在過去的文獻中，可以加入 Cox 比例風險模型的訊息來估計時間相依敏感度和特異度，並計算出一致性指標 Concordance，然而當不符合比例風險假設時，Cox 模型的使用並不恰當，此時建議使用 AFT 加速失效模型作為替代。為了補齊之前的文獻，本研究在右設限資料時，利用半母數和 AFT 模型的訊息來估計 Concordance，另外針對區間設限資料提出參數模型的一致性估計，使用一致性指標來衡量存活模型的準確性。在實務上，若試驗著重在預測準確性，那麼一致性指標可以做為模型選擇的標準。以上所有研究方法都會於模擬和實際資料分析展現。 ;In the literature, survival model based ROC incorporating the Cox proportional hazards model in the time dependent sensitivity and specificity has been derived. However, the proportional assumption may fail. In this situation, the accelerated failure time (AFT) model could be an alternative. To complement the literature, we proposed semi-parametric and parametric AFT model based ROC for right censored data and derived the corresponding concordance. In practice, for event time data, if the study aim is predictive accuracy, then the concordance could be a criterion for model selection. We evaluated the performance of the proposed method through various simulation studies. Furthermore, the new method and model selection based on concordance were applied to several data sets.
