provides a centralized location for collecting the latest influenza-related news, policy announcements, and health education materials available in Taiwan. Persons experiencing influenza-like symptoms or whose friends or relatives are experiencing influenza-like symptoms can report cases using a simple Web-based questionnaire. Attack rates and other epidemic indicators can be updated dynamically on maps or statistical plots.

To allow for more accurate aberration detection, a Bayesian hierarchical model incorporating spatial interaction, numbers of ILI visits one day ago, seasonality, weekend and holiday effects, and weather factors was applied to Taipei City’s daily ILI syndromic surveillance. Both WinBUGS and R software were used for model construction and prediction. The degree to which the posterior probability exceeded the threshold was used to quantify the probability of outbreak occurrence. In order to elucidate the epidemiologic characteristics of influenza in Taiwan, data from sentinel physician surveillance in 2005-2007 was used to apply Bayesian maximum entropy (BME) for estimating the consulting rate of influenza in each township. The spatial spreading pattern of influenza peaks was analyzed. Using the calculated age-specific P&I mortality rates and percentages of severe influenza cases, we compared epidemiological differences in seasonal versus pandemic influenza. In addition, we collected H1N1 laboratory-confirmed influenza patients in Taiwan, Mexico, and Japan for international comparison of age distributions during the 2009 influenza pandemic.

To evaluate the impact of vaccine matching and post-SARS public health efforts on influenza-associated mortalities, we used a negative binomial model to estimate three winter and annual excess influenza-associated mortalities among the elderly [pneumonia and influenza (P&I), respiratory and circulatory, and all-cause] from the 1999-2000 through the 2006-2007 influenza seasons. We obtained influenza virus sequences for the months/years in which P&I mortality was excessive, and investigated molecular variation in vaccine-mismatched influenza viruses by comparing the identity percentage of amino acids and epitopes of hemagglutinin 1 (HA1) between the circulating and vaccine strains. Results From December 1, 2009 to May 31, 2010, our citizen surveillance Web site, “Flu, where you are?” received 11,675 unique visitors. The mean duration of all visits to the Web site was 15.54 minutes. In addition, our Google gadget received 10,444 unique visits. Students comprised the major reporting population and were also the population most affected by 2009 pandemic influenza. In comparing the 2009 pandemic influenza with seasonal influenza, young adults and children were at higher risk of developing into severe (57.9%, 537/928) and fatal cases (31.4%, 11/35) than other age groups.

General speaking, the trend in geographical spreading of influenza at peaks was from northern to southern Taiwan. A Bayesian hierarchical model was applicable to the data obtained daily through syndromic surveillance in Taipei City. This decision rule detected the peaks successfully in the validation period, demonstrating that the proposed method can launch alerts for outbreak aberrations 1-2 days prior to the rise in ILI visits. Therefore, we recommend an alert for public heath action if the posterior probability is higher than 70%. After estimating influenza-associated mortality among the elderly, the results show that the higher the isolation rate of A (H3N2) and vaccine-mismatched influenza viruses, the greater the monthly P&I mortality (p<0.05) is. However, this significant positive association became negative for higher matching of A (H3N2) and in the presence of public health efforts with post-SARS effect (p<0.05). The overall mean excess P&I mortality for winters was significantly higher before than after 2003 [mean ± S.D.: 1.44 ± 1.35 vs. 0.35 ± 1.13, p = 0.04]. Further analysis revealed that vaccine-matched circulating influenza A viruses were more significantly associated with lower excess P&I mortality during post-SARS winters (i.e., 2005-2007) than during pre-SARS winters [0.03 ± 0.06 vs. 1.57 ± 1.27, p = 0.01]. Stratification by vaccine-matching and post-SARS effect showed substantial trends toward lower elderly excess P&I mortalities in winters with either mismatching vaccines during the post-SARS period or matching vaccines during the pre-SARS period. Importantly, all three excess mortalities were at their highest in May 2003, when inter-hospital nosocomial infections were peaking. Furthermore, vaccine-mismatched H3N2 viruses circulating in the years with high excess P&I mortality exhibited both lower amino acid identity percentages of HA1 between vaccine and circulating strains, and higher numbers of variations at epitope B. Conclusions Based on the findings from this bottom-up citizen surveillance, we recommend that the younger generations, who showed a higher incidence of 2009 pandemic influenza, participate in this newly developed disease surveillance system and take preventive measures in advance. The Bayesian hierarchical model not only assists in a dynamic syndromic surveillance system but also provides a stochastic probability for decision makers to evaluate needs when implementing public health intervention. In addition, a negative binomial model can be integrated with close monitoring of amino acid sequence identity and epitope variations between the vaccine and circulating strains. In combination, these three approaches provide a powerful means of estimating P&I excess mortalities and evaluating the effectiveness of public health measures. We believe that a comprehensive global influenza surveillance network, information sharing, and early epidemic alerts accompanied by quantified probabilities are valuable tools for preventing the next influenza pandemic and reducing global influenza disease burdens substantially. In the future, the integration of epidemiology, bioinformatics, virology, immunology, and clinical medicine can help both clinical and public health practitioners to understand the transmissibility, pathogenicity, virulence, and molecular variations of influenza viruses and thus to devise the most effective public health policies.