新型冠狀病毒（COVID-19）流行初期每日確診人數趨勢型態及相關因子分析-世界各國開放資料研究

Abstract

背景： 新型冠狀病毒（COVID-19）在2019年12月於中國武漢市發現且疫情迅速擴散至全球，隨著各國確診及死亡人數與日俱增，世界衛生組織於2020年3月宣布其為大流行疾病。世界各國COVID-19大流行初期之第一波疫情皆以單一或是數個小型零星地區開始爆發，各國依照其文化、經濟及地理位置等不同背景條件，制定不同的防疫政策以防堵疫情持續蔓延。本研究探討世界各國COVID-19流行初期的疫情趨勢型態及其相關背景因素。

研究方法： 本研究在時間趨勢型態分析部份，結合長期追蹤設計與時間序列設計，以151個國家為研究對象重複測量各國「每日新增確診人數7天移動平均值」(以下簡稱MA7)，並對每一個研究國家觀察了60或90天的時間序列資料；因此，本研究在時間趨勢型態分析部份結合了以上兩種研究設計的特性為「長期追蹤時間序列研究設計」(Longitudinal time series design)。 有關世界各國COVID-19流行初期MA7時間序列趨勢型態的相關因素分析，將研究的國家依據疫情趨勢型態分類，進行病例對照研究(case-control study)。研究對象以國家為單位，對151個國家開放數據進行分析，如：COVID-19每日新增確診人數、遏制和衛生指數、高齡化、國內生產總值、識字率、人口密度、肥胖盛行率、醫療資源 (醫師密度、病床密度)及地理環境 (島嶼、沿海及內陸)。統計分析運用時間序列階層群集分析法 (Time-series hierarchical clustering)，對世界各國COVID-19流行初期MA7的時間趨勢型態進行分類，並利用邏輯斯迴歸分析探討與此時間趨勢型態分類有相關的背景因素。結果： COVID-19流行初期MA7趨勢型態可歸類為「成長型」、「消退型」、「平緩消退型」。邏輯斯迴歸並以逐步迴歸校正顯示，世界各國COVID-19流行初期連續觀察60天的MA7趨勢型態分成2群集之相關因素分析顯示，較低的國內生產總值傾向於「成長型」趨勢型態(校正後勝算比=0.98，95%信賴區間0.96-1.00，p值=0.028)；較高的肥胖盛行率傾向於「成長型」趨勢型態 (校正後勝算比=1.09，95%信賴區間1.04-1.14，p值<0.001)。連續觀察60天趨勢型態3群集之多項邏輯斯逐步迴歸模式分析，較高的肥胖盛行率傾向於「成長型」趨勢型態 (校正後勝算比=1.06，95%信賴區間1.01-1.11，p值=0.010)。

結論： 世界各國在地理、經濟、文化、人口、衛生等背景因素的差異下，使得COVID-19每日新增確診人數時間序列趨勢型態有顯著不同，各國防疫政策應參考國家的特性差異來擬定。Background： Coronavirus disease 2019 (COVID-19) originated in the city of Wuhan, China in December 2019, and has rapidly widespread across the world. With the fast-growing number of cases and deaths in many countries, the World Health Organization (WHO) declared the COVID-19 outbreak a global pandemic in March 2020. The first wave of the COVID-19 pandemic outbreak starts at single or sporadic locations across each country. Prevention and control of COVID-19 are applied according to different countries' cultural, economic, and geographical characteristics. This study aims to explore each country's trend pattern of daily new cases in the early stage of the COVID-19 pandemic and explore the associated variables with this trend pattern.

Methods: In the part of time trend pattern analysis, this study combines longitudinal design and time-series design. We included 151 countries as the study subjects; the '7-day moving average of daily new cases' (MA7) in each country was measured repeatedly and observed for 60- or 90-days. Therefore, this study combines the characteristics of the above two research designs as a'Longitudinal time-series design'. In the part of exploring associated variables with the trend pattern revealed by country-level MA7 time-series data, a case-control study was conducted. The study subject is country-level and we analyze open data from 151 countries, such as MA7 of COVID-19, containment and health index, aging, Gross Domestic Product (GDP), literacy rate, population density, obesity prevalence, medical resources (physician density, hospital bed density) and geographical environment (islands, coastal and inland).Time-series cluster analysis used to reveal the trend pattern of study countries by their MA7, and then the logistic regression used to investigate associated variables with this trend pattern.Results: In the time-series hierarchical clustering, the trend pattern of MA7 at each country's early stage of the pandemic can be organized into the 'growth type', the 'decline type', and the 'smoothly decline type'. Logistic regression with stepwise selection of 2 clusters trend pattern which observed MA7 for 60 consecutive days in the early stage of the COVID-19 pandemic showed that lower gross domestic product tended to ‘growth type’ (adjusted odds-ratio=0.98, 95% confidence interval 0.96-1.00, p-value=0.028); higher obesity prevalence tended to ‘growth type’ (adjusted odds-ratio=1.09, 95% confidence interval 1.04-1.14, p-value< 0.001). Multinomial logistic regression model analysis of 3 clusters trend pattern observed for 60 consecutive days, higher obesity prevalence tended to ‘growth type’ (adjusted odds-ratio=1.06, 95% confidence interval 1.01-1.11, p-value =0.010).

Conclusions: Due to the differences in country-level background factors such as geography, economy, culture, population, and health, the time-series trend pattern of COVID-19 daily new cases varied from each other. Prevention and control policies should take countries’ heterogeneous features into account.