Time-dependent roc analysis and applications in the field of medicine

ROC analysis is a commonly used method for evaluating the accuracy of a continuous diagnostic test or a biomarker. However, when there is a time-dependency between the biomarker and the event of interest (death, disease, relapse etc.), using classical ROC analysis may not be able to estimate the true performance of the biomarker. For such cases, time-dependent ROC, an extended version of the standard ROC analysis, is developed. In this thesis, this modified ROC analysis adapted to the time dependent situations is investigated and applications are performed using datasets derived from several medical fields. A comparison between classical ROC analysis and Kaplan-Meier (KM) estimator, which is a time-dependent ROC analysis method has been made on all datasets used in this study. From this comparison, it is concluded that time-dependent ROC method is superior to the standard ROC analysis. In addition, performance of biomarkers measured at each time-point are compared. In general, the closer to the event time, the higher performance is observed. Especially, biomarkers measured at last 12 or 6 months before the event are determined to be better at classification than the earlier measurements. Nearest Neighbor Estimator (NNE), an alternative to KM estimator, is also applied on all datasets. Then, the findings obtained from these two approaches are compared. In all datasets, KM and NNE applications yielded very similar results. Although the KM-AUC values are mostly higher than the NNE-AUC values, it is more appropriate to use NNE method to evaluate the performance of a biomarker when time dependency exists in a censored data. According to the results obtained from this study, when there is a time dependency between the biomarker and the event of interest, to measure a diagnostic performance of the biomarker accurately, the time-dependent ROC analysis is recommended instead of the classical ROC analysis.