臨床研究分析經常從觀察性研究（Observational study）取得數據，探討處理變項（treatment）和感興趣事件的發生（outcome）之間的關聯性。近年的醫學研究常從大型電子化健康數據庫，或是專門用於研究之小型調查資料庫獲取資料。然而單一資料庫可能包含干擾因子資訊不完整，或是樣本數較少等研究限制，造成無法產出精準可信的研究成果。對此，Lin與Chen(2014)提出二階段統計校正（2-stage calibration，簡稱TSC）方法來應對，同時考慮大型與小型資料庫的資訊建立二階段的分析模型，並針對處理變項迴歸係數提出一估計式，可改善參數估計有偏差的問題，同時提升統計上的檢定力（power），得出更具實質意義的統計分析結果。

在醫學研究中，藉由精準預測受試者是否患病的情況可作為醫生評估更換治療方式的依據，而對於二元分類模型之預測能力常使用ROC曲線（Receiver Operating Characteristic Curve）與其曲線下面積（Area Under the ROC Curve，簡稱AUC）來評估。由於TSC方法對於結果變項的預測表現並未提及，故本研究從TSC方法建立一套新模型（命名為TSC模型），使用ROC曲線與AUC作為評估指標，探討在兩階段建立的模型與新模型之預測表現。除此之外，本研究以上述方法，分析台灣健保資料庫與成大醫院資料庫的第二期糖尿病患者資料，探討服用兩種降血糖藥物對於罹患心血管疾病的影響，校正相關病史與糖化血色素（HbA1c）等干擾因子，最終比較各模型預測心血管疾病的表現。

The confounding bias problem from missing or unmeasured confounders is one of the biggest challenge in observational studies. Lin and Chen (2014) developed a 2-stage calibration (TSC) method to obtain the estimate of treatment effect, which summarizes the confounding information from the large-scale administrative database and the small-scale survey database. The small-scale dataset contains important confounding information, in addition to the confounding variables in the large-scale dataset. However, up to our knowledge, this approach does not provide diagnostic accuracy. This paper is trying to extend TSC method for binary outcome to build new TSC logistic regression model, and use the area under the receive operating characteristic curve (AUC) to assess the diagnostic accuracy. In application, we assess the relationship between composite cardiovascular disease (CVD) and two hypoglycemic agents adjusting for the confounders, such as complication and comorbidity. The important confounder- glycated hemoglobin (HbA1c) is only obtained from the external small dataset. The proposed method is demonstrated based on the type 2 diabetes mellitus patients in Taiwan National Health Insurance Research Database and National Cheng Kung University Hospital.

Bamber, D. (1975). The area above the ordinal dominance graph and the area below the receiver operating characteristic graph. Journal of Mathematical Psychology, 12(4), 387-415.
Brumback, L. C., Pepe, M. S., & Alonzo, T. A. (2006). Using the roc curve for gauging treatment effect in clinical trials. Statistics in Medicine, 25(4), 575-590.
Heagerty, P. J., Lumley, T., & Pepe, M. S. (2000). Timedependent roc curves for censored survival data and a diagnostic marker. Biometrics, 56(2), 337-344.
Health2Sync. (2018). 什麼是糖化血色素？認識糖化血色素（hba1c）以及控制目標. https://www.health2sync.com/blog/post/20180223/toknowhba1c/.
Kim, D. H., Pieper, C. F., Ahmed, A., & ColónEmeric, C. S. (2016). Use and interpretation of propensity scores in aging research: A guide for clinical researchers. Journal of the American Geriatrics Society, 64(10), 2065-2073.
Lin, H. W., & Chen, Y. H. (2014). Adjustment for Missing Confounders in Studies Based on Observational Databases: 2-Stage Calibration Combining Propensity Scores From Primary and Validation Data. American Journal of Epidemiology, 180(3), 308-317.
Ma, S., & Huang, J. (2007). Combining multiple markers for classification using roc. Biometrics, 63(3), 751-757.
Metz, C. E. (1978). Basic principles of roc analysis. Seminars in Nuclear Medicine, 8(4), 283-298.
Pepe, M., & Pepe, P. (2003). The statistical evaluation of medical tests for classification and prediction. Oxford University Press.
Pepe, M. S., Cai, T., & Longton, G. (2006). Combining predictors for classification using the area under the receiver operating characteristic curve. Biometrics, 62(1), 221-229.