醫學研究中，常以受試者特徵曲線(Receiver Operating Characteristic curve，簡稱ROC曲線)評估生物指標對疾病預測能力。其中以Heagerty.& Zheng.(2005)提出的時間相依ROC曲線最為常見，透過分析並提高預測效率，辨別各個生物指標的預測能力，進而找出適合的模型。過去文獻使用比例風險模型(Proportional hazard model,簡稱PH或Cox模型)，或使用AFT模型(Accelerated failure time model)、PO模型(Proportional odds model)，建構時間相依ROC曲線，並獲得曲線下面積(the area under the ROC curves,簡稱AUC)，透過對各個時間下的AUC加權平均，取得一致性指標(Concordance)，並可將其推導為風險迴歸的函數，當使用不同模型時，只需代入對應的風險函數，此一致性指標在先前已被證明對預測精準度有一致性。本研究針對AUC的計算方式提出修正，透過原先的定義，將面積計算方式作調整，進而獲得更貼近真實值的結果。後續再使用過去文獻提出的各個模型，將其應用在同一筆模擬資料中，比較其中的差異與優劣，觀察一致性指標的準確性，最後以實際愛滋病資料分析，展示不同模型下的結果。

In medical research, the receiver operating characteristic curve (ROC curve) is often used to evaluate the predictive ability of biological indicators for diseases. Among them, the time-dependent ROC curve proposed by Heagerty. & Zheng.(2005) is the most common. By analyzing and improving the prediction efficiency, it can identify the predictive ability of each biological indicator, and then find a suitable model. In the past, the literature used proportional hazards model (referred to as PH or Cox model), or used AFT model (Accelerated failure time model), PO model (Proportional odds model) to construct time-dependent ROC curves, and obtained the area under the ROC curves(AUC), through the weighted average of AUC at each time, the consistency index (Concordance) can be obtained, and it can be derived as a function of risk regression. When using different models, just substitute the corresponding hazard function, this consistency metric has previously been shown to be consistent with forecast accuracy. This study proposes a revision to the calculation method of AUC. Through the original definition, the area calculation method is adjusted to obtain results that are closer to the true value. Subsequently, each model proposed in the past literature will be used, applied to the same simulation data, the differences and pros and cons will be compared, and the accuracy of the consistency indicators will be observed. Finally, the actual AIDS data will be analyzed to show the results under different models.

Bennett,S.(1983). Analysis of Survival Data by the Proportional Odds Model.{Statistics in Medicine}

Box,G. , and Cox,D.(1964). An Analysis of Transformations. \{Journal of the Royal Statistical Society Series B}

Chiou, S. H., Kang, S. and Yan, J. (2014). Fitting Accelerated Failure Time Models in Routine Survival Analysis with R Package aftgee.{Journal of Statistical Software}

Cox, D. R. (1975). Partial likelihood.{Biometrics}

Heagerty, P. J., Saha-Chaudhuri, P. and Saha-Chaudhuri, M. P. (2012). Package ‘risksetROC’: riskset ROC curve estimation from censored survival data.

Heagerty, P.J. and Zheng, Y. (2005). Survival Model Predictive Accuracy and ROC Curves.{Biometrika}

Jones, M. C. (1990). The performance of kernel density functions in kernel distribution function estimation. {Statistics and Probability Letter}

Jones, M. C. and Sheather, S. J. (1991). Using non-statistic terms to advantage in kernel-based estimation of integrated squared density derivatives.{Statistics and Probability Letter}

Kamarudin, A. N., Cox, T. and Kolamunnage-Dona, R. (2017). Time-dependent ROC curve analysis in medical research: current methods and applications.{BMC Medical Research Methodology}

Lin, Y. H. (2017). Model-base Time dependent AUC and Predictive Accuracy.{Graduate Institute of Statistics, National Central University.}

Wei, L. J. (1992). The accelerated failure time model: A useful alternative to the cox regression model in survival analysis.{Statistics in Medicine}

林威廷(2021)。風險回歸模型下時間相依~ROC~曲線。國立中央大學統計研究所碩士論文。