氣候變遷下氣象因子與空氣污染物對臺灣登革熱流行狀況與趨勢之延遲效應與閾值效應探討

Abstract

氣候變遷對全球環境與公共衛生帶來多層面的衝擊，包含蟲媒傳染病的傳播與擴散。登革熱為全球盛行蟲媒傳染病之一，主要由白線斑蚊與埃及斑蚊傳播。我國亦屬於登革熱盛行區域，曾數次於本島爆發大規模疫情，過去本土疫情主要爆發於南臺灣地區，然而近年中部與北部地區爆發頻率亦逐漸密集。目前已知氣象因子與空氣污染對於病媒的分布與傳播能力相關，然而氣候變遷如何影響各地登革熱風險與流行趨勢仍有待深入研究。 本研究納入我國22縣市1998至2020年間登革熱病例數、布氏指數、氣象與空氣污染物每日資料，並將累計本土病例數超過百例的8個縣市納入分析。以皮爾森相關係數分析（Pearson correlation coefficient analysis）觀察各項變數的共變性，並透過機器學習建造隨機森林模型（random forest model）將變數分組進行影響力排序，並從中挑選各縣市的重要變數進行延遲效應與閾值效應的分析。為了描述變數對本土登革熱或布氏指數的延遲效應與閾值效應，以非線性遞延分配模型（distributed lag non-linear model，DLNM）進行風險的延遲趨勢觀察，再進一步將其降維成暴露－延遲累加風險資料，以分段迴歸模型（segmented regression model）與線性迴歸模型進行比較，觀察是否存在有效影響模型擬合度的斷點，此斷點即為閾值，而斷點前後區段的斜率差異即為變數對風險的閾值影響。 研究結果顯示，相較於本土登革熱，氣溫指標對布氏指數的影響力與地理位置關聯性較明確，而在納入分析的8個縣市當中，相對濕度對本土登革熱影響力高於累積雨量的縣市有6個、對布氏指數影響力高於累積雨量的縣市則有7個，顯示相對濕度在臺灣對登革熱的影響力普遍高於累積雨量。氣溫對於各縣市本土登革熱的風險影響約在20天後最為顯著，此延遲狀況符合登革熱傳播周期當中兩個相關病例之間的平均間隔日數。相對濕度對本土登革熱與布氏指數的風險影響較雨量穩定，推測為每日資料尺度使雨量分布懸殊導致。無論是影響力排序或延遲效應，空氣污染物對登革熱的影響在各縣市皆存在歧異，推測與大氣中的組成分布有關。除了氣溫與相對濕度對部分縣市的布氏指數可用線性關係描述之外，所有氣象因子與空氣污染物對各縣市本土登革熱與布氏指數的延遲累加風險皆存在閾值效應。 在全球氣候變遷的趨勢之下，蟲媒傳染病的流行與擴散為當今公共衛生重要議題；我國地處熱帶與亞熱帶交界，相關應對亦刻不容緩。除了原本已知的氣象因子之外，本研究確認空氣污染物的種類對我國不同縣市登革熱風險的影響有所差異，亦針對登革熱流行與病媒風險進行比較，可供後續相關研究參考。

The impacts of global climate change on the public health are in myriads of ways, include the transmission and spread of vector-borne diseases. Dengue fever is one of the most popular vector-borne diseases worldwide, which is mainly transmitted by Aedes aegypti and Aedes albopictus. Taiwan is one of the prevalent regions of dengue, and there were several outbreaks in the past already. The local epidemics mainly broke out in the southern of Taiwan, but the frequency also grew up in the northern region in recent years. The associations between the meteorological factors and air pollutants to dengue were defined, but the influences of climate change to the risk and epidemic trend of dengue is still not clarified. We used the daily data of indigenous dengue cases, Breteau index (BI), meteorological factors and air pollutants of 22 cities and counties in Taiwan during 1998‒2020, and the eight cities and counties with over 100 cumulative indigenous dengue cases were included in the analysis. We evaluated the covariant between each variable with Pearson correlation coefficient analysis, and arranged the importances of the sorted variables with random forest model. To describe the delay effect and threshold effect of variables to dengue and BI, we used the distributed lag non-linear model (DLNM) to investigate the trend of relative risk (RR) in different lag periods, and reduced the dimensions of the outputs to get the data of exposure-overall cumulative RR. We used the segmented regression model to compared with the linear regression model and evaluated the break points, which were the threshold of the associations between variables and dengue. The effects of temperature to BI were obvious associated with the geographical locations than the effects to dengue, while the effects of relative humidity to both dengue and BI were common than the effects of rainfall. The delayed risk of temperature to dengue appeared at about 20 days later, while the length of period was also the average value of the interval length between two associated dengue cases. The effects of relative humidity to the risk of dengue and BI were stable than the effects of rainfall, and we surmised it was caused by the parameter scale of daily data. There were both differences between the air pollutants with the arrangements of importances to dengue and the delay effect of the risk, and we surmised it was caused by the distribution of composition in air. Except the associations between the temperature or relative humidity to risk of BI in few cities or counties could be described with linear regression, all of the associations between the variables to risk of dengue or BI included the threshold effect. Under the global climate change, the epidemic and spread out of vector-borne diseases became the important issue of One Health. We investigated the differences of delay effects and threshold effects of the risk caused by meteorological factors and air pollutants to dengue and the vector in difference regions of Taiwan, and it may be valuable to future studies.