建立追蹤傳染病群聚擴散過程的時空模式

Abstract

在一個流行病的地理擴散過程下，將會產生數個時空群聚，目前已經有許多統計方法用以偵測時空群聚，例如時空掃瞄統計方法，然而，僅有少數的方法探究時空群聚之間的時序關係，有鑒於此，本研究目的在於追蹤一擴散過程下的時空子群聚以及其之間的時序關係，並以臺灣高雄地區的登革熱案例作為探討對象。本研究方法首先針對任意兩個時空相近的病例，分為群聚關係與傳染關係。對於傳染關係，根據兩病例間的時空距離計算一被傳染機率，接著根據該機率值計算群聚關係之間的「共同來源機率」。所謂來源指的是感染者的感染來源，而共同來源機率則指兩者間擁有相同感染來源的機率。若該值偏高，代表兩者之間不僅於時空相近，並且有很高的機率來自相同的來源，形成時空子群聚。透過各病例間的時空關係，我們能夠追蹤其感染鏈並建立各時空子群聚之間的時序關係，並且偵測子群聚的動態行為，包括出現、消失、成長、縮減、分裂與合併。我們進一步發現子群聚的分裂地區與合併地區分別代表不同的風險型態，雖然兩者的疾病盛行率皆高於平均，但前者是由該地區環境引起的疾病風險；後者則是受到周遭地區環境影響而引起的疾病風險。最後我們建議進行流行病調查或研究時，可以同時使用時空掃瞄統計與本研究之方法，前者能夠告訴我們流行病最為嚴重的時間與空間位置，而後者則能夠進一步偵測其中的動態行為與風險形態。

Most of the space-time analyses were developed to detect space-time clusters from an epidemic outbreak, such as SaTScan. However, they failed to detect the dynamic of the diffusion process. Our objective is to propose an analytical procedure to track the dynamics of space-time clusters from a contagious diffusion process. We used locations and illness onset time of dengue fever cases as a case study to demonstrate the framework of analytical procedure. For each pair of cases who are close in time and space, we defined them as clustering and infection pairs based on their space-time distance. We assigned a probability of infection to each infection pair, and developed a measurement called 'Common Origin Probability (C.O.P.)' for each clustering pair based on the probability. The ‘origin’ was the individual or environment that infected others, and the C.O.P. represented the probability that 2 nodes infected by the same origin. Clustering pairs with high C.O.P. value are cases close in space and time and likely infected from the same origin and form space-time sub-clusters. We tracked their temporal progression based on the probability of infection, and identify different dynamic behaviors such as emergence, disappearance, growth, shrinking, splitting and merging. Areas displaying splitting and merging behaviors represent different risk patterns. The former represents places with dangerous and infections environments, and the latter represents vulnerable places surrounded by infectious environments. Identifying dynamic behaviors of sub-clusters can provide spatial and temporal insights into epidemic progression and risk patterns of disease clustering.