分析長期追蹤的資料時，常見且廣泛被使用的統計分析方法有以下三種。第一種方法為羅吉斯回歸分析(Logistic Regression)：若應變數(outcome)為連續型資料，使用特定的切點將應變數區分為二元變數，或是應變數已為二元資料，並忽略個體在不同時間的觀測值，使個體最終只有一個觀測值，最後使用羅吉斯回歸分析資料。然而，將應變數的多變量資料型態強制轉成單變量進行分析，容易導致估計上的偏誤。第二種方法為廣義估計式分析(Generalized estimating equation, GEE)，此分析方法同時考慮個體在不同時間的觀測值，應變數可為二元資料或為連續資料，透過選擇不同的工作相關矩陣(working correlation matix)，進行估計和推論。我們考慮第三種常見且可以設定隨機效果(random effect)的方法：線性混和模型分析(Linear mixed model, LMM)，若應變數原為連續資料型態，同時考慮不同時間的觀測值，亦可避免切點的不準確性。此篇論文透過臨床的長期追蹤資料，利用上述三種方法，尋找心肌梗塞與氣溫及其他變數之間的關係，並建立最佳模型，最終預期可供臨床醫生作為分析的參考。

When analyzing longitudinal data, there are three common and widely used statistical methods. If the outcome is a continuous variable, one may use a specific cutpoint to dichotomize the outcome, ignore multiple outcomes at different times and simply choose one of the outcomes. The logistic regression is then used to analyze the resulting data with only one outcome selected or generated. However, this may inevitably lead to estimation errors if the cutpoint is not chosen appropriately. To overcome the situation afore-mentioned and take the multivariate responses into account, we consider a generalized estimating equation method and a linear mixed model. The former one is suitable for continuous or categorical data while the latter method aims for continuous outcomes. We can deal with correlations between observations from individuals by implementing an appropriate working correlation matrix. We apply above three methods to find the relationship between myocardial infarction and temperature and other variables, and establish the best model in this thesis. The approaches for the analysis can be benchmark for clinicians.

[1] Beneish, M. D. Press, E. (1995). Interrelation among events of default. Contemporary accounting research, 3, 68-92

[2] Liang, K. Y. and Zeger , S. L. (1986). Longitudinal data analysis using generalized linear models. Biometrika , 73, 13-22.

[3] Zeger , S. L., Liang, K. Y., and Albert, P. S. (1988). Models for Longitudinal Data: a Generalized Estimating Equation Approach. Biometric, 45,1049-1060.

[4] Laird, N. M., and Ware, J. H. (1982). Random-effects models for longitudinal data. Biomerics, 38, 963-974.

[5] 程毅豪(2019). GEE長期追蹤資料分析在資料缺失下的模型選取問題. eNews第34期，臺北醫學大學數據處。

[6] 蘇秀媛、林其臻和謝鑫能(2011). 利用線性混和模式分析生物防治資料。Journal of data analysis, 6, 87-106.