環境低溫對於類流感急診就診之影響：時間序列分析

Impact of low ambient temperatures on influenza-like illnesses-related hospital emergency visits: a time series analysis

目標：2016年初，年假期間出現前所未有的長時間冬季低溫，流感病例數爆增。但低溫與此波流感疫情之間相關性迄今仍未有研究。本研究目標為以時間序列方法分析低溫是否為急診流感就診比例上升之預測因子。方法：以時間序列方法分析2007年1月至2016年12月疾病管制署「即時疫情監視及預警系統」大台北區類流感（influenza-like illness）急診就診人次占全部急診人次比例（下稱就診比例），並將中央氣象局台北測站氣溫紀錄納入動態迴歸模式分析。另因例假日流感病患湧向醫院急診，將每週放假天數作為干擾因子。最後以2017年1月至2017年12月實際數據對預測模式進行外部驗證。結果：大台北區類流感急診就診比例時間序列符合ARMA（1,1）模式，前一週就診比例與當週就診比例具高度相關性（自迴歸係數0.92，p＜0.001）。動態迴歸模式分析發現：前一週最低溫（係數-0.07，p=0.01）與當週放假天數（係數0.2，p＜0.001），為類流感急診就診比例之顯著預測因子。外部驗證結果顯示：含有氣溫及放假天數的時間序列模式能準確預測2017年初流感高峰時間點；全年預測值R^2達85%。結論：環境氣溫與下一週類流感急診就診比例具高度相關性，最低溫愈低，下一週類流感急診就診比例愈高。流感防治整備應變需將低溫預報納入考量。 Objectives: Taiwan had its longest recorded cold wave during the 2016 Chinese New Year holiday; a surge in influenza cases followed. To assess the effect of a cold wave on influenza activity, we used a time series model to analyze the relationship between low ambient temperatures and influenza-related emergency visits. Methods: We obtained weekly data on the proportion of influenza-like illnesses (ILIs) among emergency room visits in Taipei from January 2007 to December 2016. Data were from the Real-time Outbreak and Disease Surveillance of the Taiwan Centers for Disease Control. Ambient temperature data were from the Taiwan Central Weather Bureau. We used an autoregressive integrated moving average (ARIMA) model to analyze the association between low temperatures and ILIs. The validity of predictive models was tested against January 2017 to December 2017 data as external validation. Results: The time series of ILI-related emergency visits in Taipei was consistent with the ARIMA (1, 1) model at an autoregression coefficient of 0.92 (p ＜ 0.001). ARIMA analysis indicated that the lowest temperature in the previous week (coefficient: -0.07, p = 0.01) and the number of holiday days in the week (coefficient: 0.2, p ＜ 0.001) were significant predictors of the proportion of ILI-related emergency visits. External validation demonstrated that the time series model with temperature and holiday covariates accurately predicted the timing of the ILI surge in the 2017 winter vacation, with an R^2 of 85% for the overall fit in 2017. Conclusions: Cold waves are a predictor of increased ILI-related emergency visits. Thus, influenza preparedness measures should incorporate ambient temperature forecasts.