利用統合分析探討各國COVID-19個案死亡率

Abstract

背景 自2020年初以來，嚴重特殊傳染性肺炎（Coronavirus Disease 2019, COVID-19）成為國際重大議題。隨著全世界確診與死亡人數的迅速增加，世界衛生組織（WHO）於2020年3月11日宣布其為全球大流行。然而，雖然此疾病目前已襲擊約188個國家與地區，對全球公共衛生而言是重大挑戰，我們仍舊觀察到國家在總確診數、死亡數、住院數、以及個案死亡率（Case Fatality Rate, CFR）的差距非常明顯，凸顯了研究各國的疫情差異程度是至關重要的。 目的 本研究有兩個目的。第一個為探討各國COVID-19的個案死亡率差異是否會受到某些與公共衛生相關因素的影響，第二個為研究各國是否可以依據個案死亡率分組並探討分在同一組的國家是否有地理位置上的接近，以及觀察分組的狀況隨時間的變化。 方法與材料 資料來自於約翰霍普金斯大學冠狀病毒研究中心(Johns Hopkins University Coronavirus Resource Center)所提供的公開資源。我們收集2020年2月24日至5月25日期間周一的COVID-19數據，並每周進行一次分析。接著，進行比例統合分析（Proportion Meta-Analyses, proportion MA）以探討各國個案死亡率，並額外以統合迴歸（Meta-Regressions）與子群體分析（Subgroup Analyses）的方式，來探討公共衛生或防疫相關因子（醫療照護可及性與品質指數[HAQ]、全球衛生安全指數[GHS]、卡介苗1985年接種率與篩檢政策）是否與各國個案死亡率之間的差異有關。最後，我們利用有限混和模型（Finite Mixture Models, FMM）來探討相似個案死亡率的國家是否可以進行分組並聚集於鄰近地區。 結果 根據比例統合分析的結果，加權平均個案死亡率在固定效應模型（Fixed-Effect Model）中從3.30%迅速上升至7.02%，而在隨機效應模型（Random-Effects Model）則是1.35%上升至3.19%。 在有限混和模型中，我們發現各國可以根據個案死亡率進行分組，且同一組的國家也有聚集在相鄰地區的現象。五月底時，位於南歐、西歐的國家如義大利、法國、英國和比利時被分在高個案死亡率組別，其估計值大於12%；北美洲國家的估計個案死亡率也遠高於南美洲國家；東亞、東南亞的國家如新加坡、越南與台灣則長時間維持在較低個案死亡率的組別，估計死亡率約為0.06%至1.45%。 最後透過統合迴歸與子群體分析的結果，發現有較高水準的健康照護系統（HAQ）、以往針對生物性威脅能夠較有效率對抗（GHS）與較高卡介苗接種率的的國家傾向有較高的個案死亡率，而篩檢政策的部分則是檢測越多可疑病例與低個案死亡率有相關。 結論 加權平均個案死亡率隨著時間的增加顯示COVID-19的疫情從開始到蔓延全球越來越嚴重。 個案死亡率相似的國家有集中在附近地區的現象，如義大利、法國、西班牙與英國等歐洲國家多屬於高死亡率組別，而新加坡、越南與台灣則多維持在低死亡率組別，這可能與類似公共衛生政策、相似人口特徵與防疫手段有所相關。 另外，較高公共衛生水平、較高卡介苗接種率的國家傾向有高死亡率，而能夠篩檢更多可疑病例的國家則傾向有較低的死亡率。

Background Coronavirus Disease 2019 (COVID-19) has become a serious threat to global public health. The rapid increase in the number of confirmed cases and deaths around the world led World Health Organization (WHO) to declare it as a pandemic on March 11, 2020. Though the outbreak has now affected more than 188 countries/areas, countries faced different levels of challenges of COVID-19 in terms of infected numbers of cases, hospitalized patients and mortalities, highlighting the importance of investigations into the causes for such large disparities around the world. Objective There were two aims in the thesis. The first was to explore whether the difference in case fatality rates (CFRs) of COVID-19 would be affected by factors relating to public health and prevention policies. The second was to investigate how countries could be grouped into clusters according to COVID-19 case-fatality rates and whether countries of the same cluster were close in geography over time. Materials and Methods Data were obtained from the open sources provided by Johns Hopkins University Coronavirus Resource Center. To investigate the changes in CFRs over time, we used data reported on Mondays between February 24 and May 25. We conducted proportion meta-analyses to investigate the CFR of COVID-19 in different countries. Besides, in order to explore whether the factors relating to public health and prevention policies (Healthcare Access and Quality Index [HAQ], Global Health Security Index [GHS], Baccille Calmette-Guérin Coverage Rate [BCG Coverage Rate], Testing Policy [TP]) could affect the CFRs, we undertook meta-regressions and subgroup analyses. We conducted finite mixture models to investigate how countries with similar CFRs were forming groups and whether they were close in geography. Results The pooled estimate of CFRs given by proportion meta-analyses during the period was rising rapidly from 3.30% to 7.02% in the fixed-effect models and from 1.35% to 3.19% in the random-effects models. Countries could be divided into several groups based on their CFRs and located within near regions according to the results of finite mixture models. In May, some countries located in Southern and Western Europe such as Italy, France, United Kingdom and Belgium fell into groups with higher CFR of more than 12%. Countries in North America seemed to have higher CFR than which in South America. Countries in Eastern Asia like Singapore, Vietnam and Taiwan were in low-CFR groups with estimated CFR ranging from 0.06% to 1.45%. Countries with higher quality of healthcare system, efficient responses to an outbreak and higher BCG coverage rate in 1985 tended to have higher CFR. However, countries with more suspected cases being tested were associated with lower CFR. Conclusions The estimated CFRs increased with the spread of COVID-19 across the world. Countries of similar case fatality rates forming several groups and were even clustered within geographical regions. Some countries in Western Europe fell into high-CFR groups and Singapore, Vietnam and Taiwan remained in the low-CFR groups over time, which might be explained by their adopting similar public health policies, demographics and strategies to contain the pandemic. A higher quality level of public health and a higher coverage rate of BCG vaccination in 1985 were associated with higher CFR, and testing more suspected cases was associated with a lower CFR.